probability distribution fitting of cost overrun profiles

13
Curtin University is a trademark of Curtin University of Technology CRICOS Provider Code 00301J Professor Peter ED Love Probability Distribution Fitting of Cost Overrun Profiles Royal Institution of Chartered Surveyors Legal Research Symposium, COBRA 2010, September 11th -13 th , Las Vegas, Nevada USA

Upload: pedlove

Post on 29-Nov-2014

1.152 views

Category:

Documents


1 download

DESCRIPTION

 

TRANSCRIPT

Page 1: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Professor Peter ED Love

Probability Distribution Fitting of Cost Overrun Profiles

Royal Institution of Chartered Surveyors Legal Research Symposium, COBRA 2010, September 11th -13th, Las Vegas, Nevada USA

Page 2: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Cost Overruns: A Pervasive Problem

• Unrealistic estimate (optimum bias)

• Changes in scope

• Completion date determined before the project’s scope had been defined

• Inadequate project governance

• Inappropriate procurement method (risk allocation)

• Documentation errors

Page 3: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

The Nemesis of Cost Overruns

• Deceptive actions to ensure projects proceed

• Decision-makers are over optimistic about the outcome of planned actions

Page 4: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

The Fallacy of Cost Overruns

• Where do you measure from?

• Need to distinguish between factors that increase project cost and those affect the accuracy of estimates

• 2004 budget was $420m

• 320% cost overrun ?

Construction on time and budget

Page 5: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Comparing Apples with Oranges

• Reference class forecasting: Projects in a statistical distribution of outcomes from class of reference points

• Projects of the same ilk experience similar degrees of optimism bias and overruns

• Research has shown there is NO significance between cost overruns (% contract value) with project type, procurement etc.

Page 6: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Does Contract Size Matter?

• Larger projects experience smaller overruns (Vice versa)

• Larger projects are better managed and longer completion times provide an opportunity to facilitate cost control

Page 7: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Convenience of the Normal Distribution

• A Normal distribution is symmetric about its mean value and therefore cannot be used to accurately model left or right skewed data.

• The selection of an inappropriate statistical distribution can produce incorrect probabilities, which can adversely affect decision-making and therefore lead to negative outcomes

Page 8: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Research Approach

Probability Density Function, CDF and distribution parameters for continuous distributions were examined using the Maximum Likelihood Estimates

Goodness of Fits Test:Kolmogorov-Smirnov statistic (D):

Anderson-Darling statistic (A2):

Chi-squared statistic (χ2):

)(,

1)(max

1ii

nixF

n

i

n

ixFD

n

iini xFInxInFi

nnA

11)(1)()12(

12

k

i i

ii

E

EO

1

22

Page 9: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Results

• Mean overall cost overrun (n=276) 12.22% of contract value

• Civil engineering projects (n=115) 12.56%

• Building (n=161) 11.76%

• ANOVA revealed no significant differences between types of project, procurement method, and size (contract value)

• The likelihood that a project does not exceed a cost overrun of 12.22% is 60% (P (X < X1) = .60).

Frechet 3P

PDF

CDF

xxxf exp)(

1

x

xf exp)(

Page 10: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Distribution by Contract Value

• <$1M and $51 to $100M (Cauchy)

• $1 to $10M and >$100M (Wakeby)

The quantile function it is an alternative to the probability density or mass function, the cumulative distribution function and the characteristic function.

12

1)(

x

xfPDF =

CDF = 5.0arctan1

)(

xxF

CDF =

FFFx 1111)(

PDF <$1M

% Cost Overrun Cauchy

Percentage of Cost Overrun2520151050-5-10

Prob

abili

ty o

f Cos

t Ove

rrun

0.440.4

0.36

0.32

0.28

0.24

0.20.16

0.12

0.08

0.04

0

PDF $11-$50M

% Cost Overrun Wakeby

Percentage of Cost Overrun150100500-50-100-150

Prob

abili

ty o

f Cos

t Ove

rrun

1

0.8

0.6

0.4

0.2

0

-0.2

PDF $1-$10M

Wakeby distribution is defined by the quantile function (inverse CDF):

PDF > $100M

Page 11: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

For the 101 construction and engineering projects with a contract range of $11 to $50M at Four Parameter Burr Distribution

1

1

1

)(

k

yx

yxk

xF

k

yxxF

11)(

PDF =

CDF =

Page 12: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Contingency

• Most projects will experience cost increases from the determine of budget and contract award

• Design errors, omission and changes (identifiable risks)

• Assumption of 3 to 5% for construction contingency

• In excess of 12.22% cost contingency needed!

Page 13: Probability Distribution Fitting of Cost Overrun Profiles

Curtin University is a trademark of Curtin University of TechnologyCRICOS Provider Code 00301J

Conclusion