33928
TRANSPORT NOTES
TRANSPORT ECONOMICS, POLICY AND POVERTY THEMATIC GROUP
THE WORLD BANK, WASHINGTON, DC Transport Note No. TRN-7 January 2005
Notes on the Economic Evaluation of Transport Projects
In response to many requests for help in the application of both conventional cost benefit analysis in transport and addressing of the newer topics of
interest, we have prepared a series of Economic Evaluation Notes that provide guidance on some of issues that have proven more difficult to deal with.
The Economic Evaluation Notes are arranged in three groups. The first group (TRN-6 to TRN--10) provides criteria for selection a particular
evaluation technique or approach; the second (TRN-11 to TRN-17) addresses the selection of values of various inputs to the evaluation, and the third
(TRN-18 to TRN-26) deals with specific problematic issues in economic evaluation. The Notes are preceded by a Framework (TRN-5), that provides
the context within which we use economic evaluation in the transport sector.
The main text of most of the Notes was prepared for the Transport and Urban Development Department (TUDTR) of the World Bank by Peter Mackie,
John Nellthorp and James Laird, at the Institute for Transport Studies (ITS) , University of Leeds, UK (The draft text of Note 21 was prepared for ITS by
I.T. Transport Ltd). TUDTR staff have made a few changes to the draft Notes as prepared by ITS.
The Notes will be revised periodically and we welcome comments on what changes become necessary. Suggestions for additional Notes or for changes or
additions to existing Notes should be sent to rcarruthers@worldbank.org
RISK & UNCERTAINTY ANALYSIS
One statement that can confidently be made about any transport project is that the costs and benefits
are uncertain. An important question is `how uncertain'? By analysing the risk and uncertainty which
surrounds the project the probability of a poor outcome can be assessed. In addition, it is often
possible to identify ways in which the project can be made more robust, and to ensure that the risks
that remain are well managed. Risk and uncertainty analysis is therefore a standard component in the
project reporting requirements for World Bank projects. Risk and uncertainty analysis also features
other in World Bank tools, such as the RED model (see Note Low Volume Rural Roads).
This note reviews the general principles of risk and uncertainty analysis in transport (Section 1); and
outlines the three principal methods which may be used sensitivity analysis (Section 2), switching
values (Section 3) and Monte Carlo simulation (Section 4). Implications for risk management are
given in Section 5 and Further Reading is referenced at the end.
GENERAL PRINCIPLES OF RISK AND UNCERTAINTY ANALYSIS IN TRANSPORT
At the start, we can differentiate between risk and uncertainty. Risk is the situation where there is a
set of possible outcomes from the project, and the probability of each outcome is known (as in Figure
1(a)). Uncertainty is the situation where there is a set of possible outcomes, but the probability of
each one is not known (as in Figure 1(b)). In Figure 1(b), only the expected (or mean) outcome is
marked - estimating this is often the focal point of project appraisal in practice, however without some
understanding of the likelihood of the predicted outcome, it is worth asking: "how much confidence
can be placed in the appraisal results"?
Transport Note No. TRN-7
Pr (NPV) Pr (NPV)
0.2 ? x
probability
density function
0 min E(NPV) max 0 E(NPV)
=$40 =$128 =$216 =$128
million million million million
(a) Risk (b) Uncertainty
Figure 1. Risk vs. Uncertainty
Consider the following example. When we say that the Internal Rate of Return (IRR) of a project is
22% (see Note 1: When and how to use NPV, IRR and Adjusted IRR), what we usually mean is that
the expected outcome (strictly the `mean' IRR) is 22%. Around this is a distribution of possible
outcomes. Figure 2 illustrates this, and highlights the 12% Internal Rate of Return which is the usual
minimum criterion for acceptability of World Bank projects (see Note 1: When and how to use NPV,
IRR and Adjusted IRR). In this case there is a chance that the IRR may turn out to be less than 12%,
but that chance is very small. This type of information is very useful to decision-makers.
Probability
probability
density
function, f(IRR)
0 min E(IRR) max IRR
12%
=22%
Figure 2. A Probability Distribution on the Internal Rate of Return (IRR)
Once a project is deemed acceptable, the World Bank's usual decision criterion for projects is then to
maximise the expected NPV (see Handbook on Economic Analysis of Investment Operations (World
Bank, 1998) [i]). It is worth noting that this is not the only possible decision criterion:
a cautious, or `risk averse' approach, is to maximise the minimum outcome (called the `maxi-min'
strategy, for obvious reasons). This would favour projects which are a `sure bet', having only a
small downside risk.
more complex strategies, seeking to balance the risk across a portfolio of investment or to
minimise the `regrets' the consequences of a lower-than expected outcome are discussed in a
range of source books, including Pearce and Nash (1981) [ii].
In order to make progress with risk analysis in practice, we need to know how the probability
distribution in Figure 2 is arrived at? The IRR is determined by the flows of costs and benefits over
time, so any riskiness in particular cost or benefit items can be traced through into the IRR. In turn,
the risk in each cost or benefit item usually depends on key underlying factors. Box 1 sets out what
these factors may be.
