overconfidence modelling in financial institutions, emphasising … · ·...
TRANSCRIPT
1
Overconfidence modelling in financial institutions, emphasising
credit risk and profitability
(Modelowanie zjawiska nadmiernej pewności siebie w instytucjach finansowych, ze
szczególnym uwzględnieniem ryzyka kredytowego i dochodowości)
***Preliminary draft***
Dorota Skała1
Abstract:
This work is aimed to provide an extension to moral hazard, which is the conventional incentive of risk taking in
banks. This extension is overconfidence and stems from the area of behavioural finance. Overconfidence
stipulates that bank managers take high risks not only due to moral hazard, but also because of the cognitive bias
in the form of overconfidence. This bias makes them overestimate future returns, underestimate risks and falsely
presume some level of control over events that are external, such as macroeconomic deteriorations. The
overconfidence hypothesis is verified through a fixed effects panel data model, using financial results of 311
Western European banks from the period 1997-1H2008. The estimation shows that overconfidence has a
negative effect on bank risk and profitability, with a lag of three years. In addition, banks are demonstrated to
engage in income smoothing through loan loss provisions, but the underlying rationale does not imply that this is
driven by prudential considerations. Creating provisions is procyclical versus the economic cycle, indicating that
banks may fall short of reserves when economic conditions deteriorate.2
Keywords: Overconfidence, moral hazard, bank risk, bank profitability, Western European banks.
1 Department of Finance, Faculty of Economics and Management (WNEiZ), University of Szczecin. The author
will be grateful for all comments and suggestions at [email protected] . 2 This work is a summarised version of the doctoral thesis entitled „Ocena wyników i profili ryzyka banków
europejskich- wpływ czynników behawioralnych“ (Skala 2010), written under the supervision of Prof. dr hab.
Waldemar Tarczyński and submitted at the Faculty of Economics and Management (WNEiZ), University of
Szczecin.
2
Introduction
International financial institutions have recently witnessed one of the most severe
downturns in their history, taking into account its vast geographical scope, strong systemic
interdependencies and links to other markets and sectors worldwide. Naturally, such a
universal meltdown has resulted in a pressing need to seek its sources and even more
importantly, to apply adequate measures to prevent a re-occurrence of similar scenarios in the
future.3
This paper offers a behavioural extension to conventional banking theory, regarding
managerial incentives and motives for risk taking. The work provides a previously seldom
considered rationale within bank incentives schemes that could have contributed to excessive
risk taking by banks worldwide, and, in consequence, to the financial crisis. This motivation
is overconfidence and stems from the relatively new area of behavioural finance. To date,
underlying reasons for accepting high risk by banks have been concentrated around moral
hazard theories. As Hellman et al. (2000) state, “financial crises have become more frequent
(...) observers agree that moral hazard plays an important role in these failures” (p.148).
Overconfidence does not preclude moral hazard, but delivers an additional spur to bank
managers that face financial decisions.
The primary difference between moral hazard and the overconfidence context is the
ability and possibility of bank managers to quantify risk and future returns, as well as the
rationality of the decision regarding the level of risk. Moral hazard theories imply that agents
assume certain levels of risk knowingly and rely on other parties to bear its costs, if it brings
adverse consequences. Overconfidence does not assume the pre-calculated risk-return
formula, but implies instead that financial actions may be taken due to a cognitive bias. This
bias leads agents to miscalibrate risks, future returns, own future actions and possible
outcomes relating to the operating environment. Future risks and their variances are
underestimated, future returns overestimated and a false assumption of control over operating
environment developments comes into play, with unrealistic optimism of bank managers
exacerbating these biases.
Overconfidence has been widely studied and proven in the context of financial
markets, mostly regarding investor behaviour. A growing number of research relates also to
managerial behaviour in companies, but studies focusing on banks are rare. This work is
aimed to fill this gap, through introducing an overconfidence phenomenon into standard bank
analysis. The overconfidence hypothesis is used to construct an empirical model, verifying the
3 For further comments on the crisis, see e.g. Rajan et al. (2008) and Borio (2008).
3
influence of overconfidence proxies on bank credit risk and profitability. The model is tested
on a sample of 311 Western European banks, using financial results from 1997-2008. The
structure of this paper is as follows: Section 1 contains the theoretical background relating to
moral hazard and the most frequent frameworks, within which it is studied. Section 2 briefly
defines overconfidence and its primary facets. Section 3 outlines the main issues in bank
analysis used to construct the empirical model testing the overconfidence hypothesis. Section
4 depicts the rationale underlying the passage from theoretical findings to the model
construction, with a general description of the main dependent variables and overconfidence
proxies. Section 5 presents estimation results and robustness tests. Section 6 concludes.
1. Moral hazard and its implications for managerial risk taking
Moral hazard is part of the principal agent theory and originally appeared within
insurance research. As described by Greenbaum and Thakor (2007), moral hazard emerges
when the incentives of the principal and the agent diverge. Agents are inclined to maximise
their own utility, so if their self-interests do not converge with that of the principal, the latter
may suffer due to actions undertaken by the agent. This assumes asymmetric information,
where the principal is not able to gain full knowledge and control over the agent‟s actions.4
The contract between the principal and the agent should thus align their interests in order to
minimise moral hazard. However, contracts do not fully eliminate moral hazard, if “there is
some noise (exogenous uncertainty) that masks the agent‟s action in the final outcome”
(Greenbaum and Thakor 2007, p.31).
Moral hazard in banks is frequently mentioned, in various frameworks.5 The first area
concerns moral hazard behaviour of borrowers, who receive funding from banks in the form
of loans. Here, banks are considered the principal, who can suffer losses resulting from moral
hazard behaviour of firms, which play the role of agents.6 As assumed by Manove et al.
(2001), in the context of bank-borrower relation, moral hazard implies that “debtors have the
incentive to engage in opportunistic behaviour at their creditor‟s expense, such as asset
substitution, inadequate supply of effort, and underinvestment” (p.739). Thus, a frequent
direction of research in this field looks into the role of pledging collateral in diminishing risky
4 For a traditional specification of asymmetric information in banks, see e.g. Harris and Raviv (1979) and Leland
and Pyle (1977). 5 See e.g. Leland and Pyle (1977), Hellman et al. (2000), Kahn and Winton (2004).
6 The role of banks in loan monitoring of borrowers is theoretically explored in the classical study of Diamond
(1984). Asymmetric information between borrowers and banks, and between banks competing for the same
borrower, is analysed e.g. by Sharpe (1990).
4
behaviour of borrowers. Other studies concern equity levels in companies and their role in
minimising moral hazard versus their lenders (banks). The second large field relates to moral
hazard behaviour of financial institutions as such. In this context, it is bank managers that play
the part of agents and the role of principals may be assumed by a few parties, depending on
the setup.
The first bank specification considers managers that act together with – or in the best
interest of – shareholders. If managers and shareholders jointly succumb to moral hazard, they
engage in opportunistic bank behaviour at the expense of other parties. In the literature these
parties usually comprise bank depositors, bondholders and providers of deposit insurance, and
may be referred to as „outsiders‟. Such a setup commonly means that shareholders/managers
strive towards low capital levels in banks, simultaneously taking high risks that may bring
good profitability at the cost of an increased probability of default. If negative consequences
of risk do not materialise, the bank realises elevated profits and managers/shareholders reap
private benefits in the form of bonuses or dividend payments. This outcome does not affect
depositors, bondholders and deposit providers.
On the other hand, if a negative scenario is realised, the bank suffers losses and may
default. The cost of this default is only minimally born by shareholders, due to low equity
levels. It is the outsiders, so the funding side suppliers, that suffer most, as the bank defaults
on payments to depositors and/or bondholders. This is referred to as „gambling on
resurrection‟ by Hellman et al. (2000) and occurs when “banks choose a risky asset portfolio
that pays out high profits or bonuses if the gamble succeeds but leaves depositors, or their
insurers, with the losses if the gamble fails” (p.148). Bondholders are equally affected by
asymmetric information, as they are not able to fully observe risks taken by bank managers.
As Kahn and Winton (2004) stipulate, “the institution may have incentive to engage in risk
shifting, inefficiently increasing the overall risk of its loans, because some of the downside is
shared with debt holders, while shareholders pocket the upside” (p.2532). The existence of
deposit protection funds in most countries shields depositors from losing their money, forcing
the deposit provider to pay for the bank‟s mistakes.7 The negative aspect of deposit protection
schemes poses that if banks faced a danger of a long-term reputation loss versus its
depositors, they would take lower risks. As stipulated by Howells and Bain (2008), in case of
deposit protection banks know that their customers are protected and are thus encouraged to
take greater risks. With no deposit insurance, managers would be made more responsible for
the entrusted funds versus the general public. This pressure would probably increase their
7 A conventional view of deposit insurance is analysed in more detail by Diamond and Dybvig (1983).
5
prudency, as a personal loss of reputation would not be as easily forgotten. In this
specification there is also an important role played by the lender of last resort, usually the
central bank. In order to prevent banks from defaulting, the lender of last resort may supply
liquidity lines and/or longer-term funds, in an extreme case acquiring a share in the ailing
institution. Quite a few real life examples of such actions could be observed during the recent
financial crisis of 2007-08. Many models conventionally assume that the mere existence of
the lender of last resort drives banks to take higher levels of risk, because they rely on
liquidity support in the worst case scenario.
The second major moral hazard framework within financial institutions assumes that
the principal-agent conflict may run between bank managers and shareholders. In such a
setup, managers do not act in the best interest of shareholders and moral hazard induces them
to reap private benefits at the expense of shareholders. This also entails asymmetric
information, as managers do not inform shareholders of their actions. These actions are
similar as in the shareholders-outsiders conflict and entail taking higher risks than officially
admitted. If more aggressive risk taking results in higher earnings, bank managers profit from
this through a variety of ways, for example through links between managerial remuneration
and bank earnings or bonuses awarded for exceptional bank returns. If financial markets
positively evaluate increased profitability and bank share prices rise, then managerial stock
option plans also gain value. Non-pecuniary managerial advantages are equally important, as
they entail reputation and confidence of the financial markets, for which most managers
fiercely compete.
On the other hand, if a more aggressive risk profile of a bank leads to losses, it is the
shareholders that pay a high price, as the market may reduce its valuations and share prices
may slump. Additional capital may be necessary to cover unexpected losses, potentially
causing a dilution of old shareholder participation. Bank managers will possibly also suffer,
unless the losses are part of a more general banking slump and can thus be excused. In fact
some researchers (like e.g. Rajan 1994) suspect that banks coordinate their policies regarding
the timing of losses. In such a case, managers try to hide adverse results of their overly
aggressive risk taking, until they can be revealed among other losses and thus excused by the
market.
Regardless of whether it is the managers/shareholders versus outsiders conflict or the
managers versus shareholders specification, moral hazard is uniformly assumed to lead bank
managers to take higher risks. This work is aimed at exploring the incentives and motives of
managerial risk taking, and more specifically to examine if overconfidence may play a
6
significant role. As a result, in this work it is not necessary to determine what type of
principal-agent conflict drives moral hazard of bank managers. The examination of whether
bank managers are aggressive risk takers in cooperation with the shareholders or rather
against their will, is an important issue that should be further analysed. It lies however beyond
the scope of this work and is not crucial to overconfidence considerations. The vital part of
the moral hazard assumptions is the same for both types of conflicts and refers to agents (bank
managers) being inclined to take higher risks because the principal (deposit protection fund,
central bank and/or shareholders) is likely to cover the potential costs.
The mechanism of bank managers increasing risk exposures without knowledge of
outsiders (principals) is well described by Kahn and Winton (2004) as an “increase in loan
risk (...) through deliberate selection or through shirking on screening or monitoring of loans,
all of which are activities that are difficult for outside investors to observe” (p.2532).
Managerial risk taking in financial institutions is most frequently centred on asset substitution.
This may involve replacing safer debt securities with less stable ones (e.g. corporate bonds).
Similarly, a shift in lending targets may occur, when banks move from the secure, established
clientele to higher-risk customers. Such changes may also be performed across categories,
when safe debt securities are replaced by new loans with visibly more pronounced default
probabilities, but accordingly more elevated yields. Bank managers may choose to exert less
effort and/or accept underinvestment, thereby jeopardising sound bank performance. This can
take place through decreased loan monitoring, insufficient means accorded to credit risk
management and creating loan loss reserves that are not adequate to levels of credit risk
undertaken.
Summarising the above considerations on moral hazard, clear implications for bank
managers‟ behaviour emerge. Bank managers, in their role as agents, take actions that
maximise their own private utility and not necessarily the utility of the principals. The
principals may be either outsiders, such as deposit protection providers and bondholders, or
bank shareholders. In either case, bank managers reap private benefits from taking increased
risks, at the expense of the principals. Managerial profits include higher salaries, bonuses,
personal reputation and confidence of the financial markets. If negative consequences of risk
surface, the cost of aggressive risk taking is born either by outsiders or shareholders. The
established lag effect of bank risk taking, especially seen in credit risk, makes managers
capable of realising private gains before potential costs surface. In addition, bank managers
may be able to hide moral hazard behaviour, if experienced bank losses can be cumulated
with a sector-wide slump.
7
All of the above conclusions presume that managers engage in risk taking in order to
reap private benefits at the expense of another party. This implies that more elevated risk
profiles of banks are due to a managerial pre-calculation showing that private managerial
utility will be higher if more risks are taken. Using asset substitution as an example, it takes
place after the managers have quantified expected returns on various asset combinations. This
calculation unequivocally indicates that a more risky asset mix guarantees managers higher
private benefits than in the case of a safer combination. In parallel, managers presume, with a
high level of certainty, that they will be able to realise gains on this strategy before they are
charged with any costs (meaning for example their forced resignation) once large losses
materialise. Needless to say, the earlier realised benefits should exceed potential later costs.
Alternatively, managers anticipate that costs will be covered by other parties, such as the
deposit insurance fund or the shareholders, and their utility will not be damaged. In all cases,
the managerial decision of engaging into higher risk activities due to moral hazard is driven
by a quantifiable expectation that gains are going to exceed costs.
This work aims to introduce managerial overconfidence as an additional significant
factor feeding into bank risk taking. Overconfidence is not regarded to stand in opposition to
moral hazard, but rather to complete it by adding an important element of cognitive bias to the
equation. This error in evaluating risk and future outcomes, in various forms, creates an
alternative explanation for managerial risk taking. It may be driven not only by a conscious
pre-calculated prediction of boosting own profits, as in the conventional moral hazard context,
but also by errors in assessing own abilities, risk developments and operating environment
factors. These errors are not realised by the bank managers, who thus may not fully apprehend
the extent of risk taken.
2. Main facets of overconfidence and their application in behavioural finance
The existence of an overconfidence phenomenon has been long established in
psychological research, starting from early studies in the 1960s. In the past two decades
overconfidence has become popular among economists, especially in the area of behavioural
finance.8 Despite varying treatments of overconfidence in psychological literature, four
primary overconfidence facets emerge, consisting of miscalibration, better-than-average
effect, illusion of control and unrealistic optimism.
8 A comprehensive literature review of overconfidence studies in pschology and finance may be found in e.g.
Skala (2008). This section is extensively based on this study.
8
Miscalibration is a misalignment between confidence of an agent and his accuracy
rate. Appropriate calibration takes place “if over the long run, for all propositions assigned a
given probability, the proportion that is true is equal to the probability assigned” (Fischhoff et
al. 1977, p.552). In psychology, overconfidence has been defined as a particular form of
miscalibration, for which the assigned probability that the answers given are correct exceeds
the true accuracy of the answers. Alternatively, it is seen as “an unwarranted belief in the
correctness of one‟s answers” (Fischhoff et al. 1980, p. 108). Perfect calibration is rare and
experimental studies show that – among various professionals – it is attained only by weather
forecasters (see research quoted in Lichtenstein et al. 1982). An important role in achieving
good calibration is played by the presence (or lack) of clear, rapid feedback on accuracy rates.
Within miscalibration studies, an often studied phenomenon is the so-called “hard-easy
effect” (Lichtenstein et al. 1982). It implies that overconfidence surfaces mostly in difficult or
very difficult tasks, while easy questions frequently generate underconfidence.
Some researchers associate differing levels of overconfidence with gender issues,
which accommodates the common belief of men being more confident than women given the
same level of knowledge. Weak differences in overconfidence have been confirmed by Beyer
(1990), with stronger discrepancies found in a later study regarding masculine and feminine
tasks (Beyer and Bowden 1997). Nonetheless, some economists use gender as a proxy for
overconfidence (Barber and Odean 2001) and the financial market data seems to confirm this
assumption. Overconfidence has been questioned within psychological literature, with
researchers attributing its emergence to faulty experimental procedures (Gigerenzer et al.
1991, Juslin 1994), faulty interpretations of regression effects (Dawes and Mulford 1996) or
the existence of a random error in judgment (Soll 1996). However, the bulk of overconfidence
research in psychology, including recent works (Keren 1997, Klayman et al. 1999, Brenner
and Griffin 2004, Glaser and Weber 2007) strongly confirms its existence.
Miscalibration is the leading overconfidence facet in psychological research. The
remaining three issues (better-than-average effect, illusion of control and unrealistic
optimism) are referred to as positive illusions and studied much less frequently by
psychologists. Nonetheless, they play a significant role in behavioural finance and sometimes
they are the only overconfidence areas used by economists, possibly due to their more
straightforward application possibilities.
The first positive illusion refers to the so-called better-than-average effect. As stated
by Skala (2008): “Psychological research has established that, in general, people tend to have
an unrealistically positive view of themselves. Most of people, when comparing themselves to
9
a group (of co-students, co-workers, random participants), believe to be superior to an average
representative of that group in various fields. On the aggregate level this seems a statistical
impossibility”. In their study of the better-than-average effect, Taylor and Brown (1988)
indicate an important role played by a self-serving bias in self-assessment. The self-serving
bias makes people assign more responsibility for success and less for failure to themselves,
while others are not given the same credit. Babcock and Loewenstein (1997) also believe the
better-than-average effect to be partly caused by a self-serving bias, the existence of which
they prove in numerous experiments and find that it is pertinent and difficult to alleviate.
The second positive illusion, unrealistic optimism, or a so-called optimistic bias, is
frequently analysed in the context of the better-than-average effect and biased self-attribution.
A short definition is provided by Taylor and Brown (1988) “The future will be great,
especially for me” (p.197). This may be interpreted more broadly, as in Skala (2008):
“unrealistic optimism towards the future can be seen as an error in evaluating future events,
either in the sense of the better-than-average effect (e.g. when all or most people believe their
chances of achieving financial success are higher than the „average‟ person‟s) or in absolute
terms (when people believe their chance of winning a lottery are higher than the true
probability)”.
The last positive illusion is the so-called illusion of control. Psychological research
and common observation demonstrate that people tend to believe they are able to influence
events which in fact are governed mainly, or purely, by chance (Taylor and Brown 1988). An
extreme example of this illusion is an insistence on throwing a dice personally as if it could
then show a more favourable result. Moreover, if people expect certain outcomes and these
outcomes do occur, the participants are prone to assign them to their doing rather than luck,
and re-affirm their belief in control over a situation where the only factor is probability (Skala
2008).
Overconfidence has been increasingly included into economic models in the past two
decades, especially in the context of behaviour on financial markets and more recently in
corporate setups. Psychological definitions of overconfidence are adjusted to the financial
framework and vary from one publication to the next. In the financial market area,
overconfidence is defined as an overestimation of one‟s knowledge or precision of private
information, or the interpretation thereof. Alternatively, an underestimation of variance of
signals or volatility of asset values are also considered. The introduction of the
overconfidence concept has allowed to solve some puzzles found on the financial markets,
which previously could not be explained using standard economic theory. These are securities
10
misvaluations, excessive trading volumes and the disposition effect, i.e. a tendency to sell
well-performing stocks and to hold on to losing ones. Despite some scepticism among
economists on the existence and effect of overconfidence as such, its prevalence on financial
markets has been proven repeatedly, through methods ranging from experimental and
questionnaire studies to formal models and financial market data, described in detail in Skala
(2008).
In behavioural finance, overconfidence is often interpreted as either investors
overestimating the precision of their information (sometimes more specifically:
overestimating private signals and underestimating the public ones) as in Daniel et al. (1998),
Benos (1998) or Odean (1998). The better-than-average effect is accounted for (Odean 1998),
as traders evaluate their information as better than their peers‟. Within the overconfidence
hypothesis investors are also assumed to underestimate risk, which makes them e.g. hold
riskier portfolios. Assuming the existence of such (and similar) facets of overconfidence,
leads to various interesting results for the financial market, including: excessive trading
volumes (Odean 1998, Benos 1998), influence upon trading profitability (Kyle and Wang
1997, Benos 1998, Gervais and Odean 2001) or excessive trading volatility (Daniel et al.
1998, Chuang and Lee 2006). An important input are studies linking overconfidence with
short- and long-term asset misvaluations (Daniel et al. 2001), which have previously been
puzzling for many financial market researchers.
Behavioural theoretical models have been confirmed on large data pools, primarily
including the works of Terrance Odean and his co-authors. Investor data from large US
brokerage companies has been applied to prove the so-called disposition effect (tendency to
hold on to losing securities and sell the well-performing ones – Odean 1998), excessive
trading (Barber and Odean 2000) and detrimental effect of overconfidence upon trading
(excessive volumes and lower profitability – Barber and Odean 2001). US companies stcok
market data also proves that overconfidence affects asset mispricings and excessive trading
volumes, causing trading volatility (Chuang and Lee 2006). In consequence, the existence of
overconfidence on financial markets has been confirmed and its effects on trading volumes
and results are far from neutral.
A field less explored is the existence and possible implications of overconfidence in
the corporate finance context. Research on its implications for corporates has developed only
very recently and remains a growing field. The two main directions of overconfidence
research in the context of corporate finance are studies of merger and acquisition activities of
corporates and analyses of internal corporate financing structures.
11
Analyses of the role of overconfidence in corporate mergers and acquisitions have
been spurred by an early work of Roll (1986), who stated that managerial hubris underlying
the m&a activity may cause negative effects, such as overpaying for target firms and negative
net results for combined shareholders in terms of stock valuation. Roll (1986) underlines the
individual decision-making in m&a‟s, building a solid argument against a later theory of
Fama (1998) on behavioural anomalies cancelling out in the aggregate on an efficient market.
Acquisitions are strongly driven by company CEOs or management boards at best, which do
not have the same potential of cancelling out individual irrationalities as e.g. a population of
traders in a financial market. Empirical studies on corporate m&a‟s confirm Roll‟s
hypothesis, with Malmendier and Tate (2006) proving that overconfident managers who
“overestimate the returns they can generate in their own company” (p.1-2) are more prone to
engage in mergers and these mergers are less favourable than these undertaken by their
rational peers. Experimental results regarding behaviour of business founders (Camerer and
Loavll 1999) also confirm managerial overconfidence in the form of the better-than-average
effect.
As to the corporate financial structure, the existing overconfidence research focuses on
the optimal proportion of debt versus equity financing of new investments and a possible
over-dependence on free cash-flow in that respect (Heaton 2002). The timing of executing
managerial stock options is an innovative overconfidence proxy used by Malmendier and Tate
(2005), as overconfident managers are prone to believe in their ability to keep the share price
rising so they refrain from realising the options they hold. In this study, corporate panel data
of US companies is applied to prove that overconfident managers make investments a
function of free cash flow (more than rational peers) and the level of investment in their
companies may thus be sub-optimal. The potentially crucial implications of overconfidence
for the performance and risk profile of corporates make this field an important addition to the
standard corporate finance models.
Among the few studies of overconfidence in the specific financial institutions context,
Niu (2008) conducts a brief analysis partly based on Malmendier and Tate‟s (2008)
methodology. He demonstrates that managerial overconfidence, identified through press
portrayal of bank CEOs and managerial compensation, leads to higher risk taking in banks.
The relation emerges in an empirical sample of 108 US banks in years 1994-2004. Bank risk
taking is proxied here by the standard deviation of the bank‟s stock returns. Such a treatment
of bank risk taking is applied in empirical literature, even though it does not measure the true
underlying risk, but only the market‟s perception of particular banks expressed through share
12
price valuations. As a result, the statistical link between press portrayal and market
perceptions is likely to emerge. On the other hand, this brief study is among the few that
specifically target overconfidence in bank risk taking and paves way for further research.
This section has introduced overconfidence as an extension to moral hazard, the usual
risk-taking motivation for bank managers. This cognitive bias is complementary to moral
hazard, as it introduces a possibility of a non-recognised error in managerial evaluating of
future own performance, bank risk and results, and operating environment factors.
3. Primary factors affecting credit risk and applied in the overconfidence modelling
This Section discusses some key issues shaping credit risk and necessary for the
construction of the overconfidence model. The following treatment of credit risk factors is by
far not exhaustive – quite a few issues have been omitted, but may be found in the literature.
The focus lies on three crucial issues for overconfidence, namely loan growth, net interest
margins and loan loss provisions.
The relation between loan growth and risk is vital to the construction of an empirical
model involving risk, as many authors treat loan growth as a proxy of bank risk as a whole.
Expanding a loan portfolio in a fully saturated banking environment demands significant
effort, but its effect on bank risk is viewed by many researchers as negative. Foos et al. (2007)
make a strong assumption that “it is most likely that new loans will be granted to borrowers
that have previously been rejected, that were unknown or non-existent previously, that
demand too low loan rates (too little collateral) relative to their credit quality” and as a result
“loan growth may have an adverse impact on the overall risk of a bank” (p. 2-3).9 A similar
assumption on the relation between loan growth and risk is made by numerous authors,
including Berger and Udell (2004), which – referring to the US financial institutions – state
that “an increase in lending corresponds quite closely with the concept of an easing of lending
standards” (p.472). In their analysis of problem loans, Salas and Saurina (2002) include rate
of credit growth as one of the main drivers of problem loans, concluding from the literature
that “a rapid credit expansion is considered one of the most important causes of problem
loans” (p.209-210).10
Similarly, according to Hardy and Pazarbasioglu (1999) cases of severe
banking system distress are often preceded by especially rapid credit expansion.
