Cointegration is a confidence level that when two securities (long and short) deviate in valuation, they will revert to the mean. Here’s a more technical definition given by economists Robert Engle and Clive Granger:
"If each individual stock price series exhibits a random walk (non-stationary) but a linear combination of them is stationary, then they are said to be cointegrated.”
Cointegration is an essential part of our statistical based spread trading strategy. Because this strategy entails trading in and out of numerous spreads we can’t possibly know everything about all of the securities we trade. Instead we use the statistical history of a spread as a proxy for detailed knowledge of each security. We believe that over the long run (but definitely not the short run) the law of one price is enforced in financial markets. That is if two securities have the same payout they will have the same price. We are interested in finding spreads that when they have deviated from the law of one price in the past, consistently returned to that price. The cointegration confidence interval tells us the likelihood that we have found such a spread.
Suppose you trade spreads on fundamentals only. Can you also use cointegration? Why not? As we have said before, mash up different ideas and recipes. There is no reason why a trader can’t combine different trading approaches. We will write more about some of these statistical approaches in the future.
Written by Norm Winer. Follow me on Twitter and StockTwits.