In this episode we walk through an extreme value approach to generating multiple high impact-high uncertainty outcomes. This is the region where the rare, but often devastating, events occur, for example the landing of a Cat 5 hurricane or typhoon, the eruption of a volcano with flows on the lee-ward side of the mountain, the demise of a competitor presenting an opportunity to fill unmet demand, any kind of hard to pin down a precise supply chain disruption. The model takes very sparse data, applies several unobserved possible hypotheses with an observational model (Generalized Pareto Distribution - GPD with shape and scale parameters) which mashes together data and hypotheses and prior beliefs about hypotheses into a posterior probability ranking of hypotheses, oh my! We then use the ranking to generate 10,000 possible outcomes, here of workers compensation claims. Exploratory Data Analysis, some dynamic thinking, concept maps of our thinking, a plot, and some user opportunities to change key assumptions, all provide grist for our forecasting mill in the face of extreme circumstances. We use all of the spreadsheet tools at our disposal including some light VBA to automate the 10000 recalculations of the spreadsheet. We then play with the model, trying to break it, trying to have it answer our questions, checking assumptions, employing outside-in challenges, realizing we are not in Kansas anymore Toto.
23 окт 2024
