Firm Product Similarity and Stock Price Comovement: Evidence from China

Abstract

ABSTRACTThis article examines the effect of firm product similarity on stock price comovement. Using the financial data and annual reports of listed firms in the Chinese A-share market from January 2001 to December 2021, we find that firms with greater product similarity experience synchronized movements in their stock prices. This effect is driven by firm fundamentals, as demonstrated through major international events (Global Financial Crisis, European Sovereign Debt Crisis, and Trade Dispute between China and the U.S.) and domestic events (Two Sessions about the Deepening Overall Reform, and Central Economic Conference following the COVID-19 Outbreak). We also show that firms that release earnings announcements earlier contribute to the comovement of stock prices within their product-similarity cluster. Our findings are robust across various tests and provide insights into the dynamics of the Chinese A-share market.KEYWORDS: Firm product similaritystock price comovementexternal shockChinaJEL: G140G170G300 Disclosure statementNo potential conflict of interest was reported by the author(s).Notes1. Regression results for control variables in Tables 5, 7–9, 11 and 12 are omitted due to the limit of pages.2. In panel regressions throughout this article, the standard errors are clustered at the firm level with the year-fixed effects. In panel and Fama-MacBeth regressions, the unit of coefficients is in percentage. The t-statistics for corresponding coefficient estimates are presented within parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.3. Due to the high correlation among MOM, CLUSTER_MOM, and AREA_MOM, we only incorporate the significant one with the largest influence magnitude as control variables. In Table 3, the regression results with MOM is not displayed because it has relatively small impacts on CMVT.4. The Newey-West t-statistics are presented within parentheses.5. Guba is important investor social media in China that contains various retail investors posting their comments toward stocks and the market. We obtain Guba comments from Stock Comments Database in CNRDS (China Research Data Service), which begins from 2008 and with identifiers of positive and negative emotions.6. The Central Economic Work Conference is excluded in this test, because the event month is the last month in both the event year and our sample, therefore has no sufficient observations obtaining the estimation results.7. The reasons may be that except the sudden break and contagion of Global Financial Crisis that is prominent to be detected, the other three events are accompanied with gradual signals that occur before the event months, such as the debt crisis in Greece and the presidential memorandum signed by Donald Trump proposing levying tariffs under Special 301 Report. The Two Sessions also arouse hot discussions toward the policy directions before the conference date due to its special role, the details are in Appendix A.8. Because we focus on the instantaneous effect of major events to the link between product similarity and stock price comovement, we do not include all subsequent months of the breakpoint as Karavias, Narayan, and Joakim (Citation2023) do. Instead, we focus on the [−1,1] three months window around the event month to allow for the diffusion and delay of information before and after the shock, meanwhile control the noise of other events that may happen during the subsequent months.9. The mechanism analysis is not affected by this issue because the domestic and international events that affect firm fundamentals happens no earlier than March.10. We have 390,033 firm-month observations for each variable except ANALYST, which has 256,887 firm-month observations. This is because for most firms, data of analyst coverage starts from 2007 while data of other variables starts from 2002. To ensure the data completeness of our analysis and cover the time period as long as possible, we run baseline regressions with the original dataset with 390,033 observations, and include ANALYST separately with the dataset of 256,887 observations.Additional informationFundingThis work was supported by the Fundamental Research Funds for the Central Universities [JBK2107146]; National Natural Science Foundation of China [71903154; 71873266].