Instrumented Principal Components Analysis

This is a Python implementation of the Instrumtented Principal Components Analysis framework by Kelly, Pruitt, Su (2017).

Usage

Exemplary use of the ipca package. The data is the seminal Grunfeld data set as provided on statsmodels. Note, the fit method takes a panel of data, X, with the following columns:

1. entity id (numeric)
2. time (numeric)
3. and following columns contain characteristics.

as well as a series of dependent variables, y, of the same length as X.

import numpy as np
from statsmodels.datasets import grunfeld
data = grunfeld.load_pandas().data
data.year = data.year.astype(np.int64)

# Establish unique IDs to conform with package
N = len(np.unique(data.firm))
ID = dict(zip(np.unique(data.firm).tolist(),np.arange(1,N+1)))
data.firm = data.firm.apply(lambda x: ID[x])

# use multi-index for panel groups
data = data.set_index(['firm', 'year'])
y = data['invest']
X = data.drop('invest', axis=1)

# Call ipca
from ipca import InstrumentedPCA
regr = InstrumentedPCA(n_factors=1, intercept=False)
regr = regr.fit(X=X, y=y)
Gamma, Factors = regr.get_factors(label_ind=True)

Installing

The latest release can be installed using pip

pip install ipca

The master branch can be installed by cloning the repo and running setup

git clone https://github.com/bkelly-lab/ipca.git
cd ipca
python setup.py install

Documenation

The lastest documenation is published HERE.

Requirements

Running

With the exception of Python 3.6+, which is a hard requirement, the others are the version that are being used in the test environment. It is possible that older versions work.

• Python 3.7+:
• NumPy (1.19+)
• SciPy (1.6+)
• Numba (0.53+)
• progressbar (2.5+)
• joblib (1.0.1+)

Testing

• pandas (1.2.3+)
• scikit-learn (0.24+)
• pytest (4.3+)
• statsmodels (0.11+)

Acknowledgements

The implementation is inspired by the MATLAB code for IPCA made available on Seth Pruitt's website.

References

1. Kelly, Pruitt, Su (2017). "Instrumented Principal Components Analysis" SSRN

---

The package is still in the development phase, hence please share your comments and suggestions with us.

Contributions welcome!

- Matthias Buechner, Leland Bybee