Changes in social behavior and impacts of the COVID-19 pandemic on regional housing markets: Independence and risk.

This paper explores changes in social behavior since the start of the COVID-19 pandemic, which are characterized by reduction in relocation, mobility, and community engagement, and how the correlations between regional housing markets are affected by these changes. Because changes in mobility and engagement are the most apparent in large cities, the present study calculates the independence indicator of regional housing markets in the 50 largest metropolitan statistical areas (MSAs) in the United States and determines their relationship with Mobility and Engagement Index values. The empirical results show that as mobility and community engagement decline in a certain area, housing market fluctuations become more independent, indicating correlations between regional housing markets in the US might decrease after the COVID-19 outbreak. This paper also finds that there are more MSAs having significantly decreased in volatility since the outbreak of the pandemic. This paper provides evidence indicating that housing markets may be impacted differently by the COVID-19 pandemic than other asset markets, particularly stock markets. Changes in mobility and engagement can be used as an indicator to assess whether the correlation between regional housing markets would decline, which means that, compared with financial instruments, more factors from real aspects need to be considered when determining the changes in real estate affected by the epidemic.

Introduction

The COVID-19 pandemic has exerted unprecedented global impacts. The rapid spread of this major infectious disease has affected human health, education across multiple levels, social and financial market stability, and economic development. To prevent any opportunity that may increase the number of COVID-19 infections, national governments have implemented social distancing measures with regard to gathering in public spaces, and some have directly issued stay-at-home orders. These government-mandated closures and the reduction in social behavior that is due to fear of infection have greatly reduced economic activity, as well as interpersonal contact. In essence, the COVID-19 pandemic not only poses a major threat to public health; widespread fear and the imposition of various forms of control have had impacts on the social, educational, financial, and economic domains. The specific degree of impact to each domain requires urgent investigation. This paper focuses on how decline in relocation, going out, and community engagement affect regional housing markets during the pandemic.

The impact of the COVID-19 pandemic and epidemic prevention measures on community engagement should be massive, but how quantifying this impact is difficult. To gain insight into the economic impacts of the COVID-19 pandemic, Atkinson et al. (2020) use data from the SafeGraph Inc. to establish a Mobility and Engagement Index (MEI). On the basis of geolocation data collected from mobile devices, SafeGraph provides information on spatial behavior in an effort to help communities face the impacts of COVID-19. In constructing the index, the researchers conduct principal component analysis on the amount of time mobile phone users spend at and away from home, as well as on distance from home, comparing the results against pre-pandemic averages (set as 0). The study reports a considerable drop in the MEI in mid-March, consistent with the dramatic decline in economic activity at the time.

The unprecedented decline in the agglomeration of economic activities under the pandemic brings changes to economic variables over multiple levels, in particular, its impact on the housing market should be concerned. Previous Studies (Meen, 1999, Jones and Leishman, 2006, Gupta and Miller, 2012, Tsai, 2015, Seo and Kim, 2020) have well documented that correlations between regional housing markets are mainly attributable to regional mobility behavior that lead to co-movements and mutual convergence in housing prices across regions, phenomena considered ripple or spillover effects. The COVID-19 pandemic has not only prevented overseas travel; domestic travel has also been restricted as people are forced to work or study from home. Cross-regional housing rentals and purchases, the original product of from people traveling across regions for work or study, have also decreased substantially, potentially reducing the correlations between regional housing markets. Therefore, the present study determines whether the drop in the MEI, beyond reflecting the serious decline in the agglomeration of economic activities, affects the correlations (or lack thereof) between these markets.

Because changes to mobility and engagement are the most apparent in large cities, the top 50 metropolitan statistical areas (MSAs) in the United States are selected as the study sample. Weekly housing price returns are used to establish a regional market independence index based on the method of market connectedness proposed by Diebold and Yilmaz (2014). And the relationship between the MEI and a regional market independence index has been discovered to test whether or not when regional mobility and community engagement decrease, the housing market of that region becomes less correlated with other housing markets. Weaker correlations between regional housing markets might project a decline in the systematic risk in the entire US housing market. Thus, changes in risk are reexamined in the present study. By controlling the impact of the overall US housing market on the risk of the regional housing markets, this paper further compares the risk changes of the regional housing market before and after the epidemic. By this approach, this paper presents evidence that the correlation and risk of the US regional housing market can be affected by the COVID-19 pandemic. Moreover, the estimated results show that the epidemic’s impact on the housing market is different from that of other financial markets, which means that more and various aspects of the epidemic’s effects on the housing market need to be explored in the future.

The current literature focuses on the influence of the COVID-19 pandemic on markets’ correlation or risk, mainly by observing the financial market or commodity market phenomenon (i.e., Chevallier, 2020, Banerjee, 2021, Akhtaruzzaman et al., 2021, Umar et al., 2021, O’Donnell et al., 2021, Sifat et al., 2021, Salisu et al., 2020, Liu et al., 2022, Yarovaya et al., 2022).  Research on the housing market has primarily focused on changes in people’s relocation behavior (Cheung et al., 2021), especially in urban and suburban preferences (Duca et al., 2021, Ramani and Bloom, 2021, Liu and Su, 2021, Nygaard and Parkinson, 2021).  Therefore, The results of this study can supplement the literature.

