Hunting a COVID-19 factor

Blog post
Jun Wang, George Bonne
April 29, 2020
  • We observed severe and diverse impacts in equity markets over our study period that suggest one could think of COVID-19 as a "factor."
  • Our three methods to identify a COVID-19 factor produced one that is a linear combination of existing style and industry factors and formulated so that companies with high exposure to it may outperform under COVID-19.
  • The methods generally showed that the COVID-19 factor had positive exposure to the health care, consumer staples, size and profitability factors, and negative exposure to airlines, consumer services, leverage and beta factors.
We have noted previously how factor performance has been severely and diversely impacted by the COVID-19 pandemic. Clearly some companies have been hurt more than others, and some have even benefited. Can we identify a COVID-19 factor and quantify companies' exposure to it? In other words, can we create a metric that will provide an estimate of how positively or negatively a company might be impacted by the COVID-19 pandemic? We propose several methods to calculate such a factor. All the methods resulted in a factor that is a linear combination of existing style and industry factors of the MSCI Barra Global Equity Trading Model (GEMTR), as opposed to creating a new factor from scratch, using analysis of residuals or other methods. Our methods assume that the existing GEMTR factors sufficiently capture the effects of the pandemic across different types of companies. In all methods we define the COVID-19 factor such that stocks with high exposure to the factor are expected to outperform if markets are adversely impacted by COVID-19, while stocks with low (negative) exposure are expected to underperform. All weights presented are normalized such that their sum of absolute values is one. From simplest to most complex, we summarize the methods below.
  1. Cumulative factor returns — We calculate each factor's cumulative return over the period from Feb. 24 through March 23 and that becomes the relative weight in our COVID-19 factor.1
  2. Sensitivity to the market — We regress each factor's return against the market (world factor) return for the period from Feb. 24 through April 20, and the negative of the resulting slope coefficient becomes the relative weight in our COVID-19 factor.2
  3. Principal Component Analysis (PCA) on factor returns — We run PCA on factor returns and the first principal component becomes our COVID-19 factor, again using the period from Feb. 24 through April 20.
Method 1: Cumulative factor returns
If we assume the pandemic was the dominant driver of stock returns from Feb. 24 through March 23, and its impact was largely spanned by the existing GEMTR factors, then the relative performance of the existing factors during this period should provide a reasonable proxy for the sensitivity of stocks to the pandemic. In the table below we display the top and bottom 15 weights in a potential COVID-19 factor constructed using this method.
Top and bottom 15 style and industry factor weights using cumulative factor returns
Factor
Weight
Unnamed: 2
Factor.1
Weight.1
Factor

