"Maximize Alpha with Systematic Factor Testing" by Cheng Peng, Software Engineer at Betterment
2 likes897 views
The document outlines a systematic approach to maximizing alpha through factor testing, addressing concerns of scalability and consistency. It details backtesting results using a sample of 1,500 US equities from 2003 to 2017, analyzing factors such as momentum, quality, and volatility. Key findings suggest that while combining winning factors generally underperforms standalone low volatility strategies, careful inspection and tilting of factors can uncover potential alpha opportunities.
"Maximize Alpha with Systematic Factor Testing" by Cheng Peng, Software Engineer at Betterment
- 1. Maximize Alpha with Systematic Factor Testing By: Cheng Peng
- 2. Problem: - Given a basket of Factors, how do we extract the most alpha? - Concerns: Scalability and consistency Solution: - A systematic approach for analyzing and testing factor alpha - Key Points: - Universe Factor Tilting - Alpha Combination Techniques - Portfolio Diversification What is this about?
- 3. Backtesting Conditions Quantopian Platform - Universe: 1500 most tradeable US Equities - Timeframe: - In Sample: 01/04/2003 - 01/01/2015 (12 Years) - Out Of Sample: 01/01/2015 - 08/01/2017 (2.5 Years) - Trading Costs: - $0.0035 per share (IB Tiered) - Volume limitations of 2.5% of a minute’s trade volume, with a price impact of 0.1 - Starting balance of $ 1,000,000 - Full rebalance at start of every Month - Equal weighting with 10% constraint on each stock
- 4. Picking Factors Momentum - 1 Month / 12 Month Price Momentum Quality - Return On Equity = Net Income / Shareholder’s Equity Volatility - Standard Deviation of Daily Price in last 21 days
- 5. Picking Factors - Top / Bottom Quintiles Sharpe Beta Alpha High Momentum 0.56 1.11 0.02 Low Momentum 0.56 1.34 0.02 High ROE 0.7 1.08 0.04 Low ROE 0.58 1.29 0.02 High Volatility 0.51 1.57 0 Low Volatility 0.83 0.7 0.04 2003 - 2015
- 6. Picking Factors - Top / Bottom Quintiles Sharpe Beta Alpha High Momentum 0.56 1.11 0.02 Low Momentum 0.56 1.34 0.02 High ROE 0.7 1.08 0.04 Low ROE 0.58 1.29 0.02 High Volatility 0.51 1.57 0 Low Volatility 0.83 0.7 0.04 2003 - 2015
- 7. Combining Winning Factors Combined Ranking - Top Quintile of (Momentum Rank + ROE Rank + Low Volatility Rank) Combined Portfolio - Top Momentum Quintile + Top ROE Quintile + Bottom Volatility Quintile Cross Section - Top Momentum Quintile & Top ROE Quintile & Bottom Volatility Quintile
- 9. Sharpe Beta Alpha Avg Holdings Combined Rank 0.64 1.04 0.03 300 Combined Portfolio 0.69 0.99 0.03 700 Cross Section 0.66 0.46 0.03 15 Low Vol 0.83 0.7 0.04 All of these perform WORSE than just the standalone Low Vol factor. Let’s reinvestigate each factor more carefully. Combining Winning Factors 2003 - 2015
