Are these optimized portfolios over-fitted to historical data rendering them useless for predicting the future?
Safe Answer: Yes. Nuanced Answer: Over-fitting to historical data is a real danger to watch out for, however we think some long standing truths can be found by examining history. The optimization methodology is based on the theory that the worst market crises from the past will repeat again in different ways. Instead of assuming that asset classes have low correlation as in normal markets, the optimizer basically just looks at circumstances in history where the correlations among asset classes rose in the worst possible way. When a portfolio is generated it is stressed against market data from years like 1973, 1974, 1981, 1990, 1994, 1998, 2001, and yes 2008. 2008 was a rough market, but sometimes the most challenging markets for the optimizer are found earlier in the dataset. A portfolio that performs well in earlier crises tend to behave well in subsequent panics when compared to allocations that blew up in previous panics.
Related Questions
- Does the access date capability required by 520.10(b) only apply to historical data or does it also apply to future dates?
- Are these optimized portfolios over-fitted to historical data rendering them useless for predicting the future?
- Can the program retain significant amounts of actual historical and forecasted data?
