Instrumented Principal Component Analysis (WP) w/ Kelly and Su

with B. Kelly and Y. Su



We propose a method of factor estimation for a data panel Y by using the data tensor Z to parameterize loadings—Instrumented Principal Component Analysis. IPCA allows us to identify a model wherein factor loadings vary over both panel dimensions, which is an implication of various economic theories. Our benchmark estimator is computed virtually instantaneously using the singular value decomposition—we show the consistency and asymptotic distribution for resulting estimates. An application to international macroeconomics suggests that a nation’s import share, gross capital formation share, and overall level of GDP drive its relationship to a global growth factor, whereas population density does not.

Pruitt’s Python code

Example Matlab code folder