metadata
tags:
- finance
- nse
- stock-prediction
- xgboost
- lightgbm
- catboost
license: mit
P170 NSE Global Ensemble — Backtested
Trained: 2026-05-15 Data: 1987 symbols, 3.66M rows, 10 years (2016-2026) Features: 98 (OHLCV + technical + macro + regime)
Backtest Results (256K trades, 343 symbols, 2017-2026)
| Metric | Value |
|---|---|
| Win rate | 64.3% |
| Avg trade return | +1.12% (after costs) |
| Profit factor | 3.03 |
| Sharpe | 8.65 |
Walk-Forward (10/10 profitable years)
| Year | Win% | Sharpe |
|---|---|---|
| 2022 (bear) | 66.0% | 9.37 |
| 2020 (COVID) | 64.0% | 8.92 |
| 2017-2026 avg | 64.4% | 8.75 |
Optimal Signal Filter
Confidence >= 0.50: win=71.3%, sharpe=11.9, pf=4.3 Confidence >= 0.55: win=73.6%, sharpe=13.2, pf=4.9