Stock markets have seasonal effects, e.g., Monday-effect, weekend effect, turn-of-the-month effect, holiday effect, January effect, etc. The Halloween indicator conveys an equity return anomaly in which the period from November through April has significantly stronger returns with lower volatility than the remaining months of the year. In response to this belief, stock traders sell at the start of May and the proceeds held in risk-free assets, and buy again around Halloween.
Test on the series 1973 through 1996
Bouman & Jacobsen (1997) tested a trading strategy of tactical asset allocation based on the Halloween effect with a basic model, dummy variable, holding a market portfolio of stocks during the period October 31 through April 30 and T-bills the remaining months. Like most straightforward analyses, they simply compare the annualized returns of stock market indices of 17 countries (in local currencies) and a broad world market index (in US dollars).
The regression results show that at 10% level, a significant sell in May-effect exists in 10 of the 17 countries, i.e., Australia, Belgium, Canada, France, Germany, Ireland, Italy, Japan, Netherlands, and UK. The mean returns are generally higher for the period November through April. The Halloween Indicator strategy outperforms the buy-and-hold strategy in almost all countries except in Hong Kong and South Africa over the 24-year period, 1973 through 1996.
Beyond that, Bouman and Jacobsen also test the statistical significance of the findings above, and confirm that the Halloween strategy is substantially less risky than investing in market indices in the respective countries.
It is worth noting that Merton and Henriksson (1981) developed a non-parametric test to evaluate market timing ability of investment managers. The analysis considers that the possibility that forecasting skills are different for bull markets and for bear markets – on average, the Halloween strategy renders well when judged on its ability to time bear and bull markets.
Test on the series 1982 through 2003
In 2002, Bouman and Jacobsen test the Halloween indicator in the U.S. equity markets and found similar results over the same time period, but using futures data over the period April 1982 – April 2003.
More recently, Jacobsen & Zhang extend the research to 108 stock markets using the historical data that back further (55,425 monthly observations) and they find that the November through April returns were on average 4.52% higher than returns of the remainder of the year. The Halloween strategy was successful in beating the market more than 80% of the time when employed over a five-year horizon, and more than 90% successful in beating the market over a 10-year horizon.
Test on the series 1926 through 2008
It is not quite clear about the causes of the Halloween effect. Although an interesting study by K. Stephen Haggard and H. Douglas Witte posit that this anomaly might be driven by outliers or the “January effect”. They use monthly-value-weighted and equal-weighted stock returns over the period 1926-2008 in the regression model, and construct mean-variance efficient portfolios using the method of Britten-Jones (1999) to determine whether investing in a Halloween strategy portfolio can result in risk-adjusted returns superior to those of a buy-and-hold market portfolio; more importantly, they further the study to examine the impact of transaction costs. The results show that Halloween effect in the U.S. stock market is significant in the period 1954-2008, and the Halloween effect is robust to consideration of outliers, the January effect, and transaction costs. It is an attractive anomaly given the low number of transactions required and the easily predictable dates of those transactions. More people will argue that the greater risk-adjusted returns available by investing in a Halloween portfolio will challenge the efficient markets hypothesis.
- The Halloween Indicator: Sell in May and go away by Sven Bouman and Ben Jacobsen (November 1997)
- On Market Timing and Investment Performance I. An Equilibrium Tehory of Value of Market Forecasts by Merton, R.C. (1981)
- On Market Timing and Investment Performance II, Statistical Procedures for Evaluating Forecasing Skills by Henriksson, R.D., and R.C. Merton (1981)
- The Halloween Indicator: Everywhere and All the Time by Jacobsen and Cherry Zhang