"Algorithm aversion" and managed futures

No one believes in back-testing. No one weighs long-term track records. No one invests in new unproven models from new managers. These may be exaggerations, but it may not be far off track and represents an ongoing problem with managed futures and other systematic approaches to money management when they present results to investors

Investors seem to prefer the discretionary trader or forecaster over the systematic model or algorithm. There is always reason to be skeptical of models and a requirement for strong due diligence of a investment process. It applies to any manager or investment process, but may be even more applicable for complex strategies that may not lend itself to simple story-telling.  There are clear issues with data-snooping which require attention, but the avoidance to algorithms seems to run deeper than just healthy due diligence. I could not always put my finger on the skepticism. Perhaps it was my own failing at being able to articulate what a model does or perhaps it was a problem with the investor who did not have strong quantitative skills. Now there at least seems to be some better identification of the problem.

In a recently published paper by three Penn professors, we have a new term for why investors may avoid systematic strategies, algorithm aversion. In their paper, "Algorithm Aversion: People erroneously avoid algorithms after they err", the authors provide evidence concerning how individuals choose between human forecasters and statistical algorithms. There is a strong bias against the models. This bias exists especially after individuals see the models perform. They are less forgiving of any errors made by the model even when the models in general do better. They will quickly lose confidence in a model after it makes a mistake or after they see it work. They are more willing to accept or keep their confidence in the human forecaster. After they see an algorithm perform, they will go with the poorer forecasting human. The authors do not provide strong reasons for this aversion, but the test subjects are unforgiving with model performance versus humans.

Think about the avoidance of systematic managers in hedge fund land. By the conclusions form this research, investors will be more forgiving for a poorer performing discretionary manager. First chance an investor sees an error, he will redeem even if the long-term performance beats the human manager. If he see the algorithm perform, that is, he sees the track record, he will avoid. I am not saying this happen in all cases, but this research suggests that the bar is set higher for systematic managers than discretionary managers and it may hurt the long-term performance of an investor's portfolio. Investors should be aware of this bias before they make a portfolio decision.