beneish_m_score_high
forensic default: short
Definition
Beneish 5-factor M-Score above the manipulator threshold (-1.78). Composite of DSO, gross margin, asset quality, sales growth, depreciation.
Trigger logic
Simplified Beneish M = -4.84 + 0.92×DSRI + 0.528×GMI + 0.404×AQI + 0.892×SGI + 0.115×DEPI > -1.78. Built from facts.parquet receivables, gross profit, intangibles, revenues, depreciation.
Source: signals/build_signals.py:420
Calibration — gross stock return vs net of execution costs
Cost model: 5 bps half-spread each side, 10 × √($1M / 30d ADV) impact each side, ADV-tiered borrow (≥$50M ADV → 10 bps annual; ≥$5M → 30; <$5M → 100), scaled by holding_days/252 if short.
| Horizon | n events | Gross stock ret | Net PnL | Hit rate | r/σ net | Avg cost (bps) |
|---|---|---|---|---|---|---|
| 1d | 2,937 | +0.04% | -0.28% | +45.86% | -0.04 | 32 |
| 5d | 2,937 | +0.10% | +0.06% | +45.83% | +0.01 | 33 |
| 20d | 2,915 | +2.64% | -2.76% | +44.43% | -0.02 | 37 |
| 60d | 2,798 | +11.23% | -11.63% | +46.71% | -0.03 | 47 |
| 252d | 2,745 | +47.23% | -48.06% | +50.93% | -0.04 | 96 |
Sector breakdown (top 8 by |r/σ| at 20d)
| SIC division | n | Mean 20d | Hit rate | r/σ |
|---|---|---|---|---|
| Mining + Construction | 101 | -2.86% | +40.59% | -0.14 |
| Wholesale + Retail Trade | 378 | +4.63% | +49.21% | +0.11 |
| Services (health + legal + edu) | 77 | +1.44% | +44.16% | +0.06 |
| Manufacturing (heavy) | 1,341 | -1.11% | +42.65% | -0.06 |
| Services (business + personal) | 392 | +18.79% | +44.39% | +0.05 |
| Finance, Insurance, Real Estate | 60 | +0.81% | +48.33% | +0.03 |
| Transportation, Communications, Utilities | 62 | +0.36% | +40.32% | +0.02 |
| Manufacturing (light) | 504 | +0.33% | +46.43% | +0.01 |
Recent triggers
HD— HOME DEPOT, INC. (FY2026)LOW— LOWES COMPANIES INC (FY2026)LULU— lululemon athletica inc. (FY2026)NVDA— NVIDIA CORP (FY2026)PD— PagerDuty, Inc. (FY2026)
Caveats
- Short loses at every horizon net of costs (worst: -48.06% net at 252d).
- Academic literature claims 252d+ predictive power; we don't see it because our prices_daily lacks delistings.
- Use as watchlist flag, not tradeable signal.
Citations
- Beneish, M.D. (1999). 'The Detection of Earnings Manipulation.' Financial Analysts Journal.
- Beneish, Lee, Nichols (2013). 'Earnings Manipulation and Expected Returns.' FAJ.
Live data → /api/methodology/signal/beneish_m_score_high · MCP → pyflo_signal_methodology(signal="beneish_m_score_high")