The direct relation between loan growth and risk is proven by Foos et al. (2007). This
empirical analysis uses a very large sample of over 10,000 individual banks in the period
9 See here for further references to literature regarding the relation between loan growth and risk.
10 See for further references on links between loan growth and problem loans.
13
1997-2005, from 14 developed countries. The authors provide strong evidence that, on an
individual bank level, current loan growth leads to a peak in loan loss provisions three years
later and causes a decrease in relative- and risk adjusted interest income. This implies that
loan growth is strongly positively related to bank risk, and the assumption that new loans on
developed markets are granted at the expense of lower margins seems plausible. In addition,
loan portfolio expansion is confirmed to have a negative effect on bank capital ratios,
boosting bank risk even further. In general, the bulk of research assumes that rapid loan
growth on developed markets is positively related to higher bank risk. This presumption is
also verified in the empirical model in Section 5.
On the other hand, rapid loan growth on emerging markets, apart from the underlying
macroeconomic risk, does not necessarily translate into a higher credit risk and in some cases
may even diminish it (see Cottarelli et al. 2005, Kraft and Jankov 2005).
A link between loan growth and problem loans on developed markets is empirically
verified by Salas and Saurina (2002). Controlling for macroeconomic environment, excessive
loan growth policies bring about an increase in problem loans with a time lag of three years.
In addition, more aggressive (lower) interest margins may cause an increase in problem loans
in the future. Both of these results are similar to these found by Foos et al. (2007), who used a
different time period and bank sample. This indicates that loan growth may be targeted at
lower quality borrowers and in addition, these risky clients are not charged enough for their
higher risk levels. The negative results of such policies become visible in a three year period
after loan granting, so a lagged effect must be taken into account.
The second crucial factor shaping credit risk is the level of net interest margin (NIM).
The robust empirical results of Salas and Saurina (2002) and Foos et al. (2007) provide an
important argument in the debate regarding the role of interest margins in loan growth. In
studies of net interest margins, no consensus has been reached on its relation with risk. Some
authors claim that higher net interest margins imply that loans are granted to higher risk
customers, who are charged more for their elevated probability of default. As a result, banks
with higher NIM would be these more exposed to credit risk, but potentially charging a risk
premium for such an exposure.
The opposite view of NIM presumes that banks with higher margins take in fact lower
risks. This is caused by the fact that these players may not be willing to expand their credit
portfolios at a price of lower interest rates on new loans. This vision is coherent with the
previously discussed saturation of Western banking markets, where gaining new clients and
new loans is frequently associated with accepting higher credit risk levels. Due to visible
14
competitive pressures, banks adjust their pricing policies on both deposits and loans, in order
to increase market share.11
These pricing wars are however frequently targeting a sub-prime
clientele, which under conservative assumptions would imply charging high risk premiums.
Competing institutions minimise these premiums, ending up with new loans that are risky, but
also potentially underpriced and thus do not include a reserve cushion.
Empirical analyses of fresh bank results comprising both Salas and Saurina (2002) and
Foos et al. (2007) seem to confirm the second hypothesis on interest margins versus risk
relation. Lower margins are probably a sign of weaker asset quality and may result in future
repayment problems. In the recent past banks have been expanding loan portfolios at the cost
of quality and not charging sufficient risk premiums that would allow to cover potential loan
loss reserves in the future. These implications may also be derived from the financial crisis
analyses, where weak underlying asset quality was not matched by sufficient margins that
would offset the risk. A more detailed interpretation of various NIM implications can also be
found in Section 5, where interest margins in Western European banks are empirically
analysed. Other factors, such as macroeconomic influences, are also empirically proven to
shape bank profitability (Hardy and Pazarbasioglu 1999, Demirguc-Kunt and Detriache
1998).
The institutional memory hypothesis of Berger and Udell (2004) is linked to loan
growth and indicates that internal factors prevailing in a given organisation may strongly
influence its results, independently from external factors, such as macro considerations.
Standards within banks ease with time that elapsed since the institution last experienced a bad
loan crisis. This may result from a decreased fraction of experienced loan officers, as new
officers, especially in the expansion phases, are being hired continuously. In addition, even
experienced officers may loosen their credit evaluation standards, as the lessons learnt from
previous credit problems fade with time and herding effects from an enthusiastic, growing
market affect their behaviour. Berger and Udell (2004) confirm their institutional memory
hypothesis empirically, using data from US banks from 1980-2000. Controlling for business
cycle effects on loan growth and other supply and demand factors influencing it, the authors
demonstrate that credit standards ease and additional loans are approved as more time passes
since the last NPL crisis. As a result, loan growth should not be associated with business cycle
and herding effects only, due to the fact that the banks‟ internal behaviour may influence their
credit cycle significantly and reinforce procyclical lending patterns. This also can be linked to
overconfidence, which may be an individual trait of a given institution and not a moral hazard
11
On the liability side, this is extensively discussed by Hellman et al. (2000).
15
characteristic, common to all banks covered by deposit insurances schemes or other systemic
support (eg central bank/government funds).
A controversial theory on credit growth and credit policies is offered by Rajan (1994),
but its spirit lies close to the overconfidence approach so it is presented in more detail. The
author claims that in a rational, profit maximising world, banks should lend only to borrowers
who present positive NPV projects. As a result, all loan growth should stem uniquely from
changes in the situation of borrowers and not external „supply‟ conditions, such as e.g. a
desired level of credit expansion by banks. This seems to go against the usual practice
observed among financial institutions, which appear to increase credit supply in concordance
with their internal goals rather than as a function of improved quality of their borrowers.
Recent experiences with sub-prime mortgages in the US explicitly prove that point, as waves
of new loans appeared due to aggressive growth targets of banks rather than improved quality
of sub-prime clients.12
As a result, banks are suspected by Rajan (1994) to go clearly against
the rational profit-maximising rule. They fund negative NPV projects during good times and
reject funding of positive NPV projects in credit contraction times. The reason for this is the
short-sightedness of bank managers, who besides an earnings maximisation are also
concerned about the stock and labour market perceptions of their abilities, i.e. their reputation.
As the market is not capable of determining the true situation of a bank‟s credit portfolio or
the standing of particular borrowers, it follows only the observable earnings. As a result, bank
managers may be tempted to manipulate earnings in such a way as to create a better picture of
the bank‟s credit policy and loan portfolio. This is possible – in the short term – through a
more relaxed credit policy (extending new loans to weakly performing clients, lowering
collateral standards, attracting higher risk clients with products of high upfront fees etc.),
which brings immediate returns in terms of profitability. Inevitable losses take more time to
materialise. The management plans on realising these losses during the next overall banking
sector downturn, as “a bank‟s reputation is less sensitive to poor earnings when other banks
admit to poor earnings” (p.402). An important assumption is thus that the management or the
market value good performance during „normal‟ conditions much more than during systemic
downturns.
In addition, Rajan suspects that banks may to a certain extent coordinate their credit
policy tightening, after an adverse shock to a borrower sector was experienced. Until the
probability of an adverse shock to a sector is small, banks are forced to maintain an
12
The infamous NINJA clientele (No Income, No Job, No Assets) represented an important pool of new
customers to US banks with weak prospects for future repayment. In Rajan‟s model these clients would not find
their way into the banking system at all.
16
excessively liberal credit policy towards it. This may lead to an overinvestment in a given
industry and increases the probability of difficulties. Only after the sector situation
deteriorates visibly, banks tighten their credit policy, sometimes excessively, to cover the
previous overly lax approach. This may demonstrate that in fact the credit cycles, including
these in loan growth, are not external conditions under which banks suffer, but in fact may be
partially fuelled by their own policies. A significant assumption made in Rajan‟s model is that
“banks charge off loans or add to reserves only when such actions are anticipated by the
market and convey little new information” (p.423). As a result, studies analysing increased
loan loss provisions may be biased, in that they find no information effects. The exceptions
are new LLP‟s unexpected by the market, announced for example after a supervisory control,
where the amount of LLPs is also determined by the controlling body. Importantly, Rajan
(1994) views loan growth as a symbol of a „managerial flaw‟, coming from sources other than
macroeconomic or internal bank conditions, which lies close to the overconfidence
assumptions.
Summing up, the existing bank literature associates rapid loan growth on developed
markets with an increased credit risk, notwithstanding its importance to revenue generation.
Empirical studies prove the relation between loan growth and risk, with rapid portfolio
expansion resulting in increased repayment problems in a three year lag. In addition, new
customers are of lower quality than the existing client base, but competitive pressures refrain
banks from charging adequate risk premiums on such loans. This may explain falling net
interest margins observed in international financial institutions in recent years and indicate a
growing overall risk profile.
Loan growth is demonstrated to be positively related to macroeconomic upswings and,
in parallel, to be a factor in spurring banking crises. On an individual bank level, loan growth
depends on inside factors such as the time that has passed since the last bad loan crisis and
outside factors, such as loan growth pursued by competitors. As a result, loan growth is a
crucial factor that expresses individual bank approach towards credit risk and its assessment
of future performance. This implies that both loan growth and its pricing, in the form of net
interest margins, will be vital building blocks of the overconfidence model in Section 4.
The last crucial factor, rarely absent from credit risk analyses and also used in the
overconfidence model are loan loss provisions (LLP). The most frequently applied approach
presumes bank-specific causes for LLP fluctuations. A prevailing view in the literature
implies that loan loss reserves should cover the expected part of loan losses within a credit
portfolio, as stipulated by e.g. Kim and Santomero (1993), Cavallo and Majnoni (2001),
17
Laeven and Majnoni (2003) and Perez et al. (2006). This expected part should be derived
from historical data on loan defaults and detailed credit risk analysis of the current credit load.
Bank screening systems should update information on repayment problems continuously and
reserve levels should be adjusted through loan loss provisions to account for this feedback. If
fully unexpected loan defaults surface, only these should be covered by bank capital. This
general treatment is shared by most researchers, with bank capital serving as a buffer against
the unexpected part of every bank‟s loss distribution.
However, many authors point out that in some cases loan loss provisions may serve to
fulfil other goals as well. As already mentioned, LLPs have a direct link with net income, in
that they are a substantial entry in the profit-and-loss account. In comparison to other items,
such as net interest revenues or costs, that are more stiff and exogenous, loan loss provisions
incorporate an important discretionary component. Bank managers may more freely shape
loan loss provisions than other profitability lines, using an excuse of higher (or lower) loan
loss reserve needs in given periods. This fact lies at the origin of a substantial area of research
on earnings management, both in corporate finance and banking literature. An important role
of managerial discretion causes earnings management to be a strong potential input to the
bank overconfidence theory and because of that it is presented in detail below.
Within the scope of this work, two main directions in the field of earnings
management in banking are taken under scrutiny. The first rationale has strongly negative
implications and refers to using LLPs for earnings management in order for managers and/or
shareholders to fulfil some private benefit goals. This wide field may include incentives
coming from moral hazard, or shareholder pressure on artificially boosting share prices.
Managerial compensation schemes may make salaries dependent on earnings stability, or
herding effects may induce managers to bring their bank‟s results in line with peers. Although
the list of similar reasons for earnings management through LLPs is long, they all have an
important common denominator – the true underlying ground for LLP changes is not related
to credit risk management. This is a very important distinction that is sometimes omitted in
the literature, especially among opponents of earnings management. It is in this first area that
managerial overconfidence could be placed. If loan loss reserves are used to fulfil other goals
than prudent credit risk administration, the managers may be overconfident towards their
abilities of dealing with credit risk problems in the future, or miscalibrate the size of these
problems. Faulty assessment of potential repayment trouble leads managers to shape
provisions freely, to fulfil private benefit goals rather than assure a safe functioning of a bank,
18
under all circumstances. Such earnings management bears strong negative implications for
bank safety and optimal functioning.13
A very different treatment should however be accorded to income smoothing driven
by the crucial second motivation in earnings management. This one is caused by credit risk
considerations, basing on prudent loan loss reserve creation and is described below.
A continuing discussion divides bank analysts towards the timing of loan loss
provisions. The first view stems from an „accounting-oriented‟ field and may be referred to as
a „fair-value approach‟. According to this rationale, existing LLRs should only cover the part
of the portfolio that has already turned sour or the default of which has become obvious. As a
result, all changes in LLPs reflect only current shifts in credit risk, with no provisions (or
capital) set aside for predictable defaults. Supporters of this policy claim that it allows better
information regarding developments in the bank that take place in subsequent periods.
Outsiders receive „snapshot views‟ of a bank‟s situation in every period and on this basis they
form their judgments towards developments expected in future periods. In such a setup banks
are more concentrated on current problems and do not account for future adverse
circumstances, even if these circumstances are predictable. As an example, banks with
aggressive and rapid loan expansion do not account for foreseeable loan losses in future
periods and maintain their loan loss reserves low, until repayment difficulties surface. Such
behaviour is expected by many to result in much stronger income fluctuations, with high
profitability during economic and bank-specific upswings and more severe losses at the
bottom of the cycle, where economic pressures are exacerbated by internal LLP needs.
pressures are exacerbated by internal LLP needs.
The second view on loan loss provisioning policy advocates its forward-looking
character. Wall and Koch (2000) present this in detail, within a general discussion of the
underlying rationale for the timing of LLPs. A forward-looking perspective indicates that
LLRs should also cover expected – and not yet incurred – losses in the loan portfolio, which
are derived from historical data. Once these losses materialise, they do not affect profitability
in a significant extent, as appropriate provisions have already been made. Usually an
important part of losses materialises during an economic downturn, which is accompanied by
a slump in bank earnings. As a result, provisions created earlier soften the blow on the bank‟s
bottom-line. Subsequently, as reserve levels plunge after having been used up in times of
trouble, the economic upswing is usually accompanied by an increased need for provisions.
13
For further discussion on various facets of earnings management, see also empirical results interpretation in
Section 5.
19
This is exacerbated by the fact that the loan portfolio expands most during cycle upturns, and
a given proportion of new loans is also (on a historical basis) likely to turn sour at some point.
Thus, healthier bank earnings in the better part of the cycle are somewhat diminished by
increased provisioning, for the sake of future defaults. Researchers refer to this whole process
as “income smoothing”, as loan loss provisions diminish the volatility of the bottom line,
improving it in times of trouble and diminishing during cycle upswings.
Income smoothing is considered as a positive phenomenon in many studies, such as
Laeven and Majnoni (2003). The authors provide strong empirical support for income
smoothing, using bank data to show negative effects of inadequate provisions. They
additionally refer to a problem of capital shortages and potential credit crunch during
downturns. Recessions usually cause a deterioration of credit portfolios, increasing banks‟
risk exposures and thus their capital needs. However, at this point in the cycle capital is either
more costly or unavailable, especially to weaker institutions. This is the moment where the
income smoothing supporters and opponents go in different directions. Advocates of prudent
provisioning (and thus income smoothing), including Laeven and Majnoni (2003), indicate
that adequate loan loss reserves diminish earnings- and capital pressures on banks in times of
crisis and banks can continue their lending operations, providing financing when it is needed.
Supporters of the „fair value‟ view, criticising excessive reserves during economic boosts,
believe that the whole effect of economic downturns should be fully visible.
At the same time, many researchers stipulate that the reason for the so-called “credit
crunch” during economic downturns is not a lack of adequate LLPs, but minimum capital
requirements imposed by central banks. As the fair value view implies, banks create LLPs
only for current periods, so an exogenously driven credit deterioration results in massive
LLPs and enormous pressure on capital. If regulators let banks maintain lower capital ratios
during recessions, or allow them to fluctuate around the cycle, then a temporary capital
shortage would not necessarily imply drastic decreases in new bank lending. Such a lending
squeeze is referred to as credit crunch and causes draining of funds from an economy that
desperately needs them. The discussion on the procyclicality of capital requirements has
become even more heated after the banking turmoil of 2007/08, when banks faced severe
capital shortages and many claimed capital requirements in such circumstances should be
temporarily eased. Although this discussion is beyond the scope of this work, it needs to be
underlined that no consensus on this topic exists. There is a possibility that introducing
temporarily lower capital requirements would in fact induce more risk seeking and moral
hazard behaviour by banks that would count on such „capital breaks‟ in case of trouble. If
20
banks engaged in income smoothing through LLPs in order to maintain prudent loan loss
reserves, the pressure on capital would also be eased. A similar view is expressed by Cavallo
and Majnoni (2001), who claim that “cyclical shortages of banks‟ capital may not only be due
to the risk based regulation of bank capital but most prominently to the lack of risk based
regulation of bank‟s loan loss provisioning practices” (p.1). Referring directly to overly lax
loan loss reserve requirements, they state that “when loan loss reserves are inadequate,
expected losses will affect banks‟ capital and the impact of capital shortages on the real
economy will be magnified” (p.1). These views are an adequate representation of other
advocates of introducing risk based regulation concerning reserve practices. Such rules would
make capital regulations more efficient and minimise their procyclical effects in exacerbating
economic downturns through putting pressure on bank capital. Needless to say, the idea of
implementing additional risk-curbing guidelines upon banks is being strongly opposed by
financial institutions themselves, which interpret this as a drag on profitability.
Opponents of earnings smoothing indicate that it introduces judgmental modifications
of earnings and reduces comparability between different institutions. In opposition to these
advocates of the „fair value‟ treatment, Kim and Santomero (1993) claim that forward-
looking, conservative provisioning makes bank earnings more meaningful to analysts, as they
show underlying profitability. According to them, loan losses are a normal part of the banking
business and are certain to happen, so in a way current profitability should include a net
present value of future loan problems.
LLPs are one of the most distinctive external signs of managerial approach to risk in
financial institutions. They are easily accessible, comparable and subject to human discretion,
so may be used as a proxy for bank approach to credit risk. Few data items possess similar
traits, especially in other risk areas, such as market or liquidity risk. Although value added of
loan loss provisions would be enhanced through inside data of particular bank policies, clients
and portfolios, they provide key information even in their publicly available form. In addition,
their universal character, despite regional differences in prudential regulations, make them an
excellent tool for international comparison. As mentioned above, LLPs should ideally be
analysed in conjunction with inside bank data regarding details of past and future credit risk in
particular institutions. The shortage of this kind of data makes numerous researchers turn to
studying LLPs in relation to other bank variables, especially capital levels and income, as well
as business cycle volatilities. Hence the empirical treatment of LLPs in Section 5, basing on
recent models incorporating LLPs. In the overconfidence context, the vital question that
emerges regards not only the relations that LLPs may have with other bank variables, but
21
what drives LLP creation in the first place. This issue draws on the above theoretical debate,
in an attempt to determine if income smoothing takes place and what motives are driving it.
As already mentioned, earnings management may be pushed by a prudent approach of
conservative executives, who aim to use earnings achieved during economic boosts to build a
reserve cushion for the future. Such an explanation of income smoothing causes many
researchers to approve of these practices. On the other hand, smooth income streams may be a
fulfilment of private benefit goals of managers and show little – or none whatsoever – relation
to prudent risk management. In such a case it becomes a suboptimal policy for the bank.
Critical voices of income smoothing do not however raise this argument, focusing instead on
its inappropriate accounting treatment and current reflection of bank fair value.
The argument for overconfidence arises in the second case of income smoothing, when
it is motivated by factors other than prudent credit risk management. If bank managers are not
adjusting reserve making to expected repayment problems but using them for other goals, it
may be caused by a serious miscalibration of potential future portfolio performance. This
issue is analysed in details in Section 5. The role of LLPs in overconfidence may be largely
explained by LLP relations with income and the presence of income smoothing. In addition,
underlying reasons for income smoothing found in empirical studies may provide valuable
feedback to assessing overconfidence of bank managers and how it may be expressed through
bank policies.
Empirical evidence for income smoothing is mixed. Empirically, income smoothing is
usually understood as a positive relation between loan loss provisions and earnings. The
results on earnings management differ. Many authors confirm its existence, including Collins
et al.(1995), Cavallo and Majnoni (2001), Laeven and Majnoni (2003), Perez et al. (2006) and
Quagliariello (2007), while others reject it (such as Beatty et al. 1995, Ahmed et al. 1999).
The most recent works on European banks confirm the existence of income smoothing on
more contemporary data and further analyses should rather concentrate on its character, rather
than its existence.
There are various aspects of credit risk, which are taken into consideration in the
literature concerning the fluctuations of risk and its relations with other variables. Loan
growth is almost uniformly considered an important drive of bank risk in general, with both
theoretical and empirical studies proving this link. Loan growth is frequently facilitated by
low interest charged on new lending, which is however accorded to lower quality clients. Due
to competitive pressures, these customers are not required to pay adequate risk premiums, but
aggressively growth-oriented banks grant such loans nevertheless. In such cases, bank
22
managers display classical overconfidence through miscalibrating future loan performance
and bank risk management abilities. As a result, these two variables are a necessary input into
any model assessing origins of credit risk, even if their role in the literature remains under
discussion.
Loan loss provisions are under continuing debate as to their true underlying role in
bank results. A repeatedly established relation between pre-provisioning income and reserves
created in a given year indicate that LLPs are used in income smoothing processes. On the
positive side, the origins of this process may be prudential considerations of credit risk
managers, who use higher income periods to build reserve buffers that reflect anticipated
portfolio deteriorations of future periods. On the negative, such earnings management may be
motivated by managerial or shareholder self-interest, as these parties may derive private
benefits from a less volatile income stream displayed by their bank. Here, the potential for
moral hazard and/or overfoncifdent managerial behaviour emerges. In order to verify this, the
rationale underlying income smoothing (rather than its pure existence) has to be established.
In conclusion, the three issues of loan growth, net interest margins and origins of income
smoothing emerge as most meaningful in the area of underlying reasons of credit risk taking,
motivated also by overconfidence. They are therefore included into the empirical model
constructed in Section 4 and 5.
4. Empirical model assumptions and construction
This Section puts together all previous elements considered in this work and they are
subsequently employed to construct the empirical overconfidence model, which is verified in
Section 5. At the beginning, various applications of overconfidence in finance are
summarised, to examine their applicability into the financial institutions context. Then,
theoretical implications from the literature review on credit risk are transformed into possible
overconfidence proxies. In parallel, the main empirically verifiable relations are described,
creating the framework of the model. These links are strongly shaped by data on Western
European banks available in the Bankscope database, and by the main trends discovered in the
brief analysis of average values (not shown). Finally, two primary forms of the
overconfidence model are constructed and described, the overconfidence risk model and the
overconfidence profitability model.
As already described, overconfidence has long ceased to be regarded as a purely
psychological phenomenon, emerging only in laboratory conditions. Powerful implications of
23
various facets of overconfidence are visibly appreciated by researchers from other disciplines,
especially in the area of behavioural finance. Here, two directions of research emerge, one
focusing on financial markets and the other – still fledgling – on corporate behaviour.
Effects of overconfidence proven to exist on financial markets have already been
described. Empirical tools used to analyse overconfidence on financial markets, which are in
fact instruments for studying individual investor behaviour, cannot be easily transferred to
bank analysis. This is mainly due to a diverging set of operations and investments that are
undertaken by banks, in comparison to financial market investors. For the latter,
overconfidence may be defined as overly tight probability distributions and underestimation
of price volatilities, as well as overvaluation of own trading performance versus average
market returns. Investors are believed to overestimate the precision of (usually private)
information and, more generally, underestimate risk that leads them to hold more risky
portfolios. These traits may not be directly transferred to the banking context. In addition,
effects of overconfidence on financial markets may be directly measured through earnings
analyses of investors and comparisons of returns of individual stocks/portfolios versus market
averages. The financial institution background is much more complex and no straightforward
links exist between e.g. granting a loan in year t and earnings in year t+1, unlike in the case of
buying and selling shares. Nonetheless, the relaxed approach of behavioural finance
researchers in converting original psychological facets of overconfidence into financial
market reactions allows a similar treatment for banks, which is discussed below.
The second area of overconfidence studies relates to examining its existence and
effects in a corporate context. The current literature refers to mergers and acquisitions, as a
possible area where overconfidence may emerge, and to examining the financial structure of
companies. Malmendier and Tate (2008) assume that “overconfident managers overestimate
their ability to create value” and that they “overestimate the returns they can generate both in
their own company and by taking over other firms” (p. 22). Importantly, the authors also
specify the definition of overconfidence and bring forward an important difference between
overconfidence and other possible explanations for managerial errors. “Alternatively, one
may call a CEO who overinvests in his company and who does too many and bad mergers
simply „stupid‟ or low-skilled. Since the biased managerial decisions systematically point to
overestimation of future returns, overconfidence characterizes the type of mistake more
tightly” (p.38). Heaton (2002) presumes that overconfident managers “systematically
overestimate the probability of good performance and underestimate the probability of bad
firm performance” (p.33), while Malmendier and Tate (2005) extend this onto managerial
24
overestimation of own skills. These corporate finance measures can be more easily
incorporated into the financial institutions framework and they underlie the assumptions on
overconfidence of bank managers in this work.
Even among specific corporate analyses, overconfidence studies relating explicitly to
financial institutions are rare.14
This is most possibly due to the prevalence of the moral
hazard theory that has dominated the risk taking studies of banks. Whenever a more
aggressive risk approach appears in financial institutions, it is moral hazard that is naturally
considered the underlying ground for such managerial gambling. As demonstrated in Section
1, moral hazard does not account for possible cognitive biases in managerial calculations and
decisions regarding the risk appetite levels. Overconfidence allows for such errors, implying
that the pre-calculated high expected returns from moral hazard need not be the only answer
to aggressive risk approaches.
Before engaging into the analysis of a transfer mechanism of overconfidence facets
into bank results, a short discussion follows below, regarding the preliminary assumptions
upon the definition of „overconfident banks‟ made in this work.
The main difference between investor behaviour on financial markets and actions of
companies (and banks) is that individual investors‟ deviations from rationality may balance
out in the financial market, as stipulated by Roll (1986). Thus, on average, markets may be
assumed to be rational. As a result, the effects of overconfidence upon financial markets may
be more easily diminished, if some participants are assumed to be under- and others
overconfident. On the other hand, researchers concentrating on individual investor behaviour
may model overconfidence per se, ignoring the existence of a potential underconfident
investor for every overconfident one.
A different approach should be accorded to companies, which are rarely viewed as an
aggregated sector. A possibility that individual deviations from rationality cancel each other
out is not as easily allowed in the corporate analysis. Studies of corporate behaviour usually
examine a single-firm case, where different decisions are made and they influence outcomes
in this particular company. Thus, individual deviations from rationality in particular
institutions are not cancelled by rationality of other firms. Similarly, financial institutions in
this work are not analysed from the point of view of the banking sector as a whole. Instead,
mechanisms of decision making in single banks are taken under scrutiny, to bring out the
effect that overconfidence may have upon individual banks. Only through these individual
banks the implications for banking systems can be taken into account, but this is not the prime
14
However, see Niu (2009) for a preliminary work on bank risk taking and overconfidence.
