News and Market Efficiency in the Japanese Stock Market

Abstract

Economists have debated whether market reactions to news depart from the predictions of the efficient market hypothesis (EMH). This article uses the sentiment index and the surprise index from the Thomson Reuters MarketPsych Indices and an ARCH model to investigate the reactions of the return of the Nikkei 225 closing price to different types of news between January 5, 1998 and December 29, 2017, and whether those responses are consistent with the predictions of the EMH. Three sub-periods during the full sample are identified, and the market reactions to the news are then examined and compared among the sub-periods.

Keywords:

• News
• Stock market
• Sentiment
• Market efficiency

Notes

1 For example, Kaminsky and Schmukler (1999), Andersen et al. (2003, 2007), Bird, Du, and Willett (2017), and Narayan and Bannigidadmath (2017).

2 Japan Ratings and Investment Information (R&I), Japan Credit Rating Agency (JCR), Moody’s, and Standard and Poor’s (S&P).

3 Fitch Group, Moody’s, and Standard and Poor’s (S&P).

4 For example, Du (2018) distinguished the impact before and after the introduction of Abenomics in December 2012.

5 “History on all content dates back to the beginning of 1998 for all assets except cryptocurrencies… Reuters news is present in the entire historical news dataset, as are a host of mainstream news sources collected by MarketPsych Data. During 2005, the archive began including Internet news content collected by LexisNexis. The LexisNexis content is restricted to those from top international business news sources, top regional news sources and leading industry sources… The social media collection process is more diverse. It begins in 1998 with Internet forum and message board content. Starting in late 2008, LexisNexis social media content was added. Starting in late 2009, tweets were included. Using popularity ranks measured by incoming links, this includes generally the top 20% of blogs, microblogs, and other financial social media content. MarketPsych Data also included content from hundreds of less-popular asset-specific blogs and forums.” (Peterson, 2016).

6 Buzz refers to the total number of meanings included in the TRMI indexes.

7 Source: https://finance.yahoo.com/quote/%5EN225/history?p=%5EN225. The closing price is adjusted for splits.

8 The Augmented Dickey–Fuller test for unit root (Dickey and Fuller 1979) suggested the dependent variable was generated by a stationary process. The Durbin’s alternative test to test for serial correlation (Durbin and Watson 1950) showed that the residuals were not serially correlated. The Breusch and Pagan (1979)/Cook and Weisberg (1983) test for heteroscedasticity to test the implied presence of heteroscedasticity in the residuals. LM test for ARCH effect (Engle 1982) confirmed the existence of ARCH effects.

9 In this paper, the sentiment and news variables are assumed to be exogenous. Yet, of course, there could be a more complicated case where some feedback (on the same day) from the stock market to the media content is used to compute the sentiment proxies.

10 Source: https://finance.yahoo.com/quote/%5EN225/history?p=%5EN225. The closing price is adjusted for splits.

Additional information

Funding

This work was supported by JSPS KAKENHI Grant Number JP18K12772.