A threshold effect of COVID-19 risk on oil price returns

Abstract

Using U.S. data, we investigate how the COVID-19 pandemic influences oil price returns in an asset pricing framework. Unlike earlier studies, we consider a threshold model to allow for the possibility that COVID-19 risk may not play a role until it reaches a certain level. Based on WTI crude oil spot price data from January 2020 to December 2021, our findings show that oil returns significantly decline with the daily number of COVID-19 deaths but only if the daily death toll exceeds approximately 2100. In addition, a more severe COVID-19 pandemic can substantially increase the exposure of oil returns to various systematic risk factors, which has not been documented in previous literature.

1. Introduction

This article examines the impact of the coronavirus disease 2019 (COVID-19) pandemic on the U.S. crude oil market from a novel asset pricing angle. Consistent with Narayan (2020), we find that greater COVID-19 risk, measured by the daily number of deaths, can significantly decrease oil price returns. Furthermore, a larger number of COVID-19 deaths can substantially increase the exposure of oil price returns to various systematic risk factors. Interestingly, these effects are at work only when the daily number of deaths surpasses a threshold (approximately 2100)—if the daily death number is below the threshold, the impact of COVID-19 on oil price returns is insignificant, and systematic risk exposure is very small. This indirect threshold effect of the exposure to systematic risk has not been previously documented in the literature and cannot be detected by traditional linear asset pricing models.

How have the COVID-19 pandemic and the resulting change in social operations affected asset returns in the crude oil market, and what are the implications for oil investors? The answer is not so obvious. On the one hand, a decrease in economic activities, especially transportation, can reduce oil demand. On the other hand, lockdowns and strategic responses by oil producers can curtail oil supply.1 Furthermore, government stimulating policies can also have a complicated impact on both oil demand and supply (Ashraf, 2020). The net impact of the COVID-19 pandemic on oil prices is a popular topic, and various papers have explained it from different perspectives.

Coronavirus disease 2019, often referred to as COVID-19, is an infectious disease caused by severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2) (WHO). The virus was first detected in December 2019 in Wuhan, China, and has spread worldwide since early 2020, resulting in a global pandemic (AJMC, 2021). By the end of 2020, COVID-19 infections surpassed 83.8 million and deaths surpassed 1.8 million globally, with 20 million cases and more than 346,000 deaths in the United States (U.S.) (AJMC, 2021). Since the start of the pandemic, the virus has tremendously changed the way human society operates and has had a substantial economic impact. For example, the U.S. economy saw an unemployment rate exceeding 12.9% and gross domestic product (GDP) declining by −8.5% in 2020 (OCED, 2020). The U.S. stock market also responded to the COVID-19 pandemic with far greater volatility than during any other pandemic or epidemic in the past 120 years (Altig et al., 2020).

As the COVID-19 pandemic swept across the globe, social distancing became necessary to reduce contagion. An unprecedented change is that many people have been working from and staying at home due to self-precaution, government-enforced lockdowns, or both (OCED, 2021). In the oil market specifically, poor demand and disrupted business activities at the beginning of the pandemic led to a period of low oil prices and oil companies' disappointing financial performance. For instance, on April 20, 2020, the spot price of West Texas Intermediate (WTI) crude oil plummeted into negativity for the first time in history, closing at negative 36.98 U.S. dollars per barrel. Among others, this unprecedented event motivates us to scrutinize the impact of the pandemic on oil prices and to what degree the impact changed over time with the severity of the pandemic.

