My best guess, from your description, is that selling on the close did better because the market went up and you likely have a long strategy. You are getting the benefit of an extra up day in a bull market.
rmorse, The model is actually selling on today's close or the _next_day_open. Same for buying on close or the next day open. It's cutting the trade a tad shorter than it would otherwise be. I would've thought waiting til the next morning to sell would help (benefit from reversion trend).
So for example, if I get a signal this afternoon - I could either buy at close today or open tomorrow morning. Buying at open tomorrow seems to work better. My initial random thought is it's taking advantage of current dip/downtrend I'm targetted. (For example, TEF was a signal from yesterday to see what the model is selecting - it's finding mean reversion opportunities of stocks still below short term ma).
Similarly, when I get the sell signal in say 3 days, I could either sell at close in 3 days, or sell on open the next morning of 4th day. In that case, though, selling on close seems to work better.
I'll have to think about why - and it could just be due to peculiarities of the model - but it's not obvious to me why this should be the case. I was pretty surprised when I saw the consistency of the results. I'm thinking about why it would make that much difference over lots of trades, but seems like it does. Also, not sure if benefit comes mostly from the buy part of activity or the sell - have to investigate that.
All buy/sell at open/close combinations work, but some consistently better than others. I have to examine more to understand.