Estimates a dynamic logit model of corporate failure probability from accounting and market variables (US data 1963–2003), then documents the "distress puzzle": since 1981, stocks with the highest probability of financial distress have earned anomalously low returns despite far higher volatility, betas, and value/small-cap loadings. This is the opposite sign of a risk-based distress premium and is inconsistent with value/size effects being compensation for distress risk.
What anomaly it documents
Predictor: estimated failure probability (fitted from a dynamic logit/hazard model).
Direction:negative — high distress probability predicts low subsequent returns (a long-safe/short-distressed strategy earns positive returns/alpha).
Shape: monotone across the failure-probability distribution; strongest among small stocks, present in all size quintiles.
OSAP predictor: related to the "FailureProbability / distress" predictor family (Campbell-Hilscher-Szilagyi).
How to construct it
Estimate a dynamic panel logit of failure (following Shumway 2001, Chava-Jarrow 2004) using leverage (TLMTA), profitability (NIMTA), past excess returns (EXRET), equity volatility (SIGMA), relative market cap (RSIZE), cash holdings (CASHMTA), market-to-book (MB), and price per share; "failure" = bankruptcy, delisting, or a D ("default") credit rating. At longer horizons, persistent characteristics (market cap, market-book, volatility) gain weight.
Each month, sort firms on fitted failure probability (using only lagged data to avoid look-ahead) into ten value-weighted portfolios; form long-short portfolios going long the lowest-risk and short the highest-risk stocks.
Long safest-decile / short most-distressed-decile: +10.0%/yr raw (t = 1.9); CAPM alpha 12.4% (t = 2.3); Fama-French three-factor alpha 22.7% (distressed stocks load positively on HML and SMB, so risk adjustment worsens the puzzle).
Adding the Carhart momentum factor roughly halves the three-factor long-short alpha (≈22.7% → ≈12.0%) but it remains positive.
The failure model itself captures much of the time variation in the aggregate failure rate (e.g., 1% riskiest stocks ≈ 80 bp failure probability vs ≈ 34 bp for the next group).
Why it might work
Mispricing / limits to arbitrage: distressed stocks are volatile, lottery-like, costly to short, and dominated by small illiquid names, so they stay overpriced.
A rational distress-risk premium predicts the opposite sign, deepening rather than resolving the puzzle; the result contradicts distress as the driver of value/size premia.
Limitations and risks
The short leg concentrates in tiny, illiquid, hard-to-short names; transaction and shorting costs are severe.
The failure model needs careful point-in-time accounting/market data to avoid look-ahead bias.
Raw long-short return is only marginally significant (t = 1.9); significance rests largely on factor adjustment.
Key references
Campbell, Hilscher & Szilagyi (2008) — In Search of Distress Risk — Journal of Finance
Shumway (2001) — Forecasting Bankruptcy More Accurately: A Simple Hazard Model — Journal of Business
Dichev (1998) — Is the Risk of Bankruptcy a Systematic Risk? — Journal of Finance
Bali, Cakici & Whitelaw (2011) — Maxing Out — Journal of Financial Economics
Provenance: verified/generated from the paper's full text.
