Important Questions to Address
- In many instances both costs and benefits are driven by a few critical factors. Identifying these factors – that can be called “drivers” - is a central element in performing a good sensitivity analysis. The assumptions behind the future performance of these drivers need to be critically assessed and the sensitivity analysis can illuminate at which points in these assumptions the project is no longer worth pursuing. These values are typically called “switching values¨.
- In a sensitivity analysis, it is important to focus on “what if” situations, particularly in adverse or more stringent and acid scenarios than those assumed under a Base Case scenario. This focus will allow the analysis to illuminate those values in critical variables and assumptions that might “switch” the recommendation from Go to No Go (or vice-versa) or to reformulate project design and / or components in order to make the project viable. As there are many sources of uncertainty and numerous risks, it can be very useful to identify the key sources of uncertainty and base the analysis on those. Sometimes having too much information does not provide guidance as to the viability of the project under different situations.
- The sensitivity analysis, which is really a risk assessment, should be coherent with the Risk Matrix and other risks discussed in the project proposal.
- Many projects face significant start and implementation delays. If warranted, a simulation should be undertaken delaying start and implementation dates.
- In a CBA derived from impact evaluations – sensitivity can be done by ranking alternatives by their relative net benefit indicators (NPV, ERR or C/B) using the point estimate of their impact, and then re-compute this ratio using the lower and upper bounds of the impact estimate.
A Base Case reflects the most plausible estimates for unknown quantities and prices and should reflect the average expectation on their future behavior, particularly of cost and benefit drivers. This average expected behavior should be empirically based. The more useful sensitivity analysis are those that present few explicit and reasoned changes on the assumptions of the most critical values of the drivers of the dominant benefits and costs, many times in a “best case, worse case” situation.
- Sensitivity analysis can be “one way” or “multi-way”.
- One way sensitivity analysis allows for the variation of one critical variable at a time, holding everything else constant. Some authors call this “partial sensitivity analysis¨. This kind of analysis is most appropriately applied to situation in which the analyst believes there is one critical driver.
- Multi-way allows for simultaneous variations of more than one variable at a time. In this type of analysis, variable combinations can be combined into best case and worse case scenarios. Some projects lend themselves to simulations, where combinations of different scenarios and their probability of occurrence can be modeled and estimations of critical values can be synthesized in terms of Expectations or Expected Values. This is typically undertaken with decision trees and Monte Carlo Simulations. In these cases, means and variances can indicate project risk profiles.
- In a single factor sensitivity analysis, it is generally assumed that all other factors stay constant. In many projects, this might not be true and the variables might not be independent. In those cases, a discussion on correlation issues is warranted.
Sensitivity analyses can also help, if the feasibility of a project is very sensitive to a particular assumption on the value of a variable that is uncertain, in identifying mitigating actions that should be considered. If there is exceptional uncertainty, the project might have to be redesigned if implemented on a pilot basis.
Sensitivity analysis is most useful if the proposed scenarios and adjustments are empirically based and/or reflect the unique characteristics of the project in hand—how good and reliable is the data. As with the initial assumptions used in an analysis, ideally the scenarios are empirically based.