Gated Core With Volatility: Best All-Round Model

Back to leaderboard · 2015-01-01 to 2025-06-01 · v3_top100_monthly

Overall top-100 hits
28.17
average across horizons
Predicted top-100 avg actual rank
225.72
lower is better
Rank IC
0.0782
picks the better of two stocks 52.5% of the time
IC reliability
10.70
t-stat; IC positive 69.3% of months
not adjusted for overlapping windows

Horizon Breakdown

top 100 only
Horizon Average Top-100 Hits Predicted Top-100 Avg Actual Rank Rank IC
3m 27.65 230.76 0.0640
6m 28.35 225.15 0.0843
12m 28.50 221.24 0.0863

Run Note

test summary
Synthesis of the five-run search. Combines all four findings: 109's four signals individually positive in all 3 blocks and both halves; 110's evidence that half of them need no fundamental data so the edge is not a coverage artifact; 111's finding that volatility is additive rather than competing because it wins a membership count without ordering anything; and 112's eligibility gate, which reaches part of the ordering no continuous weight can. Model: four equal-weight date-local ranks (from_low_252d, volume, free_cash_flow_margin_pct, revenue_growth_pct) plus vol_63d at 0.75, with the deepest 5% of from_low_252d dropped each date. Result IC 0.07816 (t 10.70, positive 69.3%), hits 28.168, avg actual rank 225.72. Against strategy_000079 this improves EVERY metric the evaluator reports simultaneously: IC 0.07816 vs 0.06572 and higher in all 3 blocks and both halves (0.0943/0.0518/0.0887, 0.0761/0.0805 vs 0.0764/0.0419/0.0791, 0.0641/0.0675), hits 28.168 vs 27.965, avg actual rank 225.72 vs 232.69. Paired t vs 079 is 3.91 across 375 observations. IMPROVED as an all-round model; it is NOT the IC leader -- strategy_000112 is, at 0.08232, and that is the one to use if rank IC is the only target. Settings are not tuned: every combination of vol weight 0.75-1.0 and gate 3-10% dominates 079 on all three metrics, a plateau eight configurations wide, and this point was taken from the middle rather than the peak. Next: the search is exhausted within the current feature set -- ~9,000 rank combinations plus roughly 60 structural variants all failed to beat the 109 core, and the only thing that ever improved it was the data-coverage indicator. The coverage flag's own IC decays from 0.0754 in block 1 to 0.0133 in block 3, so the artifact is fading as the database fills. Real headroom now needs new data, not new strategy shapes: backfill fundamentals for the ~722 tickers with none, then re-test, and re-score the 11 derived price features (semivol, skew, beta, idiosyncratic vol, Amih

Run Details

strategy_test_results #232
id
232
strategy_id
strategy_000113
strategy_title
Gated Core With Volatility: Best All-Round Model
strategy_hash
88da218b7c1c6c11dd13850efffcfdd966a311b39e51ab6e5327e4e40646c773
evaluator_version
v3_top100_monthly
start_date
2015-01-01
end_date
2025-06-01
created_at
2026-07-25 14:07:07
updated_at
2026-07-25 14:07:07

Sample Top-100 Lists

2020-03-31 / 6m / 29 hits

Predicted top 100

AAPL, ABT, ADBE, AES, AFL, AIG, AMD, AMT, ANET, APTV, AVGO, AXP, BMY, CCL, CDNS, CF, CFG, CI, CMG, CNC, CNP, COP, CRM, DAL, DRI, DVN, DXC, DXCM, ETN, ETSY, EXC, EXPE, FANG, FITB, FTNT, GE, HAL, HIG, HUM, HWM, INTC, ISRG, IVZ, KEY, KLAC, KMI, KO, LEN, LKQ, LNC, LRCX, LYB, LYV, MA, MCHP, MCO, MET, MGM, MOS, MPC, MS, MSFT, NCLH, NEE, NEM, NFLX, NOW, NVDA, NWL, O, OKE, ORCL, OXY, PLD, PRU, QCOM, RCL, RF, ROST, RTX, SPG, SPGI, SRE, STT, SWKS, SYF, SYY, TER, TJX, TPR, TSLA, UAL, UNM, URI, V, VLO, VRTX, VTR, WELL, WMB