Specifically, this paper makes the following contributions to the literature on the COVID-19 pandemic: (1) this paper is one of the few studies that explores the correlation between the MSAs’ housing markets under the epidemic (2) this paper proposes that, affected by the COVID-19 pandemic, the risk changes of the housing market are different from those of other assets (3) this paper proposes and validates the reasons why correlations of the inter-regional housing market are affected by the COVID-19 pandemic (4) this paper shows that the MEI can be used as an indicator to observe the interregional correlations in the housing market. Subsequent research is suggested to explore further the fact that, after the COVID-19 outbreak, the housing market may show different volatility characteristics from other assets after being affected by actual economic activities.

Section 2 presents a review of studies on the effects of the COVID-19 pandemic, centering on articles concerning impacts on financial markets, economic activity, and the macroeconomic performance. Section 3 summarizes the research background and the methods by which the regional housing market correlations are examined. Section 4 discusses the results, and Section 5 presents the conclusion and suggestions.

Literature review

The impacts of COVID-19 pandemic

Cases of COVID-19 first appeared as early as the end of January 2020, but the United States confirmed its first cluster of cases in mid-February of the same year.1 As the number of cases increased, governments in numerous countries began implementing regional travel restrictions. Canada implemented travel restrictions and enacted provincial states of emergency in mid-March. Italy also announced city shutdowns in March. These policies do limit the spread of the virus; as Fadinger and Schymik (2020) note, working from home is relatively effective in reducing infection risk, and that regions have fewer COVID-19 cases if their industrial structure supports a large number of people working from home. Nevertheless, these various restrictions strongly affect mobility, consumer behavior, and study and work patterns, exerting substantial impacts on the economy.

A study by Fatmi (2020) explores changes to people’s daily out-of-home travel activities, in-home activities, and long-distance travel under Canada’s travel restrictions from March 24 to May 9, 2020. The number of outings decreased by more than 50%. In an investigation of how household consumption is affected by COVID-19, Baker et al. (2020a) report that after March 20, when the number of cases increased and as local governments began issuing stay-at-home orders, drastic changes in consumption patterns of households occurred, a phenomenon observed in every state. Such changes are particularly marked in states with rigorous stay-at-home orders. Fadinger and Schymik (2020) study the effects of working from home on infection risk and economic conditions in Germany, noting that city shutdown significantly lowers the regional output if a region has a smaller percentage of people who can work from home. Leger (2020) indicates that 50% of Canadians might need to work from home because of COVID-19.

From their interviews with companies of varying sizes, Bartik et al. (2020) report that COVID-19 increases the health risk of working in an office or factory, and that consequently, more employees choose to work remotely. More industries gradually switch to production activities that involve remote operations, and some companies with over a third of employees working remotely express the view that teleworking will become increasingly common after the pandemic ends.

The findings from the studies discussed thus far indicate that the impacts of the COVID-19 pandemic extends beyond 2020; even after the pandemic ends, economic structures will not remain unchanged because of changes in behavioral and industrial patterns. Baker et al. (2020b) assert that no other infectious disease has affected financial markets as profoundly as COVID-19, arguing that government bans on commercial activity and social distancing requirements constitute two pivotal factors in service-oriented economies that explain why the pandemic has such strong effects on the US stock market.

Similarly, Ludvigson et al. (2020) affirm that the COVID-19 outbreak is unprecedented and significantly affects the US economy. They quantify the effects of costly natural disasters in US history and estimate the impacts of COVID-19 on the macroeconomic performance. They then establish a costly disaster series on the basis of the economic losses from major natural disasters from 1980M1 to 2019M12 and estimate their effects on the economy and on various types of uncertainty. However, major natural disasters mostly affected specific regions and only for a short period of time, whereas the effects of COVID-19 are global in scope and have lasted a long period of time. The researchers ultimately predict a cumulative loss of up to 12.75% in industrial output and a drop in service sector employment by 17% after 10 months. Furthermore, the monthly impacts are projected to continue accumulating; for instance, if employment in the service sector falls by approximately 6% each month, the uncertainty in macroeconomic performance becomes comparably high.

Numerous studies have explored the effects of COVID-19 on financial markets and the macroeconomic performance. Ashraf (2020) and Barrero et al. (2020) study its impacts on the macroeconomic performance; Al-Awadhi et al. (2020), Cox et al. (2020), Salisu and Vo (2020), and Ding et al. (2020) study stock market reactions; Gharib et al. (2020) study the prices of gold and crude oil; and Conlon and McGee (2020) study the hedging effects of Bitcoin. However, studies addressing effects on housing markets are few, perhaps because the liquidity is relatively low and does not reflect informational changes as rapidly as other financial markets (e.g., stock markets). Stock markets and other financial markets only respond to extreme economic setbacks, but because real estate is both a consumer good and a capital good, the housing market experiences long-term impacts of changes in people’s residential behavior caused by COVID-19 and its effects on mobility, consumer behavior, and study and work patterns. Therefore, the present study centers on how changes in social behavior and reduced relocation, mobility, and community engagement during the COVID-19 pandemic affect correlations between regional housing markets. Changes in the correlation of regional housing markets result in changes in systemic risk, which is an essential factor in house price volatility behaviour and housing return volatility, especially when significant events occur (Zhu et al., 2013, DeFusco et al., 2013, Wang and Zong, 2020, Bago et al., 2021). Therefore, after verifying the correlation changes in the regional housing market, this paper will examine whether the regional housing return volatility (risk) will also change. This estimation can be compared with the current findings that financial markets and commodity markets are affected by COVID-19.

