TY - JOUR TI - Forecasting and monetary policy analysis DO - https://doi.org/doi:10.7282/T3S75JGF PY - 2016 AB - This dissertation presents new empirical evidence in two specific fields in economics: Forecasting and Monetary Policy Analysis. The dissertation comprises two separate but related papers, each of one tackles one of these two fields. The main objective aims to provide new lights and insights about specific questions that have been studied before in the literature, but using alternative methodological approaches to improve our understanding of the topics. In Chapter 2, I argue that the out-of-sample forecast performance of nonlinear models for the conditional mean has being underestimated in the literature because these models are highly parameterized and hence parameter estimation error can easily offset their predictive gains. Thus, I consider restricted versions of nonlinear models that are commonly used in forecast competition between linear and nonlinear models. The restrictions aim to reduce the number of parameters to estimate allowing the specification of parsimonious nonlinear models. This setting explores more deeply the space of nonlinear models in order to find a suitable specification able to boost the performance of this type of models. The empirical evaluation is conducted using a linear benchmark and both global and local test of forecast predictive accuracy. The main results can be summarized as follow. First, if forecast comparison between linear and nonlinear models excludes restricted nonlinear models then results are in line with previous findings. However, results change dramatically in some cases when restricted versions of nonlinear models are incorporated. In particular, I spot cases on which the mean square forecast error decreases by almost fifty percent relative to the benchmark model. These results give us new lights about the performance of nonlinear models and challenge the conventional view that the literature has about them because they show that their predictive gains may be elusive but that a simple exploration of their functional form may reveal significant predictive gains. In Chapter 3, I investigate the propagation of a foreign monetary policy shock over a small open economy, in particular over the Chilean economy. This is an joint research project with Jorge Fornero and Andres Yany from the Central Bank of Chile. Our motivation is base on the ongoing period of monetary normalization already started by the Fed. We follow Canova (2007) and compare the impulse response functions of Structural VAR models and a DSGE model tailored for the Chilean economy. We use the recursive VAR model of Sims (1980a) and an extension of the “agnostic” VAR model of Uhlig (2005) and Arias et al. (2014) for small open economies following Koop and Korobilis (2010). The results suggest that the recursive VAR model does not properly identify the shock and its implications are counterintuitive. On the contrary, beyond the quantitative differences, we find that the responses of the “agnostic” VAR model are in line qualitatively with those of the DSGE model except for output. However, the transmission of the shock to the local economy is limited but more persistent according to the DSGE model. Finally, we spot different policy implication arising from both models. According to the “agnostic” VAR model, the central bank do not need to rise its policy rate because the drop in activity offsets any burst of inflation; whereas in the DSGE model the rise in prices is partially accommodated by an increase in the policy rate. Thus, this comparison motivates an interesting discussion for the policy maker. KW - Economics KW - Economic forecasting KW - Economics, Mathematical KW - Economic policy LA - eng ER -