Co-movements in sector price indexes during the COVID-19 crisis: Evidence from the US.

This paper is an examination of co-movements between sector indexes in the United States prior to and during the COVID-19 period. Using daily data between January 2013 and July 2020, this study is the first to examine sectoral cointegration, as well as how contagion occurs from one healthcare sector to others. We find that only five sectors reacted to the shock to the healthcare sector. Our findings can assist policymakers in appropriately responding to the current crisis and tackling potential pandemics in the future. Our findings are also valuable for stockholders in terms of predicting price changes and improving portfolio diversification.

Introduction

In recent years, financial crises have multiplied, which has affected the evolution and connection of stock markets. Several studies on the contagion phenomenon have explored co-movement between sectors during periods of crisis. For instance, Phylaktis and Xia (2009) highlighted the advantages of the sectoral contagion phenomenon that provides some benefits of international diversification during crises. Similarly, Baur (2012) examined co-movement between the financial sector and the other real economy sectors in twenty-five major stock markets during the subprime crisis. Notably, the study found that all sectors and countries were exposed to adverse effects of the crisis, although some sectors (i.e., healthcare, telecommunications, and technology) realized only minor impacts. The current global COVID-19 health crisis has increased stock price co-movements, severely affecting the financial system. In response, a growing body of literature has examined the effects of COVID-19 on stock market returns (M. Akhtaruzzaman et al., 2021; Mazur et al., 2021; Topcu and Gulal, 2020; Zhang et al., 2020). These researchers employ various methods to investigate the link between the COVID-19 breakdown and global financial market performance. Other researchers have explored the effect of contagion on various asset classes—such as oil, gold, and cryptocurrency—as safe havens or hedges during the COVID-19 pandemic((Akhtaruzzaman et al., 2020a, Akhtaruzzaman et al., 2020b), 2021; Corbet et al., 2020; Devpura and Narayan, 2020; Gharib et al., 2021; Gunay, 2020; Mensi et al., 2020; Salisu et al., 2020, 2021; Sharif et al., 2020). Interestingly, the sectoral dimension of the COVID-19 pandemic has been illustrated by several recent empirical studies (Baek et al., 2020; Hanif et al., 2021; Rizwan et al., 2020), but only M. Akhtaruzzaman et al. (2021) and Shahzad et al. (2021) have focused on the connection between sector indexes during the COVID-19 period.

Furthermore, to the best of our knowledge, these studies have not examined sectoral co-movements by focusing specifically on the healthcare sector as the source of sectoral contagion during the COVID-19 outbreak. This study aims to fill this knowledge gap and use cointegration analysis to examine the shock of the healthcare sector to other sectors. In practice, the healthcare sector plays a vital role in the global economy Chen (2016). However, the analysis of co-movement between healthcare sectors and other sectors is not widely explored in the current body of literature. In this context, studies by Chen et al. (2015, 2017) confirm the importance of the dual issues of contagion and the healthcare sector. Specifically, they examined the dynamic relationship of returns in the healthcare sector among different stock markets (the US, UK, and Germany) using the continuous wavelet approach. Our work contributes to the literature in two ways: While most COVID-19-related finance research has studied the crisis effects on sectoral performance or risk, we explored the co-movements between the healthcare sector and other industries before and during the COVID-19 crisis. Moreover, we identify the most cointegrated sector indexes and explain their relationship with the healthcare sector.

The study highlights two main findings: First, we find evidence of five relationships between sectors during the COVID-19 crisis. Second, we show that contagion occurs from one healthcare sector to five other sectors. The results indicate that the healthcare sector price is an important indicator that can assist policymakers in developing early economic and healthcare policy responses, especially during this unprecedented pandemic. This finding is consistent with the existing literature (Barro, 2013; Koijen et al., 2016; Sala-I-Martin et al., 2004), suggesting that the healthcare industry is an important determinant of economic growth, making it a very attractive sector for investors (Allen, 2021). However, our results differ significantly from Akhtaruzzaman et al., 2021b, who demonstrate the central role of financial institutions while using a dynamic cross-correlation analysis, and Shahzad et al. (2021), who use network analysis to examine the interaction between sector indexes and highlight the central role of the IT sector.