25
objective of this work. Similarly to different investors on the financial market, diverging
banks may also appear in the system, with varying degrees of overconfidence. A broader
analysis of systemic consequences of overconfidence on international banking sectors is a
promising direction for future research, but remains outside the scope of this text.
In the corporate context, the research is still diversified and infrequent, but it appears
to share a common assumption. This proposition suggests that managers, seen as individual
CEOs or as the management board, have a crucial decision input into choices made by
companies. It has been explicitly expressed by Roll (1986), who in his study of corporate
mergers stated that “takeovers reflect individual decisions” (p.199). Roll allows for a
possibility that it is CEOs themselves that make crucial decisions in the company, such as the
ones regarding potential takeovers. The same assumptions are made in recent empirical work,
such as Malmendier and Tate (2005 and 2008), where the decisions of CEOs are naturally
viewed as final choices of companies that they manage. As a result, the institutional
framework of a firm, viewing the company as a complex mechanism driven by multilayered
incentives, is reformulated. It is suggested that managerial power, influences and forces of
arguments of CEOs (or management boards) may override the institutional framework
concerns. Malmendier and Tate (2008) explicitly state that they ignore the manager-
shareholder conflict, as overconfident managers may strongly believe they are acting in the
best interest of shareholders. This feeds into the moral hazard debate presented in Section 1,
where determining the kind of principal-agent conflict is not found to be crucial to managerial
risk taking, at least in this work. Malmendier and Tate go even further, explicitly stating that
no shareholder-manager conflict needs to exist, in the overconfidence setting. This broadens
the scope for bank risk analysis beyond the moral hazard setting.
Summarising the above considerations, it is presumed in this work that individual
overconfidence of bank managers is transferred on to institutions they administer. Thus the
notion „overconfident banks‟ really implies that these are banks run by overconfident
managers. Needless to say, the most straightforward way of identifying and measuring
overconfidence in bank managers (and CEOs of other firms) would be to test them under
laboratory conditions, using psychological methods extensively described in Section 1. As
this is obviously not possible, this and other works concentrate on determining
overconfidence proxies in companies that may demonstrate managerial overconfidence, or
that are most affected by it. Using theoretical underpinnings from Chapters 1 and 2 regarding
bank risk taking and established empirical relations between main variables in banking, the
next part identifies and discusses overconfidence proxies for banks put forward in this work.
26
The first step in finding adequate overconfidence proxies is defining the type of risk taking in
banks that can be empirically verified. Credit risk remains on the forefront of contemporary
bank risk considerations. The relation of credit risk with other types of risk, its role in market
risk and liquidity risk and especially its harmful input into the recent financial crisis make it a
prime candidate for overconfidence considerations. As demonstrated in descriptive statistics
above, credit risk is not represented by fully satisfactory data for the large Western European
banks sample, but many items are widely accessible. The same may not be said about market
risk and liquidity risk, where detailed treasury information is in hands of national supervisors
only. The uniformity of credit risk considerations on an international level are a further
argument. All banks face trade-offs between prudent and aggressive loan portfolio growth,
with similarly burdensome pricing pressures and the perpetual necessity of determining
adequate provisions. In addition, the personal influence of bank managers upon credit risk
management is difficult to question. Decisions regarding loan loss reserve levels, for example,
are straightforward, easy to convey and to change, and they rapidly and openly affect
bottomline results, in a predictable direction. A very different mechanism is in place for
market risk, where doubts as to managerial capabilities of understanding the underlying risk
have been raised numerous times. As a result, the link between managerial overconfidence
and decisions regarding treasury risk exposures may be biased due to the „managerial
ignorance factor‟. This factor is impossible to assess using externally available data. Such
doubts do not arise in the credit risk area, where decisions regarding e.g. the creation of low
reserves for risky customers can scarcely be explained by management unawareness.
In consequence, credit risk emerges as the most adequate area of bank risk that can be
used to analyse overconfidence. This is only partly due to data availability problems that
emerge in other primary risk areas, such as market risk, liquidity risk and operational risk.
The most important arguments supporting credit risk application in the overconfidence
context are its transparent links with managerial choices, its straightforward impact on bank
results and thus its relation to potential managerial private benefits, such as performance-
related bonuses. In addition, credit risk has been the main factor in the financial crisis of
2007/08, so its role should be analysed in detail.
The aim of this work is to analyse potential effects of managerial overconfidence on
bank risk profiles and results. This implies that a few key variables are necessary to represent
each of these elements. First, overconfidence proxies are determined. They represent
overconfident managerial behaviour in the context of financial institutions and are primary
explanatory variables in the overconfidence model. Second, representations are chosen for
27
risk and profitability in banks, forming the dependent variables of the overconfidence model.
Lastly, a short discussion presents remaining exogenous variables, which are further
interpreted in the estimation results in Section 5.
Taking into account the theoretical implications from Section 1 and 2, two
overconfidence proxies are determined. Both proxies are chosen from measures linked to
credit risk, as it has been identified above as the most reliable reflection of potential
managerial overconfidence. The first one is loan growth, frequently used in the literature as a
representation of credit risk, or even total bank risk.15
Loan growth comprises all facets of
overconfidence, reflects managerial preferences and is straightforward to interpret. If
managers believe that their abilities are high, future risk is moderate and returns are
promising, they are bound to engage in more rapid growth. This effect is exacerbated if they
overestimate their own abilities of managing risk, either towards an objective benchmark
(miscalibration) or in comparison to other bank managers (better-than-average effect). Loan
growth accurately mirrors bank (and managerial) expectancies towards the future, both of the
bank prospects and general macroeconomic environment, with a dominating positive outlook
(unrealistic optimism). In addition, expanding portfolios are assumed to remain under full
control of banks, which believe to implement the best possible credit risk management tools.
Thus, as demonstrated by the financial crisis, managers erroneously judge to be able to
control developments in their loan books and underestimate the weight of possible adverse
factors (illusion of control). These potential external issues may be economic downturns,
sectoral problems and moral hazard on the side of borrowers, which banks are evidently
unable to control. In consequence, loan growth covers most important aspects of
overconfidence and represents an adequate proxy in a bank setting.
The second overconfidence proxy is net interest margin (NIM), discussed in the
theoretical Section 3. NIM is not as straightforward as loan growth and its interpretation has
varied. In this work, NIM is viewed as a pricing level that banks impose on clients. High NIM
indicates that loans are charged elevated interest rates and/or price of funding is low. The
main assumption on NIM in this work is that it is likely to be negatively linked with
overconfidence, with low margins reflecting higher overconfidence. Managers who
underestimate risk, overestimate their risk management abilities and control over future
events, and who believe the future will be unrealistically bright, are more apt to grant loans
with lower interest rates. Due to their overconfidence biases, they less frequently predict
negative outcomes, or underestimate the cost of these outcomes. As a result, they do not
15
For references, see Section 3.
28
require sufficient risk premiums from risky clients and their NIM is lower than that generated
by their rational peers. On the other side, overconfident managers are also more likely to
engage in pricing wars on the funding side, offering inadequately high deposit rates. The drive
to expand necessitates financial means, which are likely to be more costly if acquired by
overconfident managers. Such executives overestimate future returns, so funding costs may be
regarded as a mean to achieve such high earnings. Lower interest rates on new, risky loans
and high deposit costs due to pricing competition both lead to diminished NIM. Thus reduced
NIM is most likely intended to be compensated through higher volumes, an effect that is also
verified by the overconfidence model. Summing up, NIM is a secondary overconfidence
proxy, as its links with overconfidence are not as straightforward, but the sign of the
correlation is most likely negative.
Representations of risk and profitability are the next vital step in constructing the
overconfidence model. In the area of risk, the focus remains on credit risk. Credit risk is an
adequate representation of other risks and is of a longer-term nature. Market risk is highly
volatile and its fluctuations are frequent and significant. As already discussed, market risk
exposures of a given bank do not always mirror the risk approach of its top managers, due to
complex underlying treasury instruments. Similar concerns regard liquidity risk that strongly
depends on financial market factors and may also change frequently. Both of these risks also
lack the adequate data representation that would allow a robust empirical verification.
Credit risk relatively accurately represents overall managerial approach to risk. In
addition, it does not change daily, so longer trends can be easily captured. The current quality
of an individual bank‟s loan book results from past credit decisions regarding target growth,
client base, products and pricing. The most important of these decisions, as well as the general
credit trends in the bank are decided upon by the top management, so no major disruptions
between credit risk and underlying managerial preferences are expected.
In the credit risk area, a few potential variables can be identified to accurately
represent current exposure levels. The level of non-performing loans in relation to total loans
could be a highly appropriate indication of risk. Nonetheless, the already mentioned
classification differences make this a unobvious comparison item. In addition, insufficient
data in the final data sample in this work prevents NPL usage. If a fuller data sample is
accessible, this could be a valuable proxy of risk. Similar arguments relate to reserve coverage
of non-performing loans. Loan loss reserves are not treated unanimously across different
countries and some accounting rules allow hidden reserves. More importantly, collateral data
29
is unknown, so bad loans with uneven reserve levels may simply be differently secured. Last
but not least, data availability is also weak.
As a result, this work follows similar models in empirical bank analysis that employ
loan loss provisions (LLPs) as a representation of current credit risk. The advantages of loan
loss provisions have already been discussed in descriptive statistics, so only a summary is
due. LLPs are highly comparable between banks and countries and are often viewed as a
dynamic reflection of developments within credit risk. LLPs directly affect net income, so a
risk-return link is highly exposed. LLPs should potentially concern current developments
within the loan portfolio, but as discussed in Section 3, they may be created ex ante, to build a
reserve cushion for loans that are expected to deteriorate. In addition, some LLPs may also be
made with a delay, for loans that have been poorly performing but erroneously considered as
promising. In this respect, LLPs signify a risk exposure that is of a longer term rather than
relating only to a current year, even if this is the conventional „fair view‟ approach displayed
by some researchers. Last but not least, strong data accessibility provides the final argument
for applying LLPs as a proxy for risk exposure levels in individual banks in given years.
In the area of profitability, a less complex choice background is at hand. No efficiency
indicators, similar to the DEA framework, are included in this work. The profitability ratio is
aimed to provide a picture of bank earnings that is straightforward, commonly used and easy
to interpret, so that its link with overconfidence may be explicitly established. To this aim, the
return on assets (ROA) is chosen, similarly as in numerous other works. Despite some
drawbacks, discussed in Section 5, ROA fulfils the above requirements and accurately
represents bank results. In the robustness tests in Section 5 it is replaced with ROE and the
results‟ implications do not change.
The above identification has yielded the rough structure of the overconfidence model.
Managerial overconfidence is represented by loan growth and net interest margins, reflecting
all overconfidence facets. The model is aimed to check if these overconfidence proxies affect
risk and profitability levels of individual banks, represented by loan loss provisions and ROA,
respectively. This setting implies two primary versions of the model. The first setup estimates
effects of overconfidence on risk, the second measures overconfidence repercussions on
profitability. They are respectively referred to as “the overconfidence risk model” and “the
overconfidence profitability model”.
The existing empirical research (see Section 3) suggests that negative effects of
aggressive risk policies, represented in this work through overconfidence, are likely to surface
with a time lag. As a result, both overconfidence proxies are included with time lags of up to
30
four years. The limited length of this delay is due to data restrictions only and in more
abundant data bases could be extended. Time lags allow to verify the influence of
overconfidence through time and check if its immediate effects persist in the period of up to
four years.
The overconfidence risk model comprises pre-provisions income as an additional
explanatory variable, not used in the overconfidence profitability model. Pre-provisions
income is introduced into the risk treatment for a few reasons. Its first role is to control for the
level of income that is used as a base for LLP deductions. More importantly, it allows to
check for income smoothing that has been examined at length in Section 3. Contemporary
research indicates that banks engage in earnings management through LLPs, which can be
verified using pre-LLP income in the model. The primary intention in this work is however
not establishing the existence of income smoothing (that has already been successfully
proven), but examining its causes. There is a potential role of income smoothing in the
overconfidence debate. In a thriving environment, if managers use their discretion to diminish
high earnings of their banks, in order to create more reserves for potential future downturns,
this implies that they are driven by a conservative risk approach. In such a setting, provisions
should also be positively linked to the expansion phases of the economic cycles, so that
reserve cushions are at hand when the economic environment deteriorates. If banks smooth
earnings because of prudential reasons, their overconfidence is possibly not significant. On
the other hand, if banks use LLPs in order to manage earnings because of potential private
benefits of managers, who derive advantages from diminished fluctuations of the bank‟s
bottom line, the fact of income smoothing does not imply that managers are not
overconfident. On the contrary, it strongly indicates that their assessments of future outcomes
(such as the quality of the loan portfolio) may be flawed and their risk taking is affected by
overconfidence.
Both the overconfidence risk model and the overconfidence profitability model
include additional control variables. They are aimed to stabilise internal bank factors and
operating environment issues, so that effects of overconfidence proxies on risk and
profitability can be interpreted net of other elements. The primary control variable is bank
capital, described in Section 3 as an important factor in bank evaluation. These theoretical
considerations have not unequivocally indicated the sign of the relation between capital and
risk, and capital and profitability. This is also not the aim of this work, as the background for
capital analyses should be expanded. In addition, data restrictions force the usage of total
equity as a proxy of bank capitalisation, which is merely its rough representation, as shown by
31
the Basel II analysis. If the exact relation of capital and risk/profitability were to be
established, more detailed capital calculations, including regulatory capital adequacy
treatments, would have to be applied.
Other control variables comprise bank size, commonly expressed as total assets, and
the ratio of loans to assets. The latter has been brought forward in the descriptive statistics,
where Western European banks have witnessed a visible hike in loan intensity ratios in the
studied period. This can be derived from elevated loan growth ratios and is an important sign
of a potential shift in bank activities, which have become more loan-concentrated than before.
In addition, loans/assets ratios are a further check to loan growth, as the latter does not
account for the previously existing share of loans in the balance sheet, but for its growth rate
only. External operating environment may be controlled through diverse indicators. This work
follows conventional treatments and applies a standard set, comprising GDP growth and
inflation rates, with more details discussed in Section 5.
Next, the construction of the model is presented, using theoretical overconfidence
underpinnings and contemporary issues in bank risk assessment. The overconfidence model
for banks comprises two versions, the overconfidence risk model and the overconfidence
profitability model. Both of these specifications are verified empirically in Section 5, using
Western European bank data for years 1997-2008.
5. Overconfidence modeling and estimation results
Section 4 has sketched the outline of a parsimonious overconfidence model, the
estimation of which is presented below. The main hypothesis that is to be verified by this
model states that overconfidence may affect both risk and profitability of financial
institutions. The influence of increased risk taking, driven by overconfidence, may be positive
in a short term, but is likely to turn negative in a longer term, interpreted in this setting as a
period extending over two years. This implies that although banks may not suffer adverse
consequences of overconfidence immediately and some may even profit from it, in a longer
term the negative side is likely to prevail. Importantly, the unfavourable repercussions of
earlier risk taking may appear not only in current exposure levels, but also in overall
profitability. If aggressive managerial policies boost risks and profitability simultaneously,
they should not be regarded as negative. In such a case they rather constitute a gain in a moral
hazard setting that is equivalent to a risk premium. However, if past risk taking increases
pressure on contemporary reserve levels, but at the same time diminishes profitability, it may
32
indicate that banks accept high risks without fully realising their negative implications. Such a
result may suggest that banks take risks, because they overestimate their future performance
and believe that they are able to cope with future risk better than their peers. In addition, they
may erroneously presume that they are capable of controlling external events, such as
borrower moral hazard or operating environment, and they are inclined to see the future as
unrealistically bright. These overconfidence facets may motivate managers to assume
additional risks, which however have adverse consequences for their future performance. The
secondary hypothesis that is verified states that there are differences in overconfidence effects
between potentially more and less overconfident bank groups.
In order to verify the above hypotheses, two separate models of overconfidence are
constructed and estimated. These are the overconfidence risk model and the overconfidence
profitability model respectively. In both cases, the primary model is estimated first, using the
whole sample of banks, in order to study the principal hypothesis regarding overconfidence
effects. The same model is subsequently verified using two subgroups of banks, potentially
more- and less overconfident, in order to analyse differences in overconfidence effects
between these groups. Finally, robustness tests are performed to check the stability of results
in various setups.
All variations of the overconfidence model are estimated using the panel data fixed
effects approach, on a dataset of c.311 financial institutions from Western Europe. The data
constitute an unbalanced panel, with an average period including four years of data per
bank.16
Lagged explanatory variables are integrated into the model, in order to demonstrate
delayed effects of overconfidence proxies on dependent variables of risk and profitability. The
fixed effects approach assumes that there may be time-constant, unobserved factors („fixed
effects‟) that affect particular individuals (here: banks) across various periods of time. In
particular, fixed effects imply that this unobserved heterogeneity is correlated with
explanatory variables, in addition to affecting the dependent variable.17
The application of
fixed effects estimation is particularly well suited to analysing institutions that are affected by
unobservable factors for which no data is available, but these factors are very likely to
influence the performance of such institutions. In banks and companies such features may
include corporate governance, managerial skills, moral hazard or corporate culture and
heritage, for which few quantitative measures exist. Such traits are unlikely to rapidly change
from one period to the next and they are considered to have a non-zero effect on risk and
16
Data outliers exclusions are included in the overconfidence risk model. Please note that the average four
observation periods imply in fact eight periods, as there are four time lags included. 17
The fixed effects treatment bases largely on Wooldridge (2002).
33
results. Within the financial institutions context, attitudes towards risk in particular bank
departments influence not only final risk or profitability levels, but in the first place risk
variables that shape these final results, such as e.g. loan growth or net interest margins in the
context of credit risk. In the overconfidence model, unobservable bank attitudes towards risk
and its repercussions are assumed to not only shape the final risk or profitability, but to be
correlated with explanatory variables, such as loan growth. This implies using the fixed
effects model. In order to eliminate the time invariant fixed effects, the fixed effects
transformation has been applied (Wooldridge 2002). This fixed effects transformation is also
referred to as “the within transformation”, because the time variation in both y and x is used
within each cross-sectional observation. After eliminating the unobserved fixed effect ai, it is
possible to estimate the models using pooled OLS, applying the fixed effects estimator, or the
“within” estimator. The goodness-of-fit measure in such a case, the “within” R-squared, is
interpreted as “the amount of time variation in yit that is explained by the time variation in the
explanatory variables” (Wooldridge, 2002, p.444).
Some authors in most recent empirical literature apply dynamic panel data modelling
to study interactions of bank variables, including lagged dependent variables into the
estimation. This alternative is rejected during the construction of the overconfidence model
for several reasons. Firstly, the dataset is not sufficient to allow a robust dynamic estimation.
Secondly, a high technical level of sophistication of dynamic panel data models frequently
overshadows economic conclusions that – intuitively – should also appear in more
straightforward setups. Thirdly and even more importantly, dynamic panel data models do not
always accurately capture the economic reality in which banks operate. Existence of links
between past and future values of dependent variables is not a straightforward and obvious
assumption to make, contrary to what some authors imply. Introducing dynamic treatments of
loan loss provisions, for example, implies that credit policy and loan loss reserve policy are
decided upon in the long term and not in a period-by-period setting. Some long-term trends
always exist, but a possibility of independent annual decisions on provisions should be taken
into account. Provisions are at least as much a function of rapidly changing economic
conditions and current internal situation of the bank, as a consequence of last year‟s
provisions. As a result, demonstrating a link between current and previous year‟s loan loss
provisions does not prove causality and may in fact be spurred by external economic
conditions. In consequence, no dynamic setup of dependent variables is applied in this work
and the time factor is added only to exogenous variables, in the form of time lags. This
34
simplifies the econometric procedure and makes interpretation of results much more
straightforward.
Overconfidence risk model
In this section, possible links between overconfidence proxies and risk are studied
through fixed effects panel data estimation of the proposed overconfidence risk model. The
structure of this section is as follows. First, a general model of effects of overconfidence on
risk is constructed. Using the whole bank sample, the model is estimated and discussed.
Subsequently, subgroups of potentially more and less overconfident banks are used to re-
estimate the model. Lastly, robustness tests are performed in order to verify the stability of
end results in other potential specifications and setups.
The overconfidence risk model presented below incorporates links between
overconfidence proxies (explanatory variables) and credit risk (dependent variable), taking
into account a possible delayed effect that overconfidence may have on risk exposures. The
main proxies of overconfidence, together with their origins and implications, have already
been demonstrated in Section 4. They comprise the primary variable, loan growth (LGR), and
the secondary variable, net interest margin (NIM). Both of these variables are believed to be
strongly influenced by overconfidence, so a correlation between overconfidence and
explanatory variables is assumed. Credit risk, proxied by loan loss provisions, is presumably
also shaped by overconfidence, following the theoretical treatment from Section 4, Section
3.2. The presumed correlation between unobserved fixed effects (overconfidence) and
dependent and explanatory variables implies choosing fixed effects for the estimation of the
overconfidence risk model, as described in the introductory section above.
The primary goal of the overconfidence risk model is to verify if loan growth and levels of
interest margins have an effect on risk in a short and longer term (up to four years),
controlling for internal bank characteristics and macroeconomic conditions. The general form
of the model is derived from empirical studies analysing procyclicality of loan loss provisions
(Bikker and Metzemakers 2004) and income smoothing (Laeven and Majnoni (2001), Cavallo
and Majnoni (2002), Fonseca and Gonzalez (2008)), which is however only a secondary result
in the overconfidence risk model estimation. (5.1) presents a general version of the
overconfidence risk model.
35
∑
∑
(5.1)
Where:
is the dependent variable, showing current year‟s loan loss provisions (LLP) scaled
by previous year‟s assets
is loan growth (primary overconfidence proxy) lagged by 0-4 years
is net interest margin (secondary overconfidence proxy) lagged by 0-4 years
is the explanatory variable controlling for pre-provisions operating income
(INCOME), scaled by previous year‟s assets
is a control variable for the size of the bank (total assets in current year)
is a control variable for the share of loans in total assets
is a control variable for the level of capital
is a set of three macroeconomic variables, comprising respectively: inflation, current
GDP growth and GDP growth lagged by 1 year, controlling for operating
environment
The dependent variable, LLPs, is scaled by previous year‟s total assets. Other authors also
apply a similar approach, basing on a notion that current year‟s loan loss provisions usually
refer to last year‟s loans (and thus – assets), as loans rarely default during the first year of
granting. Year-end assets also include loans granted e.g. in December, so referring LLPs to
them is not realistic. Some authors use averages from the current and previous year, to
account for the fact that new loans should statistically be evenly distributed throughout the
whole year. However, to ascertain that no important recent trends are omitted and to
accentuate all discrepancies coming from using current versus previous year‟s assets, a
simultaneous estimation is performed, via the same model scaled by current year‟s assets, as
expressed by (5.2).
∑
∑
(5.2)18
18
For description of variables, see (5.1). on the previous page.
36
The primary overconfidence proxy, in (5.1) and (5.2), is current and lagged loan
growth. LGR is the value of nominal loan growth (in %) from year to year, in line with its
conventional treatment in the literature. The secondary proxy is current and lagged net interest
margin. NIM is the net interest margin, meaning net interest revenues (interest income
reduced by interest costs) divided by total assets (averaged over previous and current year).
Both NIM and LGR include lags up to four years, in order to study effects of earlier
overconfidence on later risk. The maximum length of lags is determined by data restrictions in
the available dataset only – introducing a fifth lag drastically decreases the number of
observations and yields estimation results statistically insignificant. Extending the number of
lags would allow to account for more diversified phases of the economic cycle and potentially
improve the quality of the estimation. Nonetheless, four years of lags should suffice to
demonstrate a longer-term effect in financial institutions and other authors stipulate three
years to be the usual lag necessary for irregular loans to emerge.19
The main explanatory variable in the usual income smoothing setup is pre-provisions
operating income, scaled by previous year assets. A positive significant relation between pre-
provisions income and LLPs proves income smoothing in studied banks, when larger income
is accompanied by higher provisions. The overconfidence risk model includes pre-provisions
income (also scaled by total assets) to check for income smoothing to some extent, but
primarily to control for the level of income from which provisions are subtracted. This
follows a rationale that banks with very high profits are likely to have different levels of
provisions than banks with very weak profits. Assuming that LLPs are created in separation
from the level of overall profits achieved by a bank in a given year does not seem realistic.
Again, pre-provisions income is scaled by either lagged or current assets, showed respectively
in (5.1) and (5.2)
Income smoothing in the overconfidence risk model is treated as a side result, in the
sense that its existence has already been proven by other models.20
However, the rationale
underlying income smoothing is essential for overconfidence considerations. If banks manage
their earnings through LLPs because of prudential reasons, as proposed by many authors, the
existence of income smoothing has positive implications. This would be the case if banks
created provisions ex ante, i.e. before irregular loans start to emerge. Such prudency implies
building reserve cushions in phases of rapid loan expansion and/or during economic cycle
booms, both of which frequently run in parallel. Reserve safety nets can be utilized when
19
See e.g. Foos et al. (2007). 20
A detailed discussion of income smoothing in the literature is presented in Section 3.
37
economic environment deteriorates, loan growth slouches and irregular loans surface more
vehemently. Income smoothing motivated by prudency thus implies not only a positive
relation between LLPs and pre-provisions income, but also a positive relation between LLPs
and both current loan growth and GDP growth. Such a result means that in times of rapid loan
expansion and higher GDP growth more provisions are created. An appearance of income
smoothing without these two relations indicates that earnings management may be motivated
by reasons other than prudent ex ante provisioning policies, e.g. managerial self-interest.
Thus, the overconfidence risk model verifies reasons for income smoothing rather than only
the existence of income smoothing as such.
The main control variables include both bank-specific and economic environment
variables. Bank-specific data are size, levels of capital and loan-intensity of assets. Size is
represented by total assets (divided by 1mln, to make the coefficient transparent). Capital is
represented by total equity, habitually including retained earnings and current year‟s net
income. Although either tier 1 or total capital as implied by Basle II would be more accurate,
many banks in the dataset do not have these entries. Similarly, it would be advisable to
exclude current year‟s net income, but the equity split is also not always provided. As a result,
total equity available for all banks is applied and this ensures a consistent comparison. Capital
is incorporated as a control explanatory variable, because some authors stipulate the existence
of capital management through LLPs. However, Perez et al.(2006) claim that an appearance
of capital management in many studies may be due to the fact that models include both
current year‟s LLPs and current year‟s capital levels. The level of LLPs determines net
income and thus influences capital levels, so there may be a spurious correlation between
these two variables that causes the misinterpreted “capital management”. Once previous
year‟s equity is used, capital management disappears, as shown in the model of Perez et al.
