Monday, August 24, 2020

Rebalance timing for factor strategies - Luck or skill?

If I invest in a standard well-defined factor strategy such as value, size, momentum, quality, and low volatility, there may be the belief that that some of the construction details don't much matter. There will be a cluster around the "true" returns from a long-only factor strategy, but the construction return differences are small. There is also the belief that some of the rules will matter more than others albeit again the differences are small and may be associated with noise. For example, exclusion rules will have some impact, but other rules like when and how often a factor strategy is rebalanced will have little impact. Wrong.

A recent paper by the high quality researchers led by Corey Hoffstein at NewFound Research suggests that rebalancing is important, see "Rebalance Timing Luck: The Dumb (Timing) Luck of Smart Beta". There is a significant impact from when you rebalance and how often it is conducted. I was shocked by the cumulative effect of random rebalancing. The details matter.

Of course, I should not be that surprised since anyone who has worked closely with data will find that simple changes in rules like even mundane features of start and end dates for back-testing will matter. The art of any model-building is finding features that are stable and consistent while generating strong returns. The desire to find portfolio construction skill not luck. 

Any construction rules will create return differences so the questions are whether it matters a lot and whether there is something in these rule differences that can be exploited. Here is where the research gets interesting and also murky.  The rebalance timing over any short period can be great, and over time the cumulative difference between the best and worst rebalance timing rule seems to be large. If that is the case, then a long-short portfolio between timing periods will lead to significant return differences. Testing for these differences across strategies is less promising. They are generally not statistically significant. This work addresses some very important questions that need to be further researched to determine their value over the long-run. 

No comments: