Hi QC Community

Welcome back to the series of demo algorithms on ideas of using alternative datasets. This time, we'll be using the Brain ML Stock Ranking dataset. This dataset implemented multiple ML classifiers to estimate which quintile the stock's return will lie on, normalized into a value between -1 and 1.

In the research notebook, we're trying to see if the data has the predictability for a one-week active return. To do so, we need to check the correlation between the dataset and one-week forward active return. Then, we can check how do we make use of this correlation and make predictions.

Our demo algorithm will be using 2, 3, 5 days stock ranking prediction data, and use PCA for dimension reduction and SVM for classification on weekly active return direction as an entry signal. Position size will be based on mean-reversion and beta-neutral. Please check the embedded research notebook and backtest for details. Enjoy!

Best
Louis Szeto

Author