There is a large body of literature demonstrating that the COVID-19 pandemic or its regulation can lead to increased risks in financial markets, such as: Al-Awadhi et al. (2020), Baker et al. (2020a), Cao et al. (2021),  Engelhardt et al. (2021), Gil-Alana and Claudio-Quiroga (2020), Gormsen and Koijen (2020), Harjoto et al. (2021), Liu et al. (2020), and Phan and Narayan (2020). There are also many studies illustrating the rise in stock return volatility, for example, Corbet et al. (2020), Mazur et al. (2021), Haroon and Rizvi (2020), Sharma (2020), Zaremba et al. (2020), and Ftiti et al. (2021). Some of these studies, such as Akhtaruzzaman et al. (2021) and Corbet et al. (2020), show that stock return volatility increases due to the contagion effect between markets. However, there are few studies on housing return volatility. No literature shows that the housing market’s performance may be different from that of the stock market, so the results of this paper can supplement the lack of empirical evidence in the relevant literature.

The connections among regional housing markets and their influences on the market risk

In general, correlations are present across regional housing markets within the same country, perhaps because regional mobility behavior leads to co-movements in regional housing prices—that is, ripple or spillover effects in regional housing markets, phenomena first identified in the United Kingdom by Meen (1999), and then in multiple empirical studies that demonstrate significant correlations between UK regional markets (Cook and Thomas, 2003, Cook, 2005, Cook and Watson, 2016, Tsai, 2015). Overall, the United States has fewer studies on the correlations in regional housing markets than the United Kingdom, but some publications discussing the independence of and correlations between regional housing markets appeared in the United States after the internet bubble burst, which caused a crisis in the housing market ascribable to defaults on subprime mortgages.

These findings contribute to the determination of whether high levels of systematic risk are present in the US housing market. For example, using data from the mortgage loan company Freddie Mac from the first quarter of 1978 to the second quarter of 2008, Gupta and Miller (2012) explore the relationships between housing prices in Los Angeles, Las Vegas, and Phoenix. Miao et al. (2011) use the Standard & Poor’s CoreLogic Case–Shiller Home Price Indices to examine the dependence between housing markets in 16 US MSAs from January 1989 to June 2006. Kallberg et al. (2014) analyze the co-movement of housing prices on the basis of the Case–Shiller indices from 1992 to 2008 in 14 US cities. These studies report varying levels of evidence on correlations between housing markets in MSAs in the United States.

According to results from their analysis of data on housing indices covering 363 US cities from the first quarter of 1975 to the first quarter of 2013, Cohen et al. (2016) indicate that the divergence in housing prices between cities rose after the subprime mortgage crisis in 2007. Using housing prince index returns in 20 MSAs from January 1991 to April 2018, Tsai and Lin (2019) compare changes in housing market impacts in the United States by estimating market connectedness before, during, and after the subprime mortgage crisis. The market-associated systematic risk is correlated with the subprime mortgage crisis and the bankruptcy of Lehman Brothers, indicating that the connectedness of metropolitan housing markets may contain warning signs of market risk.

As mentioned, studies confirm that correlations between regional housing markets are related to the financial crisis (Cohen et al., 2016, Tsai and Lin, 2019). Earlier studies by Miao et al. (2011) and Canarella et al. (2012) also present evidence of structural changes in the correlations between housing markets in MSAs in the United States. However, these studies do not postulate which factors lead to these changes. Because the correlations of regional housing markets stem from mutual influence between the behavior of people in different regions, the COVID-19 pandemic exerts historically unprecedented changes in public behavior, especially with regard to social interactions, thereby greatly reducing relocation, mobility, and engagement in community activities. Therefore, the present study investigates how these behavioral changes affect correlations between regional housing markets.

Research method

Based on the literature mentioned above, the paper attempts to test whether the pandemic causes regional housing markets to become more independent. Housing price index returns for 50 MSAs in the United States are used to establish an index of regional market independence. Many studies have examined the market independence and the correlation between regional markets using market connectedness (e.g., Antonakakis et al., 2018, Tsai, 2018). To determine the connections among the regional housing markets, this study first quantifies the market connectedness based on the methodology of Diebold and Yilmaz (2014), which measured the direction of information transmission among the markets. The method is briefly described as follows.

Housing price index returns for 50 MSAs in the United States are used to estimate the Vector Autoregression (VAR) model. Let the vector  represents housing market returns of the different regions  where  is the coefficient matrix and  is the residual matrix. Then, the orthogonal shocks are obtained by estimating the variance decompositions (VDs) of the VAR model: . These shocks are followed an N-dimensional covariance-stationary datagenerating process:  and .

Contemporaneous aspects and dynamic aspects of connectedness are summarized in . Based on a generalized VD (GVD) framework proposed by Koop et al. (1996), Pesaran and Shin (1998), and Del Negro and Schorfheide, 2011, Diebold and Yilmaz, 2014 pointed out that total connectedness is robust to the ordering of the variables in the VAR model. The following briefly introduces how total connectedness and the indicator of market independence are calculated.

Let  denote the -th H-step GVD component of the VAR model it can represent the pairwise directional connectedness from  to  and the “variance decomposition matrix” can be denoted by :  where  is the standard deviation of the error term for the th equation and  is the variance matrix for the error vector. It contains a selection vector  with th element unity and zeros elsewhere and the coefficient matrix .

