Transcript Lecture 13

Updated: 22 May 2007
FINA 522: Project Finance, Risk and Risk
Analysis
Lecture Thirteen
0
WAYS TO REDUCE ERRORS
WHEN USING
MONTE CARLO SIMULATIONS
OR
EVALUATION OF REAL OPTIONS
1
Categories of Errors
1. Model Errors
•
•
•
Is excel spreadsheet done correctly?
Are all input variables in table of parameter?
Are all linkages to different parts of analysis done
correctly?
2. Assumptions Errors
•
•
•
•
Outcome of the analysis is only as good as
information used.
Use experience of experts whenever possible.
Use historical data if available.
If possible do back casting to see if model can
track what has happened previously.
2
Categories of Errors (Cont.)
3. Analytical Errors
•
Is model designed and simulation carried out in a
way that is consistent with correct approach to
analysis?
4. Interpretation Errors
•
Analysis maybe done correctly, but the output is
not correctly interpreted when providing results to
decision makers.
3
Questions that Management should
ask the Analyst
1. How are distributions obtained?
•
•
•
Distribution of all variables should not be the same.
Check distribution of historical data?
Check to see impact of changing parameters of distribution
of mean and standard deviation.
2. How sensitive are the distributional
assumptions?
•
•
Simulate the variables that actually have an impact on what
you are trying to estimate.
Tornado and sensitivity charts are useful tools in selecting
such variables.
4
Questions that Management should
ask the Analyst (Cont.)
3. What are critical success factors?
•
•
How sensitive are outcomes to the input variables and
assumptions?
Most sensitive variables should receive the most
attention.
Are the assumptions related and have their relationships
been considered?
•
•
•
Of ten variables do not operate independently.
For example, quantity demanded is negatively related to
prices.
Relationship, correlations and causations need to be
modeled appropriately.
5
Questions that Management should ask the Analyst
(Cont.)
4. Considered truncation
•
Use of triangular distribution
– Will give zero probability of the worst and best case
recurring.
•
It might be the case that the probability of the worst case
occurring is very much higher than zero.
5
10
6
Questions that Management should
ask the Analyst (Cont.)
5. How wide are the forecast results?
•
•
Results should fall within a reasonable interval.
If results are too wide, then perhaps model is not built
right or distribution of variables is not correct.
6. What are the end points and extreme values?
•
Should not suppress the extreme value estimates as
these may destroy project.
7
Questions that Management should
ask the Analyst (Cont.)
7. Are there breaks in the relationships over
time?
•
•
Historical data might not be relevant to project the
future because of changes that occurred, e.g.
change in technology.
Need to get input information from experts.
8. Do the results fall within expected
economic conditions?
•
•
•
Important to have theoretical framework.
Are relationships that come out of the data valid?
Avoid data mining.
8
Questions that Management should
ask the Analyst (Cont.)
9. What are the values at risk?
•
•
•
Monte Carlo analysis is looking at uncertainty. It tells us the
uncertainty around the point estimate.
Risk analysis has not been done yet.
One needs to know the impact that such variability has on the
losses that can arise and the cost of such outcomes.
10. How do the assumption compare to historical data
and knowledge?
•
•
Suspect distributional assumptions should be tested by
back casting.
Do assumptions of distribution fall outside historical
range? Are there valid reasons for assuming this?
9
Questions that Management should
ask the Analyst (Cont.)
11. How do results compare against
traditional analysis?
•
•
How close do the expected values of Monte
Carlo analysis compare to the base case
estimates?
If they are very different, then why are they
different?
10
Questions that Management should
ask the Analyst (Cont.)
12. Do the statistics confirm the results?
•
•
•
•
•
Crystal ball generates a series of statistics that need to be
considered?
Are the results skewed?
Skeweness of normal distribution is = 0
Check statistics on skeweness
Kurtosis for normal distribution = 3.0. A high value of
Kurtosis indicates that there is a higher probability of
recurrence in the tails of the distribution.
13. Are correct methodologies applied?
•
•
•
What was the base for selecting a particular distribution?
Bootstrap versus distribution.
Sensitivity verse tornado chats.
11
Warning Signs in Real Options
Analysis
1. Do not let real options simply over inflate
the value of a project.
•
Do not apply real options on everything, just
to those projects that have strategic options.
12
• Real options have value only if
• Traditional financial analysis can be performed and
models built
• Uncertainty exists
• The same uncertainty drives value
• Management of project has strategic options or
flexibility to either take advantage of these
uncertainties or hedge them
• Management of project has to be credible in
executing the relevant strategic options when it
becomes optional to do so.
13