README.md

DeepIPCA

This repository implements IPCA models.

Table of Contents

• Empirical Results
• Author

---

Empirical Results

Here we report the results for the number of factor with the best validation Sharpe Ratio.

• Number of Factors

Model	Optimal Number of Factors
Naive IPCA	12
Iterative IPCA	17
Iterative Deep IPCA (Linear)	14
Iterative Deep IPCA (hpSearch)	9
GD Deep IPCA (hpSearch)	18

• Sharpe Ratio

Model	Train	Valid	Test
Naive IPCA	1.04	1.17	0.47
Iterative IPCA	2.12	1.52	0.95
Iterative Deep IPCA (Linear)	2.04	1.58	0.94
Iterative Deep IPCA (hpSearch)	1.88	1.78	0.93
GD Deep IPCA (hpSearch)	2.32	1.92	0.88

• Unexplained Variation

Model	Train	Valid	Test
Naive IPCA	0.99	0.99	0.99
Iterative IPCA	0.65	0.82	0.82
Iterative Deep IPCA (Linear)	0.65	0.82	0.82
Iterative Deep IPCA (hpSearch)	0.66	0.82	0.83
GD Deep IPCA (hpSearch)

• Fama-McBeth Type Alpha (1e-03)

Model	Train	Valid	Test
Naive IPCA	1.45	3.50	4.00
Iterative IPCA	1.32	3.07	3.62
Iterative Deep IPCA (Linear)	1.30	3.11	3.71
Iterative Deep IPCA (hpSearch)	1.37	3.21	3.74
GD Deep IPCA (hpSearch)

• Weighted Fama-McBeth Type Alpha (1e-04)

Model	Train	Valid	Test
Naive IPCA	1.00	5.51	1.70
Iterative IPCA	0.68	4.73	1.23
Iterative Deep IPCA (Linear)	0.68	4.79	1.25
Iterative Deep IPCA (hpSearch)	0.69	4.95	1.27
GD Deep IPCA (hpSearch)

Author

• Luyang Chen