Despite the fact that the world is uncertain, this is often overlooked in forecasting activities which most often rely on a single case or a small set of scenarios. This blog briefly highlights some of the key practical benefits (both quantitative and qualitative) of using uncertainty approaches to forecasting.
There are many qualitative and process benefits to conserving uncertainty; at the simplest level, by doing, once is accepting that one is attempting to capture reality (as far as practical); this simple recognition has implications and knock-on effects for almost all aspects of the qualitative process, for any other approach is based on a false premise. This clarity of thought will create more process transparency, and help to expose false or biased assumptions.
From a quantitative perspective, the key benefits include:
- Capturing the effect of diversification, for example
- An under-spend in some projects can compensate for over-spend in others
- That the sum effect of using assumptions that are individually “slightly biased” (as in traditional analysis) are highly biased in aggregate or at the portfolio level
- Reflecting the impact of asymmetry
- In a static model, Success/fail processes cannot really be accounted for (at best their average may be able to be captured, but only in simple models).
- Uncertainty ranges are often inherently non-symmetric.
- Where there are non-linear (e.g. IF/MAX) elements, scenarios will often fail to capture the true frequency of their occurrence
- Reduce Transparency/Biases
- Base cases are often “arbitrary” unless possible range is made explicit.
- Base case may be optimistic (or pessimistic) e.g. in static model, assumption of single outcome for success/fail processes.
- Scenarios defined around a biased base case are still biased.
- Probability assessments for scenarios may be non-intuitive if volatility needs to be captured correctly (e.g. the correct weighted to capture the P10, P50, P90 scenarios for a normal distribution is a 30-40-30 probability weighting for each scenario, if volatility estimates are to be correct); this is known as Swanson’s rule.