# ML Factor with Meta-Label Filter (JP Growth Stocks)

- **Slug**: `ml-factor-meta-label-jp-growth`
- **Verdict**: marginal (no_edge)
- **Asset class**: equity
- **Categories**: factor, machine-learning, meta-label
- **Source**: qbuntu-analysis (computed asof 2026-09-24T02:22:16)
- **Reproducibility**: analysis-engine@1455e77, backtest-lib@17a7674, grid tier=shallow (python 3.12.10, numpy 2.4.2, pandas 3.0.0)
- **HTML**: https://qbuntu.ai/strategies/ml-factor-meta-label-jp-growth
- **JSON**: https://qbuntu.ai/v1/strategies/ml-factor-meta-label-jp-growth
- **License**: CC-BY-4.0

## Hypothesis

Applying meta-labeling (2-stage P(win) filter via LogisticRegression) to ml-factor predictions filters low-confidence signals, improving Sharpe at the cost of lower trade frequency. meta_label.py is implemented but never deployed.

## Why it didn't work

Across 48 grid combinations, median Sharpe=0.66, best=3.14. Primary failure mode: no_edge.

## Performance summary

Headline figures are the **median across the parameter grid**, not the best run.

| Metric | Value |
|---|---|
| Sharpe (median) | 0.6576 |
| Max drawdown | -30.6% |
| p-value | 0.061 |
| Backtest period | 2024-07-01 to 2025-06-30 |

Sharpe distribution across 48 grid runs: median 0.66, best 3.14 (best-case — not representative), worst -3.47, p25 -1.07, p75 2.19.

## Methodology

- Notes: {"validation": "walk-forward (3 folds, 2023-12-01 onward)", "walk_forward_definition": {"type": "single_pass", "description": "Full period backtest without walk-forward split", "verdict_basis": "full_period"}}
- Slippage model: open-on-open + variable bps fixed
- Position sizing: equal_weight
- Regime definition: N225 SMA20/SMA60 three-class
- Transaction cost bps: 5

## Target universe

Tokyo Stock Exchange Growth Market, ~495 stocks, tv >= 30M JPY

## Runs (30)

| run_id | params | sharpe | UP | FLAT | DOWN |
|---|---|---:|---:|---:|---:|
| `r-ml-factor-meta-label-jp-growth-0001` | n_short_max=10, meta_threshold=0.5, base_num_leaves=15 | 2.49 | 4.02 | 1.18 | 2.74 |
| `r-ml-factor-meta-label-jp-growth-0002` | n_short_max=10, meta_threshold=0.5, base_num_leaves=15 | -0.31 | 0 | -2.09 | 13.2 |
| `r-ml-factor-meta-label-jp-growth-0004` | n_short_max=10, meta_threshold=0.5, base_num_leaves=31 | 1.89 | 2.75 | 1.01 | 2.15 |
| `r-ml-factor-meta-label-jp-growth-0005` | n_short_max=10, meta_threshold=0.5, base_num_leaves=31 | -0.89 | -0.63 | -2.28 | 21.08 |
| `r-ml-factor-meta-label-jp-growth-0007` | n_short_max=15, meta_threshold=0.5, base_num_leaves=15 | 1.75 | 3.55 | -0.07 | 2.49 |
| `r-ml-factor-meta-label-jp-growth-0008` | n_short_max=15, meta_threshold=0.5, base_num_leaves=15 | -0.95 | -0.58 | -2.7 | 12.38 |
| `r-ml-factor-meta-label-jp-growth-0010` | n_short_max=15, meta_threshold=0.5, base_num_leaves=31 | 2 | 2.83 | 0.42 | 4.55 |
| `r-ml-factor-meta-label-jp-growth-0011` | n_short_max=15, meta_threshold=0.5, base_num_leaves=31 | 0.19 | 0.44 | -1.5 | 22.54 |
| `r-ml-factor-meta-label-jp-growth-0013` | n_short_max=20, meta_threshold=0.5, base_num_leaves=15 | 0.87 | 3.3 | -1.6 | 2.18 |
| `r-ml-factor-meta-label-jp-growth-0014` | n_short_max=20, meta_threshold=0.5, base_num_leaves=15 | -1.1 | -0.7 | -3.1 | 10.02 |
| `r-ml-factor-meta-label-jp-growth-0016` | n_short_max=20, meta_threshold=0.5, base_num_leaves=31 | 1.31 | 2.18 | -0.27 | 3.88 |
| `r-ml-factor-meta-label-jp-growth-0017` | n_short_max=20, meta_threshold=0.5, base_num_leaves=31 | -0.06 | 0.11 | -1.37 | 18.95 |
| `r-ml-factor-meta-label-jp-growth-0019` | n_short_max=10, meta_threshold=0.55, base_num_leaves=15 | 2.76 | 4.06 | 1.41 | 3.95 |
| `r-ml-factor-meta-label-jp-growth-0020` | n_short_max=10, meta_threshold=0.55, base_num_leaves=15 | -1.06 | -1.27 | -1.42 | 19.72 |
| `r-ml-factor-meta-label-jp-growth-0022` | n_short_max=10, meta_threshold=0.55, base_num_leaves=31 | 2.25 | 2.18 | 1.81 | 3.13 |
| `r-ml-factor-meta-label-jp-growth-0023` | n_short_max=10, meta_threshold=0.55, base_num_leaves=31 | 0.32 | 0.69 | -1.62 | 17.43 |
| `r-ml-factor-meta-label-jp-growth-0025` | n_short_max=15, meta_threshold=0.55, base_num_leaves=15 | 2.42 | 3.34 | 1.58 | 3.19 |
| `r-ml-factor-meta-label-jp-growth-0026` | n_short_max=15, meta_threshold=0.55, base_num_leaves=15 | -0.47 | -0.28 | -1.91 | 15.65 |
| `r-ml-factor-meta-label-jp-growth-0028` | n_short_max=15, meta_threshold=0.55, base_num_leaves=31 | 2.97 | 3.31 | 1.01 | 7 |
| `r-ml-factor-meta-label-jp-growth-0029` | n_short_max=15, meta_threshold=0.55, base_num_leaves=31 | 0.13 | 0.54 | -1.52 | 55.83 |
| `r-ml-factor-meta-label-jp-growth-0031` | n_short_max=20, meta_threshold=0.55, base_num_leaves=15 | 1.88 | 3.3 | 0.6 | 2.89 |
| `r-ml-factor-meta-label-jp-growth-0032` | n_short_max=20, meta_threshold=0.55, base_num_leaves=15 | -1.07 | -0.72 | -2.49 | 15.75 |
| `r-ml-factor-meta-label-jp-growth-0034` | n_short_max=20, meta_threshold=0.55, base_num_leaves=31 | 2.37 | 3.01 | 0.73 | 5.09 |
| `r-ml-factor-meta-label-jp-growth-0035` | n_short_max=20, meta_threshold=0.55, base_num_leaves=31 | 0.45 | 0.19 | 0.55 | 58.41 |
| `r-ml-factor-meta-label-jp-growth-0037` | n_short_max=10, meta_threshold=0.6, base_num_leaves=15 | 1.89 | 1.5 | 1.62 | 3.38 |
| `r-ml-factor-meta-label-jp-growth-0038` | n_short_max=10, meta_threshold=0.6, base_num_leaves=15 | -2.05 | -2.15 | -1.94 | 2.7 |
| `r-ml-factor-meta-label-jp-growth-0040` | n_short_max=10, meta_threshold=0.6, base_num_leaves=31 | 2.38 | 2.19 | 1.92 | 3.68 |
| `r-ml-factor-meta-label-jp-growth-0046` | n_short_max=15, meta_threshold=0.6, base_num_leaves=31 | 3.14 | 2.87 | 1.83 | 6.82 |
| `r-ml-factor-meta-label-jp-growth-0055` | n_short_max=10, meta_threshold=0.65, base_num_leaves=15 | 1.29 | 1.56 | 0.02 | 4.04 |
| `r-ml-factor-meta-label-jp-growth-0062` | n_short_max=15, meta_threshold=0.65, base_num_leaves=15 | -3.47 | -3.45 | -3.69 | -0.6 |

