Actually....~6/20 would be a confidence of 5%..... but hey, not everyone understands how confidence intervals and the Null Hypothesis work.
Source? Their win rate is publicly and clearly posted at somewhere between 52 and 55%. I can't remember the exact percentage right now. Unless you are talking about the percentage of days they are profitable. Which is something else imo.
the rebate game is all about moving faster to get your place in the queue when the price changes... so it is a natural place for machines to displace humans (as they have been doing for years) since it doesn't need a lot of intelligence, but rather a lot of speed...
I'm not positive, but I believe rebates are taken into consideration with that 52% (or so) win rate. So, if anyone can devise a trading strategy that wins around 53% of the time (loss and gain being equal) and can trade it thousands of times a day.. You too, can go years without a losing day. Easy-peasy.
The α-less independent trader is done. To make money now, you need a predictive model with α. But that's as it should be. The market doesn't owe anyone "$20 bills on the sidewalk."
OK I got sucked into this "fad" and started the machine learning course from Coursera. I'm also watching a ton of stuff on Youtube as well. It's all really fascinating. Seems very interesting, but it sure looks like everyone and there dog is jumping into this space. the number of "analysts" competing on Kaggle is over 320k now. Scary. I'll stick to discretionary trading for now.
Machine learning was in fashion in the late 1980s when I was in graduate school. I remember professsors thinking that they will become too rich with ML.When they lost their money they got around understanding that its principles apply to problems that require curve-fitting, which is exactly the opposite of what you need in trading. You will meet many in forums who think they can do confidence intervals and solve the problems. Those are usually paper traders. Read just the introduction to this blog. This is the best analysis that debunks ML.