key: cord-0886645-ihzwpxi5 authors: Zorgati, Imen; Garfatta, Riadh title: Spatial financial contagion during the COVID-19 outbreak: Local correlation approach date: 2021-09-03 journal: J Econ Asymmetries DOI: 10.1016/j.jeca.2021.e00223 sha: ae66c13f1b2993b89b5b1038ed660245be25a302 doc_id: 886645 cord_uid: ihzwpxi5 The purpose of this paper is to examine the effect of spatial proximity on financial contagion during the COVID-19 outbreak. We use the daily stock index series of Asian, American, and European countries from 1 January 2014 to 30 January 2021. Two groups of countries are considered: the first includes China and geographically close countries, namely Taiwan, Hong Kong, Singapore, India, Australia, Indonesia, Malaysia, South Korea, Singapore, Vietnam and Russia. The second group includes countries that are geographically distant from China: the United States, Brazil, Mexico, Argentina, Italy, France and Germany. Using local correlation measurement and polynomial regressions, we show that the spatial contagion effect exists between China and geographically distant countries. However, this effect is absent for geographically close countries (Taiwan, Vietnam and Hong Kong). These findings have strong implications for investors and present guidance for regulators and policymakers in understanding the true impact of the COVID-19 on financial markets. 3 approach, their results show that during COVID-19, CSR engagement has increased stock returns. Just and Echaust (2020) study the structural breaks in stock returns and focus on volatility expectations, illiquidity and correlation expectations during the COVID-19 outbreak. Using the two-regime Markov switching model, they detect a structural break between stock market returns and stock market indicators. The authors also find that stock market illiquidity does not affect stock market returns and is not related to COVID-19. Another study by Belaid et al. (2021) focuses on the consequences of the COVID-19 crisis on the interdependencies between advanced and emerging economies. The authors use daily market index data from 22 markets and show an increase in interdependence between emerging and advanced economies. The results also show an increase in the transmission of uncertainty between financial markets during the COVID-19 pandemic. Due to such crises and shocks, the contagion phenomenon has become one of the most considered topics in finance (Rigobon, 2002; Kuusk and Paas, 2010) . Several studies have defined the concept of contagion. As such, according to Eichengreen et al. (1996) , contagion is the "significant increase in the probability of a crisis in one country, conditional on the occurrence of a crisis in another country". Forbes and Rigobon (2002) present contagion as "a significant increase in cross-market linkages following a shock to an individual country (or group of countries)". For their side, Bradley et al. (2004) and Fabrizio (2014) show that spatial contagion between two financial markets X and Y exists when there is a high dependence between X and Y. Several studies have examined financial contagion (Eichengreen et al., 1996; Forbes and Rigobon, 2002; Dungey and Fry, 2009; Kenourgios et al., 2013; Kenourgios and Dimitriou, 2015; El Ghini and Saidi, 2015; Zorgati et al., 2019; Akhtaruzzaman et al., 2019) . Indeed, Kenourgios et al. (2013) investigate financial contagion as a mechanism of asymmetric propagation in equity and change markets. They use an asymmetric generalized dynamic conditional correlation (AG-DCC) model and find that conditional correlations between stock markets increase significantly during a crisis period, supporting the presence of financial contagion. Kenourgios (2014) studies contagion during the subprime and euro zone crises. They apply conditional correlation dynamics for the two periods of crisis and non-crisis. Subsequently, Kenourgios and Dimitriou (2015) investigate the contagion effects of the subprime crisis by testing different channels of financial contagion via regions and sectors of the real economy. Using the dynamic conditional correlation approach, Akhtaruzzaman and Shamsuddin (2016) investigate contagion through financial and non-financial firms using 49 J o u r n a l P r e -p r o o f 4 countries. The authors show that financial contagion is linked positively to the level of development of stock markets and the intensity of bilateral trade. Recently, Zorgati et al. (2019) investigate the existence of the financial contagion phenomenon in the context of the subprime crisis based on the copulas approach. They prove the existence of contagion effect between the U.S. and all other American countries as well as the Australian, Indian, Malaysian, Indonesian, Singaporean, and Chinese ones. Moreover, Akhtaruzzaman et al. (2019) find that Chinese and U.S. companies have transmitted more spillovers than they received during the global financial crisis. Regarding the COVID-19 crisis, Akhtaruzzaman et al. (2020) show a significant increase in conditional correlations between different stock returns. Okorie and Lin (2020) show significant fractal contagion effects of COVID-19 on the stock markets' return and volatility