This paper is organized into four major sections. Section 2 describes the data and provides descriptive statistics. Section 3 outlines and discusses our empirical results. Finally, Section 4 summarizes the main conclusions of this study.

Data and preliminary analysis

Data

We use daily data of the main sectoral index prices in the US: energy, materials, industrial (INDUST), consumer discretionary (CONS_DISCR), consumer staples (CONS_STAPLES), healthcare (HEALTH), financials, information technology (IT), communication services (COM_SERVICE), utilities, and real estate.1 The healthcare index is considered as the benchmark index to evaluate sectoral contagion before and during the COVID-19 period.

Data were collected from DataStream. We divide our sample period into two sub-periods: the pre-COVID-19 period (January 1, 2013–January 22, 2020) and the COVID-19 period (January 23, 2020–July 29, 2020). We started in January 2013 to avoid overlapping with the great recession of 2007–2012. On the other hand, we choose the start date of the COVID-19 period as the date of the first confirmed case reported by the World Health Organization.

We apply the following empirical methodology: first, we check for multivariate cointegration using Johansen's cointegration model (1988), and conduct a bivariate cointegration analysis to measure co-movements between the healthcare sector and other sectors. Once we detected the cointegration relationships, we moved to the estimation of a vector error correction model (VECM) for each cointegrated pair of sectors. Finally, we test the short-and long-run Granger causality in the VECM between sector indexes.

Descriptive statistics

Table 1 presents the descriptive statistics of the daily index prices of the 11 sectors used in this study. From this table, we can see that the average price index of all sectors increased during the COVID-19 crisis, except in the energy sector. The latter recorded an increase in its standard deviation, suggesting an increase in volatility. In contrast, we observe that the volatility of all other sectors decreased during the pandemic period. Fig. 1 shows that the most important decline in stock prices in all sectors occurred between February and March 2020. Furthermore, the kurtosis values are different from 3, which means that the distributions are leptokurtic. Moreover, the skewness values are different from 0, so the distributions are not skewed. This implies that the normality assumption is rejected for all series.

Table 1

Summary Statistics.