- 10. Picking Factors - Top / Bottom Quintiles Sharpe Beta Alpha High Momentum 0.56 1.11 0.02 Low Momentum 0.56 1.34 0.02 High ROE 0.7 1.08 0.04 Low ROE 0.58 1.29 0.02 High Volatility 0.51 1.57 0 Low Volatility 0.83 0.7 0.04 2003 - 2015
- 11. Quintiles Sharpe Beta Alpha 5 0.56 1.11 0.02 4 0.67 1.01 0.03 3 0.69 1.05 0.03 2 0.66 1.11 0.03 1 0.56 1.34 0.02 Momentum - 2003 to 2015
- 12. Quintiles Sharpe Beta Alpha 5 0.7 1.08 0.04 4 0.68 1.04 0.03 3 0.64 1.08 0.02 2 0.6 1.12 0.02 1 0.58 1.29 0.02 ROE - 2003 to 2015
- 13. Quintiles Sharpe Beta Alpha 5 0.51 1.57 0 4 0.58 1.29 0.02 3 0.62 1.11 0.02 2 0.74 0.95 0.04 1 0.83 0.7 0.04 Volatility - 2003 to 2015
- 14. Sharpe Beta Alpha Mid Momentum 0.69 1.05 0.03 High ROE 0.70 1.08 0.04 Low Vol 0.83 0.7 0.04 Picking Factors - Optimal Quintiles Repeat process to Combine Portfolios and finding Cross Sections. However, ranking Mid Momentum cannot be done with the same approach! 2003 - 2015
- 15. Combining (Actual) Winning Factors Combined Ranking (with a twist) - Top Quintile of ((- Momentum2 Rank) + ROE Rank + Low Volatility Rank) Combined Portfolios - Mid Momentum Quintile + Top ROE Quintile + Bottom Volatility Quintile Cross Section - Mid Momentum Quintile & Top ROE Quintile & Bottom Volatility Quintile
- 16. Combining (Actual) Winning Factors
- 17. All of these STILL perform WORSE than just the standalone Low Vol factor. Try Factor Tilting the Universe! Sharpe Beta Alpha Avg Holdings Combine Rank 0.67 1.14 0.04 300 Combine Portfolio 0.73 0.97 0.04 700 Cross Section 0.68 0.72 0.03 20 Low Vol 0.83 0.7 0.04 Combining (Actual) Winning Factors 2003 - 2015
- 18. Factor Tilting Universes Original Universe Factor 1 Filter Factor Universe Factor 2 Factor 3 Factor Portfolio
- 19. Factor Tilting Universes - Example Q1500US Momo Filter Momo Universe ROE Volatility Portfolio
- 20. Universe Tilt Mid Momo High ROE Low ROE High Vol Low Vol Mid Momo 0.76 0.57 0.49 0.85 High ROE 0.71 0.53 0.85 Low ROE 0.64 0.48 0.76 High Vol 0.61 0.6 0.55 Low Vol 0.74 0.82 0.68 HeatMaps - Sharpe Ratios 2003-2015
- 21. Universe Tilt Mid Momo High ROE Low ROE High Vol Low Vol Mid Momo 1.02 1.11 1.42 0.7 High ROE 1 1.46 0.69 Low ROE 1.11 1.65 0.73 High Vol 1.32 1.33 1.48 Low Vol 0.85 0.84 0.89 HeatMaps - Betas 2003-2015
- 22. Universe Tilt Mid Momo High ROE Low ROE High Vol Low Vol Mid Momo 0.05 0.01 -0.01 0.05 High ROE 0.04 0.01 0.05 Low ROE 0.03 0 0.04 High Vol 0.03 0.03 0.02 Low Vol 0.04 0.05 0.03 HeatMaps - Alphas 2003-2015
- 23. Universe Tilt Mid Momo High ROE Low ROE High Vol Low Vol Mid Momo 0.05 0.01 -0.01 0.05 High ROE 0.04 0.01 0.05 Low ROE 0.03 0 0.04 High Vol 0.03 0.03 0.02 Low Vol 0.04 0.05 0.03 HeatMaps - Best Alphas 2003-2015
- 24. Factor Tilting Universes - Example Q1500US Momo Filter Momo Universe ROE Volatility Portfolio
- 25. Universe Tilt Mid Momo High ROE Low Vol Mid Momo 0.05 0.05 High ROE 0.04 0.05 Low ROE 0.03 0.04 Low Vol 0.04 0.05 HeatMaps - Combining Alphas 2003-2015 Now what? Let’s try this again: - Combine Rankings - Combine Portfolios - Cross Section
