key: cord-0819409-xz6r58e4
authors: Bissoondoyal-Bheenick, Emawtee; Do, Hung; Hu, Xiaolu; Zhong, Angel
title: Learning from SARS: Return and Volatility Connectedness in COVID-19
date: 2020-10-16
journal: Financ Res Lett
DOI: 10.1016/j.frl.2020.101796
sha: 36d2c1eb34fa587d4718d15e25a1e9a43c4e3c26
doc_id: 819409
cord_uid: xz6r58e4

Using a sample of the G20 countries, we examine the impact of COVID-19 on stock return and volatility connectedness, and whether the connectedness measures behave differently for countries with SARS 2003 experience. We find that both stock return and volatility connectedness increase across the phases of the COVID-19 pandemic which is more more pronounced as the severity of the pandemic builds up. However, the degree of connectedness is significantly lower in countries with SARS 2003 death experience. Our results are robust to different measures of COVID-19 severity and controlling for a number of cross-country differences in economic development.

The spread of the Coronavirus Disease , which started in January 2020 has severely impacted the global financial markets over a short period of time. It is argued that no previous infectious disease, including the Spanish Flu (1900) has had such an influence on the financial markets (Baker et al., 2020) . The objective of this paper is to assess the impact of COVID-19 on the dynamic connectedness of stock return and volatility in global financial markets. More importantly, we assess whether return and volatility connectedness differ in countries with previous experience with the Severe Acute Respiratory Syndrome (SARS) in 2003. Our study is driven by both the rapid spread of COVID-19 and its severe effect on global financial markets.

We conjecture that there is greater heterogeneity in the timing and intensity of investors' responses to COVID-19 between a country (market)-pair with alike pandemic experience compared to a country pair without such prior experience. Accordingly, stock market connectedness in the former pair tends to be significantly lower than the latter pair. Our rationale is based on the behaviourist theory of investing, which incorporates elements of psychology to explain market imperfections. Ru et al. (2020) find supporting evidence for the imprint theory (see Marquis and Tilcsik, 2017) in behavioural bias of investors, such that investors with early experience on similar crises tend to react more quickly to COVID-19 than those without such imprints. Previous studies have documented that prior experience with similar events can affect individual's risk aversion (Guiso et al., 2015 and Bernie et al., 2017) and investments (e.g., Huang, 2019) . While investors should add stocks that minimise the risk of their portfolios, in practice investors choose stock which they are familiar with. A prominent example in the literature is the home bias in stock holdings (Kilka et al., 2000 , Huberman et al., 2001 and Wang et al., 2011 . Investors have richer knowledge of stocks in their home countries and hence prefer to exploit this to their advantage. This implies that perception about the situation is an important factor that drives the decision-making process (Nguyen et al., 2019) .

Our study contributes to the literature in several ways. First, we join the rapidly growing discussion of the economic impact of COVID-19 on global financial markets (Baker et al., 2020; Zaremba et al., 2020; Zhang et al., 2020) . Secondly, we contribute to the literature of connectedness and contagion by assessing the impact of COVID-19 on stock return and volatility connectedness . Thirdly, by analysing phases of the pandemic, we address how quickly return and volatility connectedness vary alongside the severity of the pandemic.

Last but not least, we contribute to the behavioural theory of markets by investigating how return and volatility connectedness change conditional on prior experience in SARS 2003 (Ru et al.. 2020 .

Our results can be summarised as follows: (1) similar to other crises, stock return and volatility connectedness increase as the severity of the pandemic builds up; (2) both return and volatility connectedness decrease among countries with experience in SARS 2003.

The remainder of the paper is as follows. Section 2 presents the data and modelling framework. Section 3 discusses the results of our empirical analysis and finally Section 4 concludes the paper.

Our sample period is from the 22 nd January 2020 to 20 nd May 2020. The start of our sample period is the day when Johns Hopkins University started to publish the daily confirmed and death case statistics in COVID-19. 2 Our initial sample includes countries in the Group of Twenty (G20), which consists of governments of 19 countries and the European Union. The choice of G20 is driven by their systemic importance to the world economy. 3 We obtain and merge our data from four sources. Firstly, we measure return and volatility connectedness using the 5-minute interval stock prices from Thomson Reuters Tick History (TRTH). Secondly, we extract daily interest rates, foreign exchange volatility and GDP growth from Datastream. Thirdly, the number of confirmed cases and death tolls of COVID-19 are obtained from the Coronavirus Resource Centre of Johns Hopkins University. Lastly, the number of deaths of each country during SARS in 2003 is collected from the website of the World Health Organisation (WHO). Our sample includes 11,696 observations over an 86-day period.

To measure return and volatility connectedness, we employ the approach of Yilmaz (2012, 2014) . More specifically, we construct the generalized total return and volatility connectedness index of each pairwise countries in G20 within a bivariate fractionally integrated Vector Autoregressive (FIVAR) model. The FIVAR model allows flexibility to capture the stationarity (or short-memory) of the stock returns as well as the long-memory behavior of the volatility. The bivariate FIVAR model can be specified as follows:

where is a vector of the stock return in country i and j in case of return connectedness analysis,

i.e. ( ) , or a vector of the two stock volatilities in case of volatility connectedness analysis, i.e. ( ) . and are respectively realized return and realized volatility, which are calculated using the 5-minute stock prices of the G20 country. 4 The error term, ( ), with * ; + as its variance-covariance matrix. is the ( ) coefficient matrix associated with , and denotes the ( ) identity matrix. L is the lag operator while is the lag order of the model. ( ) *( ) ( ) + with d denotes the memory degrees of the stock volatilities. 5 We firstly calculate the generalized forecast error variance decomposition (GFEV) matrix from Model (1) with a rolling window of 200 days and a forecast horizon of 10 days. 6 The ( ) element of the GFEV matrix at day t can be calculated as,

As shown in Do et al. (2013 Do et al. ( , 2014 , to incorporate the Diebold and Yilmaz approach in a FIVAR model, the moving coefficient matrix need to be adjusted with the long memory degree (d)

and is calculated recursively as, ∑ . We note that, , and is the identity vector with unity as its rth element.

