Machine Learning in Financial Markets
What are we trying to Predict
The action of making a prediction will change the outcome in the real world. Ie. You are in the game, like chess.
Predicting Returns is the hardest thing you can do
Some predictions do not matter for “reflexivity”.
Weather, gasoline consumption are not reflexive.
The point is to exeed the accuracy of traditional models.
What are you trying to predict?
Time horizone: 1 hour, 1 week, 1 months, 1 year
Predict just sign or magnitude?
People from outside of economics would think a prediction of sign greater than 50% would be a success. The machine learning algorithms needs to beat the human trader. Most traders are greater than 50% for mean reversion and momentum traders.