Several papers have assessed the potential nonlinearity in the impact of COVID-19 on oil prices, such as Zhang and Hamori (2021) and Narayan (2020). An undocumented perspective is that as the global commodity market has become financialized (Tang and Xiong, 2012), the oil price has been affected by not only conventional flow demand and supply shocks but also precautionary and speculative demand (Cross et al., 2022). From this speculative demand perspective, our paper examines whether there are threshold effects on the influence of COVID-19 on oil returns. Using daily data from January 2020 to December 2021 (T = 502), we innovatively bridge the standard asset pricing model with a COVID-19 threshold. Compared with earlier papers using subsample analysis (Capelle-Blancard and Desroziers, 2020; Mamaysky, 2023; Ramelli and Wagner, 2020) or structural break testing (Zhang et al., 2021; Karavias et al., 2022), our threshold model offers greater flexibility to characterize nonlinearity, if any, in the relation between COVID-19 risk and oil returns. We find two channels regarding the impact of COVID-19 on oil price returns: 1) The direct channel, in which oil returns decline with the daily number of COVID-19 deaths, but only if the number exceeds 2100 deaths per day––deemed to categorize the economy as in “the high COVID-risk regime”. 2) The indirect channel, in which oil returns display substantially higher exposure to systematic risk factors in the high COVID-risk regime than in the low COVID-risk regime. Earlier studies have not documented the indirect channel, including Narayan (2020), which uses the number of COVID-19 cases to determine the role of COVID-risk regimes in predicting oil returns from COVID news; this paper is the first to employ a nonlinear asset pricing model with multiple market systematic risk factors to study WTI oil price returns.

Applying a two-regime asset pricing model with threshold effects on COVID-19 risk, this paper examines the psychological barrier concept (Mitchell, 2001) in the context of an extremely negative exogenous shock, COVID-19. Using 502 daily observations from January 2, 2020, to December 30, 2021, our work offers empirical evidence to investors that the market's view of an asset's exposure to systematic factors can change abruptly with external shocks. Rather than offering a specific trading strategy for COVID-19 risk, our paper focuses more on the methodology of modeling the impact of extreme shocks with threshold effects. With this in mind, investors can construct corresponding trading strategies for various future exogenous shocks.

The rest of the paper is organized as follows. Section 2 briefly reviews the literature, Section 3 discusses our data, Section 4 shows the empirical results, and Section 5 concludes.

2. Literature review

There has been a growing body of literature on the impact of the COVID-19 pandemic on oil prices since 2020.2 Among those studies, Narayan (2020) utilizes a threshold regression model and finds that when the number of new COVID-19 infections surpasses 84,479, COVID-19 news exerts a larger effect on oil prices. Our paper complements Narayan (2020) from several perspectives. First, our result is consistent with that of Narayan (2020) in that the impact of COVID-19 on oil price returns varies with the severity of the pandemic. Second, in contrast to Narayan (2020), who focuses on the daily number of COVID-19 cases, we find that the daily death toll has the strongest threshold effect by comparing four different COVID-19 risk metrics.3 Third and most importantly, we show that high COVID-19 risk not only directly decreases oil price returns but also increases the sensitivity of oil prices to systematic risk factors.

In addition to the extension from Narayan (2020), our interpretation of the impact of COVID-19 on oil prices is inspired by recent studies on the financialization of oil.4 Tang and Xiong (2012) document unprecedented inflows of institutional funds into the commodity futures market since the 2000s. From this perspective, Basak and Pavlova (2016) develop a theoretical model to study how the financial market can transmit shocks to commodity prices. Cross et al. (2022) detect precautionary and speculative oil demand in addition to more conventional flow demand and supply shocks. Therefore, we raise the question of whether systematic risk factors may have a significant impact on oil returns when speculative demand becomes significant. To the best of our knowledge, this paper is the first to use a nonlinear asset pricing model to study WTI oil price returns. Using the COVID-19 pandemic as an example, our empirical analysis indeed shows that the market's view of systematic factor exposure can change abruptly with exogeneous shocks.

