Stocks & Commodities V. 22:4 (58-60): Come Here Quick, Durbin Watson by Ron McEwan

Stocks & Commodities V. 22:4 (58-60): Come Here Quick, Durbin Watson by Ron McEwan

Here’s how you can use the Durbin Watson statistic to measure the autocorrelation of two securities.

One of the most powerful statistical tools traders have at their disposal is the ability to measure the correlation between two sets of time series data.

There are many approaches to this. One is to measure the relationship of stock prices (usually the closing prices). Another method, common among portfolio analysts, is to measure the correlation of the returns (daily, weekly, or monthly) of the underlying data. The idea is that you would not want too many securities in the portfolio that are highly correlated with each other (you do not want the same kind of eggs in your basket). Yet another method is to measure the correlation of the residuals of a regression line that has been applied to the data. This is referred to as autocorrelation.

MEASURING AUTOCORRELATION

If your goal is to find autocorrelation in a database of stocks, then a simple measure of correlation will not do. While you can have high correlation between the closing prices of two stocks as well as returns of stocks (Figure 1), the level of autocorrelation, or correlation of residuals, might be very low (Figure 2).

There are times when you may want to avoid stocks that trade in a similar fashion to each other, but there are also times when autocorrelation can be helpful in developing a trading strategy.