Blog - Investing Notes

December 26, 2008 - "Sell in May" Over the Long Run

Does the conventional wisdom of "Sell in May" work over the long term? To check, we turn to the very long run dataset of Robert Shiller, which offers monthly levels of the S&P Composite Stock Index since 1871. We split the investing year into two half-years (seasons): November (October close) through April (April close), and May (April close) through October (October close). Using April and October closes for the S&P Composite Stock Index from 1871 through 2008 (276 seasons), we find that...

The following chart shows the cumulative values of $1.00 initial investments in two seasonal approaches: (1) 100% in stocks during May-October and 100% in cash (0% return) during November-April; and, (2) 100% in stocks during November-April and 100% in cash during May-October. November-April in stocks generates a much higher terminal value ($56.34 versus $3.63), with most of its outperformance occurring since about 1960.

Over the entire sample period, the average seasonal return during May-October (November-April) is 1.8% (3.6%), with standard deviation 12.7% (11.5%). The higher return during November-April is therefore not associated with higher risk.

For a different perspective on time variation of returns, we look at average seasonal returns by decade.

The next chart compares average seasonal returns by decade across the entire sample period. "Sell in May" works well since 1951, but not so well before. In fact, while November-April beats May-October by 5.5% per season from 1951-2008, it loses to May-October by 0.8% per season during 1871-1950.

The existence of long subperiods during which "Sell in May" underperforms weakens the argument for a biological/psychological explanation of seasonal returns (such as Seasonal Affective Disorder).

In summary, the conventional wisdom to "Sell in May" has worked well for U.S. stocks on average since the 1950s, but did not consistently work well before then.

For related research, see Blog Synthesis: Calendar Effects. See especially the blog entries of 12/20/07, 3/14/06 and 8/31/05.