Food and staples retailing

Weight

4.57%

Unnamed: 2

None

Factor.1

Leverage

Weight.1

-5.93%

Factor

Short interest

Weight

4.22%

Unnamed: 2

None

Factor.1

Consumer service

Weight.1

-4.02%

Factor

Industry momentum

Weight

4.00%

Unnamed: 2

None

Factor.1

Dividend yield

Weight.1

-3.65%

Factor

Profitability

Weight

3.01%

Unnamed: 2

None

Factor.1

Airlines

Weight.1

-3.08%

Factor

Size

Weight

2.61%

Unnamed: 2

None

Factor.1

Consumer durables and apparel

Weight.1

-3.05%

Factor

Biotechnology

Weight

1.88%

Unnamed: 2

None

Factor.1

Diversified financials

Weight.1

-2.93%

Factor

Telecommunication services

Weight

1.84%

Unnamed: 2

None

Factor.1

Real estate

Weight.1

-2.63%

Factor

Food and beverage and tobacco

Weight

1.80%

Unnamed: 2

None

Factor.1

Commercial and professional services

Weight.1

-2.56%

Factor

Household and personal products

Weight

1.72%

Unnamed: 2

None

Factor.1

Liquidity

Weight.1

-2.53%

Factor

Pharmaceuticals and life sciences

Weight

1.40%

Unnamed: 2

None

Factor.1

Aerospace and defense

Weight.1

-2.47%

Factor

Steel

Weight

1.32%

Unnamed: 2

None

Factor.1

Beta

Weight.1

-2.06%

Factor

Semiconductors

Weight

1.23%

Unnamed: 2

None

Factor.1

Energy equipment and services

Weight.1

-1.94%

Factor

Communications equipment

Weight

1.20%

Unnamed: 2

None

Factor.1

Insurance

Weight.1

-1.89%

Factor

Transportation non-airline

Weight

0.90%

Unnamed: 2

None

Factor.1

Thrifts and mortgage finance

Weight.1

-1.86%

Factor

IT services and software

Weight

0.77%

Unnamed: 2

None

Factor.1

Residual volatility

Weight.1

-1.70%

Factor returns are calculated over the period from Feb. 24 through March 23, 2020. GEMTR contains 22 style factors and 45 industry factors. We display the 15 most positive and 15 most negative weights in the estimated COVID-19 factor. Style factors are in blue and industry factors are in black.
Consistent with what we observed previously, the COVID-19 factor constructed in this way had positive weight on size, profitability, food and staples retailing and biotechnology, and negative weight on leverage, beta, consumer services and airlines.
Method 2: Sensitivity to the market
If we once again assume that COVID-19 was the primary driver of overall equity-market performance, this time for the period from Feb. 24 through April 20, then factors whose performance largely mirrored (or were more severe than) the overall market could be good proxies for "exposure" to a COVID-19 factor. To measure the sensitivity of each factor's performance to the market, we regress factor returns against the GEMTR market factors. The negative of the slope of the regression coefficient becomes our relative weight in the resulting COVID-19 factor. In the table below we list the resulting weights in a potential COVID-19 factor constructed via this method.
Top and bottom 15 style and industry factor weights using sensitivity to the market
Factor
Weight
Unnamed: 2
Factor.1
Weight.1
Factor

Profitability

Weight

3.43%

Unnamed: 2

None

Factor.1

Beta

Weight.1

-7.53%

Factor

Food and staples retailing

Weight

3.16%

Unnamed: 2

None

Factor.1

Short-term reversal

Weight.1

-5.94%

Factor

Short interest

Weight

2.21%

Unnamed: 2

None

Factor.1

Leverage

Weight.1

-4.04%

Factor

Pharmaceuticals and life sciences

Weight

2.05%

Unnamed: 2

None

Factor.1

Real estate

Weight.1

-3.89%

Factor

Media

Weight

1.97%

Unnamed: 2

None

Factor.1

Liquidity

Weight.1

-3.42%

Factor

Food and beverage and tobacco

Weight

1.83%

Unnamed: 2

None

Factor.1

Residual volatility

Weight.1

-3.41%

Factor

Health care equipment and suppliers

Weight

1.79%

Unnamed: 2

None

Factor.1

Insurance

Weight.1

-3.30%

Factor

Mid capitalization

Weight

1.55%

Unnamed: 2

None

Factor.1

Diversified financials

Weight.1

-3.26%

Factor

Biotechnology

Weight

1.49%

Unnamed: 2

None

Factor.1

Consumer services

Weight.1

-2.95%

Factor

Long-term reversal

Weight

1.42%

Unnamed: 2

None

Factor.1

Capital markets

Weight.1

-2.78%

Factor

Earnings variability

Weight

1.34%

Unnamed: 2

None

Factor.1

Utilities

Weight.1

-2.39%

Factor

Household and personal products

Weight

1.33%

Unnamed: 2

None

Factor.1

Thrifts and mortgage finance

Weight.1

-2.27%

Factor

Growth

Weight

1.32%

Unnamed: 2

None

Factor.1

Regional banks

Weight.1

-2.06%

Factor

IT services and software

Weight

1.17%

Unnamed: 2

None

Factor.1

Retailing

Weight.1

-1.61%

Factor

Book-to-price

Weight

1.08%

Unnamed: 2

None

Factor.1

Consumer durables and apparel

Weight.1

-1.53%

Factor returns are standardized and calculated over the period from Feb. 24 through April 20, 2020. We display the 15 most positive and 15 most negative weights in the estimated COVID-19 factor. Style factors are in blue and industry factors are in black.
We see that the weights from method 2 are similar to the weights from method 1 — the correlation of the weights is 0.67. At the stock level, the exposures of the COVID-19 factors produced from the two methods had an average cross-sectional correlation of 0.73.
Method 3: PCA on factor returns
If we assume again that the behavior of factors during the period from Feb. 24 through April 20 was driven by COVID-19, then PCA could be used to create a linear combination of factors that capture the majority of the diversity (variance) in behavior of those factors. PCA is a technique used to reduce the dimensionality of a data set while retaining most of its information. It does so by identifying "principal components," which are linear combinations of the original variables along which the variation in the data is maximized. A potential drawback of the PCA method for this application is that, because it seeks to capture as much variability as efficiently as possible, if there are factors that are highly correlated or anti-correlated to one another, it may load on one but not others if the "others" do not provide additional information. The PCA analysis showed the first principal component captured approximately 20% of the variability in factor performance among styles and industries in GEMTR. In the exhibit below we display the weights of the first principal component, which constitutes our potential COVID-19 factor from this method. We see the weights from the PCA method were similar to those from the other two methods. The correlations in weights between the PCA method and methods 1 and 2 were 0.80 and 0.77, respectively. At the stock level, the cross-sectional correlation of exposures between this method and methods 1 and 2 are 0.83 and 0.89, respectively.
Top and bottom 15 style and industry factor weights using PCA on factor returns
Factor
Weight
Unnamed: 2
Factor.1
Weight.1
Factor