## Facts (5)

Each fact has a permanent citation URL. AI agents quoting qbuntu should use these.

- **Best grid Sharpe ratio is 3.14 (p=0.0019).**
  - Value: 3.14 sharpe_ratio
  - Context: period 2024-07-01–2025-06-30; cost=5.0bps
  - Supports verdict: `marginal`
  - Cite: https://qbuntu.ai/cite/ml-factor-meta-label-jp-growth-fact-best-sharpe

- **Median Sharpe across 48 grid runs is 0.66.**
  - Value: 0.66 sharpe_ratio
  - Context: Full grid of 48 parameter combinations.
  - Supports verdict: `marginal`
  - Cite: https://qbuntu.ai/cite/ml-factor-meta-label-jp-growth-fact-median-sharpe

- **Median Sharpe in DOWN regime is 4.27 across 48 runs.**
  - Value: 4.27 sharpe_ratio
  - Context: Regime=DOWN, N225 SMA20/SMA60 classification.
  - Supports verdict: `marginal`
  - Cite: https://qbuntu.ai/cite/ml-factor-meta-label-jp-growth-fact-regime-down

- **Median Sharpe in FLAT regime is -0.40 across 48 runs.**
  - Value: -0.4 sharpe_ratio
  - Context: Regime=FLAT, N225 SMA20/SMA60 classification.
  - Supports verdict: `marginal`
  - Cite: https://qbuntu.ai/cite/ml-factor-meta-label-jp-growth-fact-regime-flat

- **Median Sharpe in UP regime is 0.88 across 48 runs.**
  - Value: 0.88 sharpe_ratio
  - Context: Regime=UP, N225 SMA20/SMA60 classification.
  - Supports verdict: `marginal`
  - Cite: https://qbuntu.ai/cite/ml-factor-meta-label-jp-growth-fact-regime-up

## Related strategies

- same_strategy: [ml-factor-lgbm-9f-jp-growth](https://qbuntu.ai/strategies/ml-factor-lgbm-9f-jp-growth)

---

Cite as: qbuntu (2026). ML Factor with Meta-Label Filter (JP Growth Stocks). https://qbuntu.ai/strategies/ml-factor-meta-label-jp-growth

Visit qbuntu.ai — the honest scorecard for quant strategies (live, promising, marginal, rejected).