| The deregulation of western economic systems, external shocks like sudden, substantial changes in the oil price and the enormous increase in the availability of large data sets in finance and marketing are three major factors that have greatly influenced dynamic econometrics in the past fifteen years. As a consequence, the question how to model dynamic economic systems plays a central role in present-day econometrics. There are several schools that each emphasize certain aspects. Certain authors stress the importance of structural models while others use reduced forms as a starting point of their analysis. Other choice problems include the choice between vector autoregressive models and mixed models (which also make use of moving averages), between deterministic and stochastic approaches to describe trends and seasons, between linear and nonlinear models, between the time domain and the frequency domain, between fixed and variable parameters, between classical statistics and systems theory, between the sampling theory approach and the Bayesian approach, and between parametric, semiparametric and nonparametric approaches. The goal of the programme is to contribute to the insight that is necessary to make several of these choices; in particular extensions of traditional models are to be explored, proper inferential procedures are to be developed and the forecasting properties of the models are to be investigated. For a description of the programme we distinguish a number of themes. 1. Bayesian econometrics. An important aim is the further development of operational Bayesian methods for the analysis of dynamic econometric models with possibbly nonstationary variables. Special attention is given to the specification of flexible prior information which implies the use of non-standard distribution theory and the use of numerical integration algorithms in order to obtain posterior and predictive results. New algorithms are being developed which make use of Monte Carlo techniques like importance sampling, Gibbs sampling and the Metropolis-Hastings method. The project includes Bayesian diagnostic analysis concerning the validity of the model assumptions. The inferential procedures are used for prediction and decision analysis in finance and marketing. Applications are in the field of stochastic trends and growth, business cycle analysis and in empirical finance. 2. Estimation of large systems. Using the earlier developed procedure for restricting the covariance matrix of the data generating process, it turns out that several large systems of equations in the fields of demand analysis, production theory, international trade and marketing can be estimated in a relatively standard way: also in cases where the dynamic characteristics of these equation systems can be described by means of a small number of adjustment parameters. A start has been made with devising a test for cross-equation parameter restrictions with reasonable small sample properties and a test for structural breaks (for various dynamic specifications). The estimation of attraction models with a very large number of categories willalso be pursued. 3. System theoretic methods in econometrics. This research project is concerned with the structure and estimation of dynamic models with special attention for concepts and techniques developed in linear system theory. The central issue is the determination of a linear dynamic model from observed time series, when there is little prior knowledge (e.g. on noise properties; lag structures; endogenous and exogenous variables). Research in this project will focus on: quality evaluation of models and choice of appropriate parameterizations; complexity and simplification of models; models with latent variables (state space models); structure and estimation of factor models. 4. Flexible dynamic econometric models. Linear vector models are in most cases too restrictive to describe varying long-term trends in economic variables. This class of models can be made more general by including nonlinear relationships; by allowing varying parameters and/or by assuming a more general error structure than the Gaussian process, e.g. moving average disturbances and/or fat tailed distributions of the error process. In this project attention is paid to the analysis of testing for parameter stability; to the estimation of long run equilibrium relations and to long run forecasting. Further, long memory behaviour in the trends and in the seasonal components of economic timeseries is investigated. Variable parameter models that stem from the structural time series approach are a topic of research. Several aspects of this research project are also parts of the projects 1 and 5. 5. Seasonality and nonlinearity in economic time series. This project concerns the development of new models for economic time series, which include explicit descriptions of seasonal variation and of nonlinear features. These models can be used for forecasting, but also for better understanding the patterns in economic data. Seasonality means that within a year or week there can be differences across means, variances and even autocorrelations. Nonlinearity means that a time series may experience various regimes with different dynamic structures. For all models, we are concerned with the representation, identification, parameter estimation, interpretation and diagnostics. While