By utilizing a threshold model, our paper also contributes to the literature that documents a time-varying impact of COVID-19 on oil prices and other asset returns. Three main approaches include analyses of moving windows, structural breaks, and subsamples. For example, using a moving-window technique, Zhang and Hamori (2021) show that the connectedness between the crude oil market and a COVID-19 index tracker fluctuates sharply during 2020. Examining the relation between crude oil futures prices and oil investor sentiment, Huang and Zheng (2020) identify a structural break during the pandemic period between December 31, 2019, and February 25, 2020.5 Karavias et al. (2022) detect a structural break in the impact of COVID-19 on stock market returns between late March 2020 and early April 2020.6 Ramelli and Wagner (2020) divide the period from January 2, 2020, to March 20, 2020, into three subsamples: incubation (January 2, 2020, through January 17, 2020), outbreak (January 20, 2020, through February 21, 2020), and fever (February 24, 2020, 2020, through March 20, 2020). They find that firm-level stock returns drop mostly during the fever period.7 Our method is different from the method used by these papers in that we estimate the threshold value of COVID-19 case/death numbers rather than identify specific days on which transitions occur. In addition, compared with subsample or structural break testing methods that can handle limited sample divisions, threshold models allow more flexible switching between the different regimes.

The threshold effect we find is also consistent with the strand of literature on psychological barriers. Mitchell (2001) notes that some specific numerical values in the financial market influence investor decision making, leading to the concept of psychological barriers. Direct applications use this concept to find such threshold numbers in financial products, including oil. For example, Narayan (2022) observes that WTI oil price decimals tend to cluster around decimal numbers closer to zero than to one. Similarly, Dowling et al. (2016) note that the Brent oil price clusters every 10 dollars. Narayan et al. (2017) take one step further to consider the nonlinear relation on such psychological barriers. They find that the impact of oil prices on the stock market changes once the oil price reaches 100 dollars per barrel. These works have identified threshold numbers in oil prices and have developed corresponding trading strategies. Our paper, on the other hand, uses the COVID-19 case/death number as an alternative example of inspecting potential psychological barriers: psychological barriers can exist not only in the prices of financial products but also among exogenous variables. During the pandemic, a psychological barrier can exist in certain counts of cases or deaths. Once a threshold is reached, people may perceive more severity from the pandemic, which may lead to consumers reducing travel plans, firms switching to work-from-home policies, and governments issuing stricter social distancing policies (Ashraf, 2020), which in the end affect oil price returns.

3. Data

Our sample period is from January 2, 2020, to December 31, 2021, including 502 daily observations. For the oil price, we use the WTI spot price,8 which is the main benchmark for the crude oil market in North America. The price data are obtained from the U.S. Energy Information Administration (EIA) website. We calculate oil price excess returns as daily percentage changes in the price in excess of the risk-free rate obtained from Kenneth R. French's website. We adopt the Chicago Board Options Exchange (CBOE) oil market volatility index OVX as a proxy for oil market risk.

For COVID-19 risk metrics, we collect data on four different variables from the COVID-19 Data Repository by the Center for Systems Science and Engineering at Johns Hopkins University, including the daily number of cases, daily number of deaths, and corresponding 7-day moving average (MA) numbers.9 All four metrics are subject to a log transformation equal to log(1 + x), where x is the raw number.10

We incorporate the five asset pricing factors from Fama and French (2015) together with the momentum factor from Carhart (1997) into our asset pricing model. These factors are obtained from Kenneth R. French's website and are referred to as MKT-RF (market factor), SMB (size factor), HML (value factor), RMW (profitability factor), CMA (investment factor), and MOM (momentum factor), respectively. In asset pricing theory, these factors represent systematic risk and can be used to price any financial asset. We use the six-factor model, augmented by OVX and COVID-19 risk factors, as a benchmark to evaluate the asset pricing performance of oil returns in the sample period. The summary statistics are shown in Table 1 .

Table 1.

Summary Statistics.

