Bone I'm not new to spread trading(did some commodity futures spread trades in 2008-2008) but I am new to stock/ETF/Index spread(pairs) trading. the "homework" I've done shows that even more than doing correlated pairs is to make sure that they are cointegrated. so I'm only trading them if the cointegration is over 90%. that and similar to what you mentioned, no pairs trades before any of the company's earnings releases
cointegration is like pregnancy. either cointegrated or no. there is no such thing as 90% cointegrated. you either reject or not reject. maybe you mean cannot reject at 90% confidence? maybe you mean 90% correlated?
NJrookie the information sources I've researched are telling me to the contrary. i.e. there are degrees/percentages of cointegration. there are multiple sites(and software available) computing this number(one site is using the Dickey Fuller formula to compute this, for example). and there are other formulas(Pearson) if you disagree with this, I guess you can take up with them. yet what I know(for now) is that a highly correlated pair is alright yet being highly cointegrated is better. and that a trader should screen for BOTH if you have traded pairs of course, and have been profitable, then feel free to respongd
I am pretty sure standard DF test is for confidence, ie you cannot reject at 5% / 10% / 1% confidence interval. I have trade pairs for over 3 years and I teach econometrics for living in a univ. njrookie
You could always make a directional trade with stop to protect your downside You can take a big loss on a hedged position all the same .. no easy money I would rather take a directional trade & manage the risk . less work to me
Cointegration is determined by a 'critical value' (a t-value in the form of a rational number) and an associated confidence level (decimal). Dickey Fuller is a test for unit roots. It does not produce output that says this pair is x% cointegrated. Yet a few sites do indeed tout what looks like a p-value and claim it's a % measure of cointegration. DF can tell you if the two data series before you might be mean reverting - which may, for example, assist correlation-based pairs trading. But that's different from using cointegration. A cointegration test (like Engle-Granger) applies DF (or ADF) to the level prices of a pair, their differences and then - if those two results are satisfactory - to their residuals. But it is possible at the end of the process to say 'this pair is more cointegrated than that one' if its t-value is more negative (and the two pairs considered are both at the same confidence level). Hope this helps.
you are trading. nothing is really cointegrated end of day unless they are same underlying with different leverage built in. forget all the academic cointegration test. it gives you a false sense of security if the test cannot reject cointegration. it is mostly lack of power or lack data in testing. if the correlation is too high, there is nothing to trade on. if too low you are cannot achieve hedging. you are managing a risk for short-term trading. the difference is not really meaningful.