(2006). This rationale is also applied in the overconfidence risk model and motivates the
inclusion of the previous year‟s equity/assets ratio.
Loan-intensity is measured through a loans/assets ratio. This control variable is added
because loan growth itself, even with time lags, does not account for the original share of
loans in assets. Banks with a very low proportion of loans are bound to have lower LLPs and
this has to be controlled for, if a pure effect of growth and margins on risk is to be established.
In the robustness tests, lagged values of loan-intensity are additionally attached to expand this
control on to previous periods and check the effect of loan-intensity on later risk, assuming
the same pace of growth and margins.
38
Last but not least, macroeconomic variables account for the economic environment in
which various banks operate. As in similar studies, these variables are the annual inflation rate
and GDP growth. Due to lag effects of economic growth, as well as the fact that some
variables are scaled by previous year‟s assets, both current and previous year‟s GDP growth
are accounted for.
Estimation results of the overconfidence risk model using the total data set bank sample
The overconfidence risk model is estimated using data from the final bank sample of
311 Western European banks from the period 1997-2008, presented in Table 4.0.21
All
financial results of banks are taken from the Bankscope database, while macroeconomic data
are OECD numbers.22
The dataset has been derived on the basis of top 500 Western European
banks (according to total assets, as of end-2007). The final set excludes outliers (1st or 5
th
percentile, depending on the variable) and banks for which no unbroken string of four years‟
data for the main variables is available. Both main versions of the model, represented by (5.1).
and (5.2), are verified.
Table 4.0a. Number of banks per country in the final data set
Country No. of banks Country No. of banks
Austria 24 Luxembourg 2
Belgium 7 Netherlands 11
Denmark 10 Norway 15
Finland 6 Portugal 7
France 51 Spain 56
Germany 19 Sweden 8
Greece 10 Switzerland 7
Iceland 1 United Kingdom 39
Ireland 8
Italy 30 Total 311
Source: Own calculations, based on bank results in Bankscope, for years 1997-2008.
Table 4.0b. Mean and standard deviations of the main variables
Variable
No.
Observ.
Mean
Stand. Dev.
LGR 1189 13.504 10.315
NIM 1189 2.093 0.988
ROA 1189 0.731 0.451
ROE 1189 11.289 6.116
Loans/Assets 1189 67.232 16.170
LLPs/Assets 1189 0.279 0.233
LLPs/Assets at t-1 1189 0.311 0.263
21
Some specifications include a lower amount of banks (309), as some variables are available for a smaller
amount of institutions. These differences are however negligible. 22
The author has assembled the database during a research period at the National Bank of Poland (NBP), the
assisstance of which is kindly appreciated.
39
Pre-prov Income/ assets 1189 1.126 0.586
Equity/assets at t-1 1189 6.699 2.915
Assets_1mln 1189 0.081 0.203
Inflation 1189 2.347 0.829
GDP growth 1189 2.509 1.250
In order to precisely estimate coefficients of particular overconfidence proxies, five
different specifications are applied. The most accurate representation of (5.1). is comprised in
Specification 1 and includes all main explanatory and control variables simultaneously.
Remaining versions include reduced forms of the general setup that verify if excluding some
groups of variables affects overall goodness of fit and significance of particular coefficients.
Specification 2 excludes net interest margin (with lags), in order to measure pure income
smoothing effects. As net interest margin and pre-provisions income are clearly positively
correlated, including them both (e.g. in Specification 1) is expected to diminish the weight of
pre-provisioning income as such, reducing it by the interest income part. Such a division
allows to precisely estimate dependence of provisions on different sorts of income separately,
but it biases pure pre-provisions income smoothing downwards. This is the reason for
introducing Specification 2, which assesses income smoothing. Specification 3 measures
effects of overconfidence proxies (with lags) without accounting for non-interest income.
Specification 4 focuses on relations between net interest margins and loan loss provisions
only. Strong potential links between loan growth and net interest margin may overshadow a
NIM-LLP dependence, if both LGR and NIM are estimated simultaneously (as in
Specification 4). In a similar vein, Specification 5 excludes loan growth, but accounts for both
interest and non-interest part of income. Estimation results of (5.1). using the total bank
sample are shown in Table 4.1.
Table 4.1. Estimation results, (5.1), total bank sample.
Dependent variable:
LLPs to Assets at t-1
Specification 1 Specification 2 Specification 3 Specification 4 Specification 5
Coef. Sign. Std. Err. Coef. Sign. Std. Err. Coef. Sign. Std. Err. Coef. Sign. Std. Err. Coef. Sign. Std. Err.
Current loan growth 0.0002 (0.00056) 0.0008 (0.00056) 0.0005 (0.00056)
Loan growth in t-1 0.0010 * (0.00055) 0.0003 (0.00056) 0.0009 * (0.00055)
Loan growth in t-2 0.0020 *** (0.00054) 0.0019 *** (0.00054) 0.0019 *** (0.00054)
Loan growth in t-3 0.0028 *** (0.00053) 0.0030 *** (0.00054) 0.0026 *** (0.00053)
Loan growth in t-4 0.0012 ** (0.00053) 0.0016 *** (0.00054) 0.0011 ** (0.00053)
Current NIM^^ 0.0763 *** (0.01534) 0.0850 *** (0.01515) 0.0862 *** (0.01519) 0.0816 *** (0.01535)
Net Interest Margin in t-1 0.0182 (0.01715) 0.0238 (0.01713) 0.0299 * (0.01731) 0.0268 (0.01736)
Net Interest Margin in t-2 0.0180 (0.01622) 0.0175 (0.01630) 0.0119 (0.01648) 0.0118 (0.01645)
Net Interest Margin in t-3 -0.0358 ** (0.01797) -0.0365 ** (0.01805) -0.0476 *** (0.01807) -0.0487 *** (0.01805)
Net Interest Margin in t-4 0.0023 (0.01686) -0.0008 (0.01691) 0.0056 (0.01710) 0.0086 (0.01713)
Pre-provisions Income /
Assets in t-1^
0.0495 *** (0.01621) 0.0951 *** (0.01541) 0.0320 ** (0.01611)
Total Assets (1m)^^^ 0.0311 (0.06537) 0.0050 (0.06639) 0.0285 (0.06568) 0.0180 (0.06431) 0.0142 (0.06423)
Loans/Assets ratio 0.0017 (0.00108) 0.0016 (0.00107) 0.0020 * (0.00108) 0.0037 *** (0.00101) 0.0035 *** (0.00102)
Equity/Assets at t-1 -0.0005 (0.00447) 0.0014 (0.00456) 0.0008 (0.00447) 0.0000 (0.00450) -0.0011 (0.00453)
Inflation^^^^ 0.0102 (0.00792) 0.0112 (0.00792) 0.0068 (0.00788) 0.0112 (0.00793) 0.0138 * (0.00803)
Current GDP growth^^^^ -0.0007 * (0.00041) -0.0005 (0.00040) -0.0006 (0.00042) -0.0006 (0.00041) -0.0007 (0.00041)
GDP growth in t-1 -0.0278 *** (0.00579) -0.0313 *** (0.00579) -0.0277 *** (0.00582) -0.0260 *** (0.00582) -0.0262 *** (0.00581)
Constant -0.0759 (0.09195) 0.0626 (0.07492) -0.0499 (0.09200) -0.0710 (0.09311) -0.0861 (0.09326)
Within R2 0.2032 0.1388 0.1944 0.1579 0.1618
40
No. of observations 1171 1189 1171 1171 1171
No. of banks 309 311 309 309 309
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over current
total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total Assets/1.000.000; ^^^^all
macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Source: Own calculations, based on Western European bank results from Bankscope for years 1997-2008.
Table 4.1. clearly demonstrates that the primary overconfidence proxy, loan growth,
has a visible and statistically significant effect on the dependent variable, loan loss provisions.
This effect is lagged and the strongest relation between loan growth and provisions emerges
for loans granted three years earlier. After this period has passed, a hike in loan loss
provisions should be expected, proving the delayed negative impact of aggressive loan
granting. The relationship for lags two and four is also significant, but weaker. This finding
provides evidence that negative effects of loan growth materialise already within two years
from granting, become the most substantial a year later and persist in the fourth year, but start
to decrease. On the other hand, current loan growth in the setup of this model does not have a
statistically significant influence on loan loss provisions. This contradicts the hypothesis of ex
ante provisioning, where more aggressive current loan growth is accompanied by higher
reserves, covering the new expanded portfolio of potentially weaker quality. Higher
provisions paired with higher growth appear only after one year has passed from granting
loans, in Specifications 1 and 3, are relatively small and on the verge of statistical
significance.23
The second overconfidence proxy, net interest margin, also demonstrates the expected
results. Current NIM is strongly positively related with current LLPs, which simply reflects its
role in the profit and loss statement. Lagged margins are mostly insignificant in all
specifications, except for net interest margin delayed by three years, which emerges as
strongly negatively linked to current provisions, in all setups. This implies that a higher NIM
three years earlier is likely to result in lower current reserves. The three year lag has already
emerged above as a moment when the positive relation between loan expansion and reserves
is distinctly the strongest. Other authors also find that three years are a typical period after
which loan defaults surface (see Section 3).
A higher interest margin demanded by banks when granting loans thus denotes
requiring a higher risk premium balancing out the weaker client quality, meaning a less
aggressive growth policy, as discussed in Sections 3 and 4. It is a very important finding and
it is very robust, persisting both with and without controlling for loan growth and non-interest
23
A relationship between current loan growth and loan loss provisions becomes significant but negative, when
LLPs are scaled by current year‟s assets – see below.
41
income. It implies that aggressive risk taking understood as granting low interest loans, or
offering high interest deposits, has a strong negative effect on bank risk levels in a longer
term. In addition, it puts into question a theory stating that low-margin banks grant loans to
more secure clients. When these loans have time to default, lower margins translate into
higher provisions, and thus imply that the original low margin clients are in fact in the high
risk category. This proves that the risk taken by banks is a result of an aggressive growth
policy, while low margins are not determined by the high quality of clients but – most likely –
by competitive pressures.
Income smoothing through LLPs is confirmed by all three specifications including
pre-provisions income, for banks included in the datasample. The strongest income smoothing
emerges in Specification 2, where net interest margin is excluded. Income smoothing is
confirmed even after controlling for current net interest margin, although understandably on a
smaller scale. In such a case, given the same level of NIM, higher non-interest income is
accompanied by higher LLPs (Specifications 1 and 5). This strong evidence proving income
smoothing in Western European banks in the period of 1998-2008 is in line with findings of
Leaven and Majnoni (2003). However, the drives of earnings management through LLPs in
the overconfidence risk model do not stem from a prudential motivation. The already
discussed lack of a link between current loan growth and current LLPs proves that income
smoothing is not driven by ex ante provisioning motives. In that sense, it should be included
into the non-prudential vein of income smoothing debate.
The last important implication of Table 4.1. concerns anti-cyclicality of provisions in
relation to economic growth, which would be an indication of prudent credit risk policies.
Conversely, economic growth of a preceding year is significantly negatively related to current
loan loss provisions in all specifications.24
This has been established in earlier literature and
means that business cycle upswings are not used by banks to expand their reserve cushions.
On the contrary, loan loss provisioning is pro-cyclical, with fewer provisions created during
economic booms. Although diminished LLPs during growth periods may be in line with
contemporary portfolio quality, it has also been recognised that most potential irregular loans
are granted in the expansion phase of the cycle. Prudency would imply to account for this fact
ex ante, taking advantage of higher profitability on the back of strong markets. Banks in this
sample do not display such prudency, implying that come economic downturn, more reserves
have to be created. This result is an important factor that should be accounted for in the
24
The fact that current GDP growth is insignificant is most likely due to using assets from the preceding year to
scale current LLPs. When current assets are used for this purpose (see below), current GDP growth becomes
significant. An additional role may played by the lag effect of GDP growth upon bank performance.
42
discussion on optimal capital adequacy regulations. Existing capital requirements are bound
to be procyclical, as long as banks are not expected to act in a more anti-cyclical way on the
loan loss reserve side.
The procyclicality of reserves versus the economic cycle is another proof for the lack
of prudential considerations in income smoothing through LLPs discussed earlier. Not only do
banks refrain from creating higher reserves when they are aggressively expanding their
loanbooks, but they also do not create reserves when the expansion phase of the economic
cycle shelters their profits. Thus, income smoothing through LLPs serves other purposes than
building up cushions for tougher times. A negative link between provisions and economic
growth provides a rationale for the existance of managerial overconfidence in the studied
period, when the unfading belief in the persistence of economic growth incited a more
aggressive risk approach.
As far as control variables relating to the internal situation of the bank are concerned,
no unexpected results surface. Both capital and size are insignificant in all setups. A positive
relation emerges between a current share of loans in total assets and amount of provisions
created in Specifications 3, 4 and 5. However, this result is not very robust and remains in
relation to loan expansion pace, as it reaches its peak in both setups without loan growth,
disappears in Specifications 1 and 2 (including loan growth) and is on the verge of
significance in Specification 3. On the other hand, a higher loan-intensity of the balance sheet
does imply higher reserves, on average. A possibility of a partly spurious relation between the
dependent variable and Loans/Assets cannot be excluded, as both numerators and
denominators are cross-related. A more interesting case is examined in the robustness tests
section, where lagged values of loan-intensity included. This allows to study the effect of loan
growth and loan-intensity of assets on later risk, delayed by up to four periods.
In order to verify the robustness of the above results of estimating (5.1). with the total
bank dataset, a re-estimation is performed, using (5.2). (where LLPs and pre-provisions
income are scaled by current year‟s assets). Results are shown in Table 4.2.
Table 4.2 Estimation results, (5.2)., total bank sample.
Dependent variable:
⁄
Specification 1
Specification 2 Specification 3 Specification 4 Specification 5
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0016 *** (0.00050) -0.0021 *** (0.00051) -0.0017 *** (0.00050)
Loan growth in t-1 0.0009 * (0.00050) 0.0003 (0.00050) 0.0008 * (0.00050)
Loan growth in t-2 0.0018 *** (0.00048) 0.0017 *** (0.00049) 0.0017 *** (0.00048)
Loan growth in t-3 0.0024 *** (0.00048) 0.0026 *** (0.00049) 0.0022 *** (0.00048)
Loan growth in t-4 0.0011 ** (0.00048) 0.0015 *** (0.00049) 0.0010 ** (0.00048)
Current NIM^^ 0.0841 *** (0.01402) 0.0950 *** (0.01363) 0.1068 *** (0.01390) 0.0974 *** (0.01437)
Net Interest Margin in t-1 0.0042 (0.01536) 0.0062 (0.01542) 0.0119 (0.01584) 0.0106 (0.01580)
Net Interest Margin in t-2 0.0188 (0.01460) 0.0187 (0.01467) 0.0106 (0.01508) 0.0106 (0.01503)
Net Interest Margin in t-3 -0.0262 * (0.01617) -0.0273 * (0.01625) -0.0493 *** (0.01653) -0.0486 *** (0.01648)
43
Net Interest Margin in t-4 -0.0048 (0.01519) -0.0014 (0.01523) 0.0119 (0.01565) 0.0091 (0.01564)
Pre-provisions Income /
Assets in t-1^
0.0492 *** (0.01609) 0.1021 *** (0.01503) 0.0407 ** (0.01654)
Total Assets (1m)^^^ 0.0563 (0.05885) 0.0287 (0.06022) 0.0529 (0.05913) 0.0006 (0.05884) 0.0039 (0.05868)
Loans/Assets ratio 0.0028 *** (0.00098) 0.0027 *** (0.00097) 0.0032 *** (0.00097) 0.0042 *** (0.00093) 0.0040 *** (0.00093)
Equity/Assets at t-1 -0.0025 (0.00400) -0.0024 (0.00412) -0.0023 (0.00402) -0.0054 (0.00412) -0.0055 (0.00411)
Inflation^^^^ 0.0178 ** (0.00734) 0.0234 *** (0.00739) 0.0116 * (0.00709) 0.0181 *** (0.00726) 0.0233 *** (0.00754)
Current GDP growth^^^^ -0.0007 * (0.00038) -0.0007 * (0.00037) -0.0005 (0.00037) -0.0007 * (0.00038) -0.0009 ** (0.00038)
GDP growth in t-1 -0.0252 *** (0.00521) -0.0282 *** (0.00525) -0.0253 *** (0.00524) -0.0256 *** (0.00532) -0.0254 *** (0.00531)
Constant -0.1453 * (0.08258) -0.0375 (0.06858) -0.1292 * (0.08282) -0.1287 * (0.08519) -0.1434 * (0.08515)
Within R2 0.2712 0.2023 0.2632 0.2043 0.2099
No. of observations 1171 1189 1171 1171 1171
No. of banks 309 311 309 309 309
Notes: ^Pre-provisions Income and Total Assets in year t; ^^Current Net Interest Margin is calculated using net interest income over current total
assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total Assets/1.000.000; ^^^^all
macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Source: Own calculations, based on Western European bank results from Bankscope for years 1997-2008.
As Table 4.2. indicates, replacing previous year‟s assets with current assets does not
largely change earlier results. The conclusions on the first overconfidence proxy, loan growth,
are not only preserved, but strengthened further. Loan growth lagged by two, three and four
years maintains its strong positive relation with current loan loss provisions, providing
evidence for negative repercussions of earlier risk taking, implying the role of
overconfidence. In addition, current loan growth becomes strongly significant and negative.
This is striking evidence for the existence of overconfidence as such and lack of prudency in
studied banks. In this estimation, a more rapid portfolio expansion is accompanied by lower
reserves, implying that banks may believe they are gaining better quality clients and no need
for higher reserve cushions exist. Another possibility is that banks knowingly engage in
higher risks and postpone provisioning until later periods, when loan defaults materialise.
This could qualify as a moral hazard behaviour that does not necessarily hurt bank results, as
long as a net effect on profitability is positive (see verification of this hypothesis in the
overconfidence profitability model).
Table 4.2. shows no changes towards the rationale underlying the second
overconfidence proxy, net interest margin. Its current part is (understandably) strongly
positively related with current LLPs, but interest margins lagged by three years continue to
display a significant negative link with current provisions. Thus, the adverse effect of
aggressive, low margins causing higher provisions with a three-year delay persists in the
modified setup of (5.2).
The direction and significance of links between reserves and economic growth are
maintained. In addition, current economic growth becomes significant almost in all setups,
possibly due to its link with current assets. Inflation emerges as positive and significant, in
line with results of other models using current year‟s dependent variables. Lastly, loan-
intensity of assets also increases both its scope and significance level, while remaining
44
positive. As mentioned above, a strong spurious relation between LLP/Assets and
Loans/Assets may additionally boost this result.
Income smoothing is again confirmed and becomes even stronger, in Specifications 2
and 5. Taken together with the negative relation between LLPs and current loan growth, and
the persisting positive one with GDP growth, earnings management through LLPs is again not
driven by prudential considerations.
In consequence, for both (5.1). and (5.2)., the overconfidence hypothesis is confirmed.
Overconfidence proxies, represented by rapid loan expansion and low margins, are shown to
have negative repercussions on risk, increasing loan loss provisioning in later periods. This
effect is most visible in the three-year lag. Income smoothing through LLPs is established, but
the prudential role that it could play in credit risk management is doubtful. A lack of a
positive link between loan expansion and contemporary reserves (5.1) proves that banks do
not engage in ex ante provisioning. Banks in the analysed data sample believed their loan
expansion was unlikely to turn sour and have not made provisions for the new, more risky
clients. The relation between current loan growth and LLPs turning significant and negative in
the second version of the model (5.2) brings further evidence of the careless approach of
banks towards loan portfolio expansion. The fact that earnings management is performed in
purposes other than building up sound reserves for the future is proven further by strong pro-
cyclicality of provisions versus the economic cycle in both estimations.
The above results have been established for the total sample of banks and confirm the
primary hypothesis that overconfidence may have an adverse effect on bank risk in a longer
term. The next section relates to verifying the secondary hypothesis that effects of
overconfidence on risk may differ across various bank groups.
Verification of the overconfidence risk model using bank subsamples
The secondary hypothesis regarding overconfidence presumes that its effects may
vary, depending on the sort of bank and its approach to risk. In order to study differences in
overconfidence effects between banks, (5.1). is re-estimated on two subgroups created from
splitting the original dataset. Due to a strong influence of loan growth on risk (proven
empirically) and theoretical assumptions regarding overconfidence proxies in banks,
presented in Section 4, average loan growth is chosen as a reference point to identifying
subgroups. Thus, financial institutions are divided into potentially more and less
overconfident, basing on a rough proxy provided by loan growth. The first group comprises
banks that display loan growth that is at or below average, and is referred to as „low growth‟.
45
The second group includes institutions with above average loan growth, referred to as „high
growth‟. The primary aim of the sub-group estimation is to verify if any significant
differences emerge between the two bank sets in all primary variables discussed above.
Results of estimating (5.1). on both subgroups are shown in Table 4.3. (low growth banks)
and Table 4.4. (high growth banks). Taking into account a high degree of similarity between
the already discussed five specifications in the overconfidence risk model and in order to
enhance clarity of presentation, in tables below only the crucial three setups are included,
most clearly showing differences between subgroups. Remaining specifications (not shown)
do not include any additional or different results.
Table 4.3. Estimation results, (5.1), low growth bank subgroup
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0003 (0.00080) -0.0005 *** (0.00081) 0.0000 (0.00081)
Loan growth in t-1 0.0012 (0.00075) 0.0014 *** (0.00076) 0.0012 (0.00075)
Loan growth in t-2 0.0020 *** (0.00070) 0.0023 *** (0.00071) 0.0020 *** (0.00071)
Loan growth in t-3 0.0025 *** (0.00067) 0.0035 *** (0.00067) 0.0023 *** (0.00067)
Loan growth in t-4 0.0010 (0.00068) 0.0021 *** (0.00067) 0.0008 (0.00068)
Current NIM^^ 0.0884 *** (0.02136) 0.1006 *** (0.02108)
Net Interest Margin in t-1 0.0013 (0.02419) 0.0034 (0.02437)
Net Interest Margin in t-2 0.0198 (0.02132) 0.0183 (0.02148)
Net Interest Margin in t-3 0.0198 (0.02328) 0.0230 (0.02344)
Net Interest Margin in t-4 0.0153 (0.02225) 0.0111 (0.02238)
Pre-provisions Income /
Assets in t-1^
0.0655 *** (0.02318) 0.0983 *** (0.02247)
Total Assets (1m)^^^ 0.0543 (0.06409) -0.0008 *** (0.06505) 0.0447 (0.06450)
Loans/Assets ratio 0.0019 (0.00143) 0.0021 *** (0.00144) 0.0019 (0.00144)
Equity/Assets at t-1 -0.0075 (0.00651) -0.0180 *** (0.00630) -0.0054 (0.00652)
Inflation^^^^ 0.0245 ** (0.00975) 0.0163 *** (0.00950) 0.0202 ** (0.00970)
Current GDP growth^^^^ -0.0015 *** (0.00050) -0.0007 *** (0.00046) -0.0014 *** (0.00050)
GDP growth in t-1 -0.0278 *** (0.00654) -0.0337 *** (0.00661) -0.0269 *** (0.00659)
Constant -0.2467 ** (0.12292) 0.0699 *** (0.10172) -0.1975 (0.12264)
Within R2 0.2912 0.2217 0.2784
No. of observations 636 651 636
No. of banks 179 181 179
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over current
total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total Assets/1.000.000; ^^^^all
macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Re-estimating (5.1). on low growth banks yields results that are only partly the
reflection of total sample findings. The primary proxy for overconfidence behaves in line with
total sample results. Loan growth lagged by two and three years maintains its positive relation
with current LLPs. Loan growth lagged by four years loses on significance in Specifications 1
and 3, which may imply that the effects of overconfidence in this group are of a shorter term
nature. Current loan growth is not related to current loan loss provisions, apart from one
specification, where the effect is negative, but its strength – low.
The second proxy of overconfidence, net interest margin, demonstrates new evidence
however. No relation is found in the low growth bank subgroup between lagged interest
margins and current loan loss provisions. This potentially indicates that a lack of aggressive
growth in the past diminishes the negative effect of low margin lending. Low growth banks
46
may experience lower margins not because of overaggressive bidding for new clients, but
because of investing into lower-yield assets, such as government bonds, which do not boost
loan loss provisions in later periods.
Income smoothing is confirmed and emerges as stronger in this subgroup than in the
total sample. As in the previous estimation, loan loss provisions are also procyclical versus
economic growth and there is a positive link to inflation. As a result, low growth banks do not
display higher prudency in terms of ex ante provisioning and their earnings management by
LLPs is driven by motives other than conservative credit risk management.
A new relation emerges in Specification 2 in terms of capital levels. Previous year‟s
capital is negatively linked to the current level of LLPs, which implies that higher capital
banks in this subsample create less provisions. This contradicts the capital management
theory, which presumes that when LLPs are used to manage capital, the relation is positive.
No capital management is detected in the total sample and the trend is opposite in the low
growth bank subgroup. Intuitively, this result seems correct as banks possessing higher levels
of capital are potentially more prudent and less likely to experience problems with higher loan
loss provisions. Nonetheless, this result on equity is not very robust in other specifications.
Table 4.4. below shows estimation results for the high growth subgroup.
Table 4.4. Estimation results, (5.1)., high growth bank subgroup.
