Академический Документы
Профессиональный Документы
Культура Документы
1
City University of Hong Kong
2
Yale University
3
University of Chicago
Motivation
Motivation
Challenges
Challenges
I How about controlling all factors proposed by the literature?
This paper
This paper
I Rather than comparing the new factor with all 300 controls, use
model selection to pick the “best model” out of those
I This weeds out useless factors from the hundreds of existing ones
This paper
This paper
Model Setup
E(rt ) − ιn γ0 = Cv λv = Cg λg + Ch λh .
Model Setup
y = |{z}
X β +ε, kβk0 ≤ s.
|{z}
n×p p×1
Sparsity
0.8
0.6
Carhart MOM
0.5 Growth in Capital Expenditure
Selection rate
0.3
0.2 Value
0.1 Size
0
0 10 20 30 40 50 60 70 80
Factor ID
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
I Lesson: Machine learning can be used for some purposes but not
others
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
Naive Method
I Single-Selection Approach
I Double-Selection Approach
Simulation Design
Simulation
1st Selection
Selection rate 1
0.5
0
50 100 150 200 250 300
Factor ID (simulated)
2nd Selection
1
Selection rate
0.5
0
50 100 150 200 250 300
Factor ID (simulated)
Total Selection
1
Selection rate
0.5
0
50 100 150 200 250 300
Factor ID (simulated)
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
Simulation
0.4 0.4
0.2 0.2
0 0
-5 0 5 -5 0 5
0.4 0.4
0.2 0.2
0 0
-5 0 5 -5 0 5
0.4 0.4
0.2 0.2
0 0
-5 0 5 -5 0 5
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
Zoo of Factors
Test Portfolios
0.8
0.7
0.6
0.5
Selection rate
0.4
0.3
0.2
0.1
0
0 50 100 150 200 250
Factor ID
None of the 248 factors, except for the Market (No. 1) and Sales to Inventory
(No. 16) are actually selected in more than 70% of the samples, and most of
the factors are selected in 5% to 30% of the subsamples, but not in the others.
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
0 0 0 0
-2 0 2 4 -2 0 2 4 -2 0 2 4 -2 0 2 4
Gross profitability Organizational Capital Betting Against Beta Quality Minus Junk
0.5 0.5 0.5 0.5
0 0 0 0
-2 0 2 4 -2 0 2 4 -2 0 2 4 -2 0 2 4
0 0 0 0
-2 0 2 4 -2 0 2 4 -2 0 2 4 -2 0 2 4
0 0 0 0
-2 0 2 4 -2 0 2 4 -2 0 2 4 -2 0 2 4
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
I One of the motivations for using our methodology is that it can help
distinguish useful from useless and redundant factors as they are
introduced in the literature.
I Over time, this should help limit the proliferation of factors.
I We test factors introduced during the year whether they are useful
or redundant relative to factors existing up to then.
Introduction Setup Using LASSO to select controls: Our Methodology Simulation Empirical
Robustness Checks