Sectoral index prices: Pre-COVID-19-period (January 1, 2013-January 22, 2020)
	COM_SERVICE	CONS_DISCR	CONS_STAPLES	ENERGY	FINANCIALS	HEALTH	INDUST	IT	MATERIALS	REAL_ESTATE	UTILITIES
Mean	95.193	316.359	312.945	407.989	181.083	312.76	341.51	237.181	343.046	491.437	285.647
Median	95.32	292.65	321.7	400.86	160.75	307.52	319.43	197.61	330.62	492.8	284.67
Maximm	138.41	498.9	433.89	534.38	272.98	475.32	497.71	477.52	450.76	716.62	443.62
Minimum	70.53	166.73	196.37	290.35	103.07	159.09	198.13	116.43	235.87	336.16	181.58
Std. Dev.	13.962	85.312	53.218	43.414	44.509	71.412	75.617	88.937	56.857	90.141	60.698
Skewness	0.637	0.363	−0.073	0.517	0.259	−0.085	0.16	0.606	0.043	0.446	0.379
Kurtosis	2.913	2.013	2.299	3.381	1.695	2.274	1.823	2.203	1.856	2.673	2.347
Jarque-Bera	125.086	115.256	39.41	93.289	151.479	42.722	114.317	161.765	101.124	69.436	76.765
Sum	175,440.6	583,050.4	576,757.6	751,923.9	333,736.1	576,417.1	629,402.1	437,123.6	632,233.2	905,717.7	526,446.5
Sum.Sq.Dev.	359,051	13,406,219	5,216,773	3,471,700	3,649,172	9,393,527	10,532,553	14,569,908	5,954,637	14,966,916	6,786,277
Sectoral index prices: COVID-19-period (January 23, 2020-July 29, 2020)
	COM_SERVICE	CONS_DISCR	CONS_STAPLES	ENERGY	FINANCIALS	HEALTH	INDUST	IT	MATERIALS	REAL_ESTATE	UTILITIES
Mean	126.8204	490.7027	403.0807	255.9677	214.0749	451.536	411.9482	463.786	392.9745	625.6058	391.9844
Median	130.57	496.555	402.5	246.7	206.725	459.525	412.93	473.05	401.18	625.31	384.48
Maximum	141.21	598.92	438.46	362.99	274.98	496.67	503.44	542.38	456.27	749.58	461.15
Minimum	99.21	349.35	330.74	153.4	154.1	341.62	293.81	343.59	278.51	449.99	292.48
Std. Dev.	11.24657	61.16602	21.74034	50.747	30.38693	30.97916	49.10671	49.19097	42.30074	64.65905	34.62291
Skewness	−0.805972	−0.37594	−0.617415	0.528358	0.743153	−1.370999	0.180243	−0.505109	−0.653923	0.024206	0.296267
Kurtosis	2.577858	2.611899	3.684157	2.690328	2.645203	4.573796	2.384708	2.466666	2.595467	2.722196	2.861378
Jarque-Bera	15.50251	3.997359	11.12689	6.770046	13.03701	55.80756	2.83932	7.286166	10.46377	0.443978	2.067577
Sum	16,993.94	65,754.16	54,012.82	34,299.68	28,686.04	60,505.83	55,201.06	62,147.33	52,658.58	83,831.18	52,525.91
Sum.Sq.Dev.	16,822.56	497,590.6	62,861.45	342,509.3	122,807.6	127,641.2	320,725.4	321,827	237,983.9	556,045.5	159,433.2

Notes: Our sample contains daily data. The period sample is divided into two sub-periods, where the pre-COVID-19 period is from 1 January 2013 to 22 January 2020 and the COVID-19 period is from 23 January 2020 to 29 July 2020.

Jarque–Bera statistic tests for the null hypothesis of Gaussian distribution.

Fig. 1

Evolution of sectoral stock prices. Notes: The above figure shows the daily prices of sectoral indexes during two sub-periods (Panel A and Panel B), where the pre-COVID-19 period is from 1 January 2013 to 22 January 2020 and the COVID-19 period is from 23 January 2020 to 29 July 2020.

To conduct a cointegration analysis, it is necessary to check whether the series of stock prices meets two conditions: non-stationarity and having the same order of integration. Thus, we applied different tests: the augmented Dickey-Fuller test (ADF) test, the Philips–Perron (PP) test, and the Kwiatkowski –Phillips–Schmidt–Shin (KPSS) test. Tables 2 and 3 present the results of the stationarity tests. We found that all series are non-stationary in level, and we tested the stationarity of the series in the first difference, where the results indicate that our series is I (1). Therefore, we can apply cointegration tests using Johansen's cointegration model (1988).

Table 2

Testing Stationarity for the Pre-COVID-19-period (January 1, 2013-January 22, 2020).

	ADF TEST		PP TEST		KPSS TEST
	In level	In first difference	In level	In first difference	In level	In first difference
COM_SERVICE	−0.22	−42.91***	−0.11	−42.97***	4.61***	0.12
CONS_DISCR	−0.25	−42.88***	−0.23	−42.89***	5.29***	0.05
CONS_STAPLES	−0.81	−42.86***	−0.76	−42.93***	5.10***	0.06
ENERGY	−2.37	−43.04***	−2.37	−43.05***	1.21***	0.082
FINANCIALS	−0.59	−43.32***	−0.59	−43.32***	5.15***	0.04
HEALTH	−0.75	−41.79***	−0.69	−41.87***	5.12***	0.06
INDUST	−0.74	−41.73***	−0.73	−41.71***	5.24***	0.03
IT	1.73	−33.47***	1.43	−44.07***	5.18***	0.36
MATERIALS	−1.42	−41.82***	−1.42	−41.82***	4.68***	0.03
REAL_ESTATE	−0.06	−41.88***	0.03	−41.93***	4.90***	0.10
UTILITIES	0.26	−32.67***	0.37	−42.48***	5.14***	0.13