using Detrended Moving Cross-Correlation Analysis (DMCA) and Detrended Cross-Correlation Analysis (DCCA) techniques. Network approaches are also applied to study the financial contagion effect. Gai and Kapadia (2010) investigate how the impact of contagion is affected by overall and idiosyncratic shocks, changes in network structure, and asset market liquidity. The authors find that the financial system is robust but fragile. Using the Maximum Entropy method, Paltalidis et al. (2015) study systemic risk and the spread of financial contagion within the euro area banking system. Recently, Chakrabarti et al. (2021) investigate the interconnections of stock markets and the contagion effect during the COVID-19 outbreak. They show the existence of contagion in the global stock markets due to the COVID-19 pandemic. The network theory increases the understanding of structural changes and their interconnections between stock markets. Guo et al. (2021) , investigate the tail risk contagion during the COVID-19 outbreak. They use the FARM-Selection approach and the time-varying financial network model. The authors show that the COVID-19 outbreak negatively influences the international financial system. Indeed, it increases tail risk spillovers for the international financial market. Moreover, using 19 international financial markets (including China, Singapore, Hong Kong, Japan, Taiwan, France, Germany, Brazil, and the U.S.), they find that COVID-19 influences the tail risk contagion in the local network system. We draw on the work of Taqqu (2004, 2005b) by studying spatial financial contagion during the COVID-19 outbreak. We consider Asian, American, and European countries over the period from 1 January 2014 to 30 January 2021. This paper extends the work J o u r n a l P r e -p r o o f 5 of Kenourgios et al. (2013) , who focused on financial contagion during the Asian crisis and used an asymmetric generalized dynamic conditional correlation (AG-DCC) model. Furthermore, it differs from Zorgati and Lakhal (2020) who investigate financial spatial contagion and compare findings by applying adjusted and local correlation approaches in the subprime crisis context. The purpose of this paper is to investigate the effect of spatial proximity on financial contagion during the COVID-19 outbreak using the local correlation approach. The contribution of this paper is twofold: first, to the best of our knowledge, there are no studies that investigate the spatial aspect on financial contagion using the local correlation approach during the period of the COVID-19 pandemic. Indeed, the local correlation approach is a nonparametric dependence measure. It differs from other approaches such as those of cointegration, copulas, GARCH, simple and adjusted correlation. Indeed, it does not require the specification of stability and crisis periods. Moreover, the spatial aspect is rarely studied in the literature. Second, we investigate the effect of spatial proximity on financial contagion by applying nonlinear forms of dependence. If X denotes the return of market X, the contagion exists when there is more dependence in the lowest quantile of the distribution X than there is in the median of this distribution using the local correlation approach. The data we consider covers the period from 1 January 2014 to 30 January 2021. We consider those countries in the same region as China. Two groups of countries are used: the first includes China and a number of geographically close countries, namely Taiwan, Vietnam, Hong Kong, Singapore, Japan, India, Indonesia, Australia, Malaysia, South Korea, Singapore and Russia. The second group comprises geographically distant countries from China: the United States, Brazil, Mexico, Argentina, Italy, France and Germany. We focus on the geographic links between markets during the COVID-19 pandemic. For more robust results, we use local correlations and polynomial regressions. Our findings show that the spatial contagion effect exists between China and geographically distant countries, while it is absent for geographically close countries (Taiwan, Vietnam and Hong Kong). The remainder of this paper is organized as follows: Section 2 presents the econometric methodology. Section 3 describes the collected data and relative descriptive statistics. Section 4 presents the results and discussion. Section 5 concludes the paper. Numerous methods have been used to test the existence of the spatial contagion. Using a simple correlation approach, Calvo and Reinhart (1996) (2002) and Chiang et al. (2007) show that this measurement of contagion is biased due to the problems of heteroscedasticity, endogenous variables and omitted variables. Forbes and Rigobon (2001) use the adjusted correlation approach to correct this bias. However, the adjusted correlation power is weak due to the short period of crisis, making it difficult to detect contagion. Recently, attention has been paid to modelling the tail of the distribution of financial markets. Bjerve and Dorsum (1993) recommend the local correlation approach as a non-parametric measure to overcome issues related to the simple and adjusted correlation approaches. This method is