My results are similar to his; much of the back-tested performance of cointegrated baskets disappears once taking into account the bid-ask spread, and cointegration is fragile. Depressing, but better to find out sooner than later. I am still working on it.