Content area
Full Text
1. Introduction
This paper provides a new hypothesis to explain the so-called “sell-in-May and go away” calendar effect in the US stock market. Bouman and Jacobsen (2002) found that stock market returns during the months of May–October significantly underperform relative to returns during the months of November–April, even after the January effect is considered. This is also called the Halloween effect because investors are advised to return to the market right after the end of October.
Our analysis suggests that most of the sell-in-May phenomenon can be attributed to an election effect tied to US national elections that occur twice in a four-year cycle (presidential, congressional and most gubernatorial[1] elections). If the sell-in-May effect is really an election-year effect, then the sell-in-May anomaly should be present in even-numbered years, when national elections are held, and absent in odd-numbered years. This is what we find. Contrary to the tenets of the sell-in-May adage, the returns over the May–October period during non-election years are economically strong and positive. Furthermore, there is no significant difference between the May–October and November–April returns of non-election years. This paper shows that most of the strongly positive November–April returns occur immediately following an election. All of these results hold even when the well-known January effect is considered. Investors would be ill-advised to follow the “sell-in-May” adage blindly, especially in non-election years.
While the sell-in-May anomaly still persists (Andrade et al., 2013), a significant and legitimate concern is that no widely accepted rationale has previously been provided for its existence. Several researchers suggest that these seasonal stock market returns could be tied to factors such as seasonal affective disorder (SAD), the weather, or daylight savings time (Kamstra et al., 2000, 2002, 2003, 2009; Cao and Wei, 2005; Garrett et al., 2005), but these arguments have not gained widespread acceptance and have been challenged by others (Pinegar, 2002; Jacobsen and Marquering, 2008, 2009; Gregory-Allen et al., 2010; Khaled and Keef, 2014).
Our hypothesis is that the sell-in-May effect is really a rolling two-year election effect. The USA holds national congressional elections on all even-numbered years, with presidential elections simultaneously taking place with every other congressional election. The results of these elections have the potential to change the direction of both...