used for data which are not jointly Gaussian. Bradley and Taqqu (2004) and Zorgati and Lakhal (2020) also use the local correlation approach. This method uses nonlinear forms of dependence and provides a better understanding on the degree of dependence between financial markets. Local correlation is a good and robust measure of dependence that allows for conclusions about the existence of spatial contagion between markets. This measure is different from other approaches such as cointegration, copulas, simple and adjusted correlation. It does not require the specification of stability and crisis periods. This method overcomes the difficulties of correlation breakdown tests, specifically when data are hand-collected. Although Stove et al. (2014) find that the local correlation approach may better reveal an asymmetry in multivariate financial distributions rather than contagion, Bradley and Taqqu (2004 , 2005a , 2005b , Inci et al. (2010) and Zorgati and Lakhal. (2020) show that local correlation is a local measure of dependence suitable for measuring financial contagion. It is a non-parametric measure used to handle nonlinear forms of dependency. Inci et al. (2010) compare international contagion patterns of different markets for a period of more than 20 years. Furthermore, Zorgati and Lakhal (2020) also use a long time frame to study spatial contagion between the markets during the subprime crisis. Let = ( , = 1,2, … ) and = ( , = 1,2, … ) the returns in two different financial markets. The local correlation approach is a dependence measure based on localizing a first order regression relation. Indeed, in a regression of Y on X, the parameters of regression and the residual variance may depend on X. This method allows for extending the links between correlation, regression slopes and the variance. Following Bradley and Taqqu (2004) , the linear regression model is presented as follows: is the slope of the regression function, 2 ( ), is the residual variance and ( ) = ( ⁄ = ). (2004) define the local correlation approach as follows: X and Y are two random variables and have finite variance. The local correlation between X and Y.at = is given by With: the standard deviation of X. To test the spatial contagion effect, Bradley and Taqqu (2004) investigate the contagion between markets using local correlation estimators. Indeed, there is a spatial contagion from market X to Y if: The choice of the 2.5% quantile depends on the notion of crisis. It can be reached when the data is heavily concentrated around the median. Although contagion may exist in this case, it may be irrelevant. We then need to study the data and the losses incurred at this quantile. Indeed, we use heavy-tailed distributions and losses may be significant. The spatial contagion effect exists when there is more dependence in the lowest quantile of the distribution X than in the median of this distribution using the local correlation approach. Nadaraya (1964) presents two non-parametric regression methods for the estimation: the kernel regression and the polynomial local regression. Using the first method, the author finds that its estimators lack robustness. Therefore, he suggests the use of the local polynomial regression method. In this study, we follow the estimation procedure of Bradley and Taqqu (2005a) based on local polynomial regression. Our analysis procedure can be summarized in three steps: Step 1: Consists of applying a local quadratic regression in order to infer 0 () x  . We use an optimal bandwidth for that regression. The solution of the local polynomial regression problem is given by: J o u r n a l P r e -p r o o f We obtain the estimator of Step 2: Consists of applying a local linear regression on the squared residuals to estimate 2 0 In this case, we use another optimal bandwidth appropriate to the second regression. We determine the estimator of residual variance at x0 using theses equations (Bradley and Taqqu (2005a): -And ifm is unbiased, Step 3: Consists of calculating the estimator of the local correlation 0 () x  and concluding whether or not the spatial contagion is present. 0 () x  is calculated as follows: Where 22 We then compute the Z-statistic, calculated as follows: We use J o u r n a l P r e -p r o o f 9 If 1 Zz   , we conclude that there is spatial contagion between the designed markets. We adopt the spatial contagion test to draw a conclusion about the spatial contagion existence between China and different markets during the COVID-19 outbreak. According to Bradley and Taqqu (2004) , the test of spatial contagion is as follows: We consider the daily series of stock indexes of Asian, American, and European countries 1 . Narayan (2015) documents that daily data is better than monthly data when the goal is to gain as much information as possible from the data. As are investigating the COVID-19 outbreak of 2019, the sampled period lasts from 1 January 2014 to 30 January 2021. We consider the first date of the pre-COVID-19 period as the 1 January 2014. Using the adjusted and unadjusted correlation approaches, our sample covers both the pre- outbreak. We calculate the daily returns from the return index. Where rt is