return forecasting by quantile regression qwafafew december 20101 larry pohlman and lingjie ma

Return Forecasting by Quantile Regression

QWAFAFEWDecember 20101

Larry Pohlman and Lingjie Ma

2

Outline

• The Math• Examples• Multivariate Model• Results

3

The Math and Code

• Model

• OLS Estimation

• QR Estimation

• R, S+, Stat, SAS

uXby

2)(minˆ bxy TiibOLS

Tii

Tiixy bxy

Tii

TiibQR bxybxy

))(1()(minˆ

4

What does QR do?

T

uT

y

uTT

y

TT

xFxxQ

FxxxF

uδxβxy

))((

)(

)(

1

11

Sample Quantile:

Conditional Quantile:

1,0,:

,yYProb

y

y

FyfQ

F

5

Example: Book to Price

6

Why QR?

A natural question is under what conditions will QR be “better” than OLS?

1. Full picture view: heterogeneity

If there is heterogeneity, then QR will provide a more complete view of the relationship between variables through the effects of independent variables across quantiles of the response distribution.

2. Robustness: fat-tail distribution

If the conditional return distribution is not Gaussian but fat-tailed, the QR estimates will be more robust and efficient than the conditional mean estimates

7

Example Price Momentum

8

Example Return on Equity

9

Multivariate Model

• Book to Price• Earnings to Price• Cashflow to Enterprise Value• Balance sheet Accruals• Return on Equity• Price Momentum (9 months)• Earnings Momentum (9 months)

10

Model Variable Plots

11

Results: Equal Weight Quintiles

12

Results: Cap Weighted Qunitiles

13

Optimized Portfolios TE=3%

14

Optimized Portfolios TE=6%

15

Conclusion

￭ Conditional mean method is still attractive

￭QR provides a full-picture distributional view

￭ Link between distribution estimates and point portfolio.