Research Theme 2: Bootstrap and Other Numerical Methods in Time Series
This research theme covers a number of projects focusing on numerical issues relating to time series econometric analysis. Bootstrap and other numerical methods, including Monte Carlo simulation, are fast becoming essential tools in modern-day time series econometric analysis. Distribution theory in time series econometrics tends to be non-standard, requiring detailed numerical calculations to be performed in order to obtain workable inferential procedures. Our research focuses on the following areas:
Bootstrap Techniques
- The use of Monte Carlo and bootstrap techniques to approximate the distributions of estimators and test statistics used in time series econometrics, with particular reference to the problem of achieving valid inference in the presence of nuisance parameters. A particular application is the issue of obtaining valid tests for unit roots and co-integration when the driving errors display non-stationary volatility effects.
- Improving upon and further developing existing methods of bootstrap and Monte Carlo approximations used in time series econometrics.
- Applying bootstrap methods to both theoretical and applied problems in univariate and multivariate time series modelling.
Numerical Methods
- The use of Markov Chain Monte Carlo methods in time series econometric modelling.
- The use of Kalman filter and particle filter methods in time series econometric modelling.
Forecast Evaluation
- Investigating the properties of alternative methods for combining probability forecasts.