The persistent statistical structure of the US input–output coefficient matrices: 1963–2007

Abstract

The paper finds evidence for the existence of a statistical structure in the US input–output coefficient (A) matrices for 1963–2007 and characterizes the identified statistical regularities. For various aspects of A matrices, we find smooth and unimodal empirical distributions (EDs) with a remarkable stability in their functional form for most of the samples. The EDs of all entries, row sums, and the entries of the (left- and right-hand) Perron–Frobenius eigenvectors are well described by fat-tailed distributions, while the EDs of column sums and eigenvalues' moduli are explained by the normal and the beta distribution. The paper provides several economic interpretations of these statistical results as well as some implications and potential uses for structural and stochastic input–output analysis.