We have often discussed in recent months how company factor exposures — most notably to market sensitivity (aka beta) — had turned topsy-turvy, a result not only of the market plunge but also as investors sorted out their expectations for relative winners and losers of the COVID pandemic. My colleague Diana Baechle recently wrote about Mortgage REITs (see here), but numerous companies and industries have seen dramatic changes in their market sensitivity exposures this year. In Exhibit 1 we show three examples for US industries. Hotels went from having an exposure just slightly below zero (about average) to a highly positive 0.6, as those stocks tanked far more than the market. In contrast, the Biotech industry fell less than the market during the downturn and saw its beta decline. Equity REITs were among the lowest beta of industries until February, when that industry’s Market Sensitivity soared and is now higher than average.
Individual stocks have displayed similar shifts. These moves have prompted clients to ask whether they should rebalance their portfolios to reflect the new reality, or wait, in case these exposures return to “normal”. Exhibit 1 suggests that none of the three industries seems to be reverting to old times, at least yet.
Exhibit 1. Market Sensitivity Exposures, Select US Industries
To look at this from a longer-term perspective, and as the issue might be reflected in a portfolio, we turned to our optimizer. We created a simple US portfolio in which we sought to minimize exposure to the Market Sensitivity factor and targeted 3% tracking error versus the Russell 1000 index. The portfolio was rebalanced monthly starting January 2007 and ending in May 2020, and had no industry, style, or turnover constraints. Our goal was to see what has happened recently to the portfolio’s turnover and compare that with history, i.e., does turnover (when unconstrained) stay high after an initial jump, suggesting exposures reverted back, or does it drop, which may mean that exposures stuck around for a while?
Our low-Market-Sensitivity portfolio did, indeed, experience one of its highest-ever levels of turnover—over 100%—in March 2020, although there have been other periods of higher turnover. An eyeballing of the monthly turnover chart seems to show that turnover does tend to settle down after a big month, but there is a positive relationship between one month’s turnover and the next, as shown in the x-y chart.
Exhibit 2. Monthly Turnover
This suggests that for this type of strategy there probably has been some reversion of the Market Sensitivity factor in the past. In other words, if a portfolio manager had higher than average turnover this month (because the factor exposures changed), one might expect a higher than average turnover next month (because the exposures changed again, possibly back to what they were two months prior). However, this was not the case this year. March’s high turnover was followed by roughly average turnover in April, and far lower turnover in May. As suggested by the chart of industry volatilities above, Market Sensitivity exposures have not yet reverted, at least for those industries. Given the economic vulnerabilities of these and other industries, it is not surprising that their exposures still reflect the sudden shift we saw in late February.
One might reasonably argue that the level of turnover also reflects the volatility environment. We have found that changing market volatility tends to be associated with bigger differences in the underlying components of active risk (factor and specific volatilities, as well as factor correlations). We have extensively documented these changes in volatility recently, particularly in the blog post “Who Blew up my Tracking Error?” (see here). We have also shown how active risk has increased over the past few months, even as total risk has leveled off (see here).
But how do we differentiate the impact of higher factor volatility from the alpha (Market Sensitivity) changes on turnover?
Luckily, my colleague, optimization expert Adrian Zymolka, has developed an effective methodology for attributing turnover to its components: those necessary to satisfy constraints (in this case the only explicit constraint was tracking error) versus that which comes from the objective function, which for our study was to maximize alpha (where alpha was low Market Sensitivity). This type of analysis is particularly useful for our current scenario, as we have recently entered a stable, but high market volatility environment. We expected high turnover when volatility shot up. If Market Sensitivity exposures had boomeranged, we would have seen high turnover again the next month. But that could have also been driven by the high and increasing levels of factor volatility and changes in correlations, so parsing the two is vital.
Over the course of our study, turnover was mainly driven by alpha, not constraints (Exhibit 3). In most months less than 5% of the total turnover was the result of the tracking error constraint (which means that more than 95% of turnover was the result of the pursuit of alpha—a very desirable outcome).
This year, active risk jumped above the 3% target in the period leading up to the February and March rebalancings (Exhibit 4), so it was clear some turnover would be needed to bring it down. But our analysis tells us that the proportion needed for risk-reduction was relatively small (about 9% of the total in February when overall turnover was high; 6% in March when we started to see big changes in Market Sensitivity exposures; and 2% in April). The vast majority of turnover, therefore, was driven by the optimization objective function and, in turn, was far lower compared with March because Market Sensitivity scores remained relatively stable.
Exhibit 3. Proportion of Turnover Driven by Constraints
Exhibit 4. Low Market Sensitivity Portfolio Active Risk vs. Market Risk, 2020
Our turnover attribution analysis indicates that big changes in Market Sensitivity scores were the major driver of high portfolio turnover for a portfolio that tilts on the factor in March. Those scores did not revert back to where they had been as market volatility plateaued in April, most likely because the COVID-driven slowdown has made some relatively permanent changes to the economics of those industries, and there is no evidence yet that this will change any time soon. This “new normal” differs from what seems to have happened in the past, when the changes in exposures seemed to be more temporary. In-depth analysis of the sources of portfolio turnover was an important key in separating Market Sensitivity scores’ impact from what was happening in the risk environment.
 In other words, we wanted a portfolio that tilted on low Market Sensitivity stocks.
 Readers may recognize this as a variation on the “unconstrained” version of the sample factor portfolios, for which we publish results and write commentary on a quarterly basis.