Short interest

Weight

3.74%

Unnamed: 2

None

Factor.1

Short-term reversal

Weight.1

-7.01%

Factor

Food and staples retailing

Weight

3.54%

Unnamed: 2

None

Factor.1

Consumer services

Weight.1

-5.78%

Factor

Industry momentum

Weight

3.03%

Unnamed: 2

None

Factor.1

Beta

Weight.1

-5.08%

Factor

Profitability

Weight

2.38%

Unnamed: 2

None

Factor.1

Real estate

Weight.1

-3.98%

Factor

Size

Weight

2.00%

Unnamed: 2

None

Factor.1

Leverage

Weight.1

-3.89%

Factor

Utilities

Weight

1.74%

Unnamed: 2

None

Factor.1

Diversified financials

Weight.1

-3.83%

Factor

Long-term reversal

Weight

1.40%

Unnamed: 2

None

Factor.1

Airlines

Weight.1

-3.38%

Factor

Food and beverage and tobacco

Weight

1.39%

Unnamed: 2

None

Factor.1

Consumer durables and apparel

Weight.1

-3.03%

Factor

Household and personal products

Weight

1.39%

Unnamed: 2

None

Factor.1

Retailing

Weight.1

-2.66%

Factor

Steel

Weight

1.30%

Unnamed: 2

None

Factor.1

Aerospace and defense

Weight.1

-2.53%

Factor

Biotechnology

Weight

1.19%

Unnamed: 2

None

Factor.1

Dividend yield

Weight.1

-2.30%

Factor

Gold

Weight

1.11%

Unnamed: 2

None

Factor.1

Automobiles and components

Weight.1

-2.13%

Factor

Regional banks

Weight

1.09%

Unnamed: 2

None

Factor.1

Earnings yield

Weight.1

-1.93%

Factor

Telecommunication services

Weight

1.01%

Unnamed: 2

None

Factor.1

Analyst sentiment

Weight.1

-1.87%

Factor

Pharmaceuticals and life sciences

Weight

0.92%

Unnamed: 2

None

Factor.1

Thrifts and mortgage finance

Weight.1

-1.63%

Factor returns are standardized and calculated over the period from Feb. 24 through April 20, 2020. We display the 15 most positive and 15 most negative weights in the estimated COVID-19 factor. Style factors are in blue and industry factors are in black.
What happened when we combined the three methods?
While each of the three methods takes a different approach to seek the same objective, they all produced similar results. Thus, since we have no a priori metric for determining which is "better" or more "correct," taking a simple average of the weights from the three seems a logical next step. The table below displays the weights in a potential COVID-19 factor after averaging the weights from the three methods and re-normalizing such that the sum of absolute values is unity.
Top and bottom 15 style and industry factor weights using a combination of the three methods
Factor
Weight
Unnamed: 2
Factor.1
Weight.1
Factor

