Compare models
21 registered · 21 with usable out-of-sample days · scope: all splits (incl. in-sample)
By cell type
| cell | horizon | models | rank IC (mean ± sd) | spread across models | LS decile | hit (alpha) |
|---|---|---|---|---|---|---|
| gru | +1d | 11 | +0.0932 ± 0.0082 | +0.0806 … +0.1118 | +1.34% | 56.1% |
| gru | +7d | 11 | +0.1268 ± 0.0173 | +0.1102 … +0.1758 | +5.88% | 61.8% |
| gru | +30d | 11 | +0.1780 ± 0.0272 | +0.1609 … +0.2587 | +19.51% | 67.1% |
| lstm | +1d | 10 | +0.0832 ± 0.0063 | +0.0749 … +0.0937 | +1.18% | 55.7% |
| lstm | +7d | 10 | +0.1153 ± 0.0075 | +0.1073 … +0.1339 | +5.11% | 61.4% |
| lstm | +30d | 10 | +0.1593 ± 0.0123 | +0.1491 … +0.1914 | +16.97% | 66.7% |
Two-stage aggregation: each model's daily rank-IC series is averaged to one score, then those scores are averaged and spread across models. So ±sd is the model-to-model (seed) spread, not day-to-day noise — that distinction is the whole point of a seeded sweep.
Note this group holds every registered model of that cell type, including any that are not part of the seeded sweep (
Note this group holds every registered model of that cell type, including any that are not part of the seeded sweep (
beta_5y, for instance, which is a gru model but was trained separately). The Paired by seed section below is the one that corresponds to compare_cells.py.Paired by seed
Same seed = same initialisation stream and the same data, so per-seed differences cancel seed-level noise. Values are lstm − gru in rank IC.
| horizon | seeds paired | mean Δ | ± SE | t | win rate | relative | verdict |
|---|---|---|---|---|---|---|---|
| +1d | 10 | -0.0097 | 0.0022 | -4.36 | 10% lstm | -10.4% | gru better |
| +7d | 10 | -0.0113 | 0.0038 | -3.00 | 0% lstm | -9.0% | gru better |
| +30d | 10 | -0.0191 | 0.0059 | -3.26 | 0% lstm | -10.7% | gru better |
These figures are all splits (incl. in-sample).
compare_cells.py selects on the validation split only — choose the val only pill above and this table reproduces its output exactly. Verified to 4 decimals against the 10-seed run: gru +0.0745/+0.0920/+0.1190 vs lstm +0.0646/+0.0810/+0.1036, paired Δ −0.0099/−0.0110/−0.0154 with t = −3.94/−4.06/−5.28. Under any other scope the numbers legitimately differ, because they cover more dates. The val-only figure is the one the GRU-over-LSTM decision was made on.Every model
| model | cell | params | data end | IC +1d | LS +1d | hit +1d | IC +7d | LS +7d | hit +7d | IC +30d | LS +30d | hit +30d | days |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| beta_5y | gru | 250,771 | 2026-10-21 | +0.0973 | +1.33% | 56.4% | +0.1280 | +5.65% | 62.2% | +0.1732 | +18.98% | 67.5% | 1,761 |