	Obs.	Mean	Std. Dev.	P10	P50	P90
Oil market data
Excess Oil Returns	502	−0.53	15.55	−3.37	0.23	2.99
CBOE OVX	502	51.85	34.82	33.36	39.47	78.00
COVID variables
[log 1+] Daily COVID-19 cases	502	10.25	3.26	6.56	11.13	12.37
[log 1+] 7-day MA cases	502	9.95	3.20	5.72	10.94	12.01
[log 1+] Daily COVID-19 deaths	502	6.54	2.23	2.52	7.16	8.13
[log 1+] 7-day MA deaths	502	6.22	2.15	1.90	6.86	7.70
Asset pricing factors
MKT-RF	502	0.10	1.66	−1.35	0.17	1.51
SMB	502	0.01	0.98	−1.08	0.01	1.18
HML	502	−0.02	1.41	−1.58	−0.10	1.74
RMW	502	0.04	0.67	−0.77	−0.01	0.86
CMA	502	0.00	0.51	−0.55	−0.02	0.61
MOM	502	−0.03	1.67	−2.06	0.12	1.74

The sample period is between January 2, 2020, and December 31, 2021, and all variables are at a daily frequency. The excess oil returns are calculated as the percentage change in the West Texas Intermediate (WTI) spot price in excess of the risk-free rate (as in MKT-RF) multiplied by 100. OVX is the Chicago Board Options Exchange (CBOE) oil market volatility index. The COVID information is from the COVID-19 Data Repository by the Center for Systems Science and Engineering at Johns Hopkins University. Daily COVID-19 cases (deaths) on date t are calculated as the difference between the cumulative number of cases (deaths) on date t and the cumulative number on the previous trading day. The 7-day MA cases (deaths) refer to 7-day moving averages. All of the COVID-19 numbers reflect a log-transformation equal to log(1 + x), where x is the raw number (note that when x = 0, i.e., when there are no new cases or deaths, log(1 + x) = 0). MKT-RF, SMB, HML, RMW, and CMA are the five asset pricing factors constructed in Fama and French (2015), and MOM is the momentum factor. For each variable, we report the number of observations, mean, standard deviation, and values of the 10th, 50th, and 90th percentiles.

4. Empirical results

This section is divided into two subsections. Section 4.1 discusses the empirical results and limitations of the benchmark linear asset pricing model, while Section 4.2 explores the empirical results using our nonlinear model with a threshold effect on COVID-19 risk.

4.1. Linear estimates

We first estimate a linear asset pricing model:

$${\mathit{Oil\; Excess\; Return}}_t=\alpha +{\mathit{X}}_t^{\prime }\mathit{\beta }+\gamma {\mathit{Covid}}_t+{\epsilon }_t,$$ (1)

where the dependent variable is oil excess returns, X _t is a vector of asset pricing factors (MKT-RF, SMB, HML, RMW, CMA, MOM), and OVX is the oil market volatility index. The Fama/French factors are the standard setup for an asset pricing model, as we examine the impact of COVID-19 on the speculative demand for oil as a financialized commodity asset (Cross et al., 2022). The inclusion of OVX is motivated by the GARCH-in-mean type of models (Rahman and Serletis, 2012), where asset returns may depend on the associated volatility. The coefficients βs summarize the corresponding exposure of oil returns to the factors in X _t. Covid _t is the daily number or 7-day moving-average number of new COVID-19 cases or deaths under the log-transformation specified in Section 3. We include the COVID-19 variable as a regressor because other recent papers have confirmed that the COVID-19 pandemic has indeed influenced oil prices (see, e.g., Zhang et al., 2021; Narayan, 2020; Narayan, 2022 in Section 2). Consequently, the coefficient γ indicates the marginal contribution of a 1% increase in the corresponding COVID-19 number to oil excess returns.