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0005 (0.00084) -0.0013 * (0.00079) 0.0008 (0.00083)
Loan growth in t-1 0.0002 (0.00087) -0.0010 (0.00085) 0.0003 (0.00087)
Loan growth in t-2 0.0016 * (0.00085) 0.0011 (0.00087) 0.0015 * (0.00085)
Loan growth in t-3 0.0025 *** (0.00088) 0.0024 *** (0.00089) 0.0025 *** (0.00088)
Loan growth in t-4 0.0007 (0.00088) 0.0008 (0.00091) 0.0007 (0.00089)
Current NIM^^ 0.0793 *** (0.02257) 0.0853 *** (0.02223)
Net Interest Margin in t-1 0.0345 (0.02468) 0.0405 * (0.02438)
Net Interest Margin in t-2 0.0133 (0.02461) 0.0134 (0.02465)
Net Interest Margin in t-3 -0.0862 *** (0.02807) -0.0884 *** (0.02808)
Net Interest Margin in t-4 -0.0077 (0.02649) -0.0103 (0.02648)
Pre-provisions Income /
Assets in t-1^
0.0340 (0.02310) 0.0819 *** (0.02182)
Total Assets (1m)^^^ -0.2378 (0.26641) -0.2577 (0.26402) -0.2429 (0.26679)
Loans/Assets ratio 0.0015 (0.00168) 0.0027 * (0.00163) 0.0018 (0.00166)
Equity/Assets at t-1 0.0098 (0.00658) 0.0136 ** (0.00669) 0.0103 (0.00658)
Inflation^^^^ -0.0028 (0.01373) -0.0051 (0.01365) -0.0055 (0.01363)
Current GDP growth^^^^ 0.0002 (0.00071) 0.0001 (0.00072) 0.0002 (0.00071)
GDP growth in t-1 -0.0226 ** (0.01102) -0.0178 (0.01097) -0.0234 ** (0.01102)
Constant 0.1017 (0.14693) 0.0188 (0.11294) 0.1167 (0.14679)
Within R2 0.1865 0.1138 0.1819
No. of observations 535 538 535
No. of banks 130 130 130
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Table 4.4. demonstrates that important differences between low and high growth bank
subgroups surface. Among high growth banks, the primary overconfidence proxy, loan
47
growth, shows overall weaker relations with loan loss provisions. The only statistically
significant, strong and permanent link emerges for loan expansion delayed by three years. A
negative relation between current loan expansion and provisions is seen in Specification 2 and
although much stronger than in the previous subgroup, it is on the verge of statistical
significance. Some links are also observed for loan growth in lag two, but equally weak
statistically. In consequence, it seems that dividing the sample with the use of loan growth
may cancel out the effect that this variable has on provisions, apart from a three year lag.
On the other hand, the second overconfidence proxy – net interest margin –
demonstrates a visible relation with loan loss provisions, proving earlier conclusions.
Contrary to the low growth group, rapidly expanding banks experience a strong negative
effect of low net interest margins lagged by three years on current loan loss provisioning
levels. This effect equals to a double of the one found in the total sample and it is highly
statistically significant. This implies a strong adverse impact of low interest margins upon
later reserve requirements of high growth banks, providing evidence that growth is achieved
at a price of quality. A higher portfolio expansion in this group translates to targeting more
risky clients that are likely to require higher provisions in the future.
As for income smoothing, it is lower in this subgroup than in both the total sample and
low growth banks. In fact, in Specification 3 income smoothing is detected only towards the
interest part of income, while non-interest income is insignificant to the level of LLPs.
Although income smoothing in Specification 2 is less pronounced in comparison to low
growth banks, it remains strongly significant.
Procyclicality of provisions versus the economic growth is less pronounced in this
group. Only GDP growth of the previous year shows any relation with current LLPs and it
disappears altogether in Specification 2. This finding may be caused by a more prudent policy
regarding anti-cyclicality of LLPs in higher growth banks, but this seems to be countered by a
negative relation between current loan growth and LLPs. However, this effect disappears in
robustness tests, when the data set is reduced to pre-2008 results, so its robustness and
stability are weak.
An unexpected positive and significant relation between previous year‟s capital and
LLPs surfaces in Specification 2. This is usually interpreted as an indication of capital
management through LLPs, more strongly so that lagged equity is used, excluding a spurious
relation between LLPs, net income and equity. Although the capital relation disappears in the
remaining two specifications, the contrast between a negative relation for low growth banks
48
and a positive one for high growth banks should be underlined. It points at different policies
regarding LLPs, with potential inclinations towards capital management among growth banks.
Results of estimating the overconfidence risk model on two subgroups bring out differences in
effects that overconfidence proxies have on risk in low- and high growth banks. In high
growth banks, although income smoothing through LLPs is observed, provision making is
less a function of current income, either interest or non-interest. In addition, lagged loan
growth is less of a drag on future reserve requirements, even if it is maintained for the three
year delay. On the other hand, net interest margin levels have diverse implications for future
provisions in high and low growth banks. Past NIM is insignificant for current reserves in low
growth banks, which may thus require adequate risk premiums for less secure clients. High
growth banks are hit by negative consequences of earlier aggressive pricing that puts pressure
on later reserve requirements. In addition, they are more prone to engage in capital
management, even if this result is not robust in all settings.
Notwithstanding the above, important similarities between these subgroups persist.
Income smoothing is found in both groups, even if in high growth banks it is visibly lower.
Negative relations between lagged loan growth and current loan loss provisions are also
confirmed for both groups. This indicates that within the studied period provisioning was
delayed until the moment of identifying irregular loans and new loans were aimed at higher
risk clients than existing ones. Both groups engage in procyclical provisioning, even if it is
much stronger for low growth banks (this result is also confirmed for high growth banks in
robustness tests). Thus, crucial differences between high growth (overconfident) and low
growth (rational) banks in this sample emerge mostly in past NIM effects on current
provisioning policies. Negative overconfidence effects of past loan growth on current risk are
present for both groups, as well as negative trends towards procyclical provisioning.
Robustness tests of the overconfidence risk model
In order to verify relations between overconfidence and risk discussed above, several
robustness tests are performed, using the overconfidence risk model. They identify the most
stable findings of the model and confirm its reliability under changing setups.
The first robustness test refers to amendments in the sample period. The final bank
sample contains data from period 1998-2008 and accounts for maximum four year lags.
However, input from 2008 is half-year data, reported for some banks only. In addition, some
banks partially included expected negative effects of the banking crisis (that started in 2007)
in their 2008 numbers, while others did not. Thus, the coherence of reporting standards
49
between countries may differ. In consequence, the first robustness test comprises re-
estimating Equations 5.1. and 5.2 on the total bank sample, excluding the 2008 results.
Results of these estimations are presented in Table 4.5. (5.1) and Table 4.6. (5.2).
Table 4.5. Estimation results, (5.1)., total bank sample, without 2008 data.
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3 Specification 4 Specification 5
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0004 (0.00057) -0.0004 (0.00055) 0.0006 (0.00057)
Loan growth in t-1 0.0012 ** (0.00056) 0.0006 (0.00056) 0.0012 ** (0.00056)
Loan growth in t-2 0.0022 *** (0.00054) 0.0023 *** (0.00054) 0.0020 *** (0.00054)
Loan growth in t-3 0.0023 *** (0.00054) 0.0024 *** (0.00055) 0.0022 *** (0.00054)
Loan growth in t-4 0.0010 * (0.00054) 0.0011 ** (0.00055) 0.0009 (0.00054)
Current NIM^^ 0.0629 *** (0.01526) 0.0737 *** (0.01500) 0.0730 *** (0.01501) 0.0658 *** (0.01523)
Net Interest Margin in t-1 0.0183 (0.01680) 0.0240 (0.01682) 0.0256 (0.01692) 0.0216 (0.01694)
Net Interest Margin in t-2 0.0131 (0.01593) 0.0135 (0.01603) 0.0057 (0.01612) 0.0046 (0.01607)
Net Interest Margin in t-3 -0.0448 ** (0.01779) -0.0456 ** (0.01791) -0.0526 *** (0.01780) -0.0539 *** (0.01774)
Net Interest Margin in t-4 -0.0177 (0.01823) -0.0221 (0.01830) -0.0230 (0.01836) -0.0185 (0.01839)
Pre-provisions Income /
Assets in t-1^
0.0544 *** (0.01637) 0.0938 *** (0.01533) 0.0413 ** (0.01630)
Total Assets (1m)^^^ 0.0865 (0.07588) 0.0743 (0.07628) 0.0818 (0.07635) 0.0682 (0.07412) 0.0643 (0.07388)
Loans/Assets ratio 0.0023 ** (0.00111) 0.0027 ** (0.00109) 0.0026 ** (0.00112) 0.0043 *** (0.00105) 0.0041 *** (0.00105)
Equity/Assets at t-1 0.0036 (0.00452) 0.0026 (0.00457) 0.0047 (0.00454) 0.0041 (0.00456) 0.0030 (0.00456)
Inflation^^^^ 0.0115 (0.00909) 0.0105 (0.00910) 0.0097 (0.00913) 0.0125 (0.00920) 0.0141 (0.00919)
Current GDP growth^^^^ -0.0411 *** (0.00677) -0.0499 *** (0.00631) -0.0383 *** (0.00676) -0.0428 *** (0.00672) -0.0454 *** (0.00677)
GDP growth in t-1 -0.0295 *** (0.00568) -0.0299 *** (0.00564) -0.0293 *** (0.00572) -0.0265 *** (0.00570) -0.0268 *** (0.00568)
Constant 0.0667 (0.10181) 0.0862 (0.07840) 0.0865 (0.10229) 0.1028 (0.10310) 0.0929 (0.10283)
Within R2 0.2478 0.2045 0.2372 0.2082 0.2145
No. of observations 1106 1121 1106 1106 1106
No. of banks 305 307 305 305 305
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Table 4.6. Estimation results, (5.2)., total bank sample, without 2008 data.
Dependent variable:
LLP/Assets in t
Specification 1 Specification 2 Specification 3 Specification 4 Specification 5
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0014 *** (0.00051) -0.0018 *** (0.00051) -0.0016 *** (0.00051)
Loan growth in t-1 0.0010 ** (0.00050) 0.0006 (0.00050) 0.0010 ** (0.00050)
Loan growth in t-2 0.0020 *** (0.00048) 0.0021 *** (0.00049) 0.0019 *** (0.00048)
Loan growth in t-3 0.0019 *** (0.00048) 0.0019 *** (0.00049) 0.0018 *** (0.00048)
Loan growth in t-4 0.0008 * (0.00048) 0.0010 ** (0.00049) 0.0007 (0.00048)
Current NIM^^ 0.0745 *** (0.01397) 0.0845 *** (0.01334) 0.0939 *** (0.01356) 0.0831 *** (0.01431)
Net Interest Margin in t-1 0.0044 (0.01493) 0.0064 (0.01495) 0.0079 (0.01528) 0.0060 (0.01526)
Net Interest Margin in t-2 0.0143 (0.01422) 0.0150 (0.01425) 0.0042 (0.01456) 0.0035 (0.01452)
Net Interest Margin in t-3 -0.0364 ** (0.01587) -0.0360 ** (0.01592) -0.0557 *** (0.01607) -0.0555 *** (0.01603)
Net Interest Margin in t-4 -0.0184 (0.01624) -0.0204 (0.01627) -0.0104 (0.01658) -0.0092 (0.01654)
Pre-provisions Income /
Assets in t-1^
0.0429 ** (0.01833) 0.1001 *** (0.01663) 0.0429 ** (0.01860)
Total Assets (1m)^^^ 0.1042 (0.06771) 0.0850 (0.06840) 0.1088 (0.06787) 0.0487 (0.06693) 0.0462 (0.06676)
Loans/Assets ratio 0.0033 *** (0.00102) 0.0032 *** (0.00100) 0.0038 *** (0.00099) 0.0046 *** (0.00094) 0.0042 *** (0.00096)
Equity/Assets at t-1 0.0006 (0.00403) -0.0002 (0.00410) 0.0012 (0.00404) -0.0014 (0.00412) -0.0019 (0.00411)
Inflation^^^^ 0.0139 * (0.00810) 0.0130 (0.00816) 0.0132 (0.00812) 0.0172 ** (0.00831) 0.0179 ** (0.00829)
Current GDP growth^^^^ -0.0379 *** (0.00604) -0.0472 *** (0.00566) -0.0362 *** (0.00601) -0.0424 *** (0.00607) -0.0442 *** (0.00610)
GDP growth in t-1 -0.0264 *** (0.00507) -0.0267 *** (0.00507) -0.0263 *** (0.00508) -0.0259 *** (0.00514) -0.0258 *** (0.00513)
Constant 0.0097 (0.09077) 0.0512 (0.06958) -0.0003 (0.09093) 0.0464 (0.09311) 0.0552 (0.09294)
Within R2 0.3177 0.2693 0.3129 0.2638 0.2688
No. of observations 1106 1121 1106 1106 1106
No. of banks 305 307 305 305 305
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
50
Table 4.5. and Table 4.6. demonstrate that the major findings regarding negative
effects of overconfidence on risk are stable and some have been reinforced in the re-
estimation excluding 2008 results. All previously found relations between the primary
overconfidence proxy, loan growth and loan loss provisions emerge. As before, earlier loan
portfolio expansion causes a hike in future loan loss provisions, with a delay of between two
and four years. Current loan growth becomes negative in both (5.1). and (5.2). results,
although it maintains statistical significance (as before) only in the current assets version.
The second overconfidence proxy, net interest margin, is confirmed to exert pressure
on reserve requirements with a three year delay. This effect is strengthened in the robustness
test, which is very important to model implications.
Income smoothing is also validated, at similar significance and strength levels. A new
relation emerges between current GDP growth and LLPs, which indicates that procyclicality
of loan loss provisions pre-2008 was even stronger. In addition, a positive link between loan-
intensity of assets and LLPs is now robust in all specifications, but remains of similar strength
and same sign as in the original results. In consequence, all previous findings regarding the
links between overconfidence proxies and risk are confirmed in the first robustness test.
Last but certainly not least, the pre-2008 estimation is statistically much better fitted,
with a higher within R2 throughout all specifications. Bank results from the first half of 2008
reflect an important turning point in both risk and profitability levels. However, they are
available for a part of the sample only and do not present a coherent model of accounting for
crisis-affected factors throughout banks and markets. In consequence, the model excluding the
2008 turbulences is bound to present a more stable picture of variable relations in banks. A
very interesting extension to this model would be to include 2008 year-end and 2009 results
for all banks, but the data to perform this was not yet available at the time of estimation.
The second robustness test relates to an important question of loan-intensity
implications on risk. Up to this point, current and lagged loan growth have been included in
the estimation, parallel to the current share of loans in total assets. There is a possibility that
past loan growth is just a representation of past loan-intensity of assets and the effects that are
found for loan growth should in fact be attributed to past levels of loans that were omitted. To
verify this hypothesis, lagged ratios of loans/assets are added to (5.1). and (5.2), and they are
re-estimated, applying the five main specifications from the original model. Each of these
setups now includes control variables for past levels of loan intensity of assets. Results are
shown in Tables 4.7. and 4.8. respectively.
51
Table 4.7. Estimation results, (5.1). with lagged loans/assets, total bank sample.
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3 Specification 4 Specification 5
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0006 (0.00059) -0.0007 (0.00057) 0.0009 (0.00057)
Loan growth in t-1 0.0009 (0.00057) 0.0002 (0.00057) 0.0009 (0.00057)
Loan growth in t-2 0.0019 *** (0.00056) 0.0018 *** (0.00057) 0.0018 *** (0.00056)
Loan growth in t-3 0.0027 *** (0.00054) 0.0029 *** (0.00055) 0.0025 *** (0.00054)
Loan growth in t-4 0.0012 ** (0.00053) 0.0016 *** (0.00054) 0.0011 ** (0.00053)
Current NIM^^ 0.0856 *** (0.01576) 0.0936 *** (0.01533) 0.0936 *** (0.01523) 0.0905 *** (0.01548)
Net Interest Margin in t-1 0.0178 (0.01757) 0.0209 (0.01754) 0.0254 (0.01758) 0.0239 (0.01764)
Net Interest Margin in t-2 0.0171 (0.01661) 0.0165 (0.01664) 0.0087 (0.01668) 0.0087 (0.01668)
Net Interest Margin in t-3 -0.0373 ** (0.01818) -0.0379 ** (0.01822) -0.0447 ** (0.01820) -0.0455 ** (0.01821)
Net Interest Margin in t-4 0.0088 (0.01710) 0.0082 (0.01713) 0.0139 (0.01721) 0.0150 (0.01724)
Pre-provisions Income /
Assets in t-1^
0.0357 ** (0.01699) 0.0897 *** (0.01601) 0.0179 (0.01646)
Loans/Assets ratio -0.0002 (0.00132) 0.0015 (0.00131) -0.0004 (0.00132) 0.0012 (0.00125) 0.0013 (0.00125)
Loans/Assets in t-1 0.0025 (0.00155) 0.0011 (0.00156) 0.0031 ** (0.00152) 0.0031 ** (0.00146) 0.0029 ** (0.00147)
Loans/Assets in t-2 0.0010 (0.00147) -0.0001 (0.00147) 0.0013 (0.00147) 0.0023 (0.00141) 0.0022 (0.00141)
Loans/Assets in t-3 0.0008 (0.00140) -0.0001 (0.00139) 0.0009 (0.00140) 0.0008 (0.00136) 0.0007 (0.00137)
Loans/Assets in t-4 -0.0006 (0.00111) -0.0016 (0.00110) -0.0008 (0.00110) -0.0017 (0.00110) -0.0016 (0.00110)
Total Assets (1m)^^^ 0.0505 (0.06566) -0.0022 (0.06662) 0.0531 (0.06578) 0.0549 (0.06458) 0.0505 (0.06470)
Equity/Assets at t-1 -0.0022 (0.00451) -0.0020 (0.00462) -0.0017 (0.00452) -0.0023 (0.00451) -0.0028 (0.00453)
Inflation^^^^ 0.0107 (0.00793) 0.0113 (0.00794) 0.0085 (0.00787) 0.0120 (0.00789) 0.0134 * (0.00799)
Current GDP growth^^^^ -0.0007 * (0.00041) -0.0004 (0.00040) -0.0007 * (0.00041) -0.0006 (0.00041) -0.0007 (0.00041)
GDP growth in t-1 -0.0276 *** (0.00578) -0.0309 *** (0.00580) -0.0275 *** (0.00579) -0.0257 *** (0.00578) -0.0259 *** (0.00578)
Constant -0.2026 * (0.11474) 0.1167 (0.08925) -0.2091 * (0.11493) -0.2192 * (0.11665) -0.2166 * (0.11666)
Within R2 0.2101 0.1421 0.2060 0.1749 0.1760
No. of observations 1171 1189 1171 1171 1171
No. of banks 309 311 309 309 309
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Table 4.8. Estimation results, (5.2). with lagged loans/assets, total bank sample.
Dependent variable:
LLP/Assets in t
Specification 1 Specification 2 Specification 3 Specification 4 Specification 5
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0017 *** (0.00052) -0.0023 *** (0.00052) -0.0018 *** (0.00052)
Loan growth in t-1 0.0010 ** (0.00051) 0.0004 (0.00052) 0.0010 ** (0.00052)
Loan growth in t-2 0.0018 *** (0.00051) 0.0017 *** (0.00051) 0.0017 *** (0.00051)
Loan growth in t-3 0.0023 *** (0.00049) 0.0024 *** (0.00050) 0.0021 *** (0.00049)
Loan growth in t-4 0.0011 ** (0.00048) 0.0014 *** (0.00049) 0.0010 ** (0.00048)
Current NIM^^ 0.0840 *** (0.01425) 0.0942 *** (0.01387) 0.1095 *** (0.01401) 0.1010 *** (0.01448)
Net Interest Margin in t-1 0.0092 (0.01582) 0.0114 (0.01587) 0.0112 (0.01618) 0.0099 (0.01615)
Net Interest Margin in t-2 0.0154 (0.01499) 0.0142 (0.01505) 0.0045 (0.01535) 0.0052 (0.01532)
Net Interest Margin in t-3 -0.0278 * (0.01642) -0.0290 * (0.01648) -0.0481 *** (0.01675) -0.0473 *** (0.01672)
Net Interest Margin in t-4 -0.0020 (0.01548) 0.0017 (0.01550) 0.0170 (0.01584) 0.0141 (0.01585)
Pre-provisions Income /
Assets in t-1^
0.0466 *** (0.01626) 0.0951 *** (0.01538) 0.0370 ** (0.01662)
Loans/Assets ratio 0.0032 *** (0.00120) 0.0040 *** (0.00121) 0.0037 *** (0.00119) 0.0036 *** (0.00115) 0.0034 *** (0.00115)
Loans/Assets in t-1 -0.0016 (0.00137) -0.0017 (0.00138) -0.0018 (0.00138) 0.0000 (0.00134) 0.0001 (0.00134)
Loans/Assets in t-2 0.0012 (0.00133) 0.0003 (0.00133) 0.0014 (0.00133) 0.0023 * (0.00130) 0.0021 * (0.00129)
Loans/Assets in t-3 0.0009 (0.00127) 0.0002 (0.00125) 0.0011 (0.00127) 0.0007 (0.00126) 0.0005 (0.00125)
Loans/Assets in t-4 -0.0006 (0.00100) -0.0014 (0.00100) -0.0009 (0.00100) -0.0018 * (0.00101) -0.0016 (0.00102)
Total Assets (1m)^^^ 0.0590 (0.05927) 0.0226 (0.06043) 0.0561 (0.05951) 0.0189 (0.05943) 0.0213 (0.05930)
Equity/Assets at t-1 -0.0022 (0.00407) -0.0015 (0.00418) -0.0020 (0.00409) -0.0063 (0.00415) -0.0064 (0.00414)
Inflation^^^^ 0.0168 ** (0.00740) 0.0218 *** (0.00744) 0.0108 (0.00712) 0.0178 ** (0.00726) 0.0227 *** (0.00756)
Current GDP growth^^^^ -0.0008 ** (0.00038) -0.0007 * (0.00037) -0.0006 (0.00037) -0.0007 * (0.00038) -0.0009 ** (0.00038)
GDP growth in t-1 -0.0252 *** (0.00522) -0.0278 *** (0.00525) -0.0253 *** (0.00524) -0.0256 *** (0.00532) -0.0255 *** (0.00531)
Constant -0.1705 (0.10383) 0.0537 (0.08376) -0.1479 (0.10397) -0.1681 (0.10734) -0.1860 * (0.10740)
Within R2 0.2736 0.2070 0.2665 0.2113 0.2159
No. of observations 1171 1189 1171 1171 1171
No. of banks 309 311 309 309 309
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
The second robustness test, comprising additional controls for past loan-intensity
levels, confirms all key findings regarding relations between overconfidence proxies and risk,
found in the original estimation. Lagged loan growth maintains its pressure on increasing
current reserves and in the current assets setup, higher loan expansion is linked with decreased
52
provisioning, proving that no ex ante reserve creation takes place. Low interest margins,
usually accompanying rapid growth, again demonstrate their negative influence on
provisioning three years later.
A new link emerges for loan-intensity lagged by one year (for (5.1).), but it seems to
take the place of previously found relation with current loans/assets that here becomes
insignificant. The relation is most probably due to scaling with previous year‟s assets that are
strongly positively linked to previous year‟s loans/assets ratio. All remaining past loan-
intensity ratios are mostly insignificant and they do not affect other relations established
earlier, which remain robust on similar strength and significance levels.25
Income smoothing
is confirmed for both equations, but due to no ex ante provisioning it is again not necessarily
linked with a more prudent reserve building. The pro-cyclical provisioning versus economic
growth is affirmed, exacerbating the non-prudential aspect of income smoothing with LLPs.
Within the studied sample, growth in GDP (current and/or lagged by one year) entails lower
provisions and demonstrates that banks in general do not create reserves on the back of
overall economic prosperity in the operating environment.
The third and last robustness test regards the subgroup estimation. Similarly as in the
first robustness test, half-year 2008 results are excluded from both groups and the model is re-
estimated. Results are demonstrated in Table 4.9. (low growth banks) and Table 4.10 (high
growth banks).
Table 4.9. Estimation results, (5.1)., low growth bank sample, without 2008 data
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0000 (0.00082) -0.0002 (0.00082) -0.0004 (0.00081)
Loan growth in t-1 0.0012 (0.00076) 0.0012 (0.00075) 0.0012 (0.00075)
Loan growth in t-2 0.0022 *** (0.00071) 0.0022 *** (0.00071) 0.0027 *** (0.00070)
Loan growth in t-3 0.0022 *** (0.00070) 0.0024 *** (0.00070) 0.0029 *** (0.00068)
Loan growth in t-4 0.0008 (0.00070) 0.0009 (0.00070) 0.0014 ** (0.00068)
Current NIM^^ 0.0859 *** (0.02100) 0.0724 *** (0.02147)
Net Interest Margin in t-1 0.0004 (0.02418) -0.0008 (0.02401)
Net Interest Margin in t-2 0.0138 (0.02102) 0.0148 (0.02087)
Net Interest Margin in t-3 0.0105 (0.02363) 0.0062 (0.02352)
Net Interest Margin in t-4 -0.0153 (0.02467) -0.0110 (0.02455)
Pre-provisions Income /
Assets in t-1^
0.0629 *** (0.02393) 0.0980 *** (0.02262)
Total Assets (1m)^^^ 0.1060 (0.07744) 0.1166 (0.07698) 0.0708 (0.07645)
Loans/Assets ratio 0.0028 * (0.00151) 0.0028 * (0.00150) 0.0028 * (0.00147)
Equity/Assets at t-1 -0.0040 (0.00660) -0.0058 (0.00659) -0.0125 ** (0.00631)
Inflation^^^^ 0.0290 ** (0.01195) 0.0301 ** (0.01187) 0.0255 ** (0.01151)
Current GDP growth^^^^ -0.0309 *** (0.00762) -0.0339 *** (0.00765) -0.0409 *** (0.00729)
GDP growth in t-1 -0.0262 *** (0.00654) -0.0272 *** (0.00650) -0.0299 *** (0.00644)
Constant -0.0902 (0.13760) -0.1176 (0.13700) 0.0607 (0.10762)
Within R2 0.3227 0.3341 0.2995
No. of observations 596 596 608
No. of banks 176 176 178
25
Some additional relations are found in Specifications 4 and 5, where lagged loan-intensities become
significant, positive in the second lag and negative in the fourth. However, both specifications exclude loan
growth and these instabilities disappear once lagged loan growth is accounted for, as in Specifications 1 and 2.
As a result, some links between LGR and loan-intensity exist and slightly influence relations between provisions
and loanbook growth. These effects are however instable and their economic impact is low.
53
Notes: ^Pre-provisions Income in year t, total assets in year t-1; ^^Current Net Interest Margin is calculated using net interest income over
current total assets; ^^^Total Assets are scaled downwards by 1mln, to achieve coherent coefficient sizes, so the final input is Total
Assets/1.000.000; ^^^^all macroeconomic indicators are OECD data; *, ** and *** note significance levels of respectively 10%, 5% and 1%.
