Essays on implications of structural parameter identification and trend misspecification in DSGE and SVAR frameworks

Abstract

This thesis consists of three self-contained essays, focusing on the consequences of structural parameter identification and trend misspecifications in DSGE and SVAR frameworks on policy implications. The purpose is to improve our understanding of estimation issues, and to recommend appropriate estimation approaches that econometricians should take to minimise distortion in policy inference. As welfare estimates are sensitive to calibrated and estimated parameter values, significant biases in structural parameter estimates may lead to biases in welfare estimates and subsequently affect policy conclusions. The first essay (Chapter 2) investigates the relationship between the bias in welfare cost of business cycles and fiscal policy estimates and the biases in structural parameter estimates due to identification problems. Assuming a simple RBC as the description of business cycles, we find the bias varies nonlinearly over the parameter space. Specifically, bias in welfare estimation is increased if we assume too high a depreciation rate of capital or too low a capital share relative to true values. We can therefore exploit this additional information by imposing priors on these two parameters to improve the estimation of welfare cost. The second essay (Chapter 3) focuses on trend misspecifications in DSGE models. The misspecification of the trend component can distort structural parameter estimates and translate into a bias in policy-relevant statistic estimates (e.g. impulse response functions, variance decompositions and estimates of policy rules, or, policy tradeoffs). This essay investigates how important this bias is to estimated policy implications within a DSGE framework. The quantitative results suggest that distortion in parameter estimates due to trend misspecification can result in significant inaccuracies in estimating statistics of interest, and it is therefore crucial to policy analysis. Particularly, a misspecified model with a deterministic-trend specification is estimated when the true process is a random walk with drift. The last essay (Chapter 4) studies the consequences of incorrect trend assumption for estimated SVARs. In particular, assumptions about a trend process determine a shock identification strategy within a SVAR. We conduct Monte Carlo simulations using a prototypical Real Business Cycle model as a Data Generating Process. Then, we quantify different sources of biases in the estimated SVAR impulse responses arising from incorrect trend assumptions. We find that, while the main source of error is a trend bias for an incorrectly-detrended variable, spillover to estimated impulse responses of a correctly-detrended variable in a system can be mainly an identification bias.