The ordinary least squares (OLS) estimates of model (1) are reported in Table 2 (without OVX) and Table 3 (with OVX), with standard errors in parentheses. According to the two tables, the direct findings are as follows. First, COVID-19 risk seems not to have any significant impact on oil excess returns, given that the COVID-19 variables are all insignificant with small t values. Second, the market factor (MKT-RF) and profitability factor (RMW) are the most significant systematic risk factors that determine oil price movements. When the market factor or the profitability factor moves up by 1%, the excess oil returns move up 1.29% or 3.43% in the same direction, respectively. These results are consistent with the conventional belief that oil prices are procyclical. Third, the coefficients of OVX are significant in all columns with different COVID-19 risk metrics, and including OVX significantly improves the model fitness from R ² = 0.033 to 0.126. Therefore, OVX is crucial to oil returns and should be included in the asset pricing model. In addition, at the bottom of Table 2 and Table 3, we report the results of the Breusch–Pagan test for heteroskedasticity together with the Breusch–Godfrey test for serial correlations of residuals in the disturbance. With or without OVX as the regressor, we can reject the assumption that the residuals are both homoskedastic and serially uncorrelated, so we report in Table 2 and Table 3 the heteroskedasticity and autocorrelation consistent (HAC) standard errors of Andrews (1991).

Table 2.

OLS Results of Augmented Asset-Pricing Models.

	(1)	(2)	(3)	(4)	(5)
	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns
Daily COVID-19 cases		0.022
		(0.213)
7-day MA cases			0.039
			(0.217)
Daily COVID-19 deaths				−0.070
				(0.312)
7-day MA deaths					−0.092
					(0.325)
MKT-RF	1.298	1.294***	1.292***	1.307***	1.309***
	(0.924)	(0.446)	(0.446)	(0.447)	(0.447)
SMB	0.843	0.838	0.834	0.857	0.863
	(0.992)	(0.928)	(0.928)	(0.929)	(0.930)
HML	−1.046	−1.046	−1.048	−1.046	−1.046
	(0.941)	(0.945)	(0.945)	(0.945)	(0.945)
RMW	3.179	3.175**	3.174**	3.188**	3.188**
	(2.826)	(1.375)	(1.375)	(1.375)	(1.375)
CMA	0.274	0.268	0.264	0.284	0.290
	(1.187)	(1.789)	(1.789)	(1.789)	(1.789)
MOM	−0.392	−0.392	−0.392	−0.393	−0.395
	(0.380)	(0.606)	(0.606)	(0.606)	(0.606)
Constant	−0.828	−1.055	−1.212	−0.374	−0.258
	(1.060)	(2.284)	(2.264)	(2.152)	(2.129)
Observations	502	502	502	502	502
R-squared	0.033	0.033	0.033	0.033	0.035
Breusch–Pagan test
p value	0.18	0.25	0.25	0.20	0.19
Breusch–Godfrey test
p value	0.00	0.00	0.00	0.00	0.00

Standard errors are reported in parentheses. “***”, “**”, and “*” indicate significance at the 1%, 5%, and 10% levels, respectively.

Table 3.

OLS Results of Augmented Asset-Pricing Models incl. OVX.

	(1)	(2)	(3)	(4)
	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns
Daily COVID-19 cases	0.002
	(0.204)
7-day MA cases		−0.011
		(0.208)
Daily COVID-19 deaths			0.018
			(0.300)
7-day MA deaths				−0.034
				(0.245)
MKT-RF	1.285***	1.287***	1.283***	1.292***
	(0.428)	(0.428)	(0.429)	(0.428)
SMB	1.355	1.358	1.352	1.368
	(0.893)	(0.893)	(0.894)	(0.893)
HML	−1.852**	−1.852**	−1.852**	−1.852**
	(0.914)	(0.914)	(0.914)	(0.914)
RMW	3.435***	3.437***	3.433***	3.442***
	(1.319)	(1.319)	(1.319)	(1.319)
CMA	0.502	0.506	0.500	0.513
	(1.716)	(1.716)	(1.715)	(1.716)
MOM	−0.849	−0.850	−0.849	−0.851
	(0.585)	(0.585)	(0.585)	(0.585)
OVX	−0.128***	−0.128***	−0.128***	−0.128***
	(0.019)	(0.019)	(0.019)	(0.019)
Constant	5.741**	5.878**	5.648**	6.137***
	(2.415)	(2.417)	(2.252)	(2.255)
Observations	502	502	502	502
R-squared	0.126	0.126	0.126	0.126
Breusch–Pagan test
p value	0.00	0.00	0.00	0.00
Breusch–Godfrey test
p value	0.00	0.00	0.00	0.00

Standard errors are reported in parentheses. “***”, “**”, and “*” indicate significance at the 1%, 5%, and 10% levels, respectively.

