Quantitative Methods – Factor Models
Overview
Factor models explain asset returns as a linear combination of systematic factors plus a stock-specific residual. The Fama-French 3-Factor Model (1992) extended CAPM by adding two well-documented risk premia: the Size premium (SMB) and the Value premium (HML), dramatically improving the explanation of cross-sectional stock returns.
Key Concepts
Evolution: CAPM → 3-Factor
| Model | Factors | Year |
|---|---|---|
| CAPM | Market (β) | 1964 |
| Fama-French 3F | Market + SMB + HML | 1992 |
| Fama-French 5F | + Profitability (RMW) + Investment (CMA) | 2015 |
| Carhart 4F | + Momentum (UMD) | 1997 |
The Three Factors
Market (MKT-RF): Rm − Rf — the return of the overall market above the risk-free rate. This is CAPM's single factor. Every stock has exposure to this.
SMB (Small Minus Big): Long small-cap stocks, short large-cap stocks. Historically, smaller companies have delivered higher returns (possibly as compensation for illiquidity and distress risk).
HML (High Minus Low): Long value stocks (high Book/Market ratio), short growth stocks. Value stocks have historically outperformed growth (possibly as compensation for financial distress risk or behavioral mispricing).
The Factor Model Equation
| Term | Name | Interpretation |
|---|---|---|
| α (alpha) | Intercept | Return unexplained by factors — manager skill or anomaly |
| β_MKT | Market beta | Sensitivity to market-wide moves |
| β_SMB | Size beta | Positive = small-cap tilt; Negative = large-cap tilt |
| β_HML | Value beta | Positive = value tilt; Negative = growth tilt |
| ε | Residual | Idiosyncratic, diversifiable risk |
Logic Implemented
- Correlated factor simulation — Multivariate normal with realistic covariance
- Stock return generation — True model + idiosyncratic noise
- OLS from scratch — Matrix algebra:
β = (X'X)⁻¹X'y - t-statistics — Statistical significance of each factor loading
- Performance attribution — Decompose average return into factor contributions
Files
factor_models_tutorial.py: Factor data generation, OLS regression, significance testing, and performance attribution.
How to Run
Financial Applications
1. Portfolio Risk Decomposition
- "How much of my hedge fund's return is beta to the market vs. true alpha?"
- Investors pay high fees for alpha; beta can be obtained cheaply via ETFs.
2. Smart Beta / Factor ETFs
- Deliberately tilting portfolio toward SMB and HML premiums
- Examples: iShares Value ETF (IVE), Dimensional Fund Advisors funds
3. Benchmark Construction
- Factor exposures define what a "fair" benchmark for a fund manager is.
- A manager who only buys small-cap value should be benchmarked to small-cap value — not the S&P 500.
4. Risk Management
- Stress-test portfolios by shocking factor exposures
- Hedge factor risks using factor ETFs or index futures
5. Research — The Zoo of Factors
- Academic literature has identified 300+ potential factors.
- Most don't survive out-of-sample. The three Fama-French factors are among the most robust.
Best Practices
- Use monthly data: Daily factor returns are noisier; monthly gives better signal-to-noise for factor regressions.
- Check R-squared: < 0.40 suggests the factors don't explain this stock well (could be a niche sector or anomaly).
- Multiple testing: With 300+ factors available, any single significant result could be spurious — use Bonferroni correction or out-of-sample tests.
- Factor stability: Betas change over time as companies grow or shift strategy — use rolling regressions to monitor.
- Real data: Download free Fama-French factor returns from Ken French's website.