Based on prices up to 19 Nov 2026 · model lgbm-h5-20261119-1545
Not investment advice
These are statistical estimates from a model trained on past prices only. They are often wrong, ignore news, company results and corporate actions, and are not recommendations to buy or sell. Past backtest results do not guarantee future results.
For the trading day after 19 Nov 2026 · NEPSE 2,628.67
Result: NEPSE closed −0.28% on 20 Nov 2026.
Last 30 calls: right on 18 (60%).
Tested on 2020–2026 (each year predicted by a model trained only on the years before): right 58.1% of days, against 48.8% for "always up" and 56.0% for "same direction as today". Experimental: an index model trained each night on NEPSE history and market breadth. Not advice.
NEPSE index closed at 2,628.67, +0.14% on the day. 280 stocks traded: 137 rose, 133 fell and 10 were unchanged. Total equity turnover was Rs 3.86 arba. Strongest sectors: Commercial Banks +0.55%, Manufacturing And Processing +0.19%. Weakest sectors: Others -1.45%, Investment -0.47%. Top gainers: BNL +6.00%, BNT +5.63%, GMLI +4.81%. Top losers: CFCL -6.53%, ECL -5.45%, PMLI -5.43%. The experimental AI model gives stocks an average 36.4% chance of closing higher in 5 trading days; 0.0% have a positive expected return.
Summary of today’s data (the AI writer was unavailable, so this is a fixed template).
Average chance up per sector, against the market average of 36.4%: green above it, red below.
The model’s real results since launch, scored after each forecast’s 5 trading days have passed. Not the backtest — shown even when it looks bad.
Past forecasts grouped by the chance up they were given, against how often those stocks really closed higher 5 trading days later. Well calibrated means the two columns are close.
| Chance up given | Forecasts | Average given | Actually rose |
|---|---|---|---|
| < 40.0% | 811 | 39.1% | 22.7% |
| 40.0%–45.0% | 1,461 | 41.6% | 37.9% |
| 45.0%–50.0% | 190 | 47.1% | 31.6% |
| 50.0%–55.0% | 22 | 50.9% | 50.0% |
| Forecast day | Live hit rate | Top 10% picks, average 5-day return | All scored stocks |
|---|---|---|---|
| 6 Nov 2026 | 74.5% | -0.06% | -1.64% |
| 5 Nov 2026 | 69.0% | +0.02% | -1.13% |
| 4 Nov 2026 | 82.2% | -1.06% | -1.67% |
| 3 Nov 2026 | 78.2% | -0.69% | -1.23% |
| 2 Nov 2026 | 67.7% | -0.28% | -0.58% |
| 30 Oct 2026 | 83.6% | -1.68% | -1.55% |
| 29 Oct 2026 | 86.3% | -2.42% | -1.97% |
| 28 Oct 2026 | 70.7% | +0.03% | -0.77% |
| 27 Oct 2026 | 42.4% | +1.65% | +0.68% |
| 16 Oct 2026 | 20.2% | +2.43% | +1.85% |
| Test year | AUC | Accuracy | Brier | Top 10% up | Rank IC |
|---|---|---|---|---|---|
| 2024 46,675 | 0.543 0.500 · Momentum 0.502 | 53.0% 50.2% | 0.249 0.252 | 52.8% 50.2% | 0.055 Momentum -0.062 |
| 2025 53,276 | 0.545 0.500 · Momentum 0.475 | 54.0% 56.3% | 0.245 0.247 | 46.2% 43.7% | 0.047 Momentum -0.072 |
| 2026 45,862 | 0.558 0.500 · Momentum 0.478 | 60.1% 59.1% | 0.241 0.244 | 44.2% 40.9% | 0.118 Momentum -0.092 |
Model / Baseline. Baselines: AUC 0.50 = coin flip; accuracy = always the more common outcome; Brier = always the training up-rate; top 10% = all stocks; momentum = last 5-day return.
Trained on 4,42,057 daily rows of 402 stocks, 2011-11-21 → 2026-11-05. Last trained 19 Nov 2026. Gradient-boosted trees (LightGBM) predict whether the close will be higher after 5 trading days. Walk-forward test: for each year, the model is trained only on earlier years and tested on that year. Probabilities are calibrated on the previous year and re-centred on the long-run share of stocks that close higher. Prices are adjusted for bonus and right shares; days with an adjustment we could not match to a recorded issue are excluded from training labels.