Notes: The above table contains the statistics of the ADF tests, the PP test and the KPSS test on both the levels and differences of variables during Pre-COVID-19-period (January 1, 2013-January 22, 2020).

Rejection of the null hypothesis at the 1%, 5% and 10% is denoted by ***, **, and * respectively.

Table 3

Testing Stationarity for the COVID-19-period (January 23, 2020-July 29, 2020).

	ADF TEST		PP TEST		KPSS TEST
	In level	In first difference	In level	In first difference	In level	In first difference
COM_SERVICE	−0.87	−17.15***	−1.19	−16.37***	0.43**		0.32
CONS_DISCR	−0.58	−7.06***	−0.51	−13.68***	0.72**		0.30
CONS_STAPLES	−1.75	−15.60***	−2.20	−15.44***	0.23***		0.19
ENERGY	−2.15	−12.28***	−2.15	−12.25***	0.37**		0.31
FINANCIALS	−1.93	−7.27***	−1.92	−14.46***	0.46**		0.25
HEALTH	−1.35	−16.06***	−1.47	−15.91***	0.49*		0.17
INDUST	−1.87	−6.79***	−1.83	−13.41***	0.31***		−13.41
IT	−0.54	−17.92***	−0.87	−17.53***	0.69*		0.27
MATERIALS	−0.54	−6.96***	−0.87	−13.34***	0.69*		0.27
REAL_ESTATE	−2.11	−6.91***	−1.32	−13.93***	0.35**		0.15
UTILITIES	−2.35	−7.21***	−2.17	−13.89***	0.48*		0.14

Notes: The above table contains the statistics of the ADF test, the PP test and the KPSS test on both the levels and differences of variables during COVID-19-period (January 23, 2020-July 29, 2020).

Rejection of the null hypothesis at the 1%, 5% and 10% is denoted by ***, **, and * respectively.

Correlation analysis between the healthcare sector and other sectors

We note that during the COVID-19 period, correlations increased with some sectors and decreased with others, which reflects signs of sectoral contagion (Table 4 ).

Table 4

Correlation analysis between the health care sector and other sectors.

Pre-COVID-19-period (January 1, 2013-January 22, 2020)
	COM_SERVICE	CONS_DISCR	CONSSTAPLES	ENERGY	FINANCIALS	INDUST	IT	MATERIALS	REAL_ESTATE	UTILITIES
Pearson correlation with health care sector	0.875	0.970	0.951	−0.313	0.944	0.957	0.952	0.917	0.932	0.942
COVID-19–period (January 23, 2020-July 29, 2020)
Pearson correlation with health care sector	0.915	0.910	0.795	0.565	0.503	0.651	0.907	0.896	0.579	0.511

Note: the above table contains the Pearson correlation coefficients between health care sector and the other sectors before and during COVID-19 period.

However, this analysis did not consider the temporal variability of interdependence. Therefore, we use the dynamic conditional correlation (DCC) model developed by Engle (2002). The author proposes a new class of multivariate GARCH estimators, which is a generalization of the Bollerslev (1990) constant conditional correlation (CCC) estimator. The dynamic correlation model allows the matrix correlation to be time-varying.