Since  denotes the fraction of variable ’s H-step forecast error variance due to shocks in variable . We can use the forecast error variance () as an indicator to observe the impact of the influence from regional market  on the housing market returns in regional market . Numerous studies (e.g., Bostanci and Yilmaz, 2020, Ameur and Louhichi, 2021) have used this framework to evaluate the correlations between asset returns. Specifically, estimations of rolling data are used to obtain dynamic, time-varying correlations, thereby determining the influence of different events on the correlations. Tsai (2019) defines the impacts on the th MSA that affect the same market’s housing price returns () as the self-influence of the housing market. Because it can be used to measure the lack of correlations between a regional housing market and other markets, it is taken as the measure of the independence indicator () in the present study. In this current paper, housing price returns are deconstructed according to whether they are self-affected or affected by other MSAs, with the proportions attributable to the self-influences used to assess the indicator of independence. Moreover, the dynamic independence indicator is assessed using rolling data, and its correlation with the index for measuring mobility and engagement are then analyzed to determine whether mobility and community engagement affect the segmentation of regional housing markets.

When the correlation between the housing markets in various regions decreases, it symbolizes that the systemic risk or contagion between markets has declined. Since the linkage risk (systematic risk) between markets is one of the factors of return volatility (risk), this paper continues to examine whether the regional housing market return volatility has changed significantly after the COVID-19 outbreak. Hence, the dynamic risk of fluctuations in housing price returns needs to be estimated. Numerous studies have used the extended model of generalized autoregressive conditional heteroskedasticity (GARCH) to estimate the risk of fluctuations in housing price returns (e.g., Li, 2012, Lee and Reed, 2014). The present study also uses the GARCH model to estimate the conditional variance of housing price returns. In addition, to control the connecting risk of influence by the national housing market, the housing price returns of the US market are added to the conditional variance formula (), where  and  denote the regional housing price returns and the national housing price returns, respectively. The model is presented as follows:  The above model can estimate the dynamic risk of fluctuations in housing price returns for each MSA, and the conditional variance () has included the factors that affect the regional housing market by the overall market.

Empirical result

The values of the independence indicators of the sample, which as mentioned comprises the 50 largest MSAs in the US, are first estimated, after which their relationship with the MEI is determined. To measure short-term changes to the housing markets more accurately, housing price data (medians of transaction prices after adjustment) with higher frequencies are used. Specifically, weekly data from January 6, 2018, to March 27, 2021 from the Zillow Transaction and Assessment Dataset.2 The measure of mobility and engagement behavior is adopted the Dallas Fed Mobility and Engagement Index, which is scaled so that the average of January-February is zero, and the lowest weekly value (week ended April 11) is -100. Weekly data from January 4, 2020 to March 27, 2021 from the MEI are used and the weekly data corresponds to the last day of the calendar week.3

Fig. 1 presents the MEI values of the 50 MSAs, with lower values indicating lower mobility and community engagement. On January 4, 2020, the starting point of the MEI data, the pandemic had not yet become widespread in the United States, and no restrictions were yet in place. The outbreak occurred at the beginning of March, with most regions imposing restrictions starting March 14. Therefore, housing price averages are calculated before and after this point. Table A.1 presents the MEI averages of the 50 MSAs, as well as the median housing prices before and after the outbreak. It is observable that the median housing prices in each region rise after the outbreak. As shown in Fig. 2, housing prices continue to climb after mid-March of 2020.

Fig. 1

MEI.

Table A.1

Descriptive statistics.

MSAs	New York		Los Angeles	Chicago		Dallas	Philadelphia
HP_be	417334.49		646814.72	235340.08		267792.05	244064.50
HP_af	456644.33		715994.30	259858.96		296693.11	273912.87
MEI	−59.06		−50.76	−54.97		−50.10	−53.91
MSAs	Houston		Washington, DC	Miami		Atlanta	Boston
HP_be	238657.88		407894.22	285779.60		247713.59	453884.71
HP_af	262890.39		445214.50	326062.81		281959.37	512226.20
MEI	−47.62		−61.08	−50.58		−43.09	−58.61
MSAs	San Francisco		Detroit	Riverside		Phoenix	Seattle
HP_be	804833.65		184852.68	373531.81		280146.81	483197.38
HP_af	877812.54		210104.04	423242.78		326339.46	551441.78
MEI	−63.38		−48.24	−39.50		−39.39	−46.51
MSAs	Minneapolis		San Diego	St. Louis		Tampa	Baltimore
HP_be	269583.12		583701.03	184169.91		227123.97	295682.69
HP_af	306450.48		644662.04	207322.54		259520.65	312270.98
MEI	−57.65		−51.12	−43.90		−38.52	−53.63
MSAs	Denver		Pittsburgh	Portland		Charlotte	Sacramento
HP_be	420964.39		169319.00	393899.23		256838.77	411441.70
HP_af	460557.80		190038.70	436942.96		291009.13	468250.61
MEI	−45.32		−44.52	−43.08		−41.57	−45.35
MSAs	San Antonio		Orlando	Cincinnati		Cleveland	Kansas city
HP_be	226601.48		260582.70	180783.23		146601.03	217115.63
HP_af	251501.26		283013.87	206063.00		168286.63	249657.02
MEI	−44.76		−40.38	−43.81		−43.02	−41.99
MSAs	Las Vegas		Columbus	Indianapolis		San Jose	Austin
HP_be	295042.10		205145.74	186179.06		1058607.43	313172.09
HP_af	320598.39		230967.69	222369.35		1130942.43	360702.33
MEI	−38.91		−44.35	−38.95		−70.59	−55.71
MSAs	Virginia Beach		Nashville	Providence		Milwaukee	Jacksonville
HP_be	248006.89		288421.77	277912.93		202819.09	246210.23
HP_af	273401.15		321552.94	320860.96		219638.57	270506.98
MEI	−40.73		−43.93	−46.99		−45.23	−34.00
MSAs	Memphis		Oklahoma city	Louisville		Hartford	Richmond
HP_be	178697.49		178639.41	186677.27		223447.77	252944.37
HP_af	212192.70		205649.78	210762.24		249439.48	281811.50
MEI	−38.39		−35.74	−42.20		−53.47	−45.82
MSAs	New Orleans		Buffalo	Raleigh		Birmingham	Salt Lake city
HP_be	203911.03		157916.57	286594.04		200708.90	325139.76
HP_af	227873.33		180824.09	314613.61		225939.11	381900.98
MEI	−40.29		−44.04	−46.60		−38.46	−35.38

Notes:  denotes the pre-outbreak mean of house prices.  denotes the post-outbreak mean of house prices.

