Nobody ever said winning alpha was going to be easy. Among the people who don’t say that, count the author of a new paper in the Rotman International Journal of Pension Management. The paper offers empirical support for what many have suspected, that the same tools that are used to detect corporate chicanery, or to evaluate governance, can be employed to earn alpha.
But the paper, by Ophir Gottlieb, of GMI Ratings, also cautions, indeed emphasizes – that this is a matter of heavy duty numbers crunching. Implementing the Forensic Alpha Model “requires enormous data collection, detailed peering, industry normalizations, forensic accounting and governance taxonomies, sophisticated measures of association and interactions, rigorous testing, and advanced supervised machine learning.”
The FAM does not have great crystal-gazing aspirations. Gottlieb says that GMI looks “just one quarter ahead in our modeling.” Nor does it have universalist claims. GMI treats each region as a separate model, and most of the paper concerns North American applications. The North American FAM was developed from in-sample data from 2002 to 2007 and tested from out-of-sample data from 2009 to 2013. [The extraordinary year 2008 GMI regards as a data set onto itself.]
Holding Periods
For that in-sample data, GMI worked with ranking of holding period returns rather than with returns as such, thus limiting volatility. Further, it focused on quarterly holding-period returns of at least 10% in absolute value, thus filtering out noise.
In developing the model, GMI started with 350 available metrics and trimmed them down to a fancy-dandy 64, where “some of these metrics had a positive and some a negative relationship to equity returns, and all had a bounded asymptotic standard error (ASE).”
It all boils down to three scores: an accounting score, a governance score, and a score for accounting-governance interaction. These scores do seem capable of predicting stock price moves over the following quarter.
In North America, “companies with FAM ranks in the worst 10% are seven times as likely as
companies the best 10% to face a 30% monthly stock drop, nine times as likely to face a 40% monthly stock drop, 14 times as likely to face a 50% monthly stock drop, and 27 times as likely to face a 60% monthly stock drop,” the paper says.
Two Drivers
There are two chief drivers of this predictive effect. One Gottlieb calls the “black swan” effect, in a nod to Nassim Taleb. The worst-rated FAM companies have a higher incidence of extraordinary dives (a fall in stock price of 30% or more within a single month) than do the higher-rated FAM companies.
There is also a driver at the other end, the positive return effect. The best 10% FAM companies achieved positive monthly returns 57% of the time, (56.7% for the best 20%) versus 54.3% for the universe. These are differences with high degrees of historical significance.