Macroeconomic applications with factor models

Abstract

This thesis utilizes factor models to test the predictions of macroeconomic theory and introduces a new model for estimating structural relations in the economy. Factor models have proven useful in overcoming limited information bias. Limited information bias occurs because the information set of the actual decision makers in the economy is larger than the information set captured by conventional empirical models (i.e. small VARs). With the help of factors we can model a large dataset by using a small model of factors that still capture the majority of aggregate dynamics in the economy. In the first chapter, joint work with Massimiliano Marcellino, we introduce a new empirical model: mixed frequency structural factor augmented VAR model. We show that in a mixed data frequency setting the model reduces aggregation bias and provides more precise estimates of factors and impulse responses, than competing models. We support this claim by means of a detailed Monte Carlo examination that also tests the new estimation procedure that we design. Finally we provide three empirical applications (monetary policy, oil and government expenditure shock) to show the usefulness of the model. In the second chapter I utilize a dynamic factor model to test the predictions of the rational inattention theory as put forward by Mackoviak et al. (2009). I first estimate a time varying parameter dynamic factor model on US post-war data on macroeconomic variables and sector prices. I identify impulse responses of three macroeconomic shocks and sector specific shocks to prices. I then regress price impulse responses, void of the influences of changing variances, on the variances of the shocks, to test the predictions of the rational inattention model over time.