Given smart beta’s popularity among investors and asset managers, people ask me about this strategy all the time. The number one question I receive is: What should an investor expect regarding returns generated from smart beta strategies (e.g. value, momentum, carry, low volatility, etc.), and how does it compare to the historical backtested results that normally accompany smart beta strategies?
Unfortunately, like the rest of you, I just don’t know. I have no crystal ball. We’d have to wait as much as 25 years to gain insights on the true return properties of the various smart betas available today. Robust backtests and theory are helpful for understanding the return properties of smart beta strategies, but they can’t completely mitigate the risks associated with unintended “data mining” - continuously searching for return predictability in a fixed historical record that inevitably mischaracterizes randomness as a systematic factor deserving of a standalone, prospective risk premium.
Even if data mining weren’t an issue, historical backtests may provide an inaccurate depiction of the future if the factor (e.g. value, momentum, carry, low volatility, etc.) in question is now publicly known by the investment community with many products available to access it. Profitable strategies that become public attract capital and that additional capital tends to push prices back to fair value. In other words, part of the observed historical risk premium reported in the backtest might already be arbitraged away and, thus, not available to investors looking to make an allocation today.
So how do we assess the impact of these critical concepts – data mining and arbitrage? Is waiting 25 years really the only answer? In order to account for the detrimental effects of data mining and arbitrage, should investors cut the historical backtested risk premium in half and use that as their forward-looking return estimate? Or is this approach too arbitrary? Sounds arbitrary to me.
While there’s no silver bullet here, I can help move the debate forward if we can answer the following questions: Is there any way to credibly run an ~25 year out-of-sample test on the most popular smart betas from the early 1990s by going back in an unbiased, documented time capsule? What were the popular smart betas ~25 years ago? How were the smart betas constructed? How do the last 25 years of out-of-sample results, i.e. smart beta returns earned by hypothetical investors from the early 1990s through 2015, compare to the backtested results that would have been available and marketed to investors in the early 1990s? Did the out-of-sample results decrease as a result of data mining and/or arbitrage? If so, by how much? What are the implications for today’s large stable of backtested only smart beta products and their prospective investors? Should today’s investors expect a similar decline (if any) over the next 25 years? I think I figured out a way to credibly implement this. Stay tuned.
A Credible, Unbiased and Documented Backtest
For this paper to provide maximum value, I need to convince you that my process for selecting/constructing the smart betas, choice of in-sample time period (backtest), and choice of out-of-sample time period (actual investor experience) is not arbitrary. In other words, I need to convince you that my approach is credible, unbiased, and documented…to the best of my ability.
My dissertation advisor and 2013 Nobel Prize recipient, Eugene Fama, plays a central role in my approach. How? Eugene Fama is unquestionably one of the leading researchers of equity security selection (i.e. smart beta) strategies. In December 1991, Eugene Fama published his Efficient Capital Markets Sequel paper (i.e. “Efficient Capital Markets: II”), which is a famous, well-respected survey paper that covered relevant topics for the market efficiency debate. In the security selection (i.e. smart beta) section of the paper, he identifies and focuses on 4 factors above and beyond general market beta that cross-sectionally predict returns:
• Equity market capitalization (“ME”, Banz 1981): small stocks outperform large stocks
• Earnings-to-price (“EP”, Basu 1977): high EP stocks outperform low EP stocks
• Debt-to-equity (“DE”, Bhandari 1988): high DE stocks outperform low DE stocks
• Book-to-market (“BEME”, Fama French 1992): high BEME stocks outperform low BEME stocks
Out of all the potential published factors (smart betas) in the historical published record at the time, Eugene Fama, the Michael Jordan of equity security selection research, chose to highlight these four as the most important, which can be thought of as one size factor and three versions of a value factor.