New research project this Sunday. I've faded a few overnight gaps unsuccessfully and realized I lack data on this. This is all manual work. I only went back two months. I will continue maintaining this dataset. Not sure I'll go further back as I pretty much answered what I was wondering about. In summary - don't fade overnight gaps blindly. They may not fill at all and even if they do it could take 20 hours. It's however interesting to note that 92 % did fill by the end of the RTH session, though. So, if an overnight gap is not filled by RTH - it could be a high probability trade.
Wonderful idea. I did stats like this years ago as well. Not on gaps, because I didn't know too much about them, but simply something like, what happens at the overnight low or high during RTH, or at the previous day high or low. The idea was, could I make a trade to fade these levels thinking that it would provide at least a 10 point profit with a 3 or 5 point stop. I looked at hundreds of days, and although there were some nuances (ie. sometimes we came within 1 point, sometimes poked 1 point, so the entry criteria wasn't perfect), the end results for all of them hovered around 40-60%. Now yes, a 50% win rate if its a 1:3 RR trade was very good, but like I say, there were nuances, but the fact that it settled around the 50% figure was disappointing. With these gap fills, I would bet that if in fact the gap fill 90% of the time, the R:R will be something like 10:1, which will almost completely wipe out any edge on its own. (ie. for a trade that ends up filling, you might have to take a 50 point drawdown just to get a 5 point win). Now don't get me wrong, collecting these stats is important, but I was just disappointed with how there was no apparent edge there if only looking at one variable. Perhaps a gap fill stat combined with another key level makes the expectancy positive, so I'm hoping this goes somewhere!
For this specific exercise, I was only interested in seeing what typically happens when the market gaps on the electronic Open. Off memory, I seemed to recall they typically fill fairly quickly, but after having done this exercise I see that I was wrong. I completed another research project on Saturday. I basically want to see when the market typically bottoms out overnight. Here's the data for one specific day type: My regular statistical model is fully automated, but it's set up for RTH. I've been doing my overnight work maually so far.