When Performance is Irrelevant (Part 4)

The motivation for this blog mini-series stems from an article on “robo-advisers” by Jaclyn N. McClellan in the Oct 2016 AAII Journal. With regard to performance comparison, she writes:

     > A major question many investors ask is, “How
     > does the performance of the robo-advisers
     > compare to that of traditional advisers?”
     > This is not a question that is easy to answer.
     >
     > Most of the robo advisory services that post
     > performance online display backtested or model-
     > based results…

That may be inaccurate because live-trading execution can significantly differ from backtesting (or based on models). They would need to present the research methodology (examples given here and here), which is usually not done.

     > Each robo-adviser has different inception
     > dates, and some don’t disclose those dates.

This is also true with regard to mutual funds, which makes it very difficult to pick up one prospectus and compare with another. Not giving the inception date seems absurd.

     > Prospective clients have to search for the
     > performance disclosures just to read the fine
     > print…

This is consistent with the mutual fund prospectus. None of this is straight-up, transparent disclosure although I’m sure their compliance departments would claim to be meeting regulatory standards.

     > Although the lack of sound performance figures
     > may seem disheartening, displaying truly
     > representative returns is difficult because
     > many investors have customized portfolios—
     > fees, allocations and rebalancing intervals can all
     > be different on an account-by-account basis.
     > Clients open accounts at different times, so
     > even if the respective risk/reward profile
     > categories don’t change, the starting value will.

The countless number of potential permutations of an investment account is another reason why it seems inappropriate to use a single historical record to measure performance.

An alternative to showing a single number for 1-year (5-year) performance, for example, would be to show a distribution of rolling 1-year (5-year) performance records. A fund that has been around for two (10) years has roughly 252 (1,260) rolling 1-year (5-year) periods that may be sampled. I could then look at the mean, standard deviation, and percentiles to make good sense out of these numbers, which constitute a much more robust sample size.

I will continue next time.