| gru_s0 | gru | 250,771 | 2026-10-21 | +0.0966 | +1.36% | 56.3% | +0.1283 | +5.96% | 61.2% | +0.1717 | +18.61% | 66.6% | 1,761 |
| gru_s1 | gru | 250,771 | 2026-10-21 | +0.0937 | +1.34% | 56.3% | +0.1241 | +5.73% | 62.0% | +0.1666 | +18.84% | 67.1% | 1,761 |
| gru_s2 | gru | 250,771 | 2026-10-21 | +0.1118 | +1.84% | 56.9% | +0.1758 | +8.13% | 62.9% | +0.2587 | +26.82% | 68.5% | 1,761 |
| gru_s3 | gru | 250,771 | 2026-10-21 | +0.0905 | +1.25% | 56.4% | +0.1160 | +5.46% | 62.2% | +0.1659 | +18.50% | 66.7% | 1,761 |
| gru_s4 | gru | 250,771 | 2026-10-21 | +0.0806 | +1.26% | 53.8% | +0.1102 | +5.32% | 60.7% | +0.1609 | +18.39% | 64.1% | 1,761 |
| gru_s5 | gru | 250,771 | 2026-10-21 | +0.0921 | +1.26% | 56.4% | +0.1196 | +5.44% | 62.3% | +0.1696 | +18.55% | 67.9% | 1,761 |
| gru_s6 | gru | 250,771 | 2026-10-21 | +0.0881 | +1.27% | 55.5% | +0.1209 | +5.62% | 60.5% | +0.1780 | +18.96% | 67.5% | 1,761 |
| gru_s7 | gru | 250,771 | 2026-10-21 | +0.0836 | +1.28% | 56.3% | +0.1180 | +5.74% | 62.3% | +0.1669 | +18.69% | 67.9% | 1,761 |
| gru_s8 | gru | 250,771 | 2026-10-21 | +0.0942 | +1.25% | 56.3% | +0.1232 | +5.53% | 61.9% | +0.1694 | +18.91% | 67.5% | 1,761 |
| gru_s9 | gru | 250,771 | 2026-10-21 | +0.0969 | +1.30% | 56.1% | +0.1305 | +6.10% | 61.9% | +0.1769 | +19.35% | 66.7% | 1,761 |
| lstm_s0 | lstm | 316,819 | 2026-10-21 | +0.0768 | +1.19% | 56.0% | +0.1135 | +5.44% | 61.9% | +0.1543 | +16.45% | 67.0% | 1,761 |
| lstm_s1 | lstm | 316,819 | 2026-10-21 | +0.0856 | +1.16% | 56.2% | +0.1172 | +5.44% | 62.0% | +0.1570 | +18.21% | 64.8% | 1,761 |
| lstm_s2 | lstm | 316,819 | 2026-10-21 | +0.0937 | +1.40% | 55.4% | +0.1339 | +5.98% | 61.3% | +0.1914 | +20.25% | 67.4% | 1,761 |
| lstm_s3 | lstm | 316,819 | 2026-10-21 | +0.0789 | +1.14% | 56.0% | +0.1092 | +5.00% | 61.9% | +0.1491 | +15.92% | 67.0% | 1,761 |
| lstm_s4 | lstm | 316,819 | 2026-10-21 | +0.0749 | +1.11% | 53.2% | +0.1099 | +4.71% | 58.2% | +0.1538 | +16.29% | 65.5% | 1,761 |
| lstm_s5 | lstm | 316,819 | 2026-10-21 | +0.0753 | +1.15% | 56.1% | +0.1073 | +4.47% | 61.9% | +0.1557 | +16.28% | 67.1% | 1,761 |
| lstm_s6 | lstm | 316,819 | 2026-10-21 | +0.0855 | +1.19% | 56.2% | +0.1129 | +4.69% | 62.0% | +0.1554 | +16.32% | 67.1% | 1,761 |
| lstm_s7 | lstm | 316,819 | 2026-10-21 | +0.0858 | +1.17% | 56.2% | +0.1157 | +5.39% | 61.9% | +0.1612 | +17.07% | 67.2% | 1,761 |
| lstm_s8 | lstm | 316,819 | 2026-10-21 | +0.0873 | +1.17% | 56.1% | +0.1193 | +5.00% | 61.7% | +0.1658 | +16.85% | 67.2% | 1,761 |
| lstm_s9 | lstm | 316,819 | 2026-10-21 | +0.0878 | +1.12% | 55.9% | +0.1144 | +4.97% | 61.4% | +0.1499 | +16.06% | 66.5% | 1,761 |