qbuntuqbuntu
marginalno_edge· equity· cross-sectional· factor· machine-learning

LightGBM Factor Model 13f-financial (JP Growth Stocks)

slug: ml-factor-lgbm-13f-jp-growth

LightGBM Factor Model 13f-financial (JP Growth Stocks) backtest. Median Sharpe across 72 parameter sets is +0.42. p-value is 0.20. Max DD is -16.2%. Period: 2024-07-01 to 2025-06-30. Best Sharpe +2.91 is not the representative figure.

Verdict: marginal. Some parameter set has a p-value under 0.10, but the median itself misses promising (Sharpe > 0.5 and p-value < 0.05). Primary reason: no_edge. It is not regime, frequency, or cost, and the median does not clear the bar.

Performance summary

Sharpe (median)
+0.42
Max DD
-16.2%
Win rate
—
Alpha (ann.)
—
p-value
0.20
backtest period: 2024-07-01 → 2025-06-30
Sharpe across 72 grid runs:median 0.42best 2.91 (not representative)worst -1.61p25–p75 -0.97–1.46

Hypothesis

Adding 4 financial features from J-Quants (sales_growth_yoy, operating_margin, operating_margin_change, eps_growth_yoy) to the ml-factor 9f baseline improves prediction accuracy and Sharpe ratio. Phase 1 showed 9f as the only promising strategy (Sharpe 1.65) but financial_store data (505 stocks, 2023-12~) was never tested.

Why it didn't work

Across 72 grid combinations, median Sharpe=0.42, best=2.91. Primary failure mode: no_edge.

Facts (5)

Best grid Sharpe ratio is 2.91 (p=0.004).
period 2024-07-01–2025-06-30; cost=5.0bps
sharpe_ratio2.91
cite ↗ /cite/ml-factor-lgbm-13f-jp-growth-fact-best-sharpevia r-ml-factor-lgbm-13f-jp-growth-0073
Median Sharpe across 72 grid runs is 0.42.
Full grid of 72 parameter combinations.
sharpe_ratio0.42
cite ↗ /cite/ml-factor-lgbm-13f-jp-growth-fact-median-sharpe
Median Sharpe in DOWN regime is 3.83 across 72 runs.
Regime=DOWN, N225 SMA20/SMA60 classification.
sharpe_ratio3.83
cite ↗ /cite/ml-factor-lgbm-13f-jp-growth-fact-regime-down
Median Sharpe in FLAT regime is -2.21 across 72 runs.
Regime=FLAT, N225 SMA20/SMA60 classification.
sharpe_ratio-2.21
cite ↗ /cite/ml-factor-lgbm-13f-jp-growth-fact-regime-flat
Median Sharpe in UP regime is 0.94 across 72 runs.
Regime=UP, N225 SMA20/SMA60 classification.
sharpe_ratio0.94
cite ↗ /cite/ml-factor-lgbm-13f-jp-growth-fact-regime-up

Runs (30)

