On Stochastic Dominance Analysis of Day‐of‐the‐week Return Patterns

Abstract

Studies show that significant differences exist among return distributions of days of the week. While these results are ubiquitous, their validity depends on the robustness of statistical procedures used. Virtually every day‐of‐the‐week study has used mean/variance analysis despite it being well documented that daily return distributions are nonnormal. This study uses stochastic dominance analysis, which is not distribution dependent, to test for a day‐of‐the‐week effect. Results indicate that the day‐of‐the‐week effect is robust and that previous findings are not artifacts deriving from violations of distributional assumptions.

Revisiting calendar anomalies in Asian stock markets using a stochastic dominance approach

“…In our study for the day-of-the-week effect, we exclude any week with fewer than five trading days. This is consistent with the study by Wingender and Groff (1989) and is, in principle, consistent with the approach in previous studies on the day-of-the-week effect. Similarly, in our test for the January effect, we only examine returns for the first twenty calendar days so as to fulfil the requirement of equal sample size.…”

Calendar anomaly in the Greek stock market: Stochastic dominance analysis

“…For US equity markets, Seyhun (1993) uses SD to evaluate the January effect while Wingender and Groff (1989) the day of the week effect. For the US bond market, Al-Khazali (2001) shows the January effect in high-yield bonds by SD.…”

The day‐of‐the‐week effect and conditional volatility: Sensitivity of error distributional assumptions

“…To circumvent the problem with departure from normality, some researchers use distribution‐free approaches. For instance, Board and Sutcliffe (1988) and Wong et al (1992) apply nonparametric tests while Wingender and Groff (1989) employ stochastic dominance analysis. Najand and Yung (1994) and Berument and Kiymaz (2001) investigate the statistical properties of returns series and then use a GARCH model assuming normal error distribution despite reporting strong evidence of leptokurtosis.…”