Adverse Excursions Part 5: Hypotheses, 09/21/07
Over the course of the decline, one of the things I did while looking at adverse excursions was to create from them a very simple risk ranking. What's interesting about risk-ranking patterns or models is that the ranking also provides a confidence indicator for a forecasting model generated from a particular pattern mash-up.
For example, let's say we're trying to forecast next week. We know these things:
- AEven in that simple example there are 4! or 24 possible combinations to test to try to find the right mix in a forecasting model. And, as we all know, coming up with a forecast out of all those combinations can be more art than science.
- B
- C
- D
With the risk rankings, I'm starting to observe how adding a new pattern to the model affects the risk of the pattern mash-up as a whole: if there's no change when I add a piece, I don't have any decisions to make - more science and less art, but if I add a pattern and the risk mix suddenly changes then I know I need to burrow down into that pattern and probably apply a bit more art to the forecast.
Which leads to the following hypotheses -
Hypotheses:
A pattern mash-up w/ a stable risk-reward profile will provide a more reliable forecast
A pattern mash-up w/ a stable risk-reward profile will be less 'risky' because the risk components have a smaller std dev
See also:
Adverse Excursions, 07/26/07
Adverse Excursions, Part 2, 08/02/07
Adverse Excursions Part 3, A Risk Reward Indicator, 08/16/07
Adverse Excursions Part 4, Monthly Return Regimes, 08/25/07
Using 'Or' Instead of 'And', 8/20/05