Fig. 2 shows the time-varying correlation between the healthcare sector and other sectors during the entire period (from January 1, 2013, to July 29, 2020). The results of the DCC show that the healthcare sector is correlated with other sectors during the COVID-19 period (from the end of 2019 through the beginning of 2020). There is considerable interdependence between these sectors and the healthcare sector, as shown in Fig. 2. Indeed, the correlation between these sectors accelerated throughout the post-crisis period. We can observe that the cross-correlations of index returns are dynamic and show upward trends related to crisis events. Therefore, the dynamic correlation estimates confirm the varying and volatile interdependencies between the sectors. After estimating the time-varying correlation, we apply cointegration tests to measure the level of this interdependence and to specify a common long-term trend.

Fig. 2

Conditional correlation between the health sector and the other sectors. Notes: The figure above represents the estimated dynamic correlations between the health sector index and the other sector indexes during the COVID-19 period (from January 23, 2020, to July 29, 2020).

Empirical results and discussion

Testing for multivariate cointegration

Johansen's multivariate cointegration test among the US sectors reveals the existence of a single long-run cointegration relationship before the COVID-19 period (see Table 5 ). These sectors were not integrated before the crisis. However, the cointegration results show evidence of five relationships between sectors during the COVID-19 crisis, which can be explained by the contagion effect. These results prove that the markets are integrated. Furthermore, these cointegrating relationships between sector indexes reflect their convergence toward a certain stable equilibrium level over the long term.

Table 5

Testing for multivariate cointegration.

Pre-COVID-19-period (January 1, 2013-January 22, 2020)
Null hypothesis	J trace	J max
r = 0	303.7518*	66.2009
r = 1	237.5508	64.5047
COVID-19-period (January 23, 2020-July 29, 2020)
r = 0	387.2194*	92.8011*
r = 1	294.4183*	69.6266*
r = 2	224.7917*	52.1832
r = 3	172.6084*	43.0205
r = 4	129.5879*	39.5865
r = 5	90.0014	34.3672

Notes: J Trace and J max indicate the statistics of Johansen's cointegration model (1988).

Rejection of the null hypothesis of no cointegration at the 5% is denoted by *.

Testing for bivariate cointegration

The healthcare index is considered the benchmark index to evaluate co-movements and the integration of other sector indexes in relation to the healthcare sector index. The bivariate cointegration test between the healthcare sector and the other sectors showed the absence of a stable equilibrium relationship before the health crisis, and hence their segmentation.

Fig. 3 shows the response of the sectors to one standard deviation shock to the healthcare sector. We observe that the standard deviation shock to the healthcare index has a noticeable impact on all sectors from the first period. These responses sharply decline until period 2 and then gradually increase.

Fig. 3

Impulsive response of sectors during COVID-19 period. Notes: The blue lines show the impulsive response of each sector index to the one standard deviation shock to health during the COVID-19 period (from January 23, 2020, to July 29, 2020), while the red lines indicate the 95 percent confidence interval.

However, the bivariate cointegration test during the COVID-19 period indicated five relationships between the healthcare index and the other sector indexes. Indeed, we did not reject the cointegration hypothesis for five pairs: health-communication services, health-consumer discretionary, health-financials, health-industrials, and health-materials (see Table 6 ), highlighting co-movements around a common tendency for the stock prices of these sectors.

Table 6

Testing for bivariate cointegration during COVID-19-period.

HEALTH - COMMUNICATION_SERVICES
Null hypothesis	J trace	J max
r = 0	17.3578*	13.1019
r = 1	4.2559*	4.2559*
HEALTH - CONS_DISCR
Null hypothesis	J trace	J max
r = 0	17.8316*	17.2738*
r = 1	0.55781	0.5578
HEALTH - FINANCIALS
Null hypothesis	J trace	J max
r = 0	16.4954*	12.7269
r = 1	3.7685	3.7685
HEALTH- INDUSTRIALS
Null hypothesis	J trace	J max
r = 0	16.1833*	9.6312
r = 1	6.5521*	6.5521
MATERIALS- HEALTH
Null hypothesis	J trace	J max
r = 0	17.0040*	12.4725*
r = 1	4.5315*	4.5316*

Notes: J Trace and J max indicate the statistics of Johansen's cointegration model (1988).

