key: cord-0875941-0cep0bnw authors: Heyden, Kim J.; Heyden, Thomas title: Market Reactions to the Arrival and Containment of COVID-19: An Event Study() date: 2020-09-02 journal: Financ Res Lett DOI: 10.1016/j.frl.2020.101745 sha: 80bc9f302d2611f8614b3bfd0851dbf7e88a7b62 doc_id: 875941 cord_uid: 0cep0bnw We study the short-term market reactions of US and European stocks during the beginning of the COVID-19 pandemic. Employing an event study, we document that stocks react significantly negatively to the announcement of the first death in a given country. While our results suggest that the announcements of country-specific fiscal policy measures negatively affect stock returns, monetary policy measures have the potential to calm markets. These reactions are either intensified or lessened by firm-specific characteristics such as tangible assets, liquidity, and institutional holdings. The methodology of our first robustness analysis relies on a similar exercise conducted by Bialkowski et al. (2008) . Note that the following paragraph is an augmented recollection of said exercise. Using the observations from the estimation window, we regress the continuous returns R it of stock i in t on the continuous index returns R t in t. Note that we use the MSCI World stock index as benchmark in order to ensure our results are not driven by the choice of index. In order to extract the firm-specific component of the variance, we employ a GARCH(1,1) (see Bollerslev, 1986 ) estimation technique, which jointly estimates the following two equations: While it denotes the firm-specific part of the returns that can not be explained by index returns, σ 2 it is the conditional variance. Considering Hwang and Pereira (2006) , we follow Bialkowski et al. (2008) and expand the event window to 500 days. Our primary measure of interest is the abnormal variance (AV ). This can be computed as the cross-sectional variance of the average forecasted residualsˆ it (over the event window with t ∈ [−10, 10]), divided by the event-independent average residuals' standard deviation: where AV it ∼ χ 2 (n 2 −n 1 +1) with n 1 (n 2 ) denoting the first (last) day of the event window. Let be the expected value of the conditional variance for stock i on day t of the event window. It corresponds to a k-step-ahead forecast based on the last available value of σ 2 it , where k denotes the number of forecasted days of the event window. <<< Insert Figure 1 around here >>> Note that forecasted residualsˆ it are merely the abnormal returns (AR), which are reported in Figure 1 . Accumulating them over the event window allows us to compare the newly computed cumulative ARs (CARs) to the ones that are reported in the manuscript's Appendix IV. Considering their close resemblance, first, the choice of benchmark appears not to play a huge role in the formation of CARs. Second, although the estimation technique employed is more sophisticated (as it considers serial dependence and thus variance clustering of error terms) the average CARs do not change by much. <<< Insert Figure 2 around here >>> Figure 2 shows the abnormal variances around the four different event dates. It is divided into US and European stocks and reports an abnormal value for each day of the event window, which is defined as t ∈ [−10, 10]. Overall, they are in line with our main findings. While the announcement of the first Coronavirus case does not appear to increase stock-market volatility, the first death spurs markets and leads to a strong increase in volatility. This supports our finding of no significantly negative stock returns after the announcement of the first case in contrast to the negative reactions after the first death. As shown in Table 1 week after the event (around 40%), but decreases only about 22% after the announcement of the fiscal measure (from 12.536 to 9.804). For European stocks, this difference is even more pronounced with a reduction of around 9% versus 26% for the announcements of fiscal and monetary measures, respectively. Given that we report a positive stock-market reaction after the announcement of the monetary measure, it was to be expected that volatility does not decrease to a pre-crisis level but calms in comparison to the fiscal measure announcements. For our second robustness check, we test whether there has been an observable reaction in economic and stock-market sentiment. For this matter, we employ the Twitter-based indices introduced by Baker et al. (2020). As Altig et al. (2020) argue, these measures are forwardlooking sentiment proxies that are particularly useful when testing hypotheses in conjunction with investing under uncertainty. One drawback is that they mostly capture US sentiment. Hence, we can only assess whether there are reactions for the US stock market. In order to analyze the abnormal reactions of the two indices, we compute the indices' percentage changes and regress them on a constant. This value is used as a benchmark to calculate abnormal changes around the event dates. Essentially, we follow MacKinley's (1997) constant-mean return model procedure, not only when computing the abnormal changes, but also when testing their significance. The choice of the length of our event window and estimation window follows that in our manuscript. Abnormal variance of US (lhs) and European (rhs) stocks resulting from the announcement of the first case. Abnormal variance of US (lhs) and European (rhs) stocks resulting from the announcement of the first death. Abnormal variance of US (lhs) and European (rhs) stocks resulting from the announcement of the first fiscal policy measure. Abnormal variance of US (lhs) and European (rhs) stocks resulting from the announcement of the first monetary policy measure. Economic (lhs) and stock-market (rhs) uncertainty index reactions to the announcement of the first case. Economic (lhs) and stock-market (rhs) uncertainty index reactions to the announcement of the first death. Economic (lhs) and stock-market (rhs) uncertainty index reactions to the announcement of the first fiscal policy measure. Economic (lhs) and stock-market (rhs) uncertainty index reactions to the announcement of the first monetary policy measure. Remark: This table reports the average abnormal index changes corresponding to the values shown in Figure 3 . Columns labeled (3) and (4) denote the event types fiscal policy measure and monetary policy measure, respectively. Economic uncertainty before and during the Covid-19 pandemic Economic Uncertainty Measures Derived from Twitter Stock market volatility around national elections Generalized autoregressive conditional heteroskedasticity Small sample properties of GARCH estimates and persistence Event studies in economics and finance