Coverage-Free Pair: Every Row Scored On Real Data

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

Overall top-100 hits
26.27
average across horizons
Predicted top-100 avg actual rank
228.54
lower is better
Rank IC
0.0712
picks the better of two stocks 52.3% of the time
IC reliability
10.48
t-stat; IC positive 69.1% 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 25.58 233.56 0.0545
6m 26.50 227.47 0.0783
12m 26.75 224.58 0.0807

Run Note

test summary
Hypothesis: establish how much of the IC edge survives when no missing-data artifact can contribute. Fundamentals are 60-78% missing so any model reading them partly scores whether a row HAS data. Evidence this is real: adding a bare coverage indicator to strategy_000109 is the ONLY single addition out of the whole feature bank that improves it in all 3 blocks and both halves, and has_market_cap alone scores IC 0.0459 t 11.89 while market cap with missing left neutral scores 0.0044 t 0.99. Change from strategy_000109: drop the two fundamental terms, keep only from_low_252d (99.5% coverage) and volume (100%), equal weight, no fitted parameter. Result IC 0.07119 (t 10.48, positive 69.1%), hits 26.275, avg actual rank 228.54. That is BELOW 109's 0.08077 but still comfortably ABOVE strategy_000079's 0.06572 with two features instead of eleven, and it is positive in all 3 blocks and both halves (0.0919/0.0466/0.0752, 0.0726/0.0696). Searching all 9,086 two/three/four-way combinations of the 22 fully covered price and volume columns returns this exact pair as the best of them -- nothing larger wins, which is the opposite of an overfit signature. The two are uncorrelated at 0.02. Conclusion: roughly 0.0712 of IC is unambiguously real and about 0.0096 of 109's edge rides on fundamental coverage. Verdict vs 109: worse on IC, better on honesty; both beat the incumbent. Next: recover hits@100, which both models give up because the volatility term that pays for top-100 overlap is absent -- test adding vol_63d back at partial weight.

Run Details

strategy_test_results #229
id
229
strategy_id
strategy_000110
strategy_title
Coverage-Free Pair: Every Row Scored On Real Data
strategy_hash
df430416965f95d086f0510d3192c1c211f490151607bad20ddcf16034afe567
evaluator_version
v3_top100_monthly
start_date
2015-01-01
end_date
2025-06-01
created_at
2026-07-25 13:55:29
updated_at
2026-07-25 13:55:29

Sample Top-100 Lists

2020-03-31 / 6m / 25 hits

Predicted top 100

AAPL, ABBV, ABT, AES, AFL, AIG, AMD, ANET, AVGO, BA, BMY, BSX, CAT, CCL, CI, CMG, CNC, CNP, COP, CVX, DAL, DG, DRI, DVN, DXC, DXCM, ETN, EXC, FANG, FCX, FITB, FTNT, FTV, GE, GILD, GM, HAL, HIG, HON, HPE, HPQ, HWM, INTC, KEY, KLAC, KMI, KR, LKQ, LLY, LNC, LOW, LRCX, LYB, LYV, MDT, MET, MGM, MO, MOS, MPC, MS, MSFT, MU, NCLH, NEE, NEM, NFLX, NKE, NOW, NVDA, ODFL, OKE, OXY, PLD, PPL, PRU, QCOM, RCL, RF, ROST, RTX, SO, SRE, SYF, SYY, TGT, TJX, TSCO, TSLA, UAL, URI, VLO, VRTX, VTR, WDC, WELL, WMB, WY, WYNN, XOM

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, AMD, AVGO, CMG, DRI, DXCM, FCX, HAL, LOW, MGM, MOS, NCLH, NKE, NOW, NVDA, QCOM, RCL, SYF, TGT, TSCO, TSLA, URI, VTR, WMB, WY

Strategy Script

df43041696
"""Generated V3 stock-ranking strategy."""
from __future__ import annotations

import numpy as np
import pandas as pd


STRATEGY_ID = "strategy_000110"
STRATEGY_TITLE = "Coverage-Free Pair: Every Row Scored On Real Data"
DESCRIPTION = """
An honesty check on strategy_000109, and on the whole IC leaderboard.

Fundamental columns are 60 to 78 percent missing, so any model that reads them
partly scores whether a row has data rather than what the data says. That is not
hypothetical here: strategy_000079's second largest IC contributor is its
log-market-cap term, and a bare has_market_cap indicator scores IC 0.0459 at
t 11.89 while market cap ranked with missing values left neutral scores 0.0044
at t 0.99. Adding a raw coverage indicator to strategy_000109 is also the only
single addition out of the entire feature bank that improves it in all three
blocks and both halves, which says the remaining headroom in that direction is
an artifact and not skill.

So this model uses only columns that are present for essentially every row:

  from_low_252d   distance held above the 252-day low   (99.5 percent coverage)
  volume          traded share volume                   (100 percent coverage)

Averaged as date-local rank percentiles, equally weighted, no fitted parameter.

Searching all 9,086 two-, three- and four-way combinations of the 22 fully
covered price and volume columns returns this pair as the best of them. Nothing
larger wins, which is the opposite of what an overfit would look like. The two
are also almost perfectly uncorrelated at 0.02, so each is carrying its own
information rather than restating the other.

The point of the run is the floor it establishes: whatever this scores is IC
that cannot be explained by missing-data patterns.
"""
FACTORS_USED = [
    "from_low_252d",
    "volume"
]
PARAMETERS = {
    "from_low_252d": 1.0,
    "volume": 1.0
}
COMPLEXITY = 2
PARENT_STRATEGY_ID = "strategy_000109"
GENERATION = 110
RANDOM_SEED = 20261100
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 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)