In looking to refine my stock market forecasting, I looked at a number of variables to try and find a relationship. I’ve laid them out below, but I’ll spare you the suspense: none of them worked.

That’s not necessarily a bad thing. There are two problems with trying to create a model to forecast stock market movements. The first is that the stock market is where people go to place bets (if you’ll excuse the gambling reference) on the future performance and the future value of stocks. Therefore, the stock market is forward-looking and it is generally considered to be a leading indicator of economic outcomes. The only way to anticipate the stock market is to anticipate business outcomes farther ahead or faster or more accurately than other traders and investors. Those people are professionals and have the resources of large firms at their disposal, so that’s just not going to happen.

The other problem is overfitting. Stock price movements contain information about the aggregate outlook for the company (which changes daily) and for the stock (which fluctuates with supply and demand). The information is a signal, but it is swamped with noise. Stock research, including fundamental and technical, is trying to identify the signal (trend) and ignore the noise. When a model becomes too complex, it is possible for it to capture noise along with (or instead of) the signal, which will lead to faulty interpretation and incorrect forecasting. So I’m happy to rule out potential relationship because it helps me avoid overfitting.

The movement of interest rates affects the financial inputs of businesses, however I found no correlation between interest rate changes and the following month stock market movement.

The change in commodity prices affects the material inputs of businesses, especially manufacturers. There was very little correlation (may explain 2% of change) between commodity price changes and the following month stock market movement. Rather, I found that stock prices and commodity prices tend to move together.

There is a theory that trading volume (supply & demand) produces technical signals. For example, high volume on rising rising prices implies increasing demand and should lead to higher prices. High volume on falling prices implies increasing supply and should lead to lower prices. However, the monthly volume has a low correlation with the following month stock market movement.

The last variable I looked at was employment / unemployment numbers, and I found almost no correlation with the following month stock market movement. Because of the time it takes to compile and publish employment statistics, the numbers are from two months ago, so it would have to influence stock returns three months later, for example through increased savings (from higher employment) or increased selling (from unemployment). This doesn’t appear to be the case.

Model so far:

    • This can be refined, given the month of the year (although it’s been less pronounced recently):
Monthly average increase % positive
January 0.95% 59.46%
February 0.92% 59.46%
March 0.57% 56.76%
April 0.89% 56.76%
May 1.39% 62.16%
June -0.43% 45.95%
July 0.67% 62.16%
August 0.68% 59.46%
September -1.62% 43.24%
October -0.15% 56.76%
November 1.46% 64.86%
December 1.83% 81.08%
 Average monthly increase 0.60%
  • Stocks are likely to rise 59% of the time. May +5%, June -10%, September -15%, October -5%, November +5%, December +15%
  • If last month was positive, this month is 5% more likely to rise (coefficient of correlation of 0.233)
  • If last month was negative, this month is 7% less likely to rise

I freely admit that it’s not a great model. I’d only expect to be right about 3 times out of 5, and that’s only about direction, not magnitude. But sometimes I think it’s better to be aware that we really don’t know, than to think we know and be surprised. It gives us more chance to protect against or prepare for risk.

Forecasting refinement – overfitting

Leave a Reply

Your email address will not be published. Required fields are marked *