Correlation Confound (Part 1)

Correlation is mentioned as a key factor in two different trading/investing contexts. In this mini-series, I’m going to describe correlation, how traders can make use of it, and a couple missing pieces (confounds) to avoid unexpected failures.

I will begin by explaining both words in the title.

Correlation is a measure of how often two variables change together. A correlation of +1 between two stocks means historically, when one stock was up 5% the other was also up 5%. A correlation of -1 means historically, when one stock was up 5% the other was down 5%. A correlation of zero means historically, no relationship between the stocks’ price changes occurred. Correlation can range from -1 to +1.

In science, a confounding variable is “an extraneous variable in an experimental design that correlates with both the dependent and independent variables.”

Ice cream [example] can better help me illustrate this. Suppose a correlation between murder rates and ice cream sales is observed. If murder rates go up (down) when ice cream sales increase (decrease) then ice cream sales drive murder rates, right? This is less likely if some other variable is also found to be correlated with murder rates. That variable would then confound our initial model. Suppose it is also observed that as seasonal temperature increases (decreases), people buy more (less) ice cream and spend more (less) time outdoors where criminals run the streets. It makes logical sense for seasonal temperatures, not ice cream sales, to affect murder rates. Seasonal temperature is a confounding variable.

In the next post I will start to explain confounding variables that prevent correlation from doing its job.