Fig. 2

Sale price.

[Insert Appendix Table 1]

Although several studies report that the COVID-19 pandemic has brought the global stock market and other financial markets to collapse (Al-Awadhi et al., 2020, Ali et al., 2020, Ashraf, 2020), this type of immediate impact does not seem to apply to housing markets. This may be attributable to the premise posited in the present study that unlike financial products, real estate has properties of both consumer goods and investment assets. Therefore, studies on housing markets should not focus solely on changes in housing prices after the start of the pandemic but should also examine how the pandemic has changed public behavior, and then determine the effects of those changes on housing markets. Therefore, the present study explores the effects of mobility and community engagement on housing markets by calculating the independence values of the markets and delineating their relationships with the MEI.

First, according to the GVD framework proposed by Diebold and Yilmaz (2014), the forecast error variance (), which represents the influence of impacts to the th MSA on the housing prices in the th MSA is obtained through variance decomposition analysis. Table 1 contains the estimated results of , providing the information of the influence of region  on region , that is , for various .

Table 1

Spillover result.

Notes: The 50 largest metropolitan statistical areas.

We take the first city (New York) as an example to illustrate the meaning of the values in Table 1. Since the influence of the first city on the second city (Los Angeles) is 4.4, and the influence of the second city on the first city is 3, the net influence of New York on Los Angeles is 1.4. Besides, the forecast error variance of 28.8 is that New York is affected by its own factors, so this value can be used to observe the fluctuations of a city that is not related to other cities, that is, as a measure of independence. Table 1 also shows the smallest city in the sample, Salt Lake City, is the most independent of all cities.

Table 1 roughly shows that, based on the 27th city (Orlando), the cities on the left (right) side of the table have a larger (smaller) impact on other cities, as the darker colors on the left side are predominant. Since the 50 MSAs are sorted by city size, the above general phenomenon shows that, in general, larger cities have a greater impact. However, this is only a general observation based on the 50 cities, and the impact of each city should be compared to its estimated . It is possible that some cities have a greater impact due to their geographic location, such as Chicago, Dallas, Houston, and San Francisco. For example, Kallberg et al. (2014) found a high degree of correlation between Chicago and San Francisco and other cities. On the other hand, the largest city, New York, does not have as much influence on other cities as the cities mentioned above. This may be due to the fact that New York has the same characteristics as London in the UK (Butler and Lees, 2006, Badarinza and Ramadorai, 2018, Webb et al., 2021),  being the financial centre or superstar city of a country is more influenced by foreign capital, resulting in most changes in housing prices are not closely related to the housing market in other regions of the country.

The paper does on to measure dynamic connectedness to estimate the dissociation of among regions in different periods. Fig. 3 presents the independence values of the 50 MSAs. Some major cities, such as Dallas and Detroit, only exhibit significant increased segmentation from other regional housing markets after July 2020. In some other major cities, such as Los Angeles and Milwaukee, market independence begins rising gradually from February and March intermarket correlations reaches a nadir in April and May, after which market independence declines again. To observe the relationship between market independence and the MEI, the correlation between these two indices is first calculated.

Fig. 3

Market independence.

Table 2 presents the causal relationships between market independence and the MEI in the 50 MSAs, which are established on the basis of panel data from the first week of 2020 to the end of the data. To obtain more precise results, estimates in Table 2 both factor in and do not factor in the heterogeneity of the regional housing markets. The results indicate that market independence affects mobility and community engagement, this may be because they are both influenced by the same behaviors of people. And whether or not housing market heterogeneity is considered, the MEI values lead market independence. When heterogeneity is taken into account, this effect becomes more significant. Those results indicate that mobility and community engagement can significantly affect market independence.

Table 2

Pairwise Granger Causality Tests.

Pairwise Granger Causality Tests
Null Hypothesis:				F-Statistic	p-value
Iso does not Granger Cause MEI				19.9950	0.0000
MEI does not Granger Cause Iso				3.0932	0.0149
Pairwise Dumitrescu Hurlin Panel Causality Tests
Null Hypothesis:		W-Stat.		Zbar-Stat.	p-value
Iso does not homogeneously cause MEI		6.6267		5.3639	0.0000
MEI does not homogeneously cause Iso		5.5629		3.1794	0.0015

Notes: This table presents the causal relationships between market independence and the MEI in the 50 MSAs.  denotes the independence indicator. Numbers in bold represents significant at 5% level.

Table 2 shows that these two variables related to people’s social behavior are mutual lead and lag. This paper continues to observe the current correlation between mobility and community engagement and market independence and the leading response of market independence in more detail. This paper uses the independence indicator as the dependent variable, the MEI and its lagged terms as the explanatory variables, and estimates the Panel regression, which is presented in Table 3. Table 3 shows that neither the current nor previous MEI can significantly affect market independence. This may be due to considerable differences in MEI values between major cities. Therefore, quantile regression is then performed to estimate how market independence at different levels is affected by the MEI (Table 4).