the daily returns and Pt is the price of the index at time t. We use local currency returns in our study 4 . uncorrelated. These results support the use of the local correlation approach. [Insert Table 1 , here] We first begin our analysis by the use of an unadjusted correlation measure. We show the changes between the market correlations during the pre-COVID-19 and the COVID-19 periods and make a conclusion about the existence of contagion. Based on the studies of Forbes and Rigobon (2001), a contagion effect exists when this coefficient increases significantly during the crisis. Mink (2015) argues that the use of local currency returns is than common currency returns (e.g., USD). Furthermore, Akhtaruzzaman and Shamsuddin (2016) find the same results for financial contagion when they use local and common currency returns. We then present in Table 3 the results associated with the adjusted correlation coefficient. We rely on the work of Forbes and Rigobon (2002) and Collins and Biekpe (2002) to test the existence of contagion. We show that the adjustment of heteroskedasticity significantly impacts on the results of the contagion tests and show the absence of contagion for all studied markets. Our results are consistent with those of Forbes and Rigobon (2002), who show the existence of interconnection between markets instead of pure contagion. The results associated with the simple and adjusted correlation approaches are inconclusive and have certain limitations. Indeed, they are not sufficiently robust to conclude on the existence of contagion. Moreover, these approaches present short-term relationships between the stock markets and do not take into account the direction of causality between them. Regarding the local correlation test, the first step was to infer 0 () x  by making a quadratic local regression. We first determine the optimal bandwidth related to this regression. Table 4 presents the optimal bandwidth h1 for China with different markets. We find that for the group of countries geographically close to China, bandwidth h1 is between 0.2150 (China/Australia) and 1.1586 (China/Taiwan), while for American and European countries, h1 is between 0.4572 (China/Argentina) and 1.0181 (China/France). Bandwidth h1 is only related to this quadratic local regression. We next apply another linear local regression using bandwidth h2 (Table 2 ). This step is important for the estimation of 0 () x  and σ 2 (x 0 ). [Insert Table 4 The next step in this study is to infer σ 2 (x 0 ). The estimation procedure of σ 2 (x 0 ) is identical to the one used in the first step. We make a local linear regression on the squared residuals to infer the residual variance σ 2 (x 0 ). We follow Ruppert et al. (1997) by using the asymptotically optimal bandwidth estimate which is suitable for this regression. [Insert Table 5 , here] The polynomial (p = 1) in this step is different from the one found in the quadratic regression (p = 2). Moreover, the bandwidth h2 differs from h1. The value of h2 is given in Table 2 Table 6 presents the estimation of residual variance for China and other countries associated with points xL and xM. We find that 2 () L x  is higher than 2 () M x  . We use these estimators to compute the estimator of local correlation. [Insert Table 6 , here] In the last step, we determine 0 () x  and draw a conclusion.regarding the presence of spatial contagion for China and the two groups of countries during the COVID-19 period. Table 7 reports the local correlations estimators associated with the xL and For the sensitivity checks, we study the spatial contagion for a different study period from 2010 to 2020. The results remain unchanged throughout the study except for Germany; we find the absence of spatial financial contagion during the COVID-19 outbreak. We conclude on the importance of the study period. These results differ from those obtained by Zorgati and Lakhal (2020) , who show the importance of spatial proximity during the subprime crisis. [Insert Table 7 , here] Our results support the fact that the COVID-19 outbreak affects stock markets around the world. This pandemic has put the world economy at risk and has quickly spread in the European, American and Asian markets. The COVID-19 has affected the economic and social conditions in Asia. It has also affected the banking sector and weakened the trade flows. According to Liu et al. (2020) , the consequences of the COVID-19 outbreak are important and directly affect stock markets around the world. It increases uncertainties and the fear of stock market investors around the world. Besides, it creates pessimistic feelings on future returns. The authors study the short-term impact of COVID-19 in 21 stock market indexes (Korea, Singapore, Japan, the U.S., Germany, Italy) over the period from 21 February 2019 to 18 March 2020. They find that COVID-19 outbreak have a negative impact on stock market indexes for 5 The critical values of the test statistic are 1.65, and 2.33 for 5%, and 1% significance levels, respectively. J o u r n a l P r e -p r o o f 14 all studied countries and that the Asian markets respond more quickly to the pandemic. This result is different from