Next, we construct the total generalized connectedness between country i and j at day t using the normalized GFEV ( ̃) as, Andersen et al. (2003) , Do et al., (2014) . ∑ , and ∑ , where is the nth 5-minute logarithmic stock return in day t. 5 In case of return connectedness analysis, d is restricted to be zero due to the stationarity of the stock returns, which makes our bivariate FIVAR model equivalent with a bivariate VAR model. We estimate our bivariate FIVAR model using Yip et al. (2017) approach. where, ̃ ( ) ( ) ∑ ( ) .

To capture the experience in SARS 2003 of each pair of G20 countries, we create a category variable SARS_death i,j , in which SARS_death i,j =0 when neither country in the pair experienced SARS death; if only one country reported deaths in SARS in 2003 in a pair, As COVID-19 spread across the world and its severity has evolved, we create a category variable COVID_stages to capture three stages of COVID-19 development in our sample. The first stage is from 22 nd January to 29 th January 2020, when COVID-19 was mostly transmitted within China. The second stage is from 30 th January to 10 th March 2020, the period over which COVID-19 gradually spreads across the world, so that the WHO declared global public health emergency on 30 th January 2020. In the third stage, the severe impact of COVID-19 was recognized and declared as a global pandemic by the WHO. We incorporate the development of COVID-19 within the following model: Furthermore, we also add three control variables in all models to consider the cross-country differences in economic development, including interest rates (proxied by 3-month government bonds), foreign exchange volatility over the previous 21 trading days and quarterly GDP growth.

Given that our dependent variable is between a pair of country, we estimate the within-pair differences for the aforementioned control variables. 

In addition to the use of discrete stages to measure the severity of COVID-19, we perform analyses using global and country-specific confirmed or death cases. Columns (1-2) and (5-6) of 

Our study provides evidence of the impact of COVID-19 pandemic on stock market connectedness. In general, connectedness in global financial markets is intensified with the rapid development of the pandemic. We also find that connectedness is considerably lower if a country experienced SARS death(s) in 2003. Our findings are associated with behavioral biases of investors in stock markets, in which investors with early experience of the similar pandemic tend to react faster and stronger to COVID-19 compared to those without such prior experience. This is because the former group of investors are more alarmed about similar risks faced in the past, while the latter group of investors tend to neglect those risks. (1). The dependent variable is Return connectedness for columns 1-2. The dependent variable for columns 3-4 is Volatility connectedness.

COVID_stage=2 is a dummy variable if the date is between 30th January to 10th March 2020, and zero otherwise. COVID_stage=3 is a dummy variable if the date is after 10th March 2020, and zero otherwise. SARS_deathi,j=1 is a dummy variable if one of a pair of countries experienced death cases in SARS, and zero otherwise. SARS_deathi,j=2 is a dummy variable if both of a pair of countries experienced death cases in SARS, and zero otherwise. Interest rate Diff is the difference of daily rate of 1-month T-bills between a pair of countries. and. Exchange Vol Diff is the difference of exchange rate fluctuation over the previous 21 trading days. GDP growth Diff is the difference in the GDP growth rate between two countries in each pair. ***, ** and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Robust standard errors cluster by countries are reported in the parentheses.

(1) ( ) are natural logarithm of the number of one plus accumulative confirmed (death) COVID-19 cases from country i and country j on each day respectively. Interest rate Diff is the difference of daily rate of 1-month T-bills between a pair of countries. and. Exchange Vol Diff is the difference of exchange rate fluctuation over the previous 21 trading days. GDP growth Diff is the difference in the GDP growth rate between two countries in each pair. ***, ** and * indicate statistical significance at the 1%, 5%, and 10% levels, respectively. Robust standard errors cluster by countries are reported in the parentheses.

(1) 

Modeling and forecasting realized volatility

The unprecedented stock market impact of COVID-19. Cambridge: NBER working paper series

What doesn't kill you will only make you more riskloving: early-life disasters and CEO behavior

Aye Corona! The contagion effects of being named Corona during the COVID-19 pandemic

The contagion effects of the COVID-19 pandemic: Evidence from gold and cryptocurrencies

Better to give than to receive: predictive directional measurement of volatility spillovers

On the network topology of variance decomposition: measuring the connectedness of financial firms

Generalized impulse response analysis in a fractionally integrated vector autoregressive model

How does trading volume affect financial return distribution?

COVID-19 and finance: Agendas for future research

The value of corporate culture

Firm-level exposure to epidemic diseases: Covid-19, SARS, and H1N1

Mark Twain's cat: investment experience, categorical thinking, and stock selection

Familiarity Breeds Investment. The Review of Financial Studies

Home Bias in International Stock Return Expectations

Imprinting: Toward a multilevel theory

The joint influence of financial risk perception and risk tolerance on individual investment decision-making. Accounting and Finance

Feverish Stock Price Reactions to COVID-19 (2020)

What do we learn from SARS-COV-1 to SARS-COV-2: evidence from global stock markets

The less you know, the more you are afraid of -a survey on risk perceptions of investment products

Dynamic spillover between commodities and commodity currencies during United States Q

Infected markets: Novel coronavirus, government interventions, and stock return volatility around the globe

Financial markets under the global pandemic of COVID-19