Within the low growth group, Table 4.9. shows that no major differences emerge
between the regular database and the pre-2008 results. The only visible and very significant
change is in pro-cyclicality of loan loss provisions that now display high connectivity with
current GDP growth, in addition to GDP lagged by one year. Other main findings are
confirmed, with negative effects of high previous loan growth on provisions, no link between
lagged NIM and provisions and income smoothing. Similarly to total sample robustness tests,
pre-2008 results show stronger within R2
for all specifications in the low growth subgroup
estimations.
Table 4.10. Estimation results, (5.1)., high growth bank sample, without 2008 data
Dependent variable:
LLP/Assets in t-1
Specification 1 Specification 2 Specification 3
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0010 (0.00085) 0.0008 (0.00086) -0.0007 (0.00082)
Loan growth in t-1 0.0007 (0.00091) 0.0007 (0.00091) -0.0002 (0.00089)
Loan growth in t-2 0.0017 * (0.00088) 0.0019 ** (0.00088) 0.0018 ** (0.00089)
Loan growth in t-3 0.0019 ** (0.00090) 0.0019 ** (0.00090) 0.0019 ** (0.00090)
Loan growth in t-4 0.0005 (0.00091) 0.0005 (0.00091) 0.0007 (0.00092)
Current NIM^^ 0.0756 *** (0.02217) 0.0677 *** (0.02258)
Net Interest Margin in t-1 0.0411 * (0.02419) 0.0341 (0.02446)
Net Interest Margin in t-2 0.0096 (0.02472) 0.0084 (0.02467)
Net Interest Margin in t-3 -0.0909 *** (0.02775) -0.0879 *** (0.02773)
Net Interest Margin in t-4 -0.0250 (0.02870) -0.0199 (0.02877)
Pre-provisions Income /
Assets in t-1^
0.0403 * (0.02338) 0.0806 *** (0.02192)
Total Assets (1m)^^^ -0.1142 (0.26584) -0.1008 (0.26524) -0.0298 (0.26375)
Loans/Assets ratio 0.0024 (0.00172) 0.0020 (0.00173) 0.0038 ** (0.00169)
Equity/Assets at t-1 0.0127 * (0.00669) 0.0124 * (0.00668) 0.0145 ** (0.00677)
Inflation^^^^ -0.0014 (0.01449) 0.0005 (0.01449) -0.0065 (0.01446)
Current GDP growth^^^^ -0.0367 *** (0.01343) -0.0392 *** (0.01347) -0.0516 *** (0.01236)
GDP growth in t-1 -0.0293 *** (0.01093) -0.0290 *** (0.01090) -0.0243 ** (0.01079)
Constant 0.2477 (0.16278) 0.2321 (0.16260) 0.0745 (0.11864)
Within R2 0.2048 0.2113 0.1505
No. of observations 510 510 513
No. of banks 129 129 129
Similarly as above, major findings for the high growth subgroup are validated.
Positive links of past loan growth with provisions persist and slightly stronger negative effects
of low interest margins are also confirmed. Income smoothing in this setting is more
accentuated than previously and becomes significant in Specification 1. Positive capital
relation to LLPs is again evident in Specification 2, pointing at capital management potential.
Importantly, procyclical provisioning emerges here in all settings and this both for current and
lagged GDP growth. This demonstrates that a seemingly more anticyclical provisioning
policy evident in the previous setup including 2008 results may have been dependent on few
2008 figures and thus not very robust. On this basis, it is doubtful that high growth banks
display higher prudency in provisioning against the economic cycle.
54
Below, the second part of the overconfidence theory relating to profitability is studied. As
discussed in Section 4, bank managers may knowingly engage in higher risks in order to reap
private benefits from increased risk premiums that outweigh potential loan loss reserve costs
and expected private costs. In such a case of pre-calculated risk/benefit analysis, a net effect
of these aggressive policies on profitability should be positive.
Overconfidence profitability model
The second part of the empirical estimation concerns relations between
overconfidence proxies and profitability. They are studied through fixed effects panel data
estimation of the proposed overconfidence profitability model. First, a general model of
effects of overconfidence on profitability is created. Using the whole bank sample, the model
is estimated and discussed. Subsequently, using the subgroups of high and low loan growth
banks, the model is re-estimated. Robustness tests check stability of end results in other
potential specifications and setups.
The overconfidence profitability model bases on numerous empirical models available
in the literature, relating to studies of revenues in international banks. The main target of the
overconfidence profitability model is to analyse the role of overconfidence proxies in shaping
final annual net income of banks. This is the reason why certain –otherwise important –
variables are omitted. They include the whole cost section, which has an obvious important
influence on bottomline figures. Nonetheless, performing a detailed cost analysis is outside
the scope of this work. Including net cost figures without accounting for e.g. type of bank,
operating environment, staffing issues, remuneration costs would not enhance the analysis. Its
aim is however not to study cost efficiency, which in itself is a broad area of empirical
research. Excluding the cost side makes interpretation of results more straightforward and the
analysis more concise, at the price of lower statistical quality. The expected within R squared
is not high, due to a restricted number of explanatory variables. Instead, the model focuses on
determining if overconfidence proxies have an effect on bottomline results and on measuring
its strength. Time lags are again introduced, in order to evaluate delayed consequences of
overconfidence on profitability. A general form of the overconfidence model for profitability
is presented by (5.3).
∑ ∑
(5.3)
Where:
is the return on average assets, calculated as Net Income/Average Assets
55
is loan growth (primary overconfidence proxy) lagged by 0-4 years
is net interest margin (secondary overconfidence proxy) lagged by 0-4 years
is a set of macroeconomic variables, comprising respectively inflation and GDP growth
(current and lagged by 1 year), controlling for operating environment
is a control variable for the level of capital
is a control variable for the share of loans in total assets
is a control variable for the size of the bank (total assets in current year)
Profitability in this setup is represented by return on assets (ROA), which is the
primary dependent variable. This is in line with other empirical models in the literature that
most frequently employ this particular ratio to show profitability. In later robustness tests,
ROA is replaced by return on equity (ROE), the second most common profitability measure.
However, ROE introduces diversification as to level of capital, which cannot be controlled for
and includes current net income, which potentially biases the ratio. ROE may display a
skewed picture, favouring low-capital and low-income banks and thus it is used only in the
robustness tests.
The main explanatory variables in (5.3) are two overconfidence proxies, loan growth
and net interest margins. As current NIM is related to current ROA, usually bring the most
significant contributor to net income, it is maintained to demonstrate reliance of banks on the
interest part of their business and for consistency reasons. The primary focus of the model lies
in lagged loan growth and lagged interest margin effects.
Similarly to the overconfidence risk model, the profitability setup controls for internal
bank characteristics. This includes size, capital and loan-intensity, to ascertain that the results
are net of any activity-specific effects. Macroeconomic variables are introduced in order to
display relations between profitability and local economic conditions and demonstrate
overconfidence net of economic environment influences. Previous year‟s GDP growth is
again included in the equation, following many implications from the literature that GDP
possibly has a lagged effect on firms, and especially on financial institutions.
Estimation results of the overconfidence profitability model using the total data set
The overconfidence profitability model is verified using the same bank dataset as in the
overconfidence risk model. Following the procedure used above, several variations of the
overconfidence profitability model (represented by (5.3)) are estimated. Specification 1 is the
most accurate reflection of (5.3), including all key variables simultaneously. Specifications 2
56
and 3 specifically target each of the overconfidence proxies separately, first loan growth
(Specification 2) and then net interest margins (Specification 3). This procedure allows to
bring forward relations between these variables and profitability. Specification 4 additionally
accounts for present and past loan loss provisions, in order to check for “net” effects of
present and past loan growth on profitability, i.e. after controlling for reserves made through
the profit and loss account in each given year. Results of estimating (5.3) on the whole bank
sample are presented in Table 4.11.
Table 4.11. Estimation results, (5.3), total bank sample.
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0016 (0.00101) -0.0005 (0.00102) -0.0014 (0.00101)
Loan growth in t-1 -0.0016 (0.00101) -0.0009 (0.00101) -0.0002 (0.00100)
Loan growth in t-2 -0.0015 (0.00099) -0.0013 (0.00098) -0.0003 (0.00097)
Loan growth in t-3 -0.0034 *** (0.00099) -0.0031 *** (0.00097) -0.0018 * (0.00097)
Loan growth in t-4 -0.0031 *** (0.00099) -0.0030 *** (0.00097) -0.0027 *** (0.00096)
Current NIM^^ 0.1214 *** (0.02744) 0.1230 *** (0.02764) 0.1729 *** (0.02793)
Net Interest Margin in t-1 -0.0346 (0.03126) -0.0246 (0.03127) -0.0205 (0.03089)
Net Interest Margin in t-2 -0.0042 (0.02976) -0.0039 (0.02975) 0.0012 (0.02967)
Net Interest Margin in t-3 0.0143 (0.03263) -0.0028 (0.03295) -0.0247 (0.03261)
Net Interest Margin in t-4 -0.0251 (0.03088) -0.0154 (0.03087) -0.0080 (0.03047)
LLPs/assets -0.5141 *** (0.06947)
LLPs/assets in t-1 0.0099 (0.06093)
LLPs/assets in t-2 0.0265 (0.05453)
LLPs/assets in t-3 0.1172 ** (0.05139)
LLPs/assets in t-4 -0.0316 (0.05335)
Total Assets (1m)^^^ 0.0477 (0.12085) 0.1690 (0.11614) 0.1144 (0.11988) 0.1428 (0.11641)
Loans/Assets ratio 0.0075 *** (0.00194) 0.0046 ** (0.00183) 0.0061 *** (0.00197) 0.0080 *** (0.00195)
Equity/Assets at t-1 0.0237 *** (0.00828) 0.0258 *** (0.00813) 0.0260 *** (0.00815) 0.0236 *** (0.00794)
Inflation^^^^ -0.0394 *** (0.01428) -0.0492 *** (0.01432) -0.0427 *** (0.01438) -0.0371 *** (0.01418)
Current GDP growth^^^^ 0.0012 (0.00073) 0.0017 ** (0.00074) 0.0015 ** (0.00076) 0.0012 (0.00074)
GDP growth in t-1 0.0363 *** (0.01054) 0.0407 *** (0.01051) 0.0388 *** (0.01062) 0.0275 ** (0.01060)
Constant 0.2113 (0.13359) 0.1092 (0.16815) 0.1091 (0.16791) -0.0001 (0.16599)
Within R2 0.0755 0.0896 0.1105 0.1757
No. of observations 1189 1171 1171 1157
No. of banks 311 309 309 305
Results of estimating (5.3) demonstrate mixed results of overconfidence effects on
profitability. The primary overconfidence proxy, loan growth, is strongly related to bank
results, but only in the third and fourth year lag. This finding provides clear evidence that
earlier loan growth decreases later profitability, at least in the bank set and time period
analysed. Current and recent loan expansion in itself is found to be insignificant for current
ROA, but a significant and positive link between ROA and loan-intensity of assets proves that
a higher share of assets does contribute to higher profitability.26
A negative relation between
lagged loan growth and current ROA remains stable throughout all settings, even if in
Specification 4 the third lag in loan growth weakens and loses on significance somewhat.
On the other hand, net interest margin affects ROA only in the current year. The latter
is an obvious and expected result, as net income is mostly shaped by contributions from
26
A more detailed analysis of loan growth in conjunction with loan intensity is provided in the robustness tests
part.
57
interest-driven activities. Importantly, lagged NIM proves insignificant for later net income,
demonstrating that the level of margins itself may affect reserve levels, but not bottomline
profits.
Specification 4 additionally takes into account loan loss provisions created currently
and in the past. Taken in conjunction with loan growth, this allows to evaluate the “net” effect
that loan growth has on profitability, keeping the levels of present and past provisions stable.
In such a setting, past loan growth is proven to maintain its negative effect on profitability,
although it is diminished for the third year lag. Instead, the level of LLPs lagged by three
years emerges as positively linked to current ROA. This provides crucial empirical
underpinnings to positive consequences of ex ante provisioning. Higher provisions made in
the past, contemporarily with loan growth, contribute to higher profits once bad loans have
had time to materialise.
Throughout all four specifications, control variables are shown to have stable and
consistent relations with profitability. As already mentioned, higher loan intensity of assets
accompanies higher ROA. Banks with a stronger capital base are shown to achieve better
profits, with a meaningful and stable effect throughout all setups. Economic control variables
are of expected signs, demonstrating a visible positive link between recent GDP growth and
ROA in all cases and current GDP and ROA in two specifications. This provides evidence for
the frequently commented lag effect of GDP growth on bank results, which has to be
transmitted through corporate and retail clients and the process may easily reach a duration of
a year.
In general, results from Table 4.11. confirm the overconfidence hypothesis in terms of
loan growth. Higher loanbook expansion causes in a longer term (three and four years) not
only an increase of risk, as demonstrated in the overconfidence risk model, but also a fall in
profitability. Thus, it is likely that the behaviour of banks is strongly influenced by
overconfidence, which causes a miscalibration of risks and a false belief in abilities of reaping
benefits from risks taken. If financial institutions consciously engaged in higher growth, their
profitability should have been positively affected. Banks underestimate adverse consequences
of rapid loan growth, which hit them in the form of higher than expected loan loss provisions.
By a given level of loan portfolio expansion, an increase in contemporary provisions has a
positive effect on later profits, so prudent banks are rewarded for their ex ante provisioning
policy.
On the other hand, overconfidence in terms of aggressive net interest margins has a
negative impact on risk in a perspective of three years, but is neutral for longer term
58
profitability. One of the explanations may be that higher volumes generated due to lower NIM
compensate for higher risk of clients. Another is that NIM is just one of bank profitability
drivers, so banks with a lower NIM may attract lower quality clients and bear a higher credit
risk, but may have strong profitability in other product areas at the same time, e.g. achieve
good returns on treasury business and have a lean cost structure. In consequence,
overconfidence expressed in overly aggressive margins does not negatively influence bank
results, as its benefits balance out the costs of higher risk, in a longer term. The section below
analyses differences in overconfidence effects on profitability between two bank subgroups.
Verification of the overconfidence profitability model using bank subgroups
In this part, (5.3) is re-estimated on two subgroups, of low and high loan growth
banks. Four specifications applied to the total sample are repeated in the subgroup verification
and its results are shown in Table 4.12. (low growth banks) and Table 4.13. (high growth
banks).
Table 4.12. Estimation results, (5.3), low growth bank sample
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0013 (0.00149) -0.0009 (0.00146) -0.0017 (0.00145)
Loan growth in t-1 -0.0026 * (0.00140) -0.0027 ** (0.00137) -0.0017 (0.00138)
Loan growth in t-2 0.0000 (0.00130) -0.0002 (0.00129) 0.0012 (0.00129)
Loan growth in t-3 -0.0026 ** (0.00122) -0.0023 * (0.00122) -0.0009 (0.00122)
Loan growth in t-4 -0.0031 ** (0.00123) -0.0030 ** (0.00125) -0.0026 ** (0.00123)
Current NIM^^ 0.0497 (0.03807) 0.0588 (0.03834) 0.1055 *** (0.03811)
Net Interest Margin in t-1 -0.0490 (0.04410) -0.0377 (0.04431) -0.0456 (0.04276)
Net Interest Margin in t-2 0.0309 (0.03900) 0.0414 (0.03907) 0.0496 (0.03787)
Net Interest Margin in t-3 0.0080 (0.04260) 0.0071 (0.04263) 0.0097 (0.04142)
Net Interest Margin in t-4 -0.0712 * (0.04027) -0.0559 (0.04070) -0.0561 (0.04026)
LLPs/assets -0.5576 *** (0.09116)
LLPs/assets in t-1 0.0288 (0.07735)
LLPs/assets in t-2 0.0067 (0.06629)
LLPs/assets in t-3 0.1431 ** (0.06493)
LLPs/assets in t-4 -0.0027 (0.06896)
Total Assets (1m)^^^ -0.0419 (0.11904) 0.0141 (0.11172) 0.0059 (0.11730) 0.0255 (0.11259)
Loans/Assets ratio 0.0009 (0.00265) -0.0018 (0.00251) -0.0008 (0.00263) 0.0017 (0.00258)
Equity/Assets at t-1 0.0345 *** (0.01157) 0.0444 *** (0.01173) 0.0461 *** (0.01186) 0.0380 *** (0.01147)
Inflation^^^^ -0.0314 * (0.01718) -0.0443 ** (0.01715) -0.0312 * (0.01764) -0.0185 (0.01743)
Current GDP growth^^^^ 0.0015 * (0.00085) 0.0026 *** (0.00089) 0.0025 *** (0.00092) 0.0019 ** (0.00090)
GDP growth in t-1 0.0390 *** (0.01215) 0.0391 *** (0.01166) 0.0377 *** (0.01198) 0.0243 ** (0.01190)
Constant 0.4521 ** (0.18165) 0.5680 ** (0.22139) 0.4688 ** (0.22304) 0.3184 (0.21749)
Within R2 0.1000 0.1132 0.1333 0.2207
No. of observations 651 636 636 625
No. of banks 181 179 179 175
Table 4.12. demonstrates that results of low-growth banks remain largely in line with
total sample figures. Loan growth lagged by three and four years remains a burden for current
profitability, although their significance and strength are lower than for the whole sample.
Conversely, in Specifications 1 and 2, loan growth of the previous year emerges as a drain on
current profitability, which is a bad sign, even if it disappears in Specification 4. This implies
that low growth banks may not be able to properly manage their loanbooks and boosting them
59
increases pressure on profits, instead of alleviating it. On the other hand, a positive effect of
lagged loan loss provisions is maintained, as indicated in Specification 4.
Specifications 1 and 3 prove weak dependency of profitability on interest income at
low growth banks – current NIM emerges as insignificant. This indicates that these banks are
relying on other sources of income and their weak management of credit portfolios may thus
be explained. Specification 3 exposes a new, negative relation between NIM lagged by four
years and profits. This implies that in this group, a higher NIM in the past negatively
influences future profitability, again pointing at potentially lower quality credit management.
Nonetheless, this variable is only weakly significant and not robust in other specifications.
As for control variables, similarly as in the total sample, low growth banks‟ profits rise
parallel to their capital levels and economic growth. Here, a much stronger link is observed
between lagged capital and current ROA, pointing at better capitalised institutions as highest
profit-achievers.
Table 4.13. Estimation results, (5.3), high growth bank sample
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0022 (0.00144) 0.0003 (0.00151) -0.0005 (0.00150)
Loan growth in t-1 -0.0018 (0.00155) 0.0006 (0.00159) 0.0010 (0.00159)
Loan growth in t-2 -0.0036 ** (0.00157) -0.0022 (0.00156) -0.0011 (0.00157)
Loan growth in t-3 -0.0052 *** (0.00162) -0.0041 ** (0.00160) -0.0028 * (0.00161)
Loan growth in t-4 -0.0038 ** (0.00165) -0.0033 ** (0.00162) -0.0030 * (0.00159)
Current NIM^^ 0.1689 *** (0.03943) 0.1756 *** (0.04056) 0.2345 *** (0.04225)
Net Interest Margin in t-1 -0.0221 (0.04423) -0.0053 (0.04447) 0.0074 (0.04574)
Net Interest Margin in t-2 -0.0202 (0.04489) -0.0313 (0.04496) -0.0436 (0.04723)
Net Interest Margin in t-3 0.0328 (0.04967) -0.0006 (0.05122) -0.0400 (0.05218)
Net Interest Margin in t-4 0.0004 (0.04780) 0.0037 (0.04830) 0.0145 (0.04848)
LLPs/assets -0.4579 *** (0.10722)
LLPs/assets in t-1 0.0125 (0.09821)
LLPs/assets in t-2 0.0916 (0.09283)
LLPs/assets in t-3 0.1219 (0.08346)
LLPs/assets in t-4 -0.0243 (0.08413)
Total Assets (1m)^^^ 0.7397 (0.48085) 1.0269 ** (0.47142) 0.9838 ** (0.48668) 0.9350 * (0.47759)
Loans/Assets ratio 0.0134 *** (0.00294) 0.0105 *** (0.00273) 0.0113 *** (0.00303) 0.0124 *** (0.00302)
Equity/Assets at t-1 0.0169 (0.01210) 0.0066 (0.01175) 0.0067 (0.01201) 0.0088 (0.01182)
Inflation^^^^ -0.0512 ** (0.02468) -0.0658 *** (0.02476) -0.0562 ** (0.02486) -0.0594 ** (0.02453)
Current GDP growth^^^^ 0.0010 (0.00132) 0.0013 (0.00126) 0.0007 (0.00129) 0.0008 (0.00127)
GDP growth in t-1 0.0308 (0.01996) 0.0283 (0.01972) 0.0320 (0.02011) 0.0250 (0.01992)
Constant 0.0845 (0.20273) -0.2775 (0.26405) -0.1676 (0.26778) -0.2425 (0.27148)
Within R2 0.0904 0.1346 0.1627 0.2097
No. of observations 538 535 535 532
No. of banks 130 130 130 130
Table 4.13. exposes important differences between two groups. High-growth banks
display a much stronger and more significant negative relation between lagged loan growth
and profitability that is especially visible in Specifications 1 and 2. In other words, their fast
growth has stronger negative repercussions on ROA in a longer term, even though they may
possibly be more proficient in managing credit risk in their portfolios than their low-growth
peers. Indeed, a much higher dependence on interest income is visible for expanding banks,
with a very strong and highly significant relation between current NIM and ROA. These
60
institutions strongly rely on their credit business to bring in revenues, which is also confirmed
by a highly significant relation between profits and current loan-intensity, a variable that was
irrelevant for low growth banks. However, lagged NIM continues to stay irrelevant for future
bottom line results, in the sense that its costs on the risk side are most likely balanced out by
gains achieved due to higher volumes and/or other products.
In terms of internal bank control variables, some discrepancies between low and high
growth banks emerge. High growth institutions show a lack of relation between capitalisation
and net income – so unlike at their low growth peers, better capitalised banks are not
necessarily stronger performers. This follows earlier conclusions on profit dependence on
loan-intensity and interest income, as more credit-inclined banks have higher pressure on their
equity, due to higher risk-weighting of loans versus e.g. Treasury bonds. In addition, for the
first time in the estimation results, bank size emerges as a positive driver of profits.
A significant change appears within the macroeconomic control variables – high
growth banks show their profitability to be independent from GDP growth effects, both
current and lagged by one year. This may be explained by a few factors, such as strong
pressure on growth, even in sluggish years. Thus a slowdown in the economy does not
automatically translate into lower growth (and profits) for these banks, as they are determined
to expand their books at all times. This may have positive and negative implications for
profitability as such, depending on banks‟ client base, expertise in managing weaker quality
clients and a buffer of loan loss reserves. No significance of GDP-ROA relation may however
also imply that when an economic revival boosts other banks‟ profitability, higher growth
institutions do not feel the upheaval as they have to fight their lagged credit quality problems.
A more detailed study of a relation between economic growth and profitability is however
beyond the scope of this text and would require an in-depth analysis, with a usage of a larger
bank dataset. The section below checks the stability of the overconfidence profitability model
in other setups, and including some additional variables.
Robustness tests of the overconfidence profitability model
In order to check stability of the above results, some modifications to the general
overconfidence model expressed by (5.3) are introduced. The first robustness test is
performed on the total sample excluding results for the first half of 2008. The rationale for
this exclusion has been presented in the overconfidence risk model. Results of estimating
(5.3) on the total sample pre-2008 are demonstrated in Table 4.14.
61
Table 4.14. Estimation results, (5.3), total bank sample, without 2008 data
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0007 (0.00107) -0.0019 * (0.00105) -0.0015 (0.00105)
Loan growth in t-1 -0.0006 (0.00106) -0.0015 (0.00105) 0.0000 (0.00104)
Loan growth in t-2 -0.0018 * (0.00102) -0.0022 ** (0.00102) -0.0007 (0.00102)
Loan growth in t-3 -0.0021 ** (0.00102) -0.0026 ** (0.00103) -0.0012 (0.00102)
Loan growth in t-4 -0.0024 ** (0.00102) -0.0026 ** (0.00104) -0.0023 ** (0.00101)
Current NIM^^ 0.1354 *** (0.02821) 0.1368 *** (0.02789) 0.1778 *** (0.02851)
Net Interest Margin in t-1 -0.0221 (0.03163) -0.0282 (0.03143) -0.0202 (0.03137)
Net Interest Margin in t-2 0.0104 (0.03015) 0.0139 (0.02995) 0.0166 (0.03017)
Net Interest Margin in t-3 -0.0039 (0.03367) 0.0070 (0.03307) -0.0276 (0.03350)
Net Interest Margin in t-4 -0.0054 (0.03441) -0.0068 (0.03411) -0.0053 (0.03405)
LLPs/assets -0.5192 *** (0.07521)
LLPs/assets in t-1 0.0181 (0.06305)
LLPs/assets in t-2 0.0093 (0.06008)
LLPs/assets in t-3 0.0965 * (0.05386)
LLPs/assets in t-4 -0.0477 (0.05521)
Total Assets (1m)^^^ 0.0786 (0.14357) -0.0017 (0.14458) 0.1359 (0.13769) 0.1324 (0.14005)
Loans/Assets ratio 0.0054 ** (0.00210) 0.0067 *** (0.00207) 0.0040 ** (0.00194) 0.0078 *** (0.00208)
Equity/Assets at t-1 0.0184 ** (0.00854) 0.0166 * (0.00866) 0.0177 ** (0.00847) 0.0176 ** (0.00835)
Inflation^^^^ -0.0162 (0.01718) -0.0149 (0.01722) -0.0194 (0.01710) -0.0129 (0.01686)
Current GDP growth^^^^ 0.0534 *** (0.01272) 0.0339 *** (0.01197) 0.0604 *** (0.01248) 0.0350 *** (0.01265)
GDP growth in t-1 0.0420 *** (0.01075) 0.0385 *** (0.01070) 0.0423 *** (0.01058) 0.0300 *** (0.01082)
Constant -0.0954 (0.19234) 0.1691 (0.14603) -0.1336 (0.19155) -0.1360 (0.191647)
Within R2 0.1109 0.0700 0.099 0.1707
No. of observations 1106 1121 1106 1094
No. of banks 305 307 305 301
Excluding data from the first half of 2008 does not significantly change earlier
conclusions. Effects of overconfidence, expressed by the loan growth proxy, on longer-term
profitability remain significant and negative, even if somewhat lower. This is possibly driven
by the fact that some banks suffered crisis-induced losses in the first half of 2008 and they
incorporate negative effects of overconfidence most acutely. Pre-2008 results are the last pre-
crisis figures, still mostly untouched by implications of earlier aggressive loan growth and the
relation is likely to be weaker.
