Using Correlation Analysis to Predict Trends

Article excerpt

Everyone wants to read tomorrow's paper today. Here's a new tool that can help you reach that goal and see if the trend really is your friend.

Have you ever wondered why profitable trading strategies undergo periods of substandard performance or a mediocre trading strategy sometimes works like tomorrow's Wall Street Journal? Correlation analysis, a form of statistical analysis, can help predict how well a relationship or pattern will work in the near future. It also can identify general market characteristics such as trending.

Correlation expresses the linear relationship between two variables. The most popular type of correlation, Pearson's correlation, measures how similar two data series called the independent and dependent variables move to each other relative to their mean. If the two series move in the same direction and proportion perfectly, the correlation is a 1. If they move in opposite directions in the same proportions perfectly, then the correlation is a -1. Typically the correlation is between -1 and 1.

Correlation analysis is a powerful tool to enhance the predictive power of intermarket analysis. Examining several intermarket relationships, we'll show how correlation analysis can be used to improve them.

Our first example uses the inverse relationship between crude oil and the dollar index to develop a simple trading system. The rules are:

If Dollar [less than] Average(Dollar,40) then buy crude at open;

If Dollar [greater than] Average(Dollar,40) then sell crude at open;

Results from Nov. 20, 1985 to Dec. 1, 1995, on one contract without slippage and commissions were:

Net profit $49,240 Average trade $339.59 Trades 145 Winners 86 Losers 59 Percent winners 59% Drawdown $11,800 Profit factor 1.97

Using correlation to filter out some of the trades, we'll only enter a new position when the 18-day correlation between the dollar and crude is less than -.1. The results using this filter are:

Net profit $42,730 Trades 76 Winners 50 Losers 26 Percent winners 66% Average trade $562.24 Drawdown $-5,350 Profit factor 2.82

Using simple Pearson's correlation, we've cut the drawdown in half while increasing the winning percentage to 65%. It also increased the profit per trade by 65%.

In our December 1995 article, we showed a pattern that buys on Monday's open when Treasury bonds are above their 26-day moving average. The current results from April 21, 1982 to Dec. 5, 1995, are:

Net profit $89,450 Trades 387 Average trade $231.14 Percent winners 55% Profit factor 1.65 Drawdown $-9,050

To improve these results we'll use a simple correlation between T-bonds and the S&P 500. Because our trades are based on the relationship between T-bonds and the S&P 500, we'll filter out trades when their link is weaker than normal. We'll only take trades when the 10-day correlation between T-bonds and the S&P 500 is greater than .40. From April 21, 1982 to Dec. 5, 1995, the results using $50 for slippage and commission are:

Net profit $83,525 Trades 227 Average trade $376.95 Percent winners 59% Profit factor ratio 2.18 Drawdown $-7,650

Filtering trades using correlation improves our average trade by 63%, while improving our winning percentage, drawdown, profit factor and win/loss ratio. We did this by filtering out about 160 trades that averaged less than $40 per trade. We could have used higher thresholds for our trigger, but that would have filtered out too many trades. For example, using an eight-day correlation and a .90 trigger produced over $500 per trade but only produced 43 trades in 13 years.

Using our simple Buy Monday when T-bonds are above their 26-day moving average pattern and an additional filter, we'll modify our rules: Buy at the open on a Monday when T-bonds are above their 26-day moving average and T-bonds closed higher than they opened Friday.

Requiring T-bonds to close higher than they opened Friday improved the performance of our original pattern. …