(Picture is below) I just want some opinions and theory discussion. Here, I've got two time series plotted (red and blue) that trade almost exactly the same -- they are the cumulative return of the stocks from time 0. For the first 20,000 cycles or so, you can see the difference in return (green) doesn't blow up and stays more or less in the same band. At one point, however, the difference just blows up. However, even though the difference relative to the initial starting point is "blown up", it's clear that the general movement of the two stocks are similar. In theory, one could re-anchor the start point and trade the pair again, right? I'm trying to understand the mechanics of the market and why cointegration or error correction models break down, even though post-breakdown in this case, the two time-series seemingly move together correctly again. Perhaps there was a shift in overall volatility, where one stock's volatility increased and the other's didn't increase to the same degree, even though their drift terms stayed the same. Or, my data is wrong; I'd like some insights from others on how they deal with such things -- looking at the pair trading journals, they seem to only dive in when pearson correlations cross a certain threshold. In some of the cases in the pair trading journal blog, it almost seemed as if the trader was trading out a spurious correlation, where there was no underlying APT/CAPM/other factorization model that suggested the two stocks paired together even really had a relationship. <img src="http://media.chartvibe.com/correlationdrop.jpg">