A Multivariate Time Series Analysis of the Data Revision Process for Industrial Production and the Composite Leading Indicator

Abstract

In this paper we try to enhance our understanding of preliminary data releases of industrial production (IP) and the composite leading indicator (CLI) by investigating several time series properties of their data revision processes. In particular, we examine moments, autocorrelation functions, and integratedness properties of IP and CLI revision processes. We also construct and estimate univariate and multivariate regression models in order to assess the “efficiency” of IP and CLI revisions. Our findings based on regressions which include both IP and the CLI suggest that multivariate information “matters” in the revision process. For example, previously available IP revisions are useful for explaining CLI revisions, suggesting that releases of the CLI do not fully incorporate newly available IP data. In addition, IP revisions can be predicted from past CLI revisions, suggesting that a kind of causal feedback characterizes the revision processes for IP and the CLI. Finally, we conduct a series of real-time forecasting experiments in order to provide further evidence that there is useful univariate as well as multivariate information in the revision processes of these variables.