Table 3

Panel regression.

Dependent variable: Iso	Coefficient	Std. error	t-Statistic	p-value
MEI_t	0.0022	0.0052	0.4257	0.6703
MEI_t−1	0.0062	0.0052	1.1981	0.2310
Constant	15.9243	0.1213	131.2610	0.0000

Notes: This table presents the panel regression estimating the influence of market independence on the MEI in the 50 MSAs.  denotes the independence indicator. Numbers in bold represents significant at 5% level.

Table 4

Quantile process estimates.

Dependent variable: Iso	Quantile	Coefficient	Std. error	t-Statistic	p-value
MEI_t	0.10	−0.0747	0.0095	−7.8720	0.0000
	0.20	−0.0918	0.0104	−8.7966	0.0000
	0.30	−0.1174	0.0113	−10.4327	0.0000
	0.40	−0.1513	0.0122	−12.3636	0.0000
	0.50	−0.1732	0.0124	−13.9180	0.0000
	0.60	−0.1993	0.0117	−16.9882	0.0000
	0.70	−0.2033	0.0125	−16.3021	0.0000
	0.80	−0.2388	0.0176	−13.5714	0.0000
	0.90	−0.2727	0.0172	−15.8643	0.0000
MEI_t−1	0.10	−0.0710	0.0095	−7.4391	0.0000
	0.20	−0.0842	0.0101	−8.3084	0.0000
	0.30	−0.0859	0.0108	−7.9728	0.0000
	0.40	−0.0788	0.0117	−6.7248	0.0000
	0.50	−0.0827	0.0121	−6.8239	0.0000
	0.60	−0.0849	0.0119	−7.1599	0.0000
	0.70	−0.1194	0.0126	−9.4531	0.0000
	0.80	−0.1335	0.0174	−7.6714	0.0000
	0.90	−0.1729	0.0180	−9.6184	0.0000

Notes: This table shows the effect of the MEI values of the current and the preceding period on the market independence indicator across various quantiles. Numbers in bold represents significant at 5% level.

Table 4 shows that the MEI values of the current and the preceding period significantly affect market independence across various quantiles. Furthermore, the market independence of higher quantiles is subject to greater influence from the MEI. Fig. 4 presents the coefficients of MEI effects on market independence at different quantiles. Both the MEI values of the current period and the preceding period significantly influence independence, with greater independence at higher quantiles subject to greater influence from the MEI. Market independence at the highest quantile is under three times the influence of that at the lowest quantile. This indicates that greater independence in a regional housing market is more likely to be caused by low mobility and community engagement.

Fig. 4

Coefficients of the quantile estimation.

This contagion in (i.e., correlations between) housing markets during the pandemic is the opposite of the contagion in typical financial markets. Studies (e.g., Cohen et al., 2016, Tsai and Lin, 2019) report that contagion during financial crises strengthens correlations between housing markets in major US cities; moreover, it increases the systematic risk in the national housing market. By contrast, the present study finds that the effect during the COVID-19 pandemic causes housing markets between major US cities to become less correlated and more independent. Whether these influences also change the risk in the US regional housing markets are discussed in the remainder of this paper.

The sampling period starting from 2018 is used to estimate housing market volatility (). Because this volatility can determine the risk that regional housing markets are affected by the performance of the national housing market, it is used to measure systematic risk in the present study.

Table 5 lists the estimated coefficients in the conditional variances, of which the  of some MSAs is significant, indicating that the risk of some cities is significantly affected by the national housing market. Fig. 5 shows the conditional variances, which are the risks of every MSA calculated using the conditional variance model, respectively. The start of the COVID-19 outbreak (March 14, 2020) is marked with a red line. The differences in risk before and after the outbreak is observable. In short, a higher risk of fluctuations in regional housing markets in the week of March 14, 2020, is noted in many of the MSAs, but that risk has decreased since then. Table 6 also presents the test statistics of the means and differences of the volatilities (i.e., pre-outbreak means minus the post-outbreak means). In 7 MSAs, the risk rises after the outbreak, but in 24 others, it declines. In essence, the risk in most regional housing prices falls significantly after the outbreak. The results here show that affected by the epidemic, the changes in the risk of the housing market are different from those of other assets.

Table 5

Estimated coefficients of conditional variance over the whole sample period.

