Our results are slightly worse than the final Voting ensemble, so we will attempt to improve them by removing the linear regression, as we did with the voting ensemble. By doing so, we can slightly improve our model, achieving an MSE of 16.16 and a Sharpe value of 0.22. Comparing it to voting, stacking is slightly better as part of an investing strategy (the same Sharpe value and a slightly better MSE), although it is unable to achieve the same level of predictive accuracy. Its results are summarized in the following table:
Metric |
SVR-KNN |
SVR-LR-KNN |
MSE |
16.17 |
16.16 |
Sharpe |
0.21 |
0.22 |
Stacking results