A-TIER technique
Efficiency-ratio forecaster — skip choppy days to lift a strategy's metrics
100k ft — What is this?
Efficiency-ratio forecaster — skip choppy days to lift a strategy's metrics
50k ft — Why care?
Instead of predicting direction, a logistic regression on 15 price features forecasts the next day's Kaufman efficiency ratio (trendiness versus chop). Trades on predicted low-efficiency days made about a third of the money, so gating them out improved nearly every metric and held on two years of out-of-sample data.
10k ft — Key takeaways
• Confirm on unseen data; here it held for two years out of sample
• Predict the regime (trend vs chop), not the direction, and use it as a trade filter
• Split existing trades by the predicted regime to validate the signal before gating
5k ft — Tools & techniques
(no tools tagged)
Kaufman efficiency ratio, algorithmic trading, logistic regression, regime gating
1k ft — Deep-dive / try it
I built a machine learning model that predicts the next day’s efficiency ratio.
Comment “MODEL” to see what it predicts every day. #machinelearning #daytrading #futurestrading #tradingstrategy #trading
Transcript excerpt
I built a machine learning model that predicts how the market will move. It improved almost every performance metric on one of my strategies. Instead of trying to predict direction, the model predicts the next day's efficiency ratio. The efficiency ratio basically measures how efficiently price moves in one direction. A high ratio means price moved clean in one direction while a low ratio means price chopped back and forth. The model makes its predictions with a logistic regression that weights 15 different measurements of how price has been moving. To see if these predictions actually mean anything, I took one of my strategies and I split each trade based off what the model predicted. And the difference was pretty clear. High efficiency days performed the best while the low efficiency days made about a third of the money per trade. So I added a gate that skips those low efficiency days entirely, and that one change improved almost every performance metric. And more importantly, that improvement held on two years of data the model's never seen. If you want to see what this model predicts every day, comment model and I'll send it to you.