The linear model (1) has been established in the asset pricing literature but features two limitations that are relevant here. First, the model assumes a constant impact of the COVID-19 pandemic on oil returns. However, earlier studies have presented empirical evidence that market participants have a time-varying response to COVID-19 risk during the pandemic. The majority of these papers have used subsample analysis (Capelle-Blancard and Desroziers, 2020; Mamaysky, 2023; Ramelli and Wagner, 2020) or a structural break test (Zhang et al., 2021; Karavias et al., 2022) that identifies specific dates of change. Alternatively, from a pandemic severity perspective, Narayan (2020) shows that a threshold effect on daily COVID-19 cases can explain the varying impact of COVID-19 on oil prices. Intuitively, when people see a low number of daily cases or deaths, they may not respond immediately by reducing their travels or other economic activities (such as oil production and activities that require crude oil as an input) but may do so when the number of cases/deaths becomes large enough to cause panic. That is, COVID-19 risk may not play a significant role in oil price returns until it reaches a certain level at which market participants start to have concerns.

Second, in the linear model (1), COVID-19 risk influences only the level of expected oil returns but not the systematic exposure of the returns. Put differently, in the model, the assumption is that regardless of the COVID-19 risk level, the sensitivity of oil prices to various systematic risk factors, which determines the magnitude of βs, is constant. This assumption seems implausible because it is reasonable to believe oil prices can be more sensitive to systematic factors when COVID-19 risk is high. Take MKT-RF as an example because it has the clearest economic interpretation among all of the asset pricing factors––as the overall stock market premium, MKT-RF reflects the prospects of future economic growth and financial wealth of market participants, at least to a large extent. When COVID risk is high, activities that were previously easy and normal may become more difficult and costly. For instance, air tickets can be more expensive due to flight cancellations or additional costs incurred, such as airline companies' costs regarding sanitization or providing personal protective equipment. Travels that were normal in the past may turn into luxuries. As a result, oil demand is more sensitive to people's wealth and beliefs of the future of the economy, leading to larger exposure to MKT-RF.

4.2. Threshold estimates

To manage the two limitations of linear regression models, we consider the following two-regime threshold model:

$${\mathit{Oil\; Excess\; Return}}_t={\alpha }_1+{\mathit{X}}_t^{\prime }{\mathit{\beta }}_{\mathbf{1}}+{\gamma }_1{\mathit{Covid}}_t+{\epsilon }_t,\text{if}{\mathit{Covid}}_t\le C,$$ (2)

$${\mathit{Oil\; Excess\; Return}}_t={\alpha }_2+{\mathit{X}}_t^{\prime }{\mathit{\beta }}_{\mathbf{2}}+{\gamma }_2{\mathit{Covid}}_t+{\epsilon }_t,\text{if}{\mathit{Covid}}_t>C,$$ (3)

where Eq. (2) describes the low COVID-risk regime in which the COVID-19 risk metric is smaller than a threshold C, and Eq. (3) represents the high COVID-risk regime in which the risk metric exceeds the threshold C. The model coefficients and the threshold C can be estimated using the method developed by Chan (1993) and Hansen (2000). Note that the threshold model we propose does not simply incorporate interaction terms between a dummy variable that identifies the high COVID-risk regime and the independent variables in Eq. (1). A key difference is that the dummy variable approach requires manually dividing the sample into high- and low-risk regimes, which can be arbitrary. Our threshold method relies on an automatic grid-search algorithm to estimate the threshold value and therefore is impartial. Our approach is also different from that of Narayan (2020), in which the coefficient of the control variable (oil price return volatility) is constant regardless of the COVID-19 pandemic risk measured by daily cases. In contrast, in our model, the coefficients of all of the variables are allowed to switch when the COVID-19 risk metric moves across the threshold. This is why we interpret our estimation results by two separate regimes: a high COVID-risk regime and a low COVID-risk regime.

