Computational Optimization Methods in Statistics, Econometrics and Finance

<p>Simulation-based methods have become increasingly popular in empirical economics during the last years. This is on the one hand due to the rapid development in computing technology which makes computationally intensive techniques applicable even for large data sets. On the other hand, particular in time series analysis solving high-dimensional optimization problems is more and more important. Examples involve stochastic volatility models, dynamic mixture models, multi-factor asset pricing models, Bayesian vector autoregressions or regime-switching models. Estimating these models typically requires sampling from complex high-dimensional distributions providing posterior distributions in a Bayesian framework or resulting in simulated likelihood functions in a likelihood-based framework. </p>
<p>Much research in Bayesian econometrics and simulation-based statistics has been devoted to the improvement of Monte Carlo techniques such as importance sampling and Markov Chain Monte-Carlo methods. Both approaches are permanently further developed by incorporating more efficient sampling algorithms and data augmentation steps for latent state variables, powerful approximations for (auxiliary) samplers as well as the use of efficient optimization and computation techniques.</p>
<p>The goal of this project is to further develop and to apply modern Monte-Carlo methods in financial econometrics and quantitative macroeconomics. A main focus is on the estimation of (multivariate) stochastic volatility models including jump components, multi-factor term structure models, dynamic mixture models as well as vector autoregressive models. We are mainly interested in comparing the performance of alternative (efficient) sampling techniques based on realistic (large) data sets as well as the derivation of model diagnostics and forecasts. A further focus will be on the development of a software package for Monte-Carlo simulation methods. This is an important step to make simulation-based methods accessible and applicable to a broader audience.</p>

<p>The project is a sub-project within the EU-RTN COMISEF ( Computational Optimization Methods in Statistics, Econometrics and Finance ) funded by the European Commission (see www.comisef.eu).</p>