Does the Halloween effect (sell in May) still hold? In the June 2013 version of their paper entitled “Are Stock Markets Really so Inefficient? The Case of the ‘Halloween Indicator'”, Hubert Dichtl and Wolfgang Drobetz investigate whether true out-of-sample results confirm that the Halloween effect persists for five total return indexes: S&P 500, DAX 30, FTSE 100, CAC 40 and Euro Stoxx 50. They consider both regression tests to compare average monthly returns and simulated trading strategies based on bootstrapping. They consider three sample periods for each index: (1) the total available sample period; (2) the subperiod since availability of a liquid investment proxy (such as a mutual fund or exchange-traded fund) to exploit the Halloween effect; and, (3) the subperiod starting January 2003 (after publication of the seminal international study of the effect). For simulated trading strategies, they assume switches from stocks to cash at the end of April and cash to stocks at the end of October incur trading friction of 0.5%. Using monthly total returns of the specified indexes as available (starting years ranging from 1965 to 1988) through December 2012, and contemporaneous one-month local interest rates as the return on cash, they find that:
- Based on monthly return estimates from regressions:
- Over the entire available sample periods, the Halloween effect is generally significant.
- Since availability of liquid investment proxies, the effect weakens substantially and is not significant for the FTSE 100, CAC 40 and Euro Stoxx 50 indexes.
- Since January 2003, the effect is no longer evident (not achieving a 95% confidence level for any index).
- Based on portfolio simulations that earn randomly selected stock index returns (cash returns) during November-April (May-October):
- Over the entire available sample periods, the Halloween strategy generates significantly higher excess returns (relative to the respective risk-free rates) than buy-and-hold for all indexes except the S&P 500 and significantly higher Sharpe ratios in all five markets.
- Since availability of liquid investment proxies, the Halloween strategy again beats buy-and-hold for all indexes except the S&P 500 based on excess returns and all five indexes based on Sharpe ratios.
- Since January 2003, buy-and-hold significantly outperforms the Halloween strategy for all five indexes based on excess returns and for all indexes except (just barely) the S&P 500 based on Sharpe ratio.
In summary, evidence does not support belief that the Halloween effect persists since publication of the seminal 2002 study of its international reach.
Cautions regarding findings include:
- Returns calculated from indexes ignores the management/administrative costs of offering a tradable asset and thereby somewhat overstate realistic performance.
- Statistical significance testing generally assumes tame return distributions. To the extent actual distributions are wild, this testing loses meaning.