ours. This is explained by the short period from 20 January 2020 to 18 March 2020 investigated that we extend in this study. The purpose of this paper is to investigate the effect of spatial proximity on financial contagion during the COVID-19 outbreak using the local correlation approach. We consider the daily stock index series of Asian, American, and European countries over the period of 1 January 2014 to 30 January 2021. As for countries that are geographically close to China (Taiwan, Vietnam, Singapore and Hong Kong), we show the non-existence of spatial contagion as these countries mitigate the spread of the COVID-19. Indeed, following the epidemic in 2003 (severe acute respiratory syndrome (SARS)), they have been on alert and acting rapidly on epidemics arising from China. As for countries that are geographically distant for China, we prove the presence of spatial contagion and the continual spread of COVID-19. Our results are consistent with those of Okorie and Lin (2020) who show the existence of significant but short-lived contagion effect on the stock markets (the U.S., Italy, France, and so on) during the COVID-19 pandemic. Furthermore, our findings are consistent with Zorgati and Lakhal (2020) , who find the presence of spatial contagion between China and the U.S. during the subprime crisis. These results have strong implications for investors, who want to diversify their portfolios internationally and hedge benefits so that portfolio managers predict and minimize market risk. Although holding an internationally diversified portfolio offers little protection against sharp declines in a market, the long-term gains from international diversification remain economically attractive. In addition, these findings have implications for policymakers, to make informed decisions about financial stability measures and decisions for different sectors of the economy. One limitation of this paper is that it did not take into consideration the African region. It would be interesting to examine the existence and intensity of financial contagion during the COVID-19 epidemic for Asian, American, European and African markets. **, and *** represent significance at 5%, and 1% levels, respectively. h1: The optimal bandwidth related to the local quadratic regression, which the polynomial (p=2). h2: The optimal bandwidth related to the second regression (local linear regression), which the polynomial (p = 1). The determination of bandwidth h1 and h2 is crucial for the estimation of We make a local linear regression on the squared residual to infer the residual variance σ 2 (x 0 ) and deduce International contagion through financial versus non-financial firms Is China a source of financial contagion? Financial contagion during COVID-19 crisis COVID− 19 and Oil Price Risk Exposure Is Gold a Hedge or Safe Haven Asset during COVID-19 Crisis? Economic Modelling Emerging and advanced economies markets behaviour during the COVID-19 crisis era Correlation curves: measures of association as functions of covariate values Framework for analyzing spatial contagion between financial markets Empirical evidence on spatial contagion between financial markets How to estimate spatial contagion between financial markets Capital flows to Latin America: is there evidence of contagion effects? World Bank Policy Research COVID-19 pandemic and global financial market interlinkages: a dynamic temporal network analysis Dynamic correlation analysis of financial contagion: evidence from Asian markets Some contagion, some interdependence: More pitfalls in tests of financial contagion Testing for asymmetric financial contagion: New evidence from the Asian crisis Contagion of the global financial crisis and the real economy: a regional analysis Contagion of Financial Crises With Special Emphasis On Cee Economies: A Metaanalysis. University of Tartu -Faculty of Economics and Business Administration Impact of COVID-19 outbreak on asymmetric multifractality of gold and oil prices Measuring stock market contagion: local or common currency returns? Emerg On estimating regression Does data frequency matter for the impact of forward premium on spot exchange rate? Stock Markets and the COVID-19 Fractal Contagion Effects Transmission channels of systemic risk and contagion in the European financial network Can corporate social responsibility protect firm value during the COVID-19 pandemic? Local polynomial variance function estimation Revisiting oil-stock nexus during COVID-19 pandemic: Some preliminary results Hedging oil price risk with gold during COVID-19 pandemic COVID-19 pandemic, oil prices, stock market, geopolitical risk and policy uncertainty nexus in the US economy: Fresh evidence from the wavelet-based approach Spatial contagion in the subprime crisis context: Adjusted correlation versus local correlation approaches Financial contagion in the subprime crisis context: a copula approach Z z   , we conclude that there is spatial contagion between the designed markets. The critical values of the test statistic are 1.65, and 2.33 for 5%, and 1% significance levels, respectively. **, and *** represent significance at the 5%, and 1% levels, respectively.