paperhedge-fundabs-factorseven-factorfactor-modelperformance-attributionfaj

Fung & Hsieh (2004) — "Hedge Fund Benchmarks: A Risk Based Approach"

The seminal paper proposing that hedge fund benchmarks be built from observable market prices (asset-based style factors), not from peer-group averages. Published in Financial Analysts Journal (forthcoming 2004).

Citation

Fung, W., & Hsieh, D. A. (2004). "Hedge Fund Benchmarks: A Risk Based Approach." Financial Analysts Journal, forthcoming (March 2004 draft).

Core contribution

Argues that the conventional approach of building hedge fund indices by averaging constituent fund returns inherits database biases (selection, survivorship, backfill, sampling, short history, opacity) that statistical cleanup cannot fix. The paper's alternative:

  1. Statistically extract common return components from hedge fund returns (return-based style / RBS factors via PCA).
  2. Replicate each RBS factor with a portfolio of liquid, exchange-traded instruments — yielding an asset-based style (ABS) factor.
  3. Fit a time-varying-beta APT model to hedge fund index returns on the ABS factors.

The resulting seven ABS factors (S&P 500, SC−LC, Δ10Y, Δ(Baa−10Y) credit spread, lookback-straddle on bonds / FX / commodities) explain up to ~90% of diversified hedge fund portfolio return variance and are traceable to decades of market history.

Key empirical findings

  • Seven factors cover 57% of TASS funds (2003-03) and 37% of HFR funds (2003-07) as their dominant risk factor.
  • Time-varying betas: alpha illusion — the 48bp/month HFRFOF intercept significant on the full 1994–2002 sample disappears in either of the two subperiods (1994–1998, 2000–2002) identified by reverse-recursive residual break detection around the LTCM crisis and the Internet-bubble peak.
  • FoF vs broad-based indices: HFRI/TASSAVG/MSCI retain significant alphas in both subperiods (40–97bp/month) while HFRFOF does not — the authors attribute the gap to inefficiency in the 500+ FoF universe.
  • Database composition diagnosed by factor exposures: HFR is equity-dominated; TASS is more balanced and contains CTAs that HFR excludes; CTI over-weights fixed-income risks relative to the underlying TASS pool.
  • Out-of-sample 2003: model predicts HFR actual 18.10% with 18.17% predicted; CTI actual 14.48% vs 12.13% predicted (largest error because asset-weighted index is least diversified).

Why it matters

  • Established the standard for hedge fund performance attribution that disentangles alternative betas from alternative alphas.
  • Provided the conceptual basis for transparent hedge fund reporting: rather than disclosing positions, funds disclose exposures to a small set of common market factors — investors can monitor portfolio-level risks daily without proprietary information leakage.
  • Motivated systemic risk surveillance by showing how trend followers, global-macro funds, and convergence traders can crowd the same trades (1994 bond-market episode) for completely different reasons, triggering simultaneous liquidation under stress.

See also