Blog - Investing Notes

January 18, 2008 - The January Barometer Retested (Updated)

As goes January, so goes the rest of the year? In their November 2007 paper entitled "How Accurate is the January Barometer?", Ben Marshall and Nuttawat Visaltanachoti examine the ability of January returns to predict February-December returns in the U.S. and other equity markets. They apply multiple robustness tests to determine the statistical and economic significance of results. Using U.S. stock return data spanning 1940-2006 and stock return data for 22 other countries and the world spanning 1970-2006, they conclude that:

• The January Barometer for rest-of-the-year stock returns in the U.S. is profitable even after adjusting for known risks and accounting for a variety of statistical challenges.
• Over the sample period, a January Barometer portfolio that is fully invested in U.S. stocks (Treasury bills) for the rest of the year after a positive (negative) January beats a buy-and-hold strategy by an average 3-4% per year.
• However, evidence against the validity of the January Barometer in other markets is strong. The differences in rest-of-the-year returns following positive and negative Januaries are statistically significant in only two of 23 other markets (Japan and Spain), and these two cases do not stand up to all robustness tests.
• The lack of explanatory power in international markets proves that the January Barometer is weaker than momentum or the Halloween indicator, both of which are just as strong internationally as in the U.S.

The following chart summarizes top-level findings from the paper for the January Barometer across a very broad sample of U.S. stocks. The statistical significance is strong and survives risk-adjustment, correction for data mining bias and other statistical robustness tests.

In summary, though lacking theoretical foundation, evidence indicates that the January Barometer works for the broad U.S. stock market. However, it does not work elsewhere.

As a separate test on an index, we use monthly returns for the S&P 500 index (based on monthly opening and closing levels) for 1950-2007 to relate the change in the index for February-December to the change for January. The following scatter plot depicts the results. The Pearson correlation between the two series is 0.30 and the R-squared statistic is 0.09, suggesting that the January return explains 9% of the return for the balance of the year. Excluding the outlier at the lower right of the distribution (1987), the R-squared statistic increases to 0.15.

Are some parts of the January return distribution more reliably predictive than others?

The next plot recasts the data in the prior one by ordering the S&P 500 index returns for January from lowest (-7.6%) to highest (+13.2) for 1950-2007. The horizontal axis is therefore not time-sequential. This view of the data suggests that the January Barometer holds mostly for intermediate values of January returns. At the extremes of the January return distribution, returns for the rest of the year do not track January results.

For related research, see Blog Synthesis: Calendar Effects.