Rejection of the null hypothesis of no cointegration at the 5% is denoted by *.

Regarding the communication services and the consumer discretionary sector, the cointegration with the healthcare sector can be explained by two points: first, the increasing demand for communication tools during the confinement, and second, the rise in discretionary expenses resulting from the lockdown and possible rise in the level of savings of households. Regarding materials, the market saw an important rise in some safe-haven assets such as gold, or essential mining materials, such as copper, due to supply shortages and lockdowns. Concerning the industrial and financial sectors, they both registered a sharp decline during confinement and then started to rise slowly.

Therefore, the birth of new relationships with the healthcare industry in response to the COVID-19 crisis reflects the contagion's occurrence from the healthcare sector to other sectors. This finding indicates the importance of the healthcare sector's impact on the progress of society as a whole Chen (2016). The findings of this study align with those of Chen et al. (2017), who highlight the role of healthcare stock as a predictor of short and long stock prices.

The results of the bivariate cointegration test enabled us to estimate the VECM for each of the above pairs of sectors. The analysis of the VECM estimation yielded two interesting results: first, we noticed the presence of a significant and negative adjustment term for all pairs under question, and second, the value of the adjustment term is significantly higher for the financial sector, suggesting a rapid mean reversion toward the healthcare sector during the COVID-19 crisis.

Table 7 shows the short-and long-run causality tests between the healthcare sector and other sectors. We can observe short-and long-run bidirectional relationships between industrial and consumer discretionary. On the other hand, there are short-and long-run relationships between healthcare and financials. Regarding materials and healthcare, we notice short-and long-run unidirectional relationships.

Table 7

Short- and Long-Run Causality in VECM.

Causality test	F-statistic
Test for Materials causes health	2.0129**
Test for Materials long-run causing health	6.6503**
Test for health causes Materials	1.1020
Test for health long-run causing Materials	1.8370
Test for Industrials causes health	1.5944
Test for Industrials long-run causing health	6.6025**
Test for health causes Industrials	0.8894
Test for health long-run causing Industrials	2.9237*
Test for communication services causes health	0.6815
Test for communication services long-run causing health	0.0387
Test for health causes communication services	0.5054
Test for health long-run causing communication services	0.0053
Test for Financial causes health care	1.4605
Test for Financial long-run causing health	8.9605***
Test for health causes Financial	0.9352
Test for health long-run causing Financial	4.6994**
Test for consumer discretionary causes health	1.8020*
Test for consumer discretionary long-run causing health	9.6476***
Test for health care index causes consumer discretionary	0.9487
Test for health care index long-run causing consumer discretionary index	2.8839*

Notes: The above table shows the results of causality tests between the health sector and the other sector indexes during COVID-19-period. It contains the statistics of Causality in VECM.

Rejection of the null hypothesis of no Causality at ***, ** and * is denoted by 1%, 5% and 10% levels of significance, respectively.

Conclusions

This study explored sectoral index co-movements in the US healthcare index during the COVID-19 pandemic. Our findings demonstrate that the pandemic has had a significant impact on the cointegration of sector indexes. This investigation also shows that sector co-movements appeared during the COVID-19 period. In line with previous studies (Baker et al., 2020; Mazur et al., 2021), we argue that lockdown impacts the tendency of sector price indexes. This research area is relevant and significant, providing a better understanding of the relationships between different sector indexes. In line with Phylaktis and Xia (2009), we can conclude that identifying sectoral contagion during this pandemic has great importance for portfolio diversification. Our findings also have policy implications for regulators, investors, and other market stakeholders. With co-movement assessment tools, policymakers can design appropriate interventions to moderate risk from the co-movement of financial assets. Understanding the sectoral co-movement and timely healthcare policy decisions during a pandemic is also critical for investors, portfolio managers, policymakers, and other market participants, allowing stakeholders to be better prepared to adopt strategies for successfully limiting risks during future health crises.