The Value of Getting Sentimental about Your Alphas
The belief that markets react to news in a causal fashion is not only the way we think about investing, it’s also one of the things that captivates us about it, i.e., the notion that we can succeed as investors through careful research and logical thought. A Mr. Spock approach, if you will.
Unfortunately, investing is not only about logic, it’s also about sentiment. In other words, you need Commander Troi—with her psionic ability to sense emotions—in the room, too.
Axioma’s ROOF Scores were developed in an attempt to bridge this gap. Sentiment indicators can be controversial and not everyone likes them. But they—and the debate they engender—have been around since the beginning.
The argument for...
If successful investing involves anticipating the anticipations of others, then sentiment absolutely needs to be part of the conversation. Humans suffer from confirmation bias; hence investors tend to under-react to news not aligned with their sentiment, and over-react to news that confirms it. Put another way, sentiment influences how we deal with the uncertainty that exists between the realities of the present and our imperfect notions of the future. Ignoring the states of mind of other investors when investing is like ignoring a loved one’s fashion tastes when buying them a new sweater. The odds of success are not good.
The argument against...
Detecting hints of ulterior outcomes, excluded (or at least resisted) by the majority of investors, is the key to successful investing. Only raw data lends itself to the kind of forensic analysis intended to expose alternative outcomes that constitute the potential for mispricing and therefore alpha. The pressure for future data points to conform to a predetermined sentimental path is behind every unfortunate market crash. Relying on sentiment when investing is like blindly ordering dinner for your guests. Yes, you chose the restaurant because you all have similar tastes. But ignorance of your friend’s allergy to peanuts will come back to haunt you.
So, two schools of thought here (who knew, right?). As the saying goes, alphas only turn profitable when they become consensus. The former may be found independently, but the latter is much more dependent on sentiment. Being there first matters, but only if everyone else comes and joins you. We therefore believe that combining the two views is worth a look.
We challenge those who say “less is more” and devise an investment strategy to back-test “more is more.” Using the short-term momentum factor return from our short-horizon fundamental factor model for the US market (i.e., AXUS4-SH), we will take on (via futures contracts) 130% or 70% of the market return depending on whether the previous trading day ended with a positive momentum or a negative one respectively. That is to say that on any given trading day, we will have a BUY or SELL signal from this alpha. The table below shows our results, both annually and cumulatively, for the past three years.
Yearly Backtest Results for US-LMS Signal Only Strategy
Note that BUY and SELL recommendations are evenly matched (347 vs. 342) over the (almost) three-year period. Over the full period the strategy underperformed its underlying market index by a cumulative 1.6% (i.e., this signal didn’t work) and only outperformed in one year (2018) out of the three. The (realized) volatility of daily active return averaged 3.9% over the period.
For our next back-test we combined the above alpha signal with the Axioma ROOF Scores for the US market. The chart below shows our original alpha on the X-axis and the daily ROOF Scores on the Y-axis. The correlation between the two series over that three-year period is -0.09. We define a BUY signal when both our Alpha and the ROOF Score for the previous day were positive (depicted with green dots), and define a SELL signal when both were negative (depicted with red dots). If there is ‘disagreement’ between the two, we define that as a HOLD signal (depicted with blue dots) and simply hold the benchmark return for that day.
US-LMS ALPHA vs. ROOF Score for Year: 01/2017 - 09/2019
The table below summarizes the results for the combined Alpha + ROOF signal. The combined signal strategy outperformed its benchmark (i.e., holding just 100% of the market return for all three years) by a cumulative 6.2%. The average realized volatility of daily active return for this strategy was just 2.6% for the period. We note that the strategy underperformed in 2017 but had a lower realized active risk that year than the Signal-only strategy.
Yearly Backtest Results for US-LMS Signal + ROOF Strategy
Note that the majority of the days saw a HOLD recommendation (359 HOLD vs. 181 BUY and 149 SELL). These were days when either the S-T Momentum told us to buy, but the ROOF scores told us the market was bearish; or days when S-T Momentum told us to SELL, but the ROOF scores told us the market was bullish. We also note that the year with the best outperformance (active return of +4.7% in 2018), was the year with the greatest number of SELL recommendations (65 vs 53 in 2017 and 31 YTD in 2019).
The two charts below show the daily active returns for both strategy (signal-only on top, Signal + ROOF below). We note that the active drawdowns for the Signal + ROOF strategy not only limited the depth of the active drawdowns, but also substantially shortened their length.
Daily Active Returns for US-LMS SIGNAL-ONLY Strategy From: 01/02/2017 To: 09/27/2019
Daily Active Returns for US-LMS SIGNAL + ROOF Strategy From: 01/02/2017 To: 09/27/2019
We repeated this back-test for all nine markets for which we compute ROOF Scores and with the exception of the UK (+6.7% vs. +7.0%) and China (-5.6% vs. -5.0%), the combined Signal+ROOF strategy outperformed the Signal-only strategy in each of the other seven markets.
While the above is by no means a comprehensive study of the matter, we do believe that investors would benefit from combining their independent alpha signal with some measure of investor sentiment via an alpha-shrinking process. This lowers the impact of false positives/negatives (every signal contains a certain amount of noise) and ensures a more optimal allocation of the risk budget by aligning it with sentiment momentum.
 I.e., depending on whether the daily Short-Term (previous 20-days) Momentum factor return was positive or negative.
 Global Developed, Global Emerging, Asia ex-Japan, Developed Europe, Japan, Australia, UK, China, and the US.