The estimation results are reported in Table 4 , where panel A presents the low COVID-risk regime and panel B presents the high COVID-risk regime. The standard errors reported in parentheses are the Andrews (1991) HAC standard errors to manage the heteroskedasticity and high-order serial correlation we detect in Section 4.1. Our first finding is that in the low COVID-19 risk regime, the number of daily COVID-19 cases or deaths does not have a significant impact on oil returns. However, in the high COVID-risk regime, more daily deaths cause oil returns to significantly decrease. As shown in column (3), a 1% increase in the daily death number can reduce excess oil returns by 6.625%, holding other variables constant. This finding suggests that COVID-19 deaths constitute a crucial determinant of oil returns but only if its number surpasses a threshold. This result is consistent with that of Mitchell (2001) in the direction that specific numeric values can affect decision making in financial markets and is similar to the empirical result of Narayan (2020).

Table 4.

Estimation Results of Augmented Asset-Pricing Threshold Models incl. OVX.

	(1)	(2)	(3)	(4)
	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns	Excess Oil Returns
Panel A: Low COVID risk
Daily COVID-19 cases	0.109
	(0.090)
7-day MA cases		0.190
		(0.276)
Daily COVID-19 deaths			0.182
			(0.302)
7-day MA deaths				0.227
				(0.254)
MKT-RF	0.721***	0.590	0.446	0.416
	(0.262)	(0.528)	(0.421)	(0.344)
SMB	1.814**	2.287*	1.440	1.113
	(0.840)	(1.361)	(0.936)	(0.745)
HML	−1.065*	−1.029	−0.119	−0.139
	(1.635)	(1.484)	(0.991)	(0.768)
RMW	4.317***	1.686	0.317	0.407
	(0.931)	(2.475)	(1.426)	(1.106)
CMA	−1.290	0.539	−0.057	−0.118
	(1.114)	(3.425)	(1.983)	(1.541)
MOM	0.243	0.015	−0.067	−0.111
	(0.244)	(0.901)	(0.574)	(0.468)
OVX	0.017	0.035	0.003	0.009
	(0.023)	(0.027)	(0.025)	(0.019)
Constant	−1.664	−2.626	−1.255	−1.643
	(1.179)	(2.332)	(2.269)	(1.870)
# Observations	138	142	357	446
Panel B: High COVID risk
Daily COVID-19 cases	0.444
	(26.392)
7-day MA cases		−1.083
		(1.277)
Daily COVID-19 deaths			−6.625*
			(3.470)
7-day MA deaths				9.753
				(13.845)
MKT-RF	1.473	2.965***	7.493***	16.015***
	(6.179)	(0.718)	(1.108)	(1.711)
SMB	2.005	1.252	0.380	−0.014
	(5.287)	(1.130)	(1.818)	(2.533)
HML	−1.940	−0.737	−5.473***	−11.380***
	(7.493)	(1.051)	(1.691)	(3.217)
RMW	4.366	3.785***	12.657***	17.581***
	(17.829)	(1.438)	(2.426)	(3.302)
CMA	0.352	−0.628	3.855	9.593***
	(5.787)	(1.857)	(2.740)	(3.488)
MOM	−0.953	−1.129*	−2.604*	−8.771***
	(6.739)	(0.655)	(1.385)	(2.713)
OVX	−0.404***	−0.326***	−0.262***	−0.348***
	(0.138)	(0.027)	(0.025)	(0.033)
Constant	11.752	25.619*	66.100**	−62.500
	(208.036)	(14.928)	(28.222)	(110.772)
# Observations	364	360	145	56
Threshold variables	Daily COVID-19 cases	7-day MA cases	Daily COVID-19 deaths	7-day MA deaths
Threshold values	10.382	10.248	7.645	7.679
F statistic	8.25	11.52	32.28	13.33
p value	0.10	0.09	0.01	0.08

Standard errors are reported in parentheses. “***”, “**”, and “*” indicate significance at the 1%, 5%, and 10% levels, respectively.

