One of the common Wall Street beliefs is that the market action during the month of January can be used as a barometer to predict how the market will perform throughout the rest of the year.
If this belief was accurate, we could develop an entire strategy around the predictive value that January had over the rest of the year. Even simpler, we could use January’s performance as a trend filter for the rest of the year. But is there any actual evidence to back up this widely held belief?
As quantitative traders, our first instinct is to test the accuracy of this belief using historic data. That is exactly what CXO Advisory Group did in a recent post. They pulled Robert Schiller’s long run sample data for the S&P Composite Stock Index dating all the way back to 1871. They also used the monthly closing data of the S&P 500 dating back to 1950.
Schiller’s S&P Composite stock data was calculated by taking the average of each day’s closing price during a given month. That makes it slightly different than the S&P 500 data, which simply uses the closing price of the last day of the month.
Testing Schiller’s S&P Composite Data
The first chart that the article discusses plots the return for each January compared to the return for the 11 month period immediately following it. What the chart shows is a wide range of data that has virtually no visual correlation.
When the author draws a best fit line through the data, we see that there is a very slight positive correlation between the January performance and the performance for the rest of the year. The article suggests that January performance accounts for 5% of the return of the rest of the year, which is a very weak correlation.
Moving forward, the author breaks the test period into three smaller periods of 47-48 years each. What this shows is that the middle period, from 1919 to 1965, had a much lower correlation than the other two periods. This is further evidence that the relationship between January and the rest of the year is unreliable.
Testing S&P 500 Data
When the same type of scatter plot is constructed using S&P 500 data, we see that the January data is responsible for 8% of the return for the rest of the year. While this is slightly better than the Schiller data, it is still a very weak correlation.
The next chart that the author provides shows us the ability of each of the twelve months to predict the returns of the 11 months that immediately follow. Using the Schiller data, we see that January is no better at predicting the next 11 months than April, May, or December. Using the S&P data, January is the best predictor, but it is still very poor.
Further analysis shows that the slight edge that January appears to have using the S&P 500 data is largely skewed towards the past data. When the S&P 500 data is separated into two different periods, the more recent data shows no edge at all of January.
The First Five Days of January
One common variation on this January theory is that the first five days of January are a good predictor for the rest of the year. Putting together another scatter plot, the author shows us that this relationship is also very weak.
Using the S&P 500 data, the first five days of January only explain 7% of the variation over the rest of the year. This relationship is also skewed towards the older data.
In addition to this January theory, CXO Advisory Group also recently covered the topic of the January Effect and proved that to be false as well. It appears that each and every January is just another month on our trading calendar.