Looking Back at Market Environment Factor Research in 2020

Following equity markets in 2020 was a wild ride, to say the least. Much attention has been given to how different securities, sectors, and regions reacted. In this article, we’re going to take a different perspective and look at things from the factor level. In rapidly changing markets like these, leveraging any static factor strategy throughout is unlikely to perform well. To be successful, a strategy must adapt quickly in real time.

To provide some context on the analysis that follows, we’ll focus on the Russell 3000 universe, looking at data from December 31, 2019 to June 1, 2020. The performance and analytics are calculated for several different factors daily. All factors are sourced from FactSet’s Quant Factor Library, which is a point-in-time database made of a wide variety of quantitative factors that are calculated and archived each night.

First, to get a feel for overall factor performance during the pandemic, we look at YTD performance across a variety of different types of factors. In the chart below, we use “F1-FN” or spread return (green) to measure the difference between the top and bottom quintile returns of each factor to determine how predictive the factor is of positive performance. In this example, factors like growth, momentum, and liquidity are the best performers while value, market sensitivity, technical, and management are the worst. The standard deviation (red) and T-Stat (purple) help put that performance in context by measuring its consistency.

A chart like this gives a quick summary of which factors were outperformers and underperformers overall, but it takes digging deeper to better understand how these factors behaved throughout the period and within different types of securities. Next, we’ll look at a time series of each factor’s spread return overlaid against the Russell 3000 benchmark’s return (blue).

The first aspect that sticks out is the notable pivot point in the market in mid to late March 2020 when the benchmark return hit a low point and began to recover. Interestingly, we can see a range of effects around this pivot on different factors. Growth—one of the top-performing factors—barely flinched and continued climbing. Other price-based factors, technical and market sensitivity, saw their spread returns peak around this time before falling to end the period as the worst-performing factors alongside value. Factors like quality and profitability saw a sharp increase as the U.S. market dropped. However, that boost faded just as quickly, replaced by a steady downtrend.

Looking more closely at the quality factor, you can see the idea of a “flight to quality” through changes in its information coefficient through time. Right as markets are dropping suddenly, there is a cluster of strong positive correlation with returns (highlighted in red). However, acting on this even a few days late would mean missing the majority of the upside and being left with the notably negative results that follow.

Alongside factor timing, it’s also important to know how factors perform across different groups of securities. Here we look at factor performance within sectors to see if any factors display inconsistent results. The size of the bubble indicates the magnitude of the positive (green) or negative (red) spread return. Growth, for example, shows consistently positive performance across nearly all sectors barring Energy and Consumer Services. Other factors, such as momentum, show a wider range of performance depending on the sector.

Having looked at each of these individual factors and their relationships with future return, we observed some mixed results. But how does it look when we put them all together? In creating a single composite factor or alpha score, we start with the most straightforward way of taking a simple average of each component. However, the performance from this approach (green) is poor and doesn't take advantage of the insight we’ve gained in our research so far.

Next, we incorporate a dynamic weighting scheme that uses trailing factor performance information to determine the factor weightings in the composite factor. This approach allows our alpha signal to automatically adjust as the efficacy of the underlying components changes. In this case, we’re using two methods. The first weights the component factors by the degree to which their best quintile outperforms the benchmark (yellow) in the previous period. That way, factors with a strong performance against the benchmark in recent periods receive more weight. The second method weights the factor components by their trailing information coefficient so that factors with more predictive power/stronger correlation with return are weighted more heavily (purple).

The performance from these dynamically-weighted alpha signals is dramatically improved due to their ability to adapt in real time. Below we can see the resulting weights attributed to each factor over time and how the alpha signal rotates into factors such as quality and market sensitivity when they’re working and out of them when performance turns.

While the attractive spread return for the best alpha factor above indicates that the top quintile outperforms the bottom, we should further assess the strength of the factor by looking across all five quintiles to confirm the relationship holds. In charting the cumulative returns of each quintile, we see this strengthens the explanatory power of the alpha factor since its quintile returns fall in order from the top to bottom quintile.

The use of these dynamic composite weightings can also be explored for differentiating between factor performance within groups or market environments. By incorporating information from recent periods, the enhanced alpha factors can react swiftly and pick up on shifting factor performance. In rapidly changing markets like we’ve observed, this can make a large difference in the results of factor research.

This article was coauthored by Julia Caffrey, CFA.