Pension funds should not rely on correlations between mean annual return and variance in annual return when deciding how much stocks and bonds to own, according to this article on which Nassim Nicholas Taleb (the Black Swan guy) is the second author. To paraphrase/oversimplify my understanding of the article greatly, the main arguments are that (1) data from the past is not a perfect predictor of the future, and (2) short term volatility is not a good measure of the risk of achieving a long term goal.

In engineering, I hear #1 all the time from people – why don’t we rely on data instead of “modeling” when trying to predict the future? Of course we do both – try to understand the underlying structure of the system we are dealing with, then use data from the past to try to confirm that we got it right, at least for the conditions that prevailed when the data were collected (and assuming the data themselves are reasonably accurate or at least any measurement error is not biased one way or the other), and then use the resulting model of the system to try to predict the future. Conditions in the future may be different than conditions in the past, and that is why we don’t “just rely on data”. If external conditions are different but the underlying structure of the system doesn’t change (much), we can come up with reasonable predictions of the future. The only true test of whether the prediction is right comes from data which will be collected in the future, but is not available today when a decision has to be made. A lot of decisions are really just playing the odds about what might work in the most likely future, or what might work across several different possible futures that collectively are very likely (a “robust” decision). The decision that is best for the single most likely condition and a group of very likely conditions may not be the same one – now you are a gambler trying to decide whether you go for the biggest possible payoff while accepting a larger chance of a loss, or whether you want to maximize your chances of a positive payoff while giving up your shot at a really big payoff. You would think the pension fund would go for the latter.

#2 makes sense to me. Variability in annual returns doesn’t matter much if you are 25 and investing money you plan to need at 65. A pension fund is a little different, because it is essentially immortal but has obligations it has to meet each year.

In the case of investment returns, the approach seems to be almost purely “data-driven” with no real understanding of the underlying system, and this leads to an existential crisis when people try to figure out what asset allocation advice to stake their future on. We understand the real economy to some extent, we think, but we don’t really seem to confidently understand how the real economy and the financial economy are related, especially over shorter time frames. So we are reduced to just describing the data, which might lead to some insights about the system but has limited predictive value. Still, examining the evidence before making a decision seems like a good idea to me. What is the alternative – guessing, wishing, praying?