Supposedly, one of the reasons so many statistical analysis have been conducted on financial markets is because that’s where the data is.
Millions of finance students have learned how to apply financial models using historical data from the real world. However the problems they solved were homework problems, not real world problems. Real world problems don’t have known inputs, and part of the trick at solving them is to figure out what the inputs could be, even if they’re different from the past. Another aspect of real world problems is you might not know for a while whether your answer is correct.
One of the underlying assumptions in both Markowitz’s Modern Portfolio Theory (MPT) and the Sharpe-Lintner Capital Asset Pricing Model (CAPM) is that we know what the expected returns and asset covariances are. In the real world we do not. Conveniently, since so much data has long been available, it has been possible to analyze MPT, CAPM, and other models using historical data. (And to solve a lot of homework problems!)
Yet the growing prevalence of historical data has led us to become lazy in two respects: first, by a tendency to populate inputs (let’s say in an asset allocation optimizer) with recent results, and second, by using the present tense when talking about the recent past. “Bonds are beating stocks,” “Value is beating growth,” “that fund/stock/asset class is not doing well.” It’s as if we’ve seen too many simple moving averages in our lives and we’ve grown habituated to thinking about the present and near future as some extrapolation of recent trends.
Solving real world problems typically doesn’t entail a single answer, but identifying a probability distribution of possible scenarios and a choice that is congruent with the client’s objectives and constraints. It will often lead to the same choice that would have been made by simply relying on recent data, which is why it’s so easy to fall captive to historical data.