Is the efficient market hypothesis day-of-the-week dependent? Evidence from the banking sector

Abstract

In this article, we propose a new hypothesis: that the efficient market hypothesis is day-of-the-week-dependent. We apply the test to firms belonging to the banking sector and listed on the NYSE. We find significant evidence that the efficient market hypothesis is day-of-the-week-dependent. Overall, for only 62% of firms, the unit root null hypothesis is rejected on all the five trading days. We also discover that when investors do not account for unit root properties in devising trading strategies, they obtain spurious profits.

Keywords:

• efficient market hypothesis
• unit root test
• structural breaks
• trading strategies

JEL Classification:

• C58
• G11
• G14

Notes

¹ Similar evidence have been documented by several other studies; see, for example, Weiß et al. (2014).

² In particular, we have in mind studies that have considered the efficiency of the banking sector (Hasan and Marton, 2003; Yeyati and Micco, 2007; Brissimis et al., 2008); determinants of risk in the banking sector (Festić et al., 2011; Fiordelisi et al., 2012); bank profitability and productivity (Nakane and Weintraub, 2005; Chen and Liao, 2011); and banking sector resilience to macroeconomic shocks (Dovern et al., 2010).

³ In a related study, Narayan et al. (2014) show that the determinants of bid-ask spread on the NYSE are day-of-the-week dependent.

⁴ Indeed, it was possible to consider data before 1998; however, this would have costed us firms. We wanted to maximize the number of banking sector firms, while at the same time maintaining a sample of time series observations sufficient to implement our proposed test. It is also possible to, and we have, undertaken results for the most recent sample period: 1998–2014. The reason we do not report these in the article is because there are obvious structural breaks in the data as a result of the 2007 Global Financial Crisis. For example, the S&P500 composite stock price index fell by 37.2% in 2009 from a 11.4% growth in 2007, followed by a further 29.8% growth in prices in 2010. Our simulation results (see Table 2) show that when the size of the break date is large, such as those characterized by years 2007, 2008 and 2010, then the power to reject the unit root null hypothesis will be high. Therefore, if we subject the test (which is already powerful) to data that are characterized by obvious structural breaks, we will almost always reject the unit root null. We avoid this situation because our goal is to demonstrate how best our test works when data are just normal and go through very normal changes in prices.

⁵ Another point about our sample size that should also be kept in mind is: here we develop a new test which ideally should be tested with data that is not contaminated by known structural breaks. If anything, with the global financial crisis, since it is a major break in the data, our test because it is a structural break test, will be more powerful. Indeed, in a Working Paper version (available upon request) we had reported results for a data set spanning the period 1998–2011. The obvious structural break of 2008 led to all statistically significant results. Therefore, we decided to exclude the global financial crisis data. In any case, this sample size should not be a controversial issue as many studies, depending on the aim and research question, have excluded the global financial crisis. Many of these papers appear in the Journal of Banking and Finance; for a recent example, see Narayan and Sharma (2011).