A systematic approach to equity factor modeling using S&P 1500 data with advanced dimensionality reduction techniques.
This project develops a comprehensive equity factor model that predicts stock returns using fundamental and technical factors. The model employs Singular Value Decomposition (SVD) for dimensionality reduction and implements a long-short hedge portfolio strategy.
- Universe: S&P 1500 stocks (filtered to 272 stocks with complete data)
- Time Period: 1995-2019 (training: 1995-2014, testing: 2015-2019)
- Methodology: Factor selection, SVD optimization, and cross-sectional trading strategy
- Performance: 11.06% annualized return with 0.83 Sharpe ratio
| Metric | Value |
|---|---|
| Annualized Return | 11.06% |
| Sharpe Ratio | 0.83 |
| Alpha vs CAPM | +5.36% |
| Alpha vs FF3 | +5.08% |
| Information Ratio | 0.47 |
- Universe Selection: Filtered S&P 1500 to 272 stocks with complete data (1995-2019)
- Data Cleansing: Applied winsorization (1st-99th percentiles) and standardization
- Factor Transformation: Converted to percentile values (0-1 range)
We developed a rigorous 4-step factor selection process:
| Selection Criteria | Description |
|---|---|
| Absolute t-statistic | Cross-sectional regression significance |
| Time-decayed t-statistic | Exponential decay weighting (48-month half-life) |
| Factor Health | Recent vs. historical performance ratio |
| Information Coefficient | Spearman rank correlation with future returns |
The top 10 factors based on composite scoring:
| Factor | Description | Expected Relationship |
|---|---|---|
| Mean Reversion | Short-term price corrections | Negative (contrarian) |
| Trend Factor | Blend of momentum and reversal patterns | Positive |
| R&D/Market Cap | Research intensity relative to market value | Mixed (quality dependent) |
| Short Interest Ratio | Short selling activity | Positive (squeeze effects) |
| Price Range (120d) | Medium-term volatility measure | Positive (momentum) |
| Price Range (20d) | Short-term volatility measure | Negative |
| Accrual | Earnings vs. cash flow difference | Negative |
| Free Cash Flow/Price | Cash generation efficiency | Positive |
| Industry-Adjusted Volatility | Peer-relative risk measure | Positive |
| R&D/Sales | Operational innovation efficiency | Negative (raw), Positive (SVD) |
We developed four proprietary factors:
- Captures short-term price reversions to historical averages
- Constructed as negative z-score of 20-60 day returns
- Based on empirical evidence from Poterba & Summers (1988)
- Quantifies economic uncertainty from forecast error volatility
- Uses 1, 3, and 12-month economic indicator horizons
- Strong predictive power during recession periods
- Textual analysis of 10-K, 10-Q, and 8-K filings
- Includes negative intensity, fog score, and polarity ratio
- Early signals of fundamental changes and management tone
- Net insider purchases and purchase-to-sale ratios
- Based on SEC Form 4 filings
- Exploits informational asymmetry advantages
Applied SVD to create orthogonal factors and reduce noise:
| SVD Factor | Beta | Primary Loadings | Interpretation |
|---|---|---|---|
| Factor 9 | 0.1378 | RD_P (-0.70), RD_SALE (+0.71), range_20 (+0.08) | Quality innovation vs. speculative R&D |
| Factor 2 | 0.1061 | RD_P (+0.65), RD_SALE (+0.66), Accrual (+0.17), FCF_P (+0.13) | Profitable growth and disciplined innovation |
| Factor 1 | 0.0317 | range_120 (+0.56), range_20 (+0.55), xret_indsize_std120 (+0.53) | Price dispersion as opportunity indicator |
| Factor 7 | -0.0003 | SIR (-0.98) | Neutralized short squeeze effects |
- Training Phase: Estimate factor betas using time-series regression (1995-2014)
- Prediction: Calculate expected returns using linear factor model
- Portfolio Construction:
- Sort stocks into deciles by predicted returns
- Long top decile, short bottom decile
- Monthly rebalancing
Expected Return: r(i,t+1) = α₀ + Σ(βₖ × Fₖ,ᵢ,ₜ)
Where:
- r(i,t+1) = predicted return for stock i at time t+1
- α₀ = intercept (alpha)
- βₖ = factor k beta coefficient
- Fₖ,ᵢ,ₜ = factor k value for stock i at time t
Email me at m9yang@uwaterloo.ca for the full dataset
- Poterba, J. M., & Summers, L. H. (1988). Mean reversion in stock prices
- Jurado, K., Ludvigson, S. C., & Ng, S. (2015). Measuring Uncertainty
- Chan, L. K., Lakonishok, J., & Sougiannis, T. (1999). Stock Market Valuation of R&D
- Cohen, L., Malloy, C., & Pomorski, L. (2012). Decoding Inside Information
... [full references in original report]