This project is with a large utility company
The focus of this exercise will be to review and advise on our approach to portfolio analysis of the production forecasts. At present we use a Monte Carlo-based approach to generate a view of the full year portfolio outcome based on P10 (high case), P50 (most likely case) and P90 (low case) estimates from the assets. Although we consider this to give us a reasonable view of the likely year end portfolio outcome, there are a few areas / issues which raise questions in my mind (activity to be carried out in bold italics):
We have only recently started using this approach (was used sporadically several years ago, then discontinued), so do not have any history on how accurate these forecasts have been. Review and advise on approach. We are reasonably comfortable that we understand the basis of the P50 numbers, far less so with the P10s and P90s as this is a new area and the assets generally do not have rigorous processes in place. Advise on how to get statistical confidence from a portfolio viewpoint, assuming that asset processes are not refined over short to medium term. To compensate for this we run scenarios assuming P10 to be P20 and P90 to be P80 etc.; I am not entirely comfortable how statistically useful this is and how we best use it to inform our analysis. Review and advise.
Whilst we have some history of accuracy of P50 forecasts over time (we are currently rebuilding this history as far as possible), we are not using this to inform our portfolio analysis in any way. I would like to understand how we can use this history to inform our projections. The model we use (using Palisade @Risk) has been developed in house. Would like to get some validation / a view on the reliability of this for our purposes.
Coming out of this review would be, I anticipate / hope, the building / refinement of the portfolio forecast model.