- 26. Sharpe Beta Alpha Holdings 0.77 1.02 0.05 140 0.63 0.89 0.02 140 0.55 0.99 0.01 140 0.8 0.8 0.04 140 Portfolio Construction - Combine Rankings Based on Factor Tilts: - Mid Momentum - (High ROE, Low Volatility) - High ROE - (Mid Momentum, Low Volatility) - Low ROE - (Mid Momentum, Low Volatility) - Low Volatility - (Mid Momentum, High ROE) 2003 - 2015
- 27. Sharpe Beta Alpha Holdings 0.78 0.87 0.05 260 0.71 0.79 0.03 260 0.65 0.87 0.02 260 0.7 0.8 0.03 260 Portfolio Construction - Combine Portfolios Based on Factor Tilts: - Mid Momentum - (High ROE, Low Volatility) - High ROE - (Mid Momentum, Low Volatility) - Low ROE - (Mid Momentum, Low Volatility) - Low Volatility - (Mid Momentum, High ROE) 2003 - 2015
- 28. Sharpe Beta Alpha Holdings 0.88 0.69 0.05 30 0.8 0.73 0.04 40 0.82 0.72 0.05 40 0.83 0.82 0.05 30 Portfolio Construction - Cross Section Based on Factor Tilts: - Mid Momentum - (High ROE, Low Volatility) - High ROE - (Mid Momentum, Low Volatility) - Low ROE - (Mid Momentum, Low Volatility) - Low Volatility - (Mid Momentum, High ROE) 2003 - 2015
- 29. Sharpe Beta Alpha Holdings 0.81 0.81 0.01 30 1.05 0.83 0.04 40 1.33 0.9 0.08 40 0.84 0.91 0.02 30 Portfolio Construction - Out Of Sample Based on Factor Tilts: - Mid Momentum - Cross Section (High ROE, Low Volatility) - High ROE - Cross Section (Mid Momentum, Low Volatility) - Low ROE - Cross Section (Mid Momentum, Low Volatility) - Low Volatility - Cross Section (Mid Momentum, High ROE) 2015 - 2017
- 30. Sharpe Beta Alpha Holdings 0.81 0.81 0.01 30 1.05 0.83 0.04 40 1.33 0.9 0.08 40 0.84 0.91 0.02 30 Portfolio Construction - Overfitted! Sharpe Beta Alpha Holdings 0.88 0.69 0.05 30 0.8 0.73 0.04 40 0.82 0.72 0.05 40 0.83 0.82 0.05 30 BEFORE AFTER
- 31. In Sample - 01/04/2003 - 01/01/2015
- 32. Out Of Sample - 01/01/2015 - 08/01/2017
- 33. Several Options (In Sample Results) - Don’t pick a particular portfolio - Hold all four portfolios - Reduce noise in portfolio - Correlation Rank - Beta Rank - Mid Momentum - Cross Section (High ROE, Low Volatility) - Low Volatility - Original Factor Bottom Quintile Sharpe Beta Alpha Holdings 0.85 0.77 0.05 100 0.88 0.75 0.06 30 0.88 0.69 0.05 30 0.83 0.82 0.05 30 Portfolio Construction - Avoid Overfitting 2003 - 2015
- 34. In Sample - 01/04/2003 - 01/01/2015
- 35. Out Of Sample - 01/01/2015 - 08/01/2017
- 36. In Sample - Benchmarked to SPY
- 37. Out of Sample - Benchmarked to SPY
- 38. Hedged Version - In Sample
- 39. Hedged Version - Out Of Sample
- 40. Takeaways: - Carefully investigate each factor before drawing conclusions - Combine factors by ranking factors, mixing them and finding cross sections - Utilize factor tilting universes to help extract hidden alphas - Avoid overfitting by holding diversified portfolios Next Steps: - Try a different set of factors and rebalance periods - Try different markets and universes Conclusion
- 41. Thank you for your time! Email: me@chengpeng.ca