Page 2 of 7
Transport Note No. TRN-7
Box 1. Sources of Risk in Transport Appraisal
Construction cost risk
It is very common for the costs of major construction projects to be underestimated in
appraisal. Underlying reasons for this include:
- engineering risks associated particularly with uncertain ground conditions (geology) and high
technology structures, equipment and methods;
- variable input prices, especially energy, labour and imported materials;
- the complexity of the project management function.
Whether these risks are directly a concern for the purchaser depends on the allocation of risk under
the contract. However, even when the risk is fully transferred to the contractor, a premium will
usually be charged to bear this risk, so the purchaser continues to face a cost. Figure 2 shows a
typical probability distribution for construction cost most of the risk is downside risk, so the
distribution is not symmetrical. So in this case, the expected (mean) value exceeds the most likely
value ($120M).
Prob.
90 120 270 $M
Figure: A Probability Distribution on Construction Cost
Operating and maintenance cost risks
These are fuelled by some of the same sources as construction risk, especially rising
real labour costs, energy prices and exchange rate risk on imports.
Demand risks `traffic risk' and `revenue risk'
Once a project is complete, the benefits derived from it depend crucially on the forecast
user demand being realised. Demand is vulnerable, however, to:
- economic shocks, including fuel price shocks and economic boom or recession;
- changing demographics;
- shifting preferences and growth of competing facilities eg. roads in competition
with rail;
- political intervention;
- random error in the forecasts.
The benefits, in the form of consumers' surplus, typically decrease as demand
decreases. Revenue decreases too, which is an issue for the financial sustainability of
projects.
Other risks
Mackie and Preston (1998) [iii] identified 21 sources of error and bias in transport
appraisal. Many of these fall under the general heading of `appraisal optimism',
suggesting that downside risks outweigh upside risks in most real situations. HM
Treasury (2002), Section 4.11 gives cautionary guidance on this.
Page 3 of 7
Transport Note No. TRN-7
SENSITIVITY ANALYSIS `WHAT IF' SCENARIOS
The simplest form of sensitivity analysis involves the creation of `what if' scenarios to reflect the
principal risks surrounding the project. Suppose, for example, that a bridge project is believed to be
subject to demand risk, because much of the forecast traffic is generated traffic (see Note 6:
Treatment of Induced Traffic and Note 7: Demand Forecasting Errors) which depends on changing
patterns of trade and residential location. The analysts responsible for forecasting demand advise that
they can only be confident in their forecasts to within +/ 30% of the expected value.
To assess the significance of this risk, the appraiser re-runs the IRR calculation for demand at +30%
and 30%, and presents these as `what if' scenarios, in Table 1.
Table 1. Sensitivity of IRR to Demand Risk for a Bridge
Scenario Demand in Year 2005, IRR
vehicle flow per day
`Low' demand 14,000 13.5%
Expected demand 20,000 16.0%
`High' demand 26,000 19.0%
Since the lowest IRR is above the 12% acceptable rate of return, this gives some confidence that the
project is robust to the demand risk identified.
SWITCHING VALUES
The World Bank's preferred approach to sensitivity analysis is to calculate switching values. These
are the values of the `risky' variables at which the IRR of the project equals the discount rate, and the
NPV=0. Table 2 gives an example for the bridge project.
Table 2. Switching Values for a Bridge
Variable Switching value
Demand (vehicle flow in 2005) -36%
Construction cost +44%
Maintenance cost (in 2005) +70%
Switching values prompt the reader to consider how (un)likely the required switch is, which gives a
handle on the robustness of the project to each variable. However, in common with basic `what if'
sensitivity analysis, switching values have one great weakness: it becomes complex to understand the
risks when several variables are acting simultaneously, rather than one at a time. To explore this,
Monte Carlo simulation provides a higher level of analysis.
MONTE CARLO SIMULATION
A leading form of `quantitative risk analysis' Vose (1996) [iv], Monte Carlo simulation produces a single
probability distribution for IRR (or NPV), based on the risk profiles for all the relevant `risky' variables.
The method works in the following way:
a) the analyst is required to define a probability distribution for each variable - demand, construction
cost, operating cost and so on. This is not as difficult as it sounds (see below).
b) the Monte Carlo procedure then samples randomly from each of the different distributions and
calculates the IRR (or NPV), many times over. By taking a very large number of samples from
each distribution, the sampling distribution is made to approximate closely the theoretical
distribution (Figure 3).
Page 4 of 7
Transport Note No. TRN-7
Theoretical distribution Sampling distribution
Figure 3. The Law of Large Numbers in Application
c) the outcome is a distribution in terms of IRR (or NPV), such as that shown in Figure 1. The more
samples are taken, the more stable the distribution becomes.
We said in (a) that Monte Carlo simulation requires information about the probability distributions for
each `risky' variable. This is more straightforward than it sounds. For example, based on our
understanding of the variable in question we might choose a `shape' for the probability distribution
from the set shown in Figure 4.
Prob. Prob.
Uniform Triangular
0 $3M $5M PVB 0 90 120 270 x
Prob. Prob.
