I'm recently playing a lot with correlation trades.. I noticed that in many cases, even when the correlation % is very high, on average one of the two tends to go up more and go down less, and so over time there will be a noticeable difference on the returns. For lack of better terms, I'll call this "relative drift" Is there any mathematically sound/consolidated practice to measure this factor?
https://www.pairtradinglab.com/analyses/X5ovm15f7pEAhKSj Does higher p-value mean pair is better CoIntegrated?
No, lower p-value is better. Also VIXY/SPY is a poor choice of pairing, since their relationship is likely to be non linear. GAT
Do a scatter plot over a longer time period (the website seems to be using like 30 observations, which is ridiculously tiny. This is quite a large and complex subject. You're going to need to do some reading. The ultimate work A bit lighter Rob
Thanks for the references. In layman terms, is Correlation the "statistical tendency for two instruments to move in the same direction" whilst CoIntegration the "statistical frequency for two overlay price lines to intersect each other"? Some pairs move well together but they do not revert to mean on ratio plot often.
I don't recognise eithier of those definitions. The first isn't bad, but the second is completely wrong. Correlation is a measure of contemporaneous linear co movement. If two things are positively correlated, then yes they will tend to move in the same direction. But we normally talk about the correlation of returns not prices; so positive correlation is more 'if X goes up, then Y is more likely than not to go up'. Cointegration is much harder to define in laymans terms without being inaccurate. In slightly more complex terms, if the price series x and y are cointegrated, then there exists some equation y_t = a + bx_t + e_t where the error term e is stationary i.e. it's not drifting and has constant expected volatility. Rob