	New York		Los Angeles	Chicago		Dallas	Philadelphia
θ_0×10^3	0.0039		0.0706	0.0151		0.0161	0.0622
θ_1	0.1470		0.0682	0.3290		0.1407	0.2199
θ_2	0.7442		−0.5802	0.4938		0.6565	0.2710
θ_3	0.0004		0.0026	0.0013		−0.0015	−0.0023
	Houston		Washington, DC	Miami		Atlanta	Boston
θ_0×10^3	0.0162		0.0034	0.0008		0.0157	0.0770
θ_1	0.1750		0.0918	−0.0290		0.0389	−0.1350
θ_2	0.6575		0.8973	1.0238		0.7445	0.5828
θ_3	−0.0013		−0.0015	−0.0004		−0.0009	−0.0032
	San Francisco		Detroit	Riverside		Phoenix	Seattle
θ_0×10^3	0.0257		0.0345	0.0276		0.0204	0.0116
θ_1	0.1032		0.0674	−0.0692		−0.1395	−0.0123
θ_2	0.7910		0.7747	0.3460		0.5567	0.9406
θ_3	−0.0012		0.0006	0.0001		−0.0008	−0.0029
	Minneapolis		San Diego	St. Louis		Tampa	Baltimore
θ_0×10^3	0.0472		0.0070	0.0321		0.0032	0.0092
θ_1	0.1660		−0.0989	0.1495		−0.0790	−0.0889
θ_2	−0.0449		1.0229	0.6203		1.0228	1.0321
θ_3	0.0009		−0.0007	−0.0011		−0.0006	−0.0009
	Denver		Pittsburgh	Portland		Charlotte	Sacramento
θ_0×10^3	0.0121		0.0300	0.0650		0.0296	0.0081
θ_1	0.1230		0.0931	0.0114		0.0510	0.0978
θ_2	0.5262		0.8090	−0.6581		0.6990	0.7634
θ_3	0.0009		−0.0044	0.0028		−0.0019	−0.0005
	San Antonio		Orlando	Cincinnati		Cleveland	Kansas city
θ_0×10^3	0.0095		0.0127	0.0162		0.0375	0.0849
θ_1	−0.0231		0.0321	0.0651		0.0647	−0.1525
θ_2	0.9072		0.7269	0.8734		0.7628	0.5372
θ_3	−0.0013		0.0002	−0.0028		−0.0026	0.0037
	Las Vegas		Columbus	Indianapolis		San Jose	Austin
θ_0×10^3	0.0037		0.0154	0.2600		0.2800	0.0371
θ_1	0.0346		−0.0967	0.0320		0.2143	0.2058
θ_2	0.8901		1.0278	−0.5809		−0.0416	0.5308
θ_3	−0.0005		−0.0015	−0.0049		0.0182	0.0008
	Virginia Beach		Nashville	Providence		Milwaukee	Jacksonville
θ_0×10^3	0.0257		0.0818	0.1200		0.1820	0.0170
θ_1	−0.0245		0.1976	0.0121		0.0872	0.2588
θ_2	0.7968		−0.2480	0.2773		0.2788	0.6040
θ_3	−0.0001		−0.0020	0.0043		0.0075	−0.0001
	Memphis		Oklahoma city	Louisville		Hartford	Richmond
θ_0×10^3	0.0333		0.0655	0.0370		0.2160	0.0280
θ_1	−0.1079		0.0493	−0.1684		0.0861	0.0294
θ_2	1.0262		0.5136	1.0197		0.0524	0.8255
θ_3	−0.0036		−0.0046	−0.0042		0.0019	−0.0029
	New Orleans		Buffalo	Raleigh		Birmingham	Salt Lake city
θ_0×10^3	0.0130		0.0505	0.0154		0.2720	0.0069
θ_1	−0.0750		−0.0872	−0.0975		−0.1196	0.0235
θ_2	1.0378		0.9239	0.9476		0.2272	0.9098
θ_3	−0.0028		0.0016	−0.0009		−0.0030	−0.0006

Notes: The table presents the estimated coefficients of regional housing price returns in conditional variance equation. Numbers in bold represents significant at 5% level. The estimated GARCH model: , , .

Fig. 5

Conditional volatility.

Table 6

Changes in conditional volatilities of regional housing markets.

	New York		Los Angeles	Chicago		Dallas	Philadelphia
VR_be×10^3	0.0414		0.0479	0.0758		0.0795	0.1211
VR_af×10^3	0.0338		0.0518	0.1082		0.0614	0.1134
diff	0.0077		−0.0039	−0.0324		0.0181	0.0078
p-value	0.0095		0.0866	0.0025		0.0139	0.3650
	Houston		Washington, DC	Miami		Atlanta	Boston
VR_be×10^3	0.0980		0.0980	0.0514		0.0723	0.1320
VR_af×10^3	0.0728		0.1031	0.0347		0.0644	0.1384
diff	0.0252		−0.0052	0.0167		0.0079	−0.0064
p-value	0.0024		0.4752	0.0000		0.0104	0.3120
	San Francisco		Detroit	Riverside		Phoenix	Seattle
VR_be×10^3	0.2261		0.2246	0.0384		0.0358	0.1749
VR_af×10^3	0.2146		0.2434	0.0388		0.0309	0.0785
diff	0.0114		−0.0188	−0.0004		0.0049	0.0964
p-value	0.3077		0.0307	0.4625		0.0016	0.0000
	Minneapolis		San Diego	St. Louis		Tampa	Baltimore
VR_be×10^3	0.0537		0.0774	0.1393		0.0855	0.2322
VR_af×10^3	0.0585		0.0516	0.1322		0.0345	0.1216
diff	−0.0048		0.0258	0.0071		0.0510	0.1107
p-value	0.0459		0.0000	0.3735		0.0000	0.0000
	Denver		Pittsburgh	Portland		Charlotte	Sacramento
VR_be×10^3	0.0346		0.259	0.0409		0.1252	0.0604
VR_af×10^3	0.0438		0.1845	0.0438		0.1015	0.0503
diff	−0.0091		0.0745	−0.0028		0.0237	0.0101
p-value	0.0000		0.0000	0.2546		0.0084	0.0092
	San Antonio		Orlando	Cincinnati		Cleveland	Kansas city
VR_be×10^3	0.0945		0.0539	0.2174		0.2225	0.1399
VR_af×10^3	0.0610		0.0533	0.2033		0.1873	0.1613
diff	0.0335		0.0006	0.0141		0.0351	−0.0213
p-value	0.0002		0.3805	0.1836		0.0047	0.0059
	Las Vegas		Columbus	Indianapolis		San Jose	Austin
VR_be×10^3	0.0492		0.3234	0.1651		0.3777	0.1412
VR_af×10^3	0.0375		0.1918	0.1596		0.3577	0.1414
diff	0.0117		0.1317	0.0055		0.0200	−0.0002
p-value	0.0000		0.0000	0.1846		0.5258	0.9903
	Virginia Beach		Nashville	Providence		Milwaukee	Jacksonville
VR_be×10^3	0.1096		0.0764	0.1745		0.3024	0.1385
VR_af×10^3	0.1151	0.0733		0.1837		0.3077	0.0948
diff	−0.0055		0.0031	−0.0092		−0.0053	0.0436
p-value	0.0000		0.4642	0.0512		0.5966	0.0009
	Memphis		Oklahoma city	Louisville		Hartford	Richmond
VR_be×10^3	0.4353		0.1431	0.2389		0.2553	0.1881
VR_af×10^3	0.3569		0.1209	0.2038		0.2502	0.1481
diff	0.0785		0.0222	0.0351		0.0051	0.0400
p-value	0.0275		0.0019	0.0581		0.3602	0.0001
	New Orleans		Buffalo	Raleigh		Birmingham	Salt Lake city
VR_be×10^3	0.2597		0.2963	0.1164		0.3037	0.0923
VR_af×10^3	0.1563		0.3888	0.0777		0.2951	0.0840
diff	0.1034		−0.0925	0.0387		0.0085	0.0083
p-value	0.0000		0.0000	0.0000		0.3103	0.0001

