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

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)

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

The lastest documenation is published HERE.

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+)

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

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

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