My case differs in that the large number of stocks in my portfolio, all S&P500, exhibited volatility that did not show up in the S&P500 -- not even close. I emphasize "large number" because the more of the S&P500 in the portfolio, the less departure from it can be explained by chance. Perhaps the momentum algorithm was ironically picking up on which stocks had the most short term, intraday, volatility and this translated into the strange interday pattern -- somehow. I have difficulty with this explanation because it seemed to me the intrAday pattern was a pure, big time, loser. Why would intrAday volatility strategy yield a portfolio with returns that are the inverse of intErday?
Randomness? Making money intrrA-day requires skill, but being on the right side of an overnight gap is just luck. I don’t want to offend any swing traders, I am generalizing here to make a point.
Like I told newwurldmn, "chance" or, to use your word "randomness" doesn't seem likely here due to 4 factors which occurred in conjunction: The sample size of the S&P500 symbols was 70. The one day gain (5/7)of the portfolio was 7 times that of the S&P500 The one day loss (5/6) of the portfolio was 10 times that of the S&P500 The distribution of gains/losses across the 70 stocks didn't concentrate unreasonably (ie: "luckily") on a few. It appears that the short term (minutes) momentum strategy was picking stocks for longer-term (days) volatility -- and that during the purchase it was on a downward trend of that longer term volatility (which is another factor to consider -- although that could just be the bid/ask spread since these were market orders).
Short term often will result in crumps, meaning all produce like wise profits. Scalping is often made by "noise" with/against trend, what I do for much smaller profits. I know by my own testing, especially forward testing, many fills will not happen or slippage occurs to tune of 50% of trades.
I'm not the guy to advise you on the math of this, but people analyze correlations within their portfolios quite seriously for precisely this reason. Your algo may well be picking a subset of 70 S&P500 companies that are more correlated than the index average. I assume you have already thought about this, but if you are like me you might have lazily skipped doing the analysis so far. Apologies if you already properly handled this.
Yes, indeed, I did lazily skip doing the analysis because, while it is obvious -- by definition -- why a short term strategy would, over the short term, pick stocks that were conditionally correlated (ie: the condition being the short term strategy) it didn't make sense to me why they would be correlated on longer term volatility. I'll go get the eigenvectors now.