In-Depth Evaluation of Our High-Performing Models
We reviewed binary classification stock predictions made by our top-performing models. Notably, 56% of these predictions correctly anticipated the market’s direction—just a bit above the 50% expected by random chance. We simulated investing $1,000 on every “Up” prediction, the overall return on investment (ROI) came out to only about 1.75%. Over the same time period, if you had invested the money in $SPY you would have made approximately 2.5%. This modest gain highlights that even with a slight edge in prediction accuracy, the real-world benefit can be limited. The interactive chart below illustrates these results, showing how individual trade gains and losses over time tend to follow the broader trends of the market and hit big when market is going up.
We analyzed parameter choices frequently found in our top-performing models. In the tables below, Usage Rate represents the percentage of total occurrences of a parameter value that contributed to a top-performing model.
Value Range | Usage Rate |
---|---|
25-29 | 16.32% |
30-34 | 15.04% |
20-24 | 13.16% |
15-19 | 7.19% |
10-14 | 3.28% |
5-9 | 1.27% |
Value Range | Usage Rate |
---|---|
700-799 | 11.16% |
800-899 | 10.83% |
900-999 | 9.28% |
1000-1099 | 7.88% |
1100-1199 | 6.82% |
Below is an ordered list of the most impactful features (highest to lowest), along with their approximate contribution scores. Each feature is briefly defined for clarity: