The lackluster performance of certain momentum and trend-following models in the fourth quarter of 2018 has prompted new discussions by some of the major proponents of momentum investing. In particular, Corey Hoffstein of NewFound Research and Gary Antonacci of Portfolio Management Associates have proposed different models for managing specification risk in momentum investing. This topic is especially relevant to Stillpoint clients because we use similar models to those utilized by Hoffstein and Antonacci.
What is specification risk? Specification risk is the name for the possibility that the specific parameters of an investing style like momentum fail to capture the style's expected benefits while other unused parameters do. Weaker returns and both higher volatility and drawdowns are the key ways in which specification risk is made real. Notably, specification risk is not style risk—the risk of investing in momentum or some other style versus other styles.
When considering how to manage specification risk in momentum, Hoffstein and Antonacci are both focused on the issue of "lookback." Central to the style of momentum investing is the specific timeframe used to evaluate the performance of a particular asset. The "lookback" is the timeframe of past performance utilized to determine whether the asset exhibits sufficient momentum to justify investing in it for a period of time in the future.
For instance, in his Global Equities Momentum system (GEM), Antonacci utilizes a 12-month lookback to determine whether to invest in equities or bonds for the ensuing month. If the performance of the S&P500 stock index fund exceeds that of a fund holding short-term treasuries (T-bills) over the previous 12-months, the momentum of the stock index fund justifies holding it for the next month. If the reverse is true, the portfolio assets are invested in an intermediate term bond fund for the following month.
As set forth in his Dual Momentum (2014), Antonacci's choice of using a 12-month lookback is based on the significant academic research on momentum investing and also on making GEM as tax-efficient as possible. It is not based on the theory that lookbacks of varying lengths are equivalent and demonstrate the benefits of momentum investing equally.
By contrast, Hoffstein claims that utilizing simply a 12-month lookback creates too much specification risk on shorter timeframes, generating sufficient fragility to justify diversifying portions of a portfolio into shorter lookbacks. In this way, Hoffstein advocates devoting portions of the portfolio to momentum strategies with lookbacks ranging from 1 to 12 months as he provides performance data for the previous 10 years for lookbacks ranging from 6-12 months. By diversifying within the general style of momentum investing, Hoffstein proposes his model of managing specification risk.
Instead of agreeing with Hoffstein's momentum models, however, Antonacci proposes a model for managing specification risk by diversifying externally, by including a percentage of the portfolio in an intermediate-term bond fund. This would be the same bond fund (AGG) utilized under Antonacci's GEM model if stocks had exhibited weak momentum over the 12-month lookback relative to T-bills.
To manage specification risk, Antonacci has proposed devoting 10% of a portfolio to a constant position in AGG, while the remaining 90% is invested in GEM. Such an approach reduces the volatility of the portfolio and its drawdowns and provides for better risk adjusted returns. If GEM’s 12-month lookback has not performed as well as models with shorter lookbacks, the constant position in AGG almost always provides a positive total return on an annualized basis at low volatility. The same principle applies when allocating a portion of the portfolio to any asset or investment style that does not exhibit whipsaw losses or lag in acting on buy and sell signals that momentum portfolios demonstrate from time to time.
Which model does a better job of managing specification risk for momentum investing both in general and on key timeframes? As Antonacci recently stated in a statement on Twitter replying to one by Hoffstein,
[n]o one is denying that specification risk exists. The questions are what to do about it and at what opportunity cost. Having 7 lookback models is one way. Using outside diversification is another.
Leaving aside the question of whether it is beneficial to use both methods, commentators are split on which approach works better. See here and here. While I also acknowledge benefits with Hoffstein's approach, I am not convinced that the burdens of his internal approach justify utilizing it when a simpler external approach like Antonacci's manages specification risk with fewer burdens and solves other problems as well.
No doubt both models succeed to some extent in managing specification risk. In Hoffstein's case, utilizing more lookback models helps to manage the risk that any one fails due to statistical noise. Of course, such failure is more likely over a shorter time frame than on a longer one.
However, the risk of such diversification is missing out on relying more on the best performing lookback. The more time passes the greater the cost of these missed opportunities will become apparent to investors, regardless of whether they are comforted in the short term. There are also additional losses that derive from shorter lookbacks. Shorter lookbacks are more likely to generate false or mis-timed signals for buying and selling.
Hoffstein's blogpost on the topic raises some additional issues as well. Its title is not precise, "Fragility Case Study: Dual Momentum GEM." In the abstract, it's a difficult proposition to term GEM "fragile," particularly when Antonacci's GEM counters well the obvious fragility of buy-and-hold approaches to stocks. How exactly is GEM fragile? Is GEM fragile in the sense that Nassim Taleb has described?
Hoffstein defines as "fragile" a strategy that causes portfolio returns to be highly sensitive to the specification of its models where even slight variation of model parameters leads to dramatically different return profiles. But Hoffstein's conclusion that GEM is fragile is itself driven by observations of statistical noise from short timeframes that he sees as significant for his purposes. Whether this constitutes fragility is not at all clear, and if the term signifies as Hoffstein uses it, how useful is it and to which investors is it useful?
Possibly, it is Hoffstein's misperception of fragility in the GEM model that leads him to trust more in randomness than expected value to create his model of managing specification risk. In this regard, Antonacci correctly notes that "[t]he crux of Corey's argument is that all lookbacks are equal and any differences among them are statistical noise." In doing so, Hoffstein de-emphasizes major trends in the data that favor a 12-month lookback over shorter lookbacks, as is well documented by the research that drives the actual parameters of Antonacci's GEM.
In this regard, Hoffstein relies too easily on the gambler's fallacy as a basis to defeat a simple claim that expected value wins out over time. Instead of relying on data, Hoffstein theorizes that "[t]he unfortunate reality is that these performance differences are not expected to mean-revert." (Emphasis mine) If that is true, it would be more useful to see it in the data rather than as a theory or expectation, preferably in rolling or randomized key time frames.
Hoffstein oversells his conclusions in other ways as well. His edge of diversifying by lookback is "in being vaguely wrong rather than precisely wrong. The former is annoying. The latter can be catastrophic." I laud Hoffstein's emphasis on risk management, but would it be "catastrophic" to utilize GEM's 12-month lookback for broad equity indexes even on a limited timeframe? The limited data set in Hoffstein's blogpost indicates that it would not be, nor does Antonacci's more extensive data.
Similarly, it's likely overstating things to claim that "[t]he larger problem at hand is that none of us have a hundred years to invest" and that "[i]n reality, most investors have a few decades." Of course, I agree that "none of us have a hundred years to invest," but why does it take 100 years to overcome the perceived fragility of GEM? Nor am I at all sure that a "few decades" is all most investors have, unless most investors don't begin investing until later middle age.
I respect and regularly follow the work of both Hoffstein and Antonacci, and I expect to do so in the future. But I cannot fathom the circumstances under which I would manage specification risk through Hoffstein's approach and not through Antonacci's. In my view, GEM is not sufficiently fragile to justify the internal diversification for which Hoffstein advocates. External diversification is the way to do it.