Actual top 100

AAP, AAPL, ABMD, ADBE, ADSK, ALB, ALGN, AMD, AMP, AMZN, APD, APH, APTV, ATVI, AVGO, BBY, BWA, CARR, CDNS, CE, CHRW, CMG, CMI, CPRT, CRM, CTAS, CTLT, CTSH, DD, DE, DHI, DHR, DRI, DXCM, EBAY, EMN, ETSY, EXPE, FCX, FDX, GLW, GPS, HAL, HD, IDXX, IPGP, JCI, KMX, LEG, LEN, LH, LOW, MAS, MCHP, MGM, MOS, NCLH, NKE, NOW, NSC, NVDA, NWS, NWSA, ORLY, PAYC, PH, PHM, PNR, POOL, PVH, PWR, PYPL, QCOM, QRVO, RCL, ROK, ROL, SHW, SIVB, SNPS, SWK, SWKS, SYF, TDG, TEL, TER, TGT, TMO, TSCO, TSLA, TT, UPS, URI, VAR, VIAC, VTR, WHR, WMB, WST, WY

Matched tickers

AAPL, ADBE, AMD, APTV, AVGO, CDNS, CMG, CRM, DRI, DXCM, ETSY, EXPE, HAL, LEN, MCHP, MGM, MOS, NCLH, NOW, NVDA, QCOM, RCL, SWKS, SYF, TER, TSLA, URI, VTR, WMB

Strategy Script

88da218b7c
"""Generated V3 stock-ranking strategy."""
from __future__ import annotations

import numpy as np
import pandas as pd


STRATEGY_ID = "strategy_000113"
STRATEGY_TITLE = "Gated Core With Volatility: Best All-Round Model"
DESCRIPTION = """
The synthesis of this search. Four findings, combined:

  strategy_000109  four signals individually positive in all three blocks and
                   both halves, equally weighted, beat the incumbent on IC by
                   23 percent
  strategy_000110  two of the four need no fundamental data, so the edge is not
                   a missing-data artifact
  strategy_000111  volatility is additive rather than competing, because it wins
                   a membership count without ordering anything
  strategy_000112  an eligibility gate reaches a part of the ordering that no
                   continuous weight can

The model is the four robust ranks, plus trailing volatility at three quarter
weight, with the deepest 5 percent of drawdowns each date made ineligible.

  from_low_252d              distance held above the 252-day low
  volume                     traded share volume
  free_cash_flow_margin_pct  cash conversion
  revenue_growth_pct         top-line growth
  vol_63d                    63-day trailing volatility, weight 0.75
  gate                       lowest 5 percent of from_low_252d dropped

The two settings are not tuned. Every combination of volatility weight from 0.75
to 1.0 and gate from 3 to 10 percent beats strategy_000079 on rank IC in all
three blocks and both halves, on hits@100, and on average actual rank at the same
time. That is a plateau eight configurations wide, so this point was taken from
the middle of the region rather than its peak, and the choice inside the region
changes almost nothing.

This is not the highest IC on the board; strategy_000112 is, at 0.0823, and that
is the model to use if rank IC is the only thing being optimised. This one is
the model to use if hits@100 still has to hold up, because it is the strongest
result that improves every metric the evaluator reports at once.
"""
FACTORS_USED = [
    "from_low_252d",
    "volume",
    "free_cash_flow_margin_pct",
    "revenue_growth_pct",
    "vol_63d"
]
PARAMETERS = {
    "from_low_252d": 1.0,
    "volume": 1.0,
    "free_cash_flow_margin_pct": 1.0,
    "revenue_growth_pct": 1.0,
    "vol_63d": 0.75
}
GATE_PERCENTILE = 0.05
COMPLEXITY = 6
PARENT_STRATEGY_ID = "strategy_000112"
GENERATION = 113
RANDOM_SEED = 20261130
CREATED_AT = "2026-07-25T00:00:00+00:00"


def _date_group(df):
    if "trade_date" in df.columns:
        return df["trade_date"]
    return pd.Series(0, index=df.index)


def _rank_pct(df, column):
    """Date-local rank percentile centred on zero; missing values stay neutral."""
    if column not in df.columns:
        return pd.Series(0.0, index=df.index)
    clean = df[column].replace([np.inf, -np.inf], np.nan)
    ranked = clean.groupby(_date_group(df)).rank(pct=True)
    return ranked.fillna(0.5) - 0.5


def eligibility_filter(df):
    """Drop the deepest 5 percent of drawdowns each date; the rest stay eligible."""
    return _rank_pct(df, "from_low_252d") > (GATE_PERCENTILE - 0.5)


def calculate_score(df):
    score = pd.Series(0.0, index=df.index)
    for column, weight in PARAMETERS.items():
        score = score + weight * _rank_pct(df, column)
    total = sum(abs(w) for w in PARAMETERS.values()) or 1.0
    score = score / total
    return score.replace([np.inf, -np.inf], np.nan).fillna(0.0)