I admit it; I'm an amateur at the stock market. I have a set of 4 machine learning models for trying to predict S&P500 stocks that will make a move of at least plus or minus .5% on the next trading day. I use Interactive Brokers to get the data and place the trades (mostly paper trading at this point). I enter positions at the open of normal trading hours and exit them at the normal close. Its very rare that I force it to exit early; I've been burned too many times by panic exits. I mainly use market orders for entry and exit, with the entry orders using IB's adaptive algo.
I've been running some of these models for almost a year, and results have been interesting enough to make me think I'm on to something…the models are pretty good at finding stocks that will make a big move (not always right about the direction though). More often than not, at least one of the models will have predictions that will make over 1% total on a given day. One of the models would have given a YoY increase of about 30%. But, of course, the catch is that all of that is based off of the gains you would see from the official open and closing price.
Going by official open/close prices, my best model over the past month would have done over 8%, but the paper trades couldn't even break 3% Good predictions aren't worth anything if the entry prices are shit. I say all of this to ask this question: Am I on to something, or am I wasting my time? The predictions I generate seem to be good, so is the divergence between the "official" open prices and what I get something I can solve with more sophisticated trade placing logic, or is it basically a crap shoot?
Submitted August 10, 2019 at 07:41PM by dgerdem