Food and staples retailing

Weight

4.06%

Unnamed: 2

None

Factor.1

Beta

Weight.1

-5.29%

Factor

Short interest

Weight

3.67%

Unnamed: 2

None

Factor.1

Short-term reversal

Weight.1

-5.06%

Factor

Profitability

Weight

3.18%

Unnamed: 2

None

Factor.1

Leverage

Weight.1

-5.00%

Factor

Industry momentum

Weight

2.83%

Unnamed: 2

None

Factor.1

Consumer services

Weight.1

-4.60%

Factor

Food and beverage and tobacco

Weight

1.81%

Unnamed: 2

None

Factor.1

Real estate

Weight.1

-3.79%

Factor

Size

Weight

1.81%

Unnamed: 2

None

Factor.1

Diversified financials

Weight.1

-3.62%

Factor

Biotechnology

Weight

1.64%

Unnamed: 2

None

Factor.1

Consumer durables and apparel

Weight.1

-2.75%

Factor

Household and personal products

Weight

1.61%

Unnamed: 2

None

Factor.1

Airlines

Weight.1

-2.57%

Factor

Pharmaceuticals and life sciences

Weight

1.58%

Unnamed: 2

None

Factor.1

Residual volatility

Weight.1

-2.43%

Factor

Telecommunication services

Weight

1.37%

Unnamed: 2

None

Factor.1

Dividend yield

Weight.1

-2.39%

Factor

Steel

Weight

1.14%

Unnamed: 2

None

Factor.1

Liquidity

Weight.1

-2.39%

Factor

Communications equipment

Weight

0.97%

Unnamed: 2

None

Factor.1

Aerospace and defense

Weight.1

-2.34%

Factor

IT services and software

Weight

0.92%

Unnamed: 2

None

Factor.1

Insurance

Weight.1

-2.34%

Factor

Semiconductors

Weight

0.92%

Unnamed: 2

None

Factor.1

Retailing

Weight.1

-2.09%

Factor

Gold

Weight

0.82%

Unnamed: 2

None

Factor.1

Thrifts and mortgage finance

Weight.1

-2.08%

We display the 15 most positive and 15 most negative weights in a combined COVID-19 factor. Style factors are in blue and industry factors in colored black.
With our COVID-19 factor defined, we can also examine how it performed and measure any portfolio's exposure to it. The exhibit below displays the performance of the GEMTR market factor (for reference) and our combined COVID-19 factor, as defined by the weighted combination of its input factor returns, over the period from Jan. 1 through April 20. By construction, the factor performed well and was strongly anti-correlated with the market over this period. As a result, the performance of a COVID-19 factor may have been useful in estimating market sentiment toward the pandemic.
Performance of a potential COVID-19 factor vs. the GEMTR market factor
We also examined the exposures of a handful of factor and sector indexes to our COVID-19 factor. We display these results in the table below. As expected, the consumer staples and quality indexes had high exposure to our COVID-19 factor while the consumer discretionary index had low exposure.
MSCI indexes exposures to a potential COVID-19 factor
Unnamed: 0
Feb. 24, 2020
April 20, 2020
Unnamed: 0

MSCI ACWI Consumer Discretionary Index

Feb. 24, 2020

-0.16

April 20, 2020

0.06

Unnamed: 0

MSCI ACWI Diversified Multiple-Factor Index

Feb. 24, 2020

0.21

April 20, 2020

0.09

Unnamed: 0

MSCI ACWI Momentum Index

Feb. 24, 2020

0.4

April 20, 2020

0.09

Unnamed: 0

MSCI ACWI Quality Index

Feb. 24, 2020

0.48

April 20, 2020

0.4

Unnamed: 0

MSCI ACWI Minimum Volatility (USD) Index

Feb. 24, 2020

0.5

April 20, 2020

0.1

Unnamed: 0

MSCI ACWI Consumer Staples Index

Feb. 24, 2020

0.85

April 20, 2020

0.62

In the wake of the COVID-19 pandemic and the large and diverse impact on stocks and factors, we explored three ways in which one could construct a COVID-19 factor that is a linear combination of existing GEMTR factors. The methods assumed that the impacts of COVID-19 were already captured by the GEMTR factors, and the factors produced by the three methods were all similar to one another and produced a factor that was anti-correlated to the market returns during the crisis period. While it is difficult to argue how to define a "true" COVID-19 factor, we believe the results presented could provide a reasonable estimate to a portfolio's sensitivity to the impacts of COVID-19.
Further Reading
The coronavirus market impact spreads globally The coronavirus epidemic: Implications for markets Factors separated fact from fiction The FaCS report
George Bonne
George Bonne
Jun Wang
Jun Wang
Vice President, MSCI Research

Subscribe today
to have insights delivered to your inbox.

1We use Feb. 24 through March 23, as that is period over which we saw the most severe negative impact on markets. We scale (standardize) the returns to each factor by dividing by its volatility to facilitate comparison across factors.2We use the period from Feb. 24 through April 20 for methods 2 and 3. We use a longer history for these methods because we are seeking to capture the variability (up and down) in factor performance as opposed to just the drawdown as in method 1.

The content of this page is for informational purposes only and is intended for institutional professionals with the analytical resources and tools necessary to interpret any performance information. Nothing herein is intended to recommend any product, tool or service. For all references to laws, rules or regulations, please note that the information is provided “as is” and does not constitute legal advice or any binding interpretation. Any approach to comply with regulatory or policy initiatives should be discussed with your own legal counsel and/or the relevant competent authority, as needed.