Other variables estimated in the pre-2008 sample display no visible discrepancies in
comparison to previous verifications. Macroeconomic variables have expected signs, with
both current and lagged economic growth positively related to profitability. A faster
transmission mechanism between economic upturns and positive bank results seems to have
been in place before the banking crisis. A high share of loans in the balance sheet continues to
boost profitability, while the relation with capital weakens but remains positive and
significant. Specification 4 again indicates a link between loan loss provisions and loan
growth and significance of loan growth is weak in this setup controlling for lagged reserve
making. Nonetheless, LLPs lagged by three years re-emerge as an important positive drive of
later profitability, repeatedly demonstrating profitability gains of earlier prudence.
Statistical value of this estimation is similar as in the full sample, with relatively low
R2 values. Again, this is understandable, given a lack of some drivers of profitability in the
model, such as cost levels and non-interest parts of income.
62
A second robustness test is another replication from the robustness tests in the
overconfidence risk model and accounts for past loan-intensity of assets. Four lags of
loans/assets ratio are added to the general specification of the overconfidence profitability
model expressed by (5.3) Results of this estimation are shown by Table 4.15.
Table 4.15. Estimation results, (5.3) including lagged loans/assets, total bank sample
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0001 (0.00104) -0.0012 (0.00103) -0.0009 (0.00102)
Loan growth in t-1 -0.0008 (0.00104) -0.0017 (0.00104) -0.0002 (0.00102)
Loan growth in t-2 -0.0012 (0.00102) -0.0015 (0.00102) -0.0003 (0.00101)
Loan growth in t-3 -0.0033 *** (0.00098) -0.0036 *** (0.00100) -0.0021 ** (0.00098)
Loan growth in t-4 -0.0032 *** (0.00097) -0.0031 *** (0.00099) -0.0029 *** (0.00095)
Current NIM^^ 0.1387 *** (0.02786) 0.1323 *** (0.02752) 0.1862 *** (0.02801)
Net Interest Margin in t-1 -0.0165 (0.03188) -0.0263 (0.03178) -0.0100 (0.03137)
Net Interest Margin in t-2 -0.0027 (0.03024) 0.0004 (0.03015) 0.0042 (0.03009)
Net Interest Margin in t-3 -0.0118 (0.03311) 0.0092 (0.03290) -0.0306 (0.03267)
Net Interest Margin in t-4 0.0077 (0.03113) -0.0077 (0.03111) 0.0122 (0.03056)
LLPs/assets -0.5183 *** (0.06901)
LLPs/assets in t-1 -0.0227 (0.06107)
LLPs/assets in t-2 0.0047 (0.05432)
LLPs/assets in t-3 0.0949 * (0.05123)
LLPs/assets in t-4 -0.0341 (0.05341)
Current Loans/Assets 0.0016 (0.00240) 0.0043 * (0.00239) 0.0008 (0.00226) 0.0036 (0.00235)
Loans/Assets in t-1 0.0038 (0.00276) 0.0036 (0.00277) 0.0034 (0.00263) 0.0034 (0.00270)
Loans/Assets in t-2 0.0029 (0.00268) 0.0020 (0.00267) 0.0018 (0.00254) 0.0036 (0.00261)
Loans/Assets in t-3 0.0044 * (0.00255) 0.0029 (0.00252) 0.0039 (0.00247) 0.0044 * (0.00249)
Loans/Assets in t-4 -0.0001 (0.00201) -0.0018 (0.00197) 0.0010 (0.00199) -0.0003 (0.00197)
Total Assets (1m)^^^ 0.1718 (0.11956) 0.0775 (0.12095) 0.2362 ** (0.11673) 0.1991 * (0.11596)
Equity/Assets at t-1 0.0217 *** (0.00821) 0.0202 ** (0.00838) 0.0227 *** (0.00815) 0.0192 ** 0.007982
Inflation^^^^ -0.0396 *** (0.01431) -0.0385 *** (0.01429) -0.0468 *** (0.01425) -0.0330 ** (0.01411)
Current GDP growth^^^^ 0.0013 * (0.00075) 0.0012 (0.00073) 0.0016 ** (0.00074) 0.0010 (0.00073)
GDP growth in t-1 0.0395 *** (0.01053) 0.0362 *** (0.01053) 0.0423 *** (0.01045) 0.0269 ** (0.01049)
Constant -0.3898 * (0.20888) 0.0031 (0.16025) -0.3777 * (0.21084) -0.4882 ** (0.20320)
Within R2 0.1306 0.0855 0.1064 0.1961
No. of observations 1171 1189 1171 1157
No. of banks 309 311 309 305
The second robustness test of controlling for past loan intensities of assets
demonstrates stability of previous conclusions regarding loan growth. A higher pace of past
credit expansion retains its adverse influence over later profitability throughout all
specifications. Other variables are of parallel strength and significance levels towards ROA as
in previous verifications, with no impact made by previous NIM levels and strong positive
spurs provided by capitalisation and economic growth variables. On the other hand,
accounting for earlier lags of loan intensity takes away all statistical significance from current
loans/assets. They emerge as weakly significant in Specification 2, while loan intensity lagged
by three years exerts a minimal ROA boost in Specifications 1 and 4, taking away
significance from the current indicator. A highly likely hypothesis is that changes in loan
intensity ratios are small over the years and including five annual values distributes the effect
of a relatively stable ratio onto all five variables, reducing its original statistical strength.
Thus, a general share of loans in total assets positively influences profitability of banks in
general, and in banks engaged in rapid credit expansion in particular. However, it is rather the
63
pace of growth in loans as such and their potential weak quality that negatively affect
profitability and should be associated with overconfidence.
The next robustness test concerns the subgroup estimation. Again, 2008 data is
excluded from the two subgroup samples and (5.3) is re-estimated on both. Results are
showed in Table 4.16. (low growth group) and Table 4.17. (high growth group).
Table 4.16. Estimation results, (5.3), low growth bank group, no 2008 data
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth -0.0012 (0.00153) -0.0018 (0.00154) -0.0022 (0.00152)
Loan growth in t-1 -0.0028 ** (0.00142) -0.0025 * (0.00143) -0.0020 (0.00144)
Loan growth in t-2 -0.0002 (0.00135) -0.0004 (0.00134) 0.0011 (0.00134)
Loan growth in t-3 -0.0022 (0.00131) -0.0022 * (0.00131) -0.0010 (0.00130)
Loan growth in t-4 -0.0028 ** (0.00132) -0.0027 ** (0.00130) -0.0025 * (0.00129)
Current NIM^^ 0.0701 * (0.03953) 0.0679 * (0.03920) 0.1120 *** (0.03909)
Net Interest Margin in t-1 -0.0370 (0.04550) -0.0489 (0.04508) -0.0404 (0.04396)
Net Interest Margin in t-2 0.0452 (0.03955) 0.0389 (0.03945) 0.0553 (0.03833)
Net Interest Margin in t-3 0.0221 (0.04447) 0.0193 (0.04422) 0.0208 (0.04335)
Net Interest Margin in t-4 -0.0702 (0.04642) -0.0786 * (0.04597) -0.0715 (0.04548)
LLPs/assets -0.5764 *** (0.09925)
LLPs/assets in t-1 0.0196 (0.08024)
LLPs/assets in t-2 -0.0357 (0.06949)
LLPs/assets in t-3 0.1320 * (0.06796)
LLPs/assets in t-4 -0.0267 (0.07278)
Total Assets (1m)^^^ -0.0406 (0.14572) -0.1092 (0.14544) -0.0310 (0.13703) 0.0159 (0.14062)
Loans/Assets ratio 0.0005 (0.00283) 0.0012 (0.00281) -0.0007 (0.00270) 0.0036 (0.00281)
Equity/Assets at t-1 0.0424 *** (0.01242) 0.0286 ** (0.01207) 0.0408 *** (0.01222) 0.0342 *** (0.01203)
Inflation^^^^ 0.0047 (0.02249) 0.0043 (0.02199) -0.0027 (0.02220) 0.0197 (0.02215)
Current GDP growth^^^^ 0.0365 ** (0.01433) 0.0373 *** (0.01386) 0.0445 *** (0.01396) 0.0217 (0.01401)
GDP growth in t-1 0.0402 *** (0.01230) 0.0398 *** (0.01232) 0.0397 *** (0.01192) 0.0269 ** (0.01229)
Constant 0.2268 (0.25893) 0.3209 (0.20180) 0.2886 (0.25791) 0.1039 (0.25250)
Within R2 0.1312 0.1070 0.1144 0.2184
No. of observations 596 608 596 587
No. of banks 176 178 176 172
Re-estimating (5.3) on the group of low growth banks using pre-2008 figures proves
stability of the model and its earlier interpretation. Overconfidence in loan growth negatively
affects profitability, net interest margins are insignificant in most part, but current NIM is now
related to ROA in all specifications, even if in some it is on the verge of statistical
significance. Highly capitalised banks continue to be the best profit makers in this subgroup,
while loan intensity proves again to be of little impact. The value of current economic growth
variable is boosted as a driver of ROA, while lagged GDP growth stays also statistically
reliable. Higher provisions lagged by three years demonstrate benefits of prudency once again
in this subgroup.
Table 4.17. Estimation results, (5.3), high growth bank group, without 2008 data
Dependent variable:
ROA
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0001 (0.00158) -0.0022 (0.00154) -0.0006 (0.00157)
Loan growth in t-1 0.0011 (0.00169) -0.0015 (0.00166) 0.0016 (0.00169)
Loan growth in t-2 -0.0027 (0.00163) -0.0044 *** (0.00165) -0.0014 (0.00165)
Loan growth in t-3 -0.0024 (0.00167) -0.0039 ** (0.00169) -0.0017 (0.00166)
Loan growth in t-4 -0.0027 (0.00169) -0.0033 * (0.00174) -0.0027 (0.00167)
64
Current NIM^^ 0.1897 *** (0.04118) 0.1870 *** (0.04020) 0.2410 *** (0.04285)
Net Interest Margin in t-1 0.0003 (0.04493) -0.0141 (0.04430) 0.0092 (0.04678)
Net Interest Margin in t-2 -0.0046 (0.04592) 0.0078 (0.04529) -0.0223 (0.04905)
Net Interest Margin in t-3 -0.0135 (0.05155) 0.0076 (0.04968) -0.0474 (0.05313)
Net Interest Margin in t-4 0.0265 (0.05331) 0.0261 (0.05236) 0.0291 (0.05353)
LLPs/assets -0.4700 *** (0.11472)
LLPs/assets in t-1 0.0245 (0.10249)
LLPs/assets in t-2 0.1152 (0.11207)
LLPs/assets in t-3 0.0652 (0.08890)
LLPs/assets in t-4 -0.0342 (0.08669)
Total Assets (1m)^^^ 0.7646 (0.49376) 0.6554 (0.49697) 0.7990 * (0.48080) 0.7748 (0.48658)
Loans/Assets ratio 0.0088 *** (0.00320) 0.0122 *** (0.00314) 0.0080 *** (0.00290) 0.0104 *** (0.00319)
Equity/Assets at t-1 -0.0011 (0.01243) 0.0084 (0.01271) -0.0011 (0.01218) 0.0026 (0.01231)
Inflation^^^^ -0.0399 (0.02691) -0.0359 (0.02718) -0.0472 * (0.02678) -0.0424 (0.02664)
Current GDP growth^^^^ 0.0744 *** (0.02494) 0.0176 (0.02330) 0.0829 *** (0.02453) 0.0555 ** (0.02492)
GDP growth in t-1 0.0411 ** (0.02030) 0.0354 * (0.02031) 0.0366 * (0.01968) 0.0310 (0.02034)
Constant -0.4085 (0.30234) 0.1030 (0.22059) -0.5032 * (0.29847) -0.3953 (0.31216)
Within R2 0.1619 0.0734 0.1431 0.2043
No. of observations 510 513 510 507
No. of banks 129 129 129 129
Robustness of the estimation results regarding high growth banks is lower, as the pre-
2008 sample suggests some changed features in comparison to the group including 2008
results. Here, dependence between economic growth and profitability re-emerges, to varying
degrees including lagged and/or current GDP. This proves that earlier results are not very
stable and that profitability of high growth institutions also depends on economic fluctuations.
Possible distortions originating from including or excluding crisis-driven figures may play a
role in non-significance of macroeconomic variables in the 2008 setup. Loan intensity within
this group remains statistically similar, confirming a dependence of these institutions on their
credit-driven business. This is further exacerbated by a visibly higher and more significant
reliance of bottomline profits on current net interest margins, which are a primary source of
income.
On the other hand, high growth banks pre-2008 display a striking inter-dependence
between NIM and LGR in the profitability setup. Although lagged loan growth shows a high
and significant relation with profitability, when estimated separately (Specification 2), the link
disappears when NIM is accounted for simultaneously (Specification 1 and 4). This indicates
that a mixture between possibly lower NIM and higher LGR cancels out the negative effect of
loan growth. This remains in line with the earlier hypothesis that negative implications (in the
shape of provisions) caused by an aggressive growth policy may be counterbalanced by
benefits derived from increased volumes. However, this is valid until economic growth
persists and banks are profitable, so in the pre-crisis period of up to end-2007. Come
economic downturn and more intense pressure from lagged defaulted loans, a strong negative
effect of higher growth materialises on profitability.
This could potentially be even more visible in a comparison between samples as of
end-2007 and end-2008, however these last figures are unavailable at the time of the
estimation. Despite that, the estimation results reveal that negative implications of past loan
65
growth on later profitability are likely to be weak among high growth banks, until economic
conditions are favourable and no slouch in growth is experienced. It is thus during the
business and economic cycle downturns that adverse repercussions of overconfidence make
themselves most painful for high growth banks.
Last but not least, a robustness test is performed on the profitability ratio itself,
replacing the primary dependent variable with the second most-popular profitability indicator,
Return on Equity, ROE. To account for this, (5.3) is reformulated and a second profitability
specification is demonstrated by Equation (5.4).
∑
∑
(5.4)
Results of estimating (5.4) on the whole sample of banks are presented in Table 4.17.
Table 4.17. Estimation results, Equation 4.6., total bank sample
Dependent variable:
ROE
Specification 1 Specification 2 Specification 3 Specification 4
Coef. Std. Err. Coef. Std. Err. Coef. Std. Err.
Current loan growth 0.0146 (0.01721) -0.0042 (0.01738) 0.0014 (0.01704)
Loan growth in t-1 0.0043 (0.01699) -0.0049 (0.01741) 0.0157 (0.01685)
Loan growth in t-2 -0.0086 (0.01651) -0.0088 (0.01702) 0.0061 (0.01647)
Loan growth in t-3 -0.0549 *** (0.01637) -0.0620 *** (0.01698) -0.0369 ** (0.01642)
Loan growth in t-4 -0.0435 *** (0.01639) -0.0480 *** (0.01701) -0.0391 ** (0.01626)
Current NIM^^ 2.0040 *** (0.46520) 1.8687 *** (0.46264) 2.7477 *** (0.47227)
Net Interest Margin in t-1 -0.5082 (0.52621) -0.7275 (0.52708) -0.5326 (0.52228)
Net Interest Margin in t-2 -0.1520 (0.50058) -0.1439 (0.50169) -0.0206 (0.50154)
Net Interest Margin in t-3 -0.4927 (0.55443) -0.0960 (0.55011) -0.7987 (0.55126)
Net Interest Margin in t-4 -0.2460 (0.51950) -0.4684 (0.52066) -0.1155 (0.51513)
LLPs/assets -8.2249 *** (1.17445)
LLPs/assets in t-1 0.8862 (1.03008)
LLPs/assets in t-2 0.1307 (0.92199)
LLPs/assets in t-3 1.5027 * (0.86879)
LLPs/assets in t-4 -0.6760 (0.90196)
Total Assets (1m)^^^ 3.2041 (2.01737) 2.2319 (2.08288) 5.0061 ** (1.95799) 3.5463 * (1.96817)
Loans/Assets ratio 0.0207 (0.03316) 0.0579 * (0.03348) 0.0133 (0.03084) 0.0519 (0.03290)
Equity/Assets at t-1 -0.5886 *** (0.13722) -0.6616 *** (0.14275) -0.5773 *** (0.13711) -0.6300 *** (0.13416)
Inflation^^^^ -0.6987 *** (0.24191) -0.5828 ** (0.24612) -0.7939 *** (0.24149) -0.6158 ** (0.23980)
Current GDP growth^^^^ 0.0323 ** (0.01275) 0.0155 (0.01260) 0.0398 *** (0.01255) 0.0273 ** (0.01245)
GDP growth in t-1 0.4773 *** (0.17864) 0.5099 *** (0.18166) 0.5459 *** (0.17715) 0.3205 * (0.17915)
Constant 14.2676 *** (2.82568) 13.4399 *** (2.30248) 13.7534 *** (2.83494) 12.5663 *** (2.80635)
Within R2 0.1017 0.0728 0.0772 0.1586
No. of observations 1171 1189 1171 1157
No. of banks 309 311 309 305
The stability of the overconfidence profitability model is reaffirmed in the above
robustness test. The central conclusions are confirmed and overconfidence proxied by loan
growth preserves its negative influence over later profitability, measured here by ROE. No
new relations emerge in this setup, apart from an understandable spurious relation between
lagged equity and ROE, which is bound to be negative by construction. Loan intensity
emerges with weak significance, indicating that a technical relation between this and ROA
66
might have been present before, which was partially proven by a low statistical reliability of
lagged loan intensities in the previous robustness test. Relations between economic growth
and profitability are restated, underlining the lagged GDP as a primary driver of bank profits
in this measurement. Some setups indicate size of assets as a factor boosting end results, but
this link disappears in other variations.
Using empirical data from above 300 Western European banks in the period 2002-
2008 with up to four years in lags, the verification of the overconfidence risk model has
confirmed the primary hypothesis that managerial overconfidence may have an effect on
subsequent bank risk and profitability levels.
First, overconfidence proxied by aggressive loanbook growth and low net interest margin
policies is proven to exert pressure on loan loss reserve needs in later periods. A three year
delay is the most significant turning point, where these effects are the strongest. This remains
in line with findings of other authors, claiming this period to be crucial for emergence of non-
performing loans. This negative impact of overconfidence proxies on risk is proven to be
robust across various settings. In addition, banks in the analysed sample and time period are
found to have engaged in income smoothing, increasing their provisioning levels as their pre-
provisions income expanded. Despite this, higher loan growth was not accompanied by higher
contemporary reserve making and, in some settings, it was even paired with lower LLPs. In
addition, banks in the sample have engaged in pro-cyclical provisioning versus the economic
cycle, increasing their vulnerability to economic downturns. Taken together, this indicates
that income smoothing is not motivated by prudential reasons stemming from ex ante credit
risk management or using economic prosperity to build up reserve cushions. Instead, earnings
management is potentially used to fulfil other aims, which according to the literature may
include private benefit goals on the side of bank management.
Second, overconfidence proxied by loan growth is also proven to undermine later
profitability, with a lag of three and four years. This result is robust across specifications and
samples, proving that the aggressive risk taking of banks may be due to overconfidence
effects. Managers are inclined to overestimate their risk management skills, future bank
performance and operating environment conditions. On the other hand, probability of adverse
occurrences within a bank and in the economy is underestimated, with an underlying
psychological certainty that “the future will be great, especially for me” (Weinstein, 1980).
The secondary hypothesis regarding differences in overconfidence effects between
more- and less overconfident banks is also confirmed. A subsample estimation brought
forward differences between high growth (potentially more overconfident) and low growth
67
(less overconfident) banks, exposing negative repercussions of aggressive net interest margins
on risk in the overconfident banks group, as well as their overconfidence expressed by low
provisioning levels accompanying high growth. In addition, adverse consequences of high
growth policy on profitability are significantly more visible for more overconfident banks.
6. Conclusions
The main aim of this work was an empirical study of the risk profiles and profitability of
Western European banks, and an assessment of the effect that overconfidence may have on
both of these areas. This goal was meant to be achieved through the empirical verification of
the primary hypothesis stating that overconfidence has a significant influence on bank risk
exposures and profitability levels that is especially visible in a longer term. The secondary
hypothesis stated that effects of overconfidence may diverge, between banks identified as
more- or less overconfident.
The theoretical considerations presented in this work, in conjunction with the empirical
verification of the overconfidence risk model and overconfidence profitability model, confirm
both the primary and the secondary hypothesis. As a result, the research objective of this work
is fulfilled.
Section 1 has demonstrated that although moral hazard has to date been the prevailing
framework for analysing risk taking motivation in banks, it includes some shortcomings.
These deficiencies relate, among others, to the underlying implication that managers assume
risk knowingly and rationally assess it. CEOs are presumed to be able to quantify the scope of
risk and its consequences on the performance of the bank and their private benefits before the
risk is taken. Overconfidence, discussed in Section 2, supplements this rationale with a
possibility that bank managers may be inclined to take risks due to their cognitive biases.
They underestimate risks taken and overestimate own future performance, due to
miscalibration and better-than-average effects. They may erroneously believe that they are
able to control events which are beyond their influence, such as borrower behaviour, his
moral hazard or developments in underlying asset markets. In addition, they underestimate
probabilities of adverse occurrences in the macroeconomic environment, displaying
unrealistic optimism.
The contemporary banking literature presented in Section 3 has established that one of
crucial drives of bank risk is loan growth, examined in the credit risk context. Rapid
expansion is frequently paired with aggressive pricing. Setting ambitious growth targets on
68
developed Western Europe banking markets frequently implies expanding onto client groups
that are of higher risk than the existing client base. Nonetheless, pressure on profitability
prevents banks from charging sufficient risk premiums on these new, risky loans. Banks are
proven to engage in income smoothing, but the authors are not unanimous as to its
implications, due to diverging motivations for income smoothing. If earnings management
through loan loss provisions is performed out of prudential reasons, it could have positive
consequences for later bank results. A stronger agreement exists in the area of pro-cyclical
provisioning, which is demonstrated to exacerbate adverse consequences of economic crises,
if these occur. Bank capital levels and capital requirements are not commonly believed to
solve the moral hazard problem expressed through asset substitution. Recent studies
demonstrate that imposing higher capital requirements on banks may in fact boost their risk
taking, instead of curbing it. As a result, capital emerges as an important control factor in risk
and profitability analyses of banks, but the sign of its relations with the other variables is not a
given.
In Section 4, the theoretical considerations on overconfidence and bank risk taking
have been collected. This has allowed to create the empirically verifiable model of
overconfidence in banks, comprising the two main forms, the overconfidence risk model and
the overconfidence profitability model. The primary proxy for overconfidence has emerged in
the form of loan growth, supplemented by net interest margin as the secondary proxy. Key
dependent variables for risk and profitability models have been identified as loan loss
provisions to assets and return on assets, respectively.
Empirical verification of the overconfidence model is carried out in Section 5. The
fixed effects model for panel data is presented as the most appropriate econometric method to
validate the assumptions, given the Western European data set of 311 banks. The estimation
has proven that both overconfidence proxies have an effect on bank risk exposures. This
effect materialises and is mostly visible in a three year lag, showing that overconfidence has a
delayed adverse influence. In addition, the primary overconfidence proxy has a strong
negative effect on bank profitability three and four years later. This vital result indicates that
managerial risk taking may be driven by an underestimation of risk, overestimation of own
risk management capabilities and of future earnings, as well as unrealistic optimism and
faulty belief in personal control over external outcomes. In consequence, the primary
hypothesis of this work is confirmed, using the given data set.
The secondary hypothesis is also validated in Section 5, which demonstrates that
overconfidence may have different effects on potentially less and more overconfident banks.
69
The most visible differences in the risk setting suggest that less overconfident banks may be
pricing risk more adequately and requiring higher risk premiums from more risky clients.
More overconfident banks are hit by adverse results of more aggressive pricing in a three year
lag, an effect which does not exist in the less overconfident bank group. The profitability
setup brings forward negative consequences of loan growth on profitability delayed by three
and four years, experienced especially by more overconfident banks.
The additional propositions are estimated in the overconfidence risk model. The
results suggest that banks in the sample have engaged in income smoothing, but it was not
driven by conservative risk management. Loan loss provisions have been demonstrated to
remain largely independent of current loan growth, and even negatively related to it in some
settings. This proves that banks do not engage in ex ante reserve making and income
smoothing detected in the results is potentially driven by managerial private benefits. The
results relating to the second proposition exacerbate the lack of prudential considerations in
bank risk reserve making. Loan loss provisions are shown to be strongly procyclical versus
economic growth, implying that no reserves are made on the back of strong, growing markets.
As a result, when recessions appear, banks suffer increased pressure on profits coming not
only from lower revenues but also from increased needs of creating additional loan loss
reserves. Thus, the additional propositions strengthen the rationale of potential managerial
overconfidence in banks. Banks exploit loan loss provisions in order to smooth earnings and
this is driven by non-prudential reasons, separate from credit risk management considerations.
In addition, periods of high economic growth are not used to puff up existing reserves, as
managers may erroneously believe the boom to last unrealistically long.
This work has demonstrated that moral hazard may not be the unique rationale that can
be applied to account for risk taking in financial institutions. Overconfidence, stemming from
behavioural finance, may enhance the existing moral hazard framework, allowing for
managerial biases that influence decision making. The effects of overconfidence on risk and
profitability of banks that have been identified in this work have been negative. They surface
with a delay of three to four years and have different implications for less and more
overconfident bank groups.
This study is a preliminary step to including overconfidence into future bank research.
Despite some existing work, the field remains to be explored in more detail. Possible
extensions could include applying other bank data sets, most particularly with end-2008 and
2009 results, as well as extending the geographical scope to account for the US banks. In such
a case, some overconfidence analysis tools commonly applied in corporate finance could be
70
used, including delays in managerial stock option plan executions. The overconfidence model
put forward in this work could be enhanced by using other variables, including detailed loan
portfolio data, should it become more available. Furthermore, the second strand of
overconfidence research in corporate finance relating to mergers and acquisitions could also
be applied to the banking context. This is particularly interesting in view of catastrophic
results of recent bank mergers on international markets, such as the ABN Amro takeover by
Royal Bank of Scotland.
