November 24th, 2011, 4:18 pm
Ok, can I ask one thing ? Have you checked that the stock prices are actually I(1) before doing the cointegration ? ie are you unable to reject a unit root in the individual prices ? I find it strange that you are able to reject a unit root on the pair with so few crosses of the mean. I guess on 4 years worth of data we are talking like 1000 data points, so would have thought critical level at 1% on adf statistic would be something like ~ -3.5. That might be wrong as I don't have my adf stats with me at the mo. But I know it's pretty strict. Back when I did these sorts of analysis, sometimes I'd get a successful cointegration because one of the stocks had risen and then tanked. So the success / mean reversion etc was coming from that one name, and if you test it alone you couldn't say it was I(1), and hence it wouldn't actually make a good trade although the pair would pass the adf test.
Last edited by
blade on November 23rd, 2011, 11:00 pm, edited 1 time in total.