The threshold values for four different COVID-19 risk metrics are shown at the bottom of panel B. We perform a bootstrap-based F test to test for the existence of threshold effects and report the p values. For all four COVID-19 risk metrics, the threshold effect is significant at the 10% level; particularly for the daily death number, the threshold effect is significant at the 5% level. Therefore, the best threshold is when the daily death toll reaches approximately 2100 deaths per day, where log(1 + 2100) = 7.65.

Furthermore, oil returns display almost no systematic exposure in the low COVID-risk regime but considerably high systematic exposure in the high COVID-risk regime. For example, in column (3), the market (MKT-RF) beta is 7.493 when COVID risk is high, much larger than the market beta of 0.446 when COVID risk is low. This market beta is also much higher than the linear estimates in Table 2 and Table 3, which can be viewed as an average of the low and high COVID-risk regimes. The same conclusion applies to almost all other systematic risk factors. In summary, we find that COVID-19 risk can not only depress oil returns but also strongly raise the sensitivity of oil returns to various systematic risk factors. A direct implication from a risk management perspective is that oil investors should rebalance their portfolios to offset the increased systematic risk exposure from oil assets when COVID-19 risk is high.

In addition, it is interesting to note that in column (3) (also in column (4)), the constant term in the high COVID-risk regime is positive and significant. This result means that investors with a long position in oil assets can receive positive abnormal returns when COVID risk is high. From a practitioner's perspective, fund managers can benefit from investing in oil assets during high COVID risk periods because abnormal returns are usually the most important criterion to evaluate the investment performance of asset management companies. However, these abnormal returns may not translate into actual investment profits because they (a) are obtained from in-sample fit and (b) require that the daily number of COVID deaths be incorporated in the asset pricing benchmark. In other words, one needs to acknowledge that asset returns are expected to be low in periods of high COVID risk. With this in mind, investing in oil assets can generate returns that are better than expectations, but the actual returns can still be negative.

5. Conclusions

Our paper aims to characterize the relation between oil prices and common systematic risk factors in an established asset pricing model and examine how COVID-19 pandemic risk can influence this relation. By employing a two-regime threshold model and exploring U.S. data between January 2, 2020, and December 31, 2021, we find that when the COVID-19 death number exceeds a threshold of approximately 2100 deaths per day, more deaths significantly decrease oil price returns, and the returns become more sensitive to common systematic risk factors.

Our results have not been previously documented and may spur future research in two broad aspects. First, from an energy economics perspective, future research could examine the potential threshold effects of other exogenous shocks on energy prices if such effects are suggested by economic theories or intuitions. Second, from a finance perspective, our finding that the COVID-19 pandemic can raise the exposure of oil prices to various systematic risk factors indeed sheds important light for investors. We recommend that investors take into account the possibility that the exposure of their portfolios to systematic risks can change abruptly through external shocks. With this in mind, investors can better rebalance their portfolios to manage risk.

CRediT authorship contribution statement

Yiguo Sun: Conceptualization, Formal analysis, Methodology, Software, Supervision, Writing – review & editing. Delong Li: Data curation, Writing – original draft, Writing – review & editing. Chenyi Suo: Data curation, Formal analysis, Investigation, Software, Writing – original draft. Yu Wang: Formal analysis, Validation, Writing – review & editing.

Appendix A. Supplementary data

Supplementary material