dc.description.abstract | The aim of this thesis is the development and the application
of econometric models with time-varying parameters in a policy
environment.
The popularity of these methods has run in parallel with advances in
computing power, which has made feasible estimation methods that
until the late ‘90s would have been unfeasible. Bayesian methods, in
particular, benefitted from these technological advances, as sampling
from complicated posterior distributions of the model parameters became
less and less time-consuming. Building on the seminal work by Carter
and Kohn (1994) and Jacquier, Polson, and Rossi (1994), bayesian
algorithms for estimating Vector Autoregressions (VARs) with drifting
coefficients and volatility were independently derived by Cogley and
Sargent (2005) and Primiceri (2005).
Despite their increased popularity, bayesian methods still suffer from
some limitations, from both a theoretical and a practical viewpoint.
First, they typically assume that parameters evolve as independent
driftless random walks. It is therefore unclear whether the output
that one obtains from these estimators is accurate when the model
parameters are generated by a different stochastic process. Second, some
computational limitations remain as only a limited number of time series
can be jointly modeled in this environment. These shortcomings have
prompted a new line of research that uses non-parametric methods to
estimate random time-varying coefficients models. Giraitis, Kapetanios,
and Yates (2014) develop kernel estimators for autoregressive models
with random time-varying coefficients and derive the conditions under
which such estimators consistently recover the true path of the model
coefficients. The method has been suitably adapted by Giraitis,
Kapetanios, and Yates (2012) to a multivariate context.
In this thesis I make use of both bayesian and non-parametric methods,
adapting them (and in some cases extending them) to answer some of
the research questions that, as a Central Bank economist, I have been
tackling in the past five years. The variety of empirical exercises proposed
throughout the work testifies the wide range of applicability of these
models, be it in the area of macroeconomic forecasting (both at short
and long horizons) or in the investigation of structural change in the
relationship among macroeconomic variables.
The first chapter develops a mixed frequency dynamic factor model
in which the disturbances of both the latent common factor and of
the idiosyncratic components have time varying stochastic volatility.
The model is used to investigate business cycle dynamics in the euro
area, and to perform point and density forecast. The main result is
that introducing stochastic volatility in the model contributes to an
improvement in both point and density forecast accuracy.
Chapter 2 introduces a nonparametric estimation method for a large
Vector Autoregression (VAR) with time-varying parameters. The
estimators and their asymptotic distributions are available in closed
form. This makes the method computationally efficient and capable
of handling information sets as large as those typically handled by
factor models and Factor Augmented VARs (FAVAR). When applied
to the problem of forecasting key macroeconomic variables, the method
outperforms constant parameter benchmarks and large Bayesian VARs
with time-varying parameters. The tool is also used for structural
analysis to study the time-varying effects of oil price innovations on
sectorial U.S. industrial output.
Chapter 3 uses a bayesian VAR to provide novel evidence on changes
in the relationship between the real price of oil and real exports in
the euro area. By combining robust predictions on the sign of the
impulse responses obtained from a theoretical model with restrictions
on the slope of the oil demand and oil supply curves, oil supply and
foreign productivity shocks are identified. The main finding is that from
the 1980s onwards the relationship between oil prices and euro area
exports has become less negative conditional on oil supply shortfalls
and more positive conditional on foreign productivity shocks. A general
equilibrium model is used to shed some light on the plausible reasons for
these changes.
Chapter 4 investigates the failure of conventional constant parameter
models in anticipating the sharp fall in inflation in the euro area in 2013-
2014. This forecasting failure can be partly attributed to a break in the
elasticity of inflation to the output gap. Using structural break tests
and non-parametric time varying parameter models this study shows
that this elasticity has indeed increased substantially after 2013. Two
structural interpretations of this finding are offered. The first is that the
increase in the cyclicality of inflation has stemmed from lower nominal
rigidities or weaker strategic complementarity in price setting. A second
possibility is that real time output gap estimates are understating the
amount of spare capacity in the economy. I estimate that, in order
to reconcile the observed fall in inflation with the historical correlation
between consumer prices and the business cycle, the output gap should
be wider by around one third. | en_US |