Last but certainly not least, an analysis with significant added value would be a
comparison of overconfidence implications between developed and developing market banks.
This draws on existing studies of overconfidence per se, which attempt to establish if
overconfidence depends on the operating environment and is to some extent exacerbated by
this environment, or if it is an inherent personal trait that is independent of regional
influences.
In conclusion, the financial market downturn of 2007/08 has demonstrated that
existing conventional studies of bank risk taking do not suffice to account for recent
developments, especially on developed markets. Thus, new assumptions should be considered
that allow to extend existing banking models by accounting for new feedback from
behavioural finance.
71
References27
1. Ahmed, A.S., Takeda, C., Thomas, S. (1999) Bank loan loss provisions: a
reexamination of capital management, earnings management and signalling effects,
Journal of Accounting and Economics, 28, p.1-25.
2. Alicke M.D., Klotz M.L., Breitenbecher D.L., Yurak T.J., Vredenburg D.S. (1995),
Personal contact, Individuation and the Better-Than-Average Effect, “Journal of
Personality and Social Psychology”, Vol. 68, No. 5, p. 804-825.
3. Altunbas, Y., Carbo, S., Gardener, E., Molyneux, P. (2007) Examining the
Relationships between Capital, Risk and Efficiency in European Banking, European
Financial Management, 13(1), p.49-70.
4. Ayton P., McClelland A. (1997), How Real is Overconfidence?, “Journal of Behavioral
Decision Making”, Vol. 10, No. 3, p. 279-285
5. Babcock L., Loewenstein G. (1997), Explaining Bargaining Impasse: The Role of Self-
Serving Biases, “Journal of Economic Perspectives”, Vol. 11, No. 1, p. 109-126.
6. Barber B., Odean T. (2000), Trading Is Hazardous to Your Wealth: The Common Stock
Investment Performance of Individual Investors, “Journal of Finance”, Vol. 55, No. 2,
p. 773-806.
7. Barber B., Odean T. (2001), Boys Will Be Boys: Gender, Overconfidence, and Common
Stock Investment, “The Quarterly Journal of Economics”, Vol. 116, No. 1, p. 261-292.
8. Bauer, P., Berger, A.N., Ferrier, G., Humphrey, D. (1998) Consistency Conditions for
Regulatory Analysis of Financial Institutions: A Comparison of Frontier Efficiency
Methods, Journal of Economics and Business, 50, p.85-114.
9. Ben-David I., Graham J.R., Harvey C.R. (2007), Managerial Overconfidence and
Corporate Policies, “AFA 2007 Chicago Meetings Paper”, November,
http://ssrn.com/abstract=890300
10. Benoit J-P., Dubra J. (2007), Overconfidence?, “MPRA Paper”, No. 6017,
http://mpra.ub.uni-muenchen.de/6017/
11. Benos A.V. (1998), Aggressiveness and survival of overconfident traders, “Journal of
Financial Markets”, Vol. 1, No. 3-4, p. 353-383.
12. Beatty, A., Chamberlain, S., Magliolo, J. (1995) Managing Financial Reports of
Commercial Banks: The Influence of Taxes, Regulatory Capital, and Earnings,
Journal of Accounting Research, 33(2), p.231-261.
13. Berger, A.N., Udell, G.F. (1990) Collateral, loan quality and bank risk, Journal of
Monetary Economics, 25, p.21-42.
14. Berger, A.N., Herring, R. J., Szego, G.P. (1995), The role of capital in financial
institutions, “Journal of Banking and Finance”, 19, p. 393-430.
15. Berger, A.N., Humphrey, D. (1997) Efficiency of financial institutions: International
survey and directions for future research, European Journal of Operational Research,
98(2), p.175-212.
27
This reference list includes positions that have not been directly used in this work. They constitute a broader
reference list for an interested reader.
72
16. Berger, A.N., Udell, G.F. (2004) The institutional memory hypothesis and the
procyclicality of bank lending behavior, Journal of Financial Intermediation, 13, p.
458-495.
17. Bernardo A., Welch I. (2001), On the Evolution of Overconfidence and Entrepreneurs,
“Journal of Economics and Management Strategy”, Vol. 10, No. 3, p. 301-330.
18. Beyer P. (1990), Gender Differences in the Accuracy of Self-Evaluations of
Performance, “Journal of Personality and Social Psychology”, Vol. 59, No. 5, p. 960-
970.
19. Beyer S., Bowden E. (1997), Gender Differences in Self-Perceptions: Convergent
Evidence from Three Measures Accuracy and Bias, “Personality and Social
Psychology Bulletin”, Vol. 23, No. 2, p. 157-172.
20. Besanko, D., Kanatas, G. (1996), The regulation of bank capital: Do capital standards
promote bank safety? “Journal of Financial Intermediation”, 5, p. 160–183.
21. Bester, H. (1985) Screening vs Rationing in Credit Markets with Imperfect
Information, American Economic Review, 75 (4), p.850-855.
22. Bhattacharya, S. (1982), Aspects of Monetary and Banking Theory and Moral Hazard,
“Journal of Finance”, 37, p. 371-384.
23. Biais B., Hilton D., Mazurier K., Pouget S. (2005), Judgemental Overconfidence, self-
monitoring and trading performance in an experimental financial market, “Review of
Economic Studies”, Vol. 72, No. 251, p. 287-312.
24. BIS (1988) International Convergence of Capital Measurement and Capital
Standards, Basel Committee on Banking Supervision (July 1988), as amended.
25. BIS (1996) Amendment to the capital accord to incorporate market risks, Basel
Committee on Banking Supervision.
26. BIS (2006) International Convergence of Capital Measurement and Capital
Standards, Basel Committee on Banking Supervision, as amended.
27. Blum, J (2008), Why ‘Basel II’ may need a leverage ratio restriction, “Journal of
Banking and Finance”, 32, 1699–1707.
28. Boot, A.W. (2000) Relationship Banking: What Do We Know?, Journal of Financial
Intermediation, 9, p. 7-25.
29. Borio, C. (2008) The Financial Turmoil of 2007-?: A Preliminary Assessment and
Some Policy Considerations, BIS working paper no.251.
30. Brenner L.A., Koehler D.J., Liberman V., Tversky A. (1996), Overconfidence in
Probability and Frequency Judgments: A Critical Examination, “Organizational
Behavior and Human Decision Processes”, Vol. 65, No. 3, p. 212-219.
31. Brenner L., Griffin D. (2004), Perspectives on Probability Judgment Calibration, in:
D.J. Koehler, N. Harvey, (eds.) Blackwell Handbook of Judgement and Decision
Making, Blackwell Publishing, Oxford.
32. Calem, P., Rob, R. (1999), The Impact of Capital-Based Regulation on Bank Risk-
Taking, “Journal of Financial Intermediation”, 8, p. 317-352.
33. Camerer C., Lovallo D. (1999), Overconfidence and Excess Entry: An Experimental
Approach, “American Economic Review”, Vol. 89, No. 1, p. 306-318.
34. Campbell, T., Chan, Y-S., Marino, A. (1992), An incentive-based theory of bank
regulation, “Journal of Financial Intermediation”, 2, p.255-276.
73
35. Cavallo, M., Majnoni, G. (2001) Do Banks Provision for Bad Loans in Good Times?
Empirical Evidence and Policy Implications, World Bank Policy Research Working
Paper, 2619.
36. Chuang W.-I., Lee B.-S. (2006), An empirical evaluation of the overconfidence
hypothesis, “Journal of Banking and Finance”, Vol. 30, No. 9, p. 2489-2515.
37. Collins, J.H., Shackelford, D.A., Wahlen, J.M. (1995) Bank Differences in the
Coordination of Regulatory Capital, Earnings and Taxes, Journal of Accounting
Research, 33(2), p.263-291.
38. Cottarelli, C., Dell‟Ariccia, G., Vladkova-Hollar, I. (2005) Early brids, late risers, and
sleeping beauties: Bank credit growth to the private sector in Central and Eastern
Europe and in the Balkans, Journal of Banking and Finance, 29, p.83-104.
39. Daniel K., Hirshleifer D., Subrahmanyam A. (1998), Investor Psychology and Security
Market Under- and Overreactions, “Journal of Finance”, Vol. 53, No. 6, p. 1839-
1885.
40. Daniel K., Hirshleifer D., Subrahmanyam A. (2001), Overconfidence, Arbitrage, and
Equilibrium Asset Pricing, “Journal of Finance”, Vol. 56, No. 3, p. 921-965.
41. Dawes R., Mulford M. (1996), The False Consensus Effect and Overconfidence:
Flaws in Judgment or Flaws in How We Study Judgment?, “Organizational Behavior
and Human Decision Processes”, Vol. 65, No. 3, p. 201-211.
42. Deaves R., Luders E., Schroder M. (2005), The Dynamics of Overconfidence:
Evidence from Stock Market Forecasters, “Discussion Paper”, No. 05-83, Centre for
European Economic Research (ZEW), Mannheim.
43. De Bondt W.F. (1998), A portrait of the individual investor, “European Economic
Review”, Vol. 42, No. 3-5, p. 831-844.
44. De Long B.J., Shleifer A., Summers L.H., Waldmann R. (1990), Noise Trader Risk in
Financial Markets, “Journal of Political Economy”, Vol. 99, No. 4, p.703-738.
45. De Long, B.J., Shleifer, A., Summers, L.H., Waldmann, R. (1991,) The survival of
noise traders in financial markets, “Journal of Business”, Vol. 64, No. 1, p. 1-19.
46. Demirguc-Kunt, A., Detragiache, E. (1998) The Determinants of Banking Crises in
Developing and Developed Countries, IMF Staff Papers, 45(1), p. 81-109.
47. Diamond, D.W., Dybvig, P.H. (1983) Bank runs, Deposit Insurance and Liquidity,
Journal of Political Economy, 91, p.401-419.
48. Diamond, D.W. (1984) Financial Intermediation and Delegated Monitoring, Review
of Economic Studies, 51, p.393-414.
49. Diamond, D.W., Rajan, R.G. (2000), A Theory of Bank Capital, “Journal of Finance”,
55, p.2431-2465.
50. Diamond, D.W., Rajan, R.G. (2001),Bank liquidity.
51. Dunning D., Meyerowitz J.A., Holzberg A.D. (1989), Ambiguity and Self-Evaluation:
The Role of Idiosyncratic Trait Definitions in Self-Serving Assessments of Ability,
“Journal of Personality and Social Psychology”, Vol. 57, No. 6, p. 1082-1090.
52. Fama E., (1998), Market efficiency, long-term returns, and behavioral finance,
“Journal of Financial Economics”, Vol. 49, No. 3, p. 283–306.
74
53. Fischhoff B., Lichtenstein S., Slovic P. (1977), Knowing with Certainty: The
Appropriateness of Extreme Confidence, “Journal of Experimental Psychology”, Vol.
3, No. 4, p. 552-564.
54. Fischhoff B., Koriat A., Lichtenstein S. (1980), Reasons for Confidence, “Journal of
Experimental Psychology: Human Learning and Memory”, Vol. 6, No. 2, p. 107-118.
55. Foos, D., Norden, L., Weber, M. (2007), Loan Growth and Riskiness of Banks,
University of Mannheim Working Paper 142.
56. Friesen G., Weller P.A. (2006), Quantifying cognitive biases in analyst earnings
forecasts, “Journal of Financial Markets”, Vol. 9, No. 4, p. 333-365.
57. Furlong, F.T., Keeley, M.C. (1989), Capital Regulation and Bank Risk Taking: A
Note, “Journal of Banking and Finance”, 13, p.883-891.
58. Gennotte, G., Pyle D. (1991), Capital Controls and Bank Risk, “Journal of Banking
and Finance”, 15, p. 805-824.
59. Gervais S., Odean T. (2001), Learning to Be Overconfident, “The Review of Financial
Studies”, Vol. 14, No. 1, p. 1-27.
60. Gigerenzer G., Hoffrage U., Kleinbolting H. (1991), Probabilistic Mental Models: A
Brunswikian Theory of Confidence, “Psychological Review”, Vol. 98, No. 4, p. 506-
528.
61. Glaser M., Weber M. (2005), Overconfidence of Professionals and Lay Men:
Individual Differences Within and Between Tasks?,
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=712583
62. Glaser M., Weber M. (2007), Overconfidence and Trading Volume, “The Geneva Risk
and Insurance Review”, Vol. 32, No. 1, p. 1-36.
63. Goel, A.M., Thakor, A.V. (2008) Overconfidence, CEO Selection, and Corporate
Governance, Journal of Finance, 53 (6), p.2737-2784.
64. Gorton, G., Rosen, R. (1995), Corporate Control, Portfolio Choice, and the Decline of
Banking, “Journal of Finance”, 50, p. 1377- 1420.
65. Greenbaum Thakor (2007)
66. Greene, W.H. (2007) Econometric Analysis, eds. Pearson, Prentice Hall, Upper Saddle
River, NJ.
67. Griffin D., Tversky A. (1992), The Weighing of Evidence and the Determinants of
Confidence, “Cognitive Psychology”, Vol. 24, No. 3, p. 411- 435.
68. Griffin D., Varey C. (1996), Towards a Consensus on Overconfidence,
“Organizational Behavior and Human Decision Processes”, Vol. 65, No. 3, p. 227-
231.
69. Hardy, D.C., Pazarbaşioğlu, C. (1999) Determinants and Leading Indicators of
Banking Crises: Further Evidence, IMF Staff Papers, 46(3), p.247-258.
70. Harris, M., Raviv, A. (1979) Optimal Incentive Contracts with Imperfect Information,
Journal of Economic Theory, 20, p.231-259.
71. Heaton J.B. (2002), Managerial Optimism and Corporate Finance, “Financial
Management”, Vol. 31, No. 2, p. 33-45.
72. Hellman, T.F., Murdock, K.C., Stiglitz, J.E. (2000), Liberalization, Moral Hazard in
Banking, and Prudential Regulation: Are Capital Requirements Enough? “American
Economic Review”, 90, p. 147-165.
75
73. Hilary G., Menzly L. (2006), Does Past Success Lead Analysts to Become
Overconfident? “Management Science”, Vol. 52, No. 4, p. 489-500.
74. Hirshleifer D., Luo G. Y. (2001), On the survival of overconfident traders in a
competitive securities market, “Journal of Financial Markets”, Vol. 4, No. 1, p. 73-84.
75. Howells, P., Bain, K. (2008) The Economics of Money, Banking and Finance, FT
Press.
76. Iwanicz-Drozdowska (2005) Zarządzanie finansowe bankiem, Polskie Wydawnictwo
Ekonomiczne, Warszawa.
77. Jeitschko, T.D., Jeung, S.D. (2005), Incentives for risk-taking in banking – A unified
approach, “Journal of Banking and Finance”, 29, p. 759-777.
78. Jimenez, G., Saurina, J. (2004) Collateral, type of lender and relationship banking as
determinants of credit risk, Journal of Banking and Finance, 28, p.2191-2212.
79. Juslin P. (1994), The Overconfidence Phenomenon as a Consequence of Informal
Experimenter-Guided Selection of Almanac Items, “Organizational Behavior and
Human Decision Processes”, Vol. 57, No. 2, p. 226-246.
80. Juslin P., Olsson H., Bjorkman M. (1997), Brunswikian and Thurstonian origins of
bias in probability assessment: On the interpretation of stochastic components of
judgements, “Journal of Behavioral Decision Making”, Vol. 10, No. 3, p. 189-209.
81. Kahn, C., Winton, A. (2004) Moral Hazard and Optimal Subsidiary Structure for
Financial Institutions, Journal of Finance, 59 (6), p.2531-2575.
82. Kahneman D., Lovallo D. (1993), Timid Choices and Bold Forecasts: A Cognitive
Perspective on Risk Taking, “Management Science”, Vol. 39, No. 1, p. 17-31.
83. Kahneman D., Tversky A. (1982), Heuristics and biases, in: D. Kahneman, P. Slovic,
A. Tversky (eds.) Judgement Under Uncertainty: Heuristics and Biases, Cambridge
University Press, Cambridge.
84. Keeley, M.C., Furlong, F.T. (1990), A reexamination of the mean-variance analysis of
bank capital regulation, “Journal of Banking and Finance”, 14, p. 69-84.
85. Keren G. (1991), Calibration and probability judgements: Conceptual and
methodological issues, “Acta Psychologica”, Vol. 77, No. 3, p. 217-273.
86. Keren G. (1997), On The Calibration of Probability Judgments: Some Critical
Comments and Alternative Perspectives, “Journal of Behavioral Decision Making”,
Vol. 10, No. 3, p. 269-278.
87. Kim, D., Santomero, A.M. (1993) Forecasting Required Loan Loss Reserves, Journal
of Economics and Business, 45, p.315-329.
88. Koehler D.J., Brenner L., Griffin D. (2002), The Calibration of Expert Judgment:
Heuristics and Biases Beyond the Laboratory, in: T. Gilovich, D. Griffin and D.
Kahneman (eds.), Heuristics and Biases: The Psychology of Intuitive Judgement,
Cambridge University Press, Cambridge.
89. Kraft, E., Jankov, L. (2005) Does speed kill? Lending booms and their consequences
in Croatia, Journal of Banking and Finance, 29, p.105-121.
90. Kyle A.S., Wang F.A. (1997), Speculation Duopoly with Agreement to Disagree: Can
Overconfidence Survive the Market Test?, “Journal of Finance”, Vol. 52, No.5, p.
2073-2090.
76
91. Laeven, L., Majnoni, G. (2003) Loan loss provisioning and economic slowdowns: too
much, too late?, Journal of Financial Intermediation, 12(2), p. 178-197.
92. Langer E.J., Roth J. (1975), Heads I Win, Tails It’s Chance: The Illusion of Control as
a Function of the Sequence of Outcomes in a Purely Chance Task, “Journal of
Personality and Social Psychology”, Vol. 32, No. 6, p. 951-955.
93. Langer E.J. (1975), The Illusion of Control, “Journal of Personality and Social
Psychology”, Vol. 32, No. 2, p. 311-328.
94. Leland, H., Pyle, D. (1977) Informational Assymetries, Financial Structure and
Financial Intermediation, Journal of Finance, 32, p.371-387.
95. Lepetit, L., Nys, E., Rous, P., Tarazi, A. (2008) Bank income structure and risk: An
empirical analysis of European banks, Journal of Banking and Finance, 32, p.1452-
1467.
96. Lichtenstein S., Fischhoff B., Philips L. (1982), Calibration of probabilities: The state
of the art to 1980, in: D. Kahneman, P. Slovic and A. Tversky (eds.) Judgement Under
Uncertainty: Heuristics and Biases, Cambridge University Press, Cambridge.
97. Lozano-Vivas, A., Pastor, J., Hasan, I. (2001) European Bank Performance Beyond
Country Borders: What Really Matters?, European Finance Review, 5, p.141-165.
98. Lundeberg M., Fox P., Punccohar J. (1994), Highly Confident but Wrong: Gender
Differences and Similarities in Confidence Judgments, “Journal of Educational
Psychology”, Vol. 86, No. 1, p. 114-121.
99. Łuniewska, M., Tarczyński, W. (2006) Metody wielowymiarowej analizy
porównawczej na rynku kapitałowym, eds. Wydawnictwo Naukowe PWN, Warszawa.
100. Maciejovsky B., Kirchler E. (2002), Simultaneous over- and underconfidence:
Evidence from experimental asset markets, “Journal of Risk and Uncertainty”, Vol.
25, No. 1, p. 65-85.
101. Malmendier U., Tate G. (2005), CEO Overconfidence and Corporate Investment,
“Journal of Finance”, Vol. 60, No. 6, p. 2661-2700.
102. Malmendier U., Tate G. (2008), Who makes acquisitions? CEO Overconfidence and
the Market’s Reaction, Journal of Financial Economics.
103. Manove, M., Padilla, A.J. (1999) Banking (conservatively) with optimists, RAND
Journal of Economics, 30(2), p. 324-350.
104. Manove, M., Padilla, A.J., Pagano, M. (2001) Collateral versus project screening: a
model of lazy banks, RAND Journal of Economics, 32(4), p.726-744.
105. March J., Shapira Z. (1987), Managerial Perspectives on Risk and Risk Taking,
“Management Science”, Vol. 33, No. 11, p. 1404-18.
106.Mishkin, F.S. (2007), The economics of Money, Banking and the Financial Markets,
Pearson, Boston.
107.Morrison, A.D., White, L. (2005), Crises and Capital Requirements in Banking,
“American Economic Review”, 95, p. 1548-1572.
108.Niu, J. (2009) CEO Overconfidence and Bank Risk Taking, Working Paper, Simon
Fraser University.
109.Odean, T. (1998) Are Investors Reluctant to Realize Their Losses?, Journal of
Finance, 53 (5), p.1775-1798.
77
110.Odean T. (1998), Volume, Volatility, Price and Profit When All Traders Are Above
Average, “Journal of Finance”, Vol. 53, No. 6, p. 1887-1934.
111.Odean T. (1999), Do Investors Trade Too Much?, “American Economic Review”,
Vol. 89, No. 5, p. 1279-1298.
112.Oskamp S. (1965), Overconfidence in Case-Study Judgments, “Journal of Consulting
Psychology”, Vol. 29, No. 3, p. 261-265.
113.Pawlowska, M. (2003) Wpływ fuzji i przejęć na efektywność w sektorze banków
komercyjnych w Polsce w latach 1997-2001, Bank i Kredyt, February.
114.Perez, D., Salas, V., Saurina, J. (2006) Earnings and Capital Management in
Alternative Loan Loss Provision Regulatory Regimes, Bank of Spain Working Paper,
0614.
115.Presson P.K., Benassi V.A. (1996), Illusion of Control: A Meta-Analytic Review,
“Journal of Social Behavior and Personality”, Vol. 11, No. 3, p. 493-510.
116.Quagliariello, M. (2007) Banks‘ riskiness over the business cycle: a panel analysis on
Italian intermediaries, Applied Financial Economics, 17, p.119-138.
117.Rajan, R. G. (1994) Why bank credit policies fluctuate – a theory and some evidence,
Quarterly Journal of Economics, May, p. 399-441.
118. Rajan, R.G., Kashyap, A.K., Stein, J.C. (2008) Rethinking Capital Regulation, Paper
prepared for Federal Reserve Bank of Kansas City symposium on “Maintaining
Stability in a Changing Financial System”.
119.Ritter, L., Silber, W., Udell, G. (1999) Principles of Money, Banking, and Financial
Markets, eds. Addison Wesley Longman Inc., New York.
120.Roll R. (1986), The Hubris Hypothesis of Corporate Takeovers, “Journal of
Business”, Vol. 59, No. 2, p. 197-216.
121.Russo J., Schoemaker P. (1992), Managing Overconfidence, “Sloan Management
Review” Vol. 33, No. 2, p. 7-17.
122.Salas, V., Saurina, J. (2002) Credit Risk in Two Institutional Regimes: Spanish
Commercial and Savings Banks, Journal of Financial Services Research, 22(3), p.203-
224.
123.Santos, J.A.C (2001), Bank Capital Regulation in Contemporary Banking Theory: A
Review of the Literature, “Financial Markets, Institutions & Instruments”, 10, p. 41-
84.
124.Schumpeter, J. (1939) Business Cycles, McGraw-Hill, New York.
125.Sharpe, S.A. (1990) Asymmetric Information, Bank Lending and Implicit Contracts:
A Stylized Model of Customer Relationships, Journal of Finance, 45 (4), p.1069-1087.
126. Shefrin H., Statman M. (1985), The Disposition to Sell Winners Too Early and Ride
Losers Too Long: Theory and Evidence, “Journal of Finance”, Vol. 40, No. 3, p. 777-
790.
127. Skala, D. (2008) Overconfidence in Psychology and Finance – an Interdisciplinary
Literature Review, Bank i Kredyt, Vol. 39, No 4, p. 33-50.
128. Skala, D. (2010) Ocena wyników i profile ryzyka banków europejskich – wpływ
czynników behawioralnych; Doctoral thesis, unpublished manuscript, University of
Szczecin.
78
129. Soll J.B. (1996), Determinants of overconfidence and miscalibration: The roles of
random error and ecological structure, “Organizational Behavior and Human
Decision Processes”, Vol. 65, No. 2, p. 117-137.
130. Statman M., Thorley S., Vorkink K. (2003), Investor overconfidence and trading
volume, “AFA 2004 San Diego Meetings Paper”, March,
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=168472
131. Stiglitz, J. E., Weiss, A. (1981) Credit Rationing in Markets with Imperfect
Information, American Economic Review, 71(3), p.393-410.
132. Svenson O. (1981), Are We All Less Risky and More Skilful Than Our Fellow
Drivers?, “Acta Psychologica”, Vol. 47, No. 2, p. 143-48.
133. Tarczyński, W., Mojsiewicz, M. (2001) Zarządzanie ryzykiem, PWE, Warszawa.
134. Taylor S., Brown J.D. (1988), Illusion and Well-Being: A Social Psychological
Perspective on Mental Health, “Psychological Bulletin”, Vol. 103, No. 2, p. 193-210.
135. Thakor, A.V. (1996) Capital Requirements, Monetary Policy, and Aggregate Bank
Lending: Theory and Empirical Evidence, “Journal of Finance”, 51, p. 279-324.
136. VanHoose, D. (2007) Theories of bank behavior under capital regulation, “Journal
of Banking and Finance”, 31, p. 3680-3697.
137. Wall, L.D., Koch, T.W. (2000) Bank Loan-Loss Accounting: A Review of Theoretical
and Empirical Evidence, Economic Review, Federal Reserve Bank of Atlanta, Second
Quarter.
138. Wallsten T.S. (1996), An Analysis of Judgment Research Analyses, “Organizational
Behavior and Human Decision Processes”, Vol. 65, No. 3, p. 220-226.
139. Weinstein N.D. (1980), Unrealistic Optimism About Future Life Events, “Journal of
Personality and Social Psychology”, Vol. 39, No. 5, p. 806-20.
140. Wooldridge, J.M. (2001) Econometric Analysis of Cross-Section and Panel Data,
MIT Press.
141. Wooldridge, J.M. (2002) Introductory Econometrics: A Modern Approach, eds.
South Western College Publishing, 2nd
edition.
142. Zaleska, M. (2008) Współczesna bankowość, Tom 1. (ed.) Difin, Warszawa.