Reporting/Market Focus from the June/July 2007 Edition of the SGS Newsletter

The problem using statistical models to simulate or forecast economic activity, or to simulate or predict default risks, is tied largely to simplifying assumptions that have to be made, and to inaccurate numbers that get plugged into the models. When the limiting factors are not fully appreciated, the results of that modeling can have such a limited relationship to the real world as to be not only useless, but also financially dangerous to those relying on them.
Consider econometric forecasting. Some decades back, the Wharton School developed a wonderful model that simulated the U.S. economy, with something like 30,000 equations that had to be solved simultaneously. The model’s value was in answering "What if?" types of questions. For example, if taxes were raised, what would that do to the economy next year? While the model could provide an indication of net impact, it could not predict the level of economic activity with any accuracy.
Nonetheless, this new tool soon was called into service as a forecasting model — which it was not — going well beyond the "What if?" If one wanted to forecast interest rate levels a year out, one first had to forecast what inflation would be the next year. If the inflation assumption were off, so also would be the interest rate forecast. To the extent the model was accurate, its forecasts could only be as good as the underlying assumptions made by the forecaster. Further, if the forecaster knew what the next year’s inflation rate was going to be, he or she really did not need the model to figure out what would happen to interest rates.
The lack of forecast accuracy, however, did not dissuade the U.S. Government, the Federal Reserve and Corporate America from adopting such models for economic and business forecasting. The use of similar modeling today is so widespread that it partially explains the closeness of some consensus forecasts (often wrong) that come out of U.S. industry.
There have been ways of forecasting key data with reasonable accuracy, using single-equation formulas to predict a certain factor, where the formulas used leading indicators with enough lead-team to avoid having to rely on underlying assumptions. The growing problem with such models, however, has been the rapid deterioration in recent years of the quality of government statistics that would be used in such modeling.
Then there are simplifying assumptions. Financial models that do not allow for or otherwise assume away the possibilities of a major economic downturn, massive sell-offs in equities and the U.S. dollar, or significant spikes in long-term interest rates, have potential vulnerability to downside surprises, at such time as the economic and financial environments eventually unfold in a not so happy manner.