Although our results are better than linear regression, we can further improve them by removing the linear regression, thus, leaving the base learners as follows:
base_learners = [('SVR', SVR()), ('KNN', KNeighborsRegressor())]
This further improves the MSE, reducing it to 15.71. If we utilize this model as a trading strategy, we can achieve a Sharpe value of 0.21; considerably better than simple linear regression. The following table summarizes our results:
Metric |
SVR-KNN |
SVR-LR-KNN |
MSE |
15.71 |
16.22 |
Sharpe |
0.21 |
0.22 |
Voting ensemble results