Accuracy — lstm_s3
validation only · 2.3M
Summary
| horizon | days | token-days | rank IC | t | LS decile | hit (alpha) | hit (price) | MAE (alpha) | MAE (price) |
|---|---|---|---|---|---|---|---|---|---|
| +1d | 265 | 179k | +0.0588 | 11.8 | +0.73% | 55.7% | 54.3% | 0.0356 | 5.783096 |
| +7d | 265 | 178k | +0.0725 | 14.1 | +1.83% | 61.0% | 61.2% | 0.0934 | 16.445183 |
| +30d | 265 | 177k | +0.1001 | 17.1 | +6.04% | 61.4% | 69.4% | 0.2082 | 33.258805 |
By month
| month | IC +1d | hit +1d | LS +1d | IC +7d | hit +7d | LS +7d | IC +30d | hit +30d | LS +30d | days |
|---|---|---|---|---|---|---|---|---|---|---|
| 2026-01 | +0.061 | 56.1% | +0.74% | +0.071 | 63.2% | +0.26% | -0.036 | 47.3% | -11.58% | 28 |
| 2025-12 | +0.031 | 56.3% | +0.70% | +0.040 | 60.1% | +2.73% | +0.071 | 63.7% | +4.76% | 31 |
| 2025-11 | +0.060 | 51.2% | +1.49% | +0.058 | 45.8% | +4.23% | +0.059 | 62.3% | +5.97% | 30 |
| 2025-10 | +0.050 | 59.1% | +0.58% | +0.044 | 65.9% | +0.23% | +0.138 | 46.1% | +13.90% | 31 |
| 2025-09 | +0.043 | 58.8% | +0.21% | +0.027 | 71.3% | -0.84% | +0.031 | 85.1% | +0.27% | 30 |
| 2025-08 | +0.070 | 49.8% | +0.69% | +0.080 | 49.2% | +2.08% | +0.115 | 57.9% | +9.13% | 31 |
| 2025-07 | +0.078 | 53.5% | +0.85% | +0.117 | 53.8% | +4.12% | +0.137 | 51.2% | +11.33% | 31 |
| 2025-06 | +0.053 | 59.8% | +0.38% | +0.065 | 74.5% | +0.08% | +0.130 | 64.0% | +5.28% | 30 |
| 2025-05 | +0.091 | 57.5% | +1.04% | +0.176 | 67.9% | +3.86% | +0.291 | 78.6% | +15.94% | 23 |
t-stats at +7d and +30d are optimistic: consecutive windows overlap, so the effective sample size is closer to the number of distinct days than the number of rows. +1d is the clean read. Long-short spreads are before trading costs and slippage, and the tail of the top-1000 by volume is thin.