Normal Complex
0 x' E(x) x'' x 0 4.5 6 9 18 20
Figure 4. Alternative `Shapes' for Probability Distributions
For example, we might adopt a triangular distribution for construction cost. Then by specifying the
maximum cost, the minimum cost and the `modal' cost (120 in the Figure), the distribution is fully
defined. Alternatively, suppose we judge that demand is normally distributed around the expected
flow E(x). We can then fix the distribution by setting the 95% confidence limits on the distribution x'
and x''.
The simulation itself is carried out by specialised computer software, however, this software is readily
available as an add-in to conventional spreadsheet applications and reasonably easy to use. The
results will enable the appraiser to report not only the mean value of the project IRR, but also the
distribution of the IRR (as in Figure 1) and the cumulative distribution which is useful for representing
the probability of meeting the desired rate of return (Figure 5). In this example, there is a 79%
probability that the IRR will exceed 12%.
Page 5 of 7
Transport Note No. TRN-7
Prob.
1.0
0.21
0 12% 39%
Figure 5. Cumulative Distribution on IRR
Further advice on Monte Carlo simulation is available in the Handbook on Economic Analysis of
Investment Operations (World Bank, 1998) [i] and in the many textbooks on the market, such as
Vose [iv].
An example of the application of this form of risk analysis to a toll road project in Colombia is shown in
Box 1
Box 1. Risk Assessment in the El Vino-Tobiagrande-Salgar Toll Road Project
This project included the construction of a new 68.4 km highway from Tobiagrande to
Salgar, including 19.5 kms of tunnels and 2.9 kms of bridges. This would significantly
improve communications between the two largest population centers of the Colombia
Bogota and Medellin. This raised significant problems in assessing not only the likely
construction costs but also the rates of traffic growth and newly generated traffic
associated with such a major improvement to the infrastructure of the country. To
handle this, the economic evaluation included a Monte Carlo simulation of the effect of
the estimated risk profiles for these three elements.
In addition, risk analysis was performed to test the effect on the economic feasibility of
the project with probabilistic changes in the assumptions related to traffic growth,
construction and maintenance costs, and generated traffic. Traffic volume was varied
according to major deviations in economic growth during the analysis period, assuming a
beta distribution of economic (and consequently, traffic) growth rates between -0.2 and
1.5, with a median of 0.75. Generated traffic also is varied assuming a beta distribution
with changes in the percentage of generated traffic from 0 to 1.2 of the estimated values
and a median of about 0.9. Construction and maintenance costs were varied following a
triagular distribution with a maximum of 1.7 and a minimum of 0.95 of the estimated
costs. A Monte-Carlo simulation of these risks yields a distribution of economic rates of
return with a probability of 25 percent of being lower than 12 percent in the case of not
considering the passenger-time benefits and of about 11 percent in the case of including
those benefits.
Colombia Toll Road Concession Project, Project Appraisal Report, Report 17986, June
1986
(http://imagebank.worldbank.org/servlet/WDSContentServer/IW3P/IB/1999/06/03/0000
09265_3980709144721/Rendered/PDF/multi_page.pdf)
Page 6 of 7
Transport Note No. TRN-7
RISK MANAGEMENT
A first step towards effective risk management is risk registration, which includes:
recognition that a particular set of risks are relevant in planning and implementating the project;
identification of these risks, and at a very basic level what they are linked to (eg. construction cost
of high-technology bridges on a road project; toll road revenues linked to demand growth, in
which future income growth and car ownership are key factors; etc.)
Registering risks allows all parties within a project to remain aware of risks, and to monitor
development of known risks as a project progresses. The risk register might contain quantitative
information about risk if any is available and even a description of the type of probability
distribution it is believed to follow.
More pro-actively, the project planner can think about prevention, control and transfer of risks.
Prevention may be possible by, for example, choosing tried-and-tested solutions in place of riskier
innovative ones. Another measure which can play a major role in preventing risk is a comprehensive
maintenance programme, which is both designed and implemented to ensure that both level of service
delivered by transport assets and the costs of operation remain in line with the project plan (see Note
8: Treatment of Maintenance). Risk control may be exercised by setting up monitoring systems
which can alert project managers to emerging problems. Risk transfer can be used at the contract
drafting stage to balance risk and reward for each partner in the contract including the public sector
and private partners [v].
FURTHER READING
[i] The World Bank (1998), Handbook on Economic Analysis of Investment Operations, Chapter 12.
The World Bank, Washington, DC. [Available online at:
http://www.worldbank.org/education/economicsed/project/projwork/Handbook.pdf]
[ii] Pearce DW and Nash CA (1981), The Social Appraisal of Projects: a text in cost-benefit analysis.
Macmillan, London.
[iii] Mackie PJ and Preston JM (1998), `Twenty-one sources of error and bias in transport project
appraisal', Transport Policy 5(1), pp1-7.
[iv] Vose (1996), Quantitative Risk Analysis. Wiley, New York.
[v] HM Treasury (2001), Management of risk a strategic overview (`the Orange book'). The
Stationery Office, London.
Page 7 of 7