Notes: The table presents the test statistics of the means and differences in conditional volatilities. Numbers in bold represents significant at 5% level.

Conclusion

The severe effects of COVID-19 on people’s lives and on the global economy, both overall and with regard to financial markets, are the subject of numerous studies. Specifically, the impacts of the pandemic on connections among regional housing markets receive comparatively scant scholarly attention, perhaps because liquidity is lower in housing markets, which means that relatively few affected transaction data can be obtained.

Under the condition of limited data, we cannot directly infer the impact on housing markets from the impact of the COVID-19 pandemic on the stock markets that previous studies well documented. This is because, real estate possesses characteristics of both capital goods and consumer goods, unlike stocks, which are purely investment assets. Therefore, the changes in study and work patterns might cause the effects of the pandemic on mobility and consumer behavior and should be incorporated to estimate the changes in residential behavior caused by these effects and the long-term consequences on housing markets. It is with this consideration that the present study investigates how changes in social behavior and reduced relocation, mobility, and community engagement during the COVID-19 pandemic affect segmentation between regional housing markets.

On the basis of the literature review, it can be inferred that the limited regional mobility under the COVID-19 pandemic leads to substantial reductions in housing rentals and purchases for school- or work-related reasons, potentially resulting in weaker correlations between regional housing markets. This is why the present study analyzes the relationship between housing market independence and changes in mobility and engagement. This paper puts forward a viewpoint: The COVID-19 outbreak has caused a reduction in relocation, mobility, and community engagement that may affect the performance of the housing market in various regions, resulting from the different characteristics of housing markets compared to other asset markets. Consequently, the primary purpose of this paper is to study whether the reduction in relocation, mobility, and community engagement caused by the COVID-19 outbreak will also reduce the relationship between the inter-regional housing markets.

The 50 largest MSAs in the United States are selected as the sample because changes in mobility and engagement are the most apparent in large cities. To reiterate, the market independence of these housing markets are calculated, after which the values are compared with MEI values to determine their relationships. The present results indicate that when mobility behavior and community engagement in a certain region decrease, its housing market fluctuations become more independent. Notably, this effect is nonlinear; greater independence in a regional housing market is more likely to be caused by low mobility and community engagement.

The correlation between the housing markets in various regions decreases, indicating the systemic risk or contagion between markets has declined. Since the linkage risk between markets is one of the factors of return volatility, this paper continues to examine whether the regional housing market return volatility has changed significantly after the COVID-19 outbreak; the results show that the risk in most regional housing prices falls significantly after the outbreak. Impacts on housing markets ascribable to the changes in connection caused by COVID-19 are the opposite of those reported in past studies. The contagion during the financial crisis contributed to stronger correlations between housing markets in the MEI in the United States. It increased the systematic risk in the entire US housing market. By contrast, the present results indicate that the influence of the COVID-19 pandemic caused the housing markets of major US cities to become less correlated and more independent, which also indicates significant reductions in the systematic risk in regional housing prices.

The present results suggest under the COVID-19 pandemic, changes in public behavior, such as working or studying from home, may drive structural changes to correlations between regional housing markets.

While there is a large body of literature documenting that financial and commodity market risks are affected by the COVID-19, questions remain as to what causes these effects. In this paper, we present a view to illustrate and validate that the correlation between regional housing markets might vary due to changes in social behavior (as measured by the MEI). Thus, in addition to showing that changes in housing market risk due to the epidemic are different from changes in risk for other assets, this paper contributes to the literature by presenting an instrumental variable that can be used to observe the impact of the COVID-19 on the housing market. Although the decline in the MEI may be a short term phenomenon, the COVID-19 epidemic may continue to change and other infectious diseases may affect people’s social behavior outside the home in the future, so the reasons for the COVID-19’s impact on different markets are worth exploring.

CRediT authorship contribution statement

I-Chun Tsai: Conceptualization, Methodology, Investigation, Data curation, Writing – original draft, Writing – review & editing.

Uncited References

Diebold and Yilmaz (2012)

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.