runparamssharpeUPFLATDOWN
feature_set=9f-base, num_leaves=7, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0001
n_short: 10
num_leaves: 7
feature_set: "9f-base"
+1.26+1.83-0.28+3.25
feature_set=9f-base, num_leaves=7, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0002
n_short: 10
num_leaves: 7
feature_set: "9f-base"
+1.62+2.41-1.44+10.04
feature_set=9f-base, num_leaves=7, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0005
n_short: 15
num_leaves: 7
feature_set: "9f-base"
+1.28+1.81-1.25+13.37
feature_set=9f-base, num_leaves=7, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0007
n_short: 20
num_leaves: 7
feature_set: "9f-base"
-0.75+0.28-2.20+0.17
feature_set=9f-base, num_leaves=15, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0010
n_short: 10
num_leaves: 15
feature_set: "9f-base"
+2.25+4.02+0.44+2.73
feature_set=9f-base, num_leaves=15, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0013
n_short: 15
num_leaves: 15
feature_set: "9f-base"
+1.17+3.55-0.83+0.95
feature_set=9f-base, num_leaves=15, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0016
n_short: 20
num_leaves: 15
feature_set: "9f-base"
+0.62+3.30-1.76+0.90
feature_set=9f-base, num_leaves=31, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0019
n_short: 10
num_leaves: 31
feature_set: "9f-base"
+2.09+2.75+0.60+3.74
feature_set=9f-base, num_leaves=31, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0020
n_short: 10
num_leaves: 31
feature_set: "9f-base"
+0.38+0.73-1.64+21.08
feature_set=9f-base, num_leaves=31, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0023
n_short: 15
num_leaves: 31
feature_set: "9f-base"
+0.46+0.73-1.36+22.54
feature_set=9f-base, num_leaves=31, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0026
n_short: 20
num_leaves: 31
feature_set: "9f-base"
+0.22+0.47-1.37+18.95
feature_set=13f-financial, num_leaves=7, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0029
n_short: 10
num_leaves: 7
feature_set: "13f-financial"
+2.43+3.82-2.42+30.27
feature_set=13f-financial, num_leaves=7, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0032
n_short: 15
num_leaves: 7
feature_set: "13f-financial"
+2.38+4.02-3.08+22.01
feature_set=13f-financial, num_leaves=7, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0035
n_short: 20
num_leaves: 7
feature_set: "13f-financial"
+1.66+3.21-3.36+26.35
feature_set=13f-financial, num_leaves=15, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0038
n_short: 10
num_leaves: 15
feature_set: "13f-financial"
+1.80+2.72-1.80+11.01
feature_set=13f-financial, num_leaves=15, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0041
n_short: 15
num_leaves: 15
feature_set: "13f-financial"
+1.86+2.91-1.60+4.36
feature_set=13f-financial, num_leaves=15, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0044
n_short: 20
num_leaves: 15
feature_set: "13f-financial"
+1.41+2.52-2.30+4.04
feature_set=13f-financial, num_leaves=31, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0047
n_short: 10
num_leaves: 31
feature_set: "13f-financial"
+1.99+2.90-2.05+27.21
feature_set=13f-financial, num_leaves=31, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0050
n_short: 15
num_leaves: 31
feature_set: "13f-financial"
+1.94+2.64-1.42+15.73
feature_set=13f-financial, num_leaves=31, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0053
n_short: 20
num_leaves: 31
feature_set: "13f-financial"
+1.79+2.95-3.01+14.57
feature_set=12f-quality, num_leaves=7, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0055
n_short: 10
num_leaves: 7
feature_set: "12f-quality"
+2.04+1.52+1.73+3.53
feature_set=12f-quality, num_leaves=7, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0058
n_short: 15
num_leaves: 7
feature_set: "12f-quality"
+1.41+0.39+1.44+2.98
feature_set=12f-quality, num_leaves=7, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0061
n_short: 20
num_leaves: 7
feature_set: "12f-quality"
+1.02+0.27+0.77+2.66
feature_set=12f-quality, num_leaves=15, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0064
n_short: 10
num_leaves: 15
feature_set: "12f-quality"
+1.95+1.33+2.69+1.75
feature_set=12f-quality, num_leaves=15, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0067
n_short: 15
num_leaves: 15
feature_set: "12f-quality"
+1.84+1.39+2.75+1.02
feature_set=12f-quality, num_leaves=15, n_short=20
r-ml-factor-lgbm-13f-jp-growth-0070
n_short: 20
num_leaves: 15
feature_set: "12f-quality"
+1.43+1.57+2.11-0.14
feature_set=12f-quality, num_leaves=31, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0073
n_short: 10
num_leaves: 31
feature_set: "12f-quality"
+2.91+1.40+3.80+3.92
feature_set=12f-quality, num_leaves=31, n_short=10
r-ml-factor-lgbm-13f-jp-growth-0074
n_short: 10
num_leaves: 31
feature_set: "12f-quality"
+0.59+1.67-2.97+13.34
feature_set=12f-quality, num_leaves=31, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0077
n_short: 15
num_leaves: 31
feature_set: "12f-quality"
+0.63+1.35-2.07+8.32
feature_set=16f-all, num_leaves=7, n_short=15
r-ml-factor-lgbm-13f-jp-growth-0085
n_short: 15
num_leaves: 7
feature_set: "16f-all"
-1.61-0.14-3.21-0.54

Target universe

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

Related strategies

variant:ml-factor-lgbm-12f-jp-growthvariant:ml-factor-lgbm-16f-jp-growthvariant:ml-factor-lgbm-9f-jp-growth
source: qbuntu-analysis · kabu STATION REST API + J-Quants daily bars
computed asof: 2026-09-24T02:22:15
view as Markdown ↗view as JSON ↗license: CC-BY-4.0

Related strategies