A Multi-Factor Residual-Based Trading Strategy

1
A Multi-Factor Residual-Based Trading Strategy

• Finance 453
• Adrian Helfert
• Terry Moore
• Kevin Stoll
• Ben Thomason
• February 26, 2004

2
Agenda

• CAPM Roots
• Our Multi-factor Model
• Our Trading Strategy
• Our Results
• Next Steps

3
Is the CAPM Dead?

• The CAPMs beta does not work well for all
  securities
• Fama and French found 3 factors described asset
  returns better than the basic CAPM

4
An Intuitive Multi-Factor Model

• We chose the following risk factors
• CAPM market risk premium
• The square of the market risk premium
• US dollar returns
• GS Commodity Index returns
• US long-term govt. bond returns
• Change in the term structure

5
Estimating a Better Pricing Model

• Dow Jones Industrial 30 large cap, liquid
  stocks
• In-sample daily returns 1/1/94-12/31/02
• Out-of-sample 1/1/03-1/31/04
• Linear regression for in-sample period
• R-squared range from 3 (BS) to 52 (GE)
• Significant t-stats
• Residuals show negative autocorrelation

6
Screens

• Rank residual factors (or expected variance) in
  ascending order, rebalancing weekly
• Ten lowest form Portfolio 1 (long)
• Ten highest form Portfolio 3 (short)
• Screen 1 sum of last 5 days residuals
• Screen 2 sum of last 30 days residuals
• Screen 3 5 day moving avg 30 day moving avg
• Screen 4 5 day moving avg 10 day moving avg
• Screen 5 expected variance (GARCH)
• Screen 6 change in expected variance

7
Screens 1 2 Sum Previous Residuals

• Low residuals signal underperformance to risk
  factors
• Stock will catch up when investors digest news
• High residuals signal outperformance to risk
  factors
• Stock should correct downward
• Negative autocorrelations in our residuals
  support this theory

8
Screens 3 4 Difference Between Moving
Averages of Previous Residuals

• Technical reversal
• Stocks tend to track longer term trend relative
  to the market
• Profit-taking may cause near-term
  underperformance
• Dip-buying may cause near-term outperformance

9
Screens 5 6 Expected Variance (GARCH)

• Use residuals to estimate expected variances
• Low variance stocks are rewarded by investors
• High variance stocks are penalized by investors
• Reductions in variance are positive
• Increases in variance are negative

10
In-Sample Results

• We discarded Screens 2, 4 and 6
• - Results were similar, but not as good
  as 1,3 5

11
Scoring System

• Screen 1
• Portfolio 1 scores 5, Portfolio 3 scores -4
• Screen 3
• Portfolio 1 scores 3, Portfolio 3 scores -3
• Screen 5
• Portfolio 1 scores 3, Portfolio 3 scores -2
• Add scores for each week, sort and repeat process
  for next week

12
Out-of-Sample Results

• Total scoring screen significantly
  underperforms in the out-of-sample
  year -23.7 return

13
Next Steps

• Test different stocks
• Estimate a rolling pricing model instead of fixed
  historical time period
• Optimize scoring (weighting) instead of
  subjective scoring
• Factor trading costs and slippage costs
  explicitly into model
• Test a 2-day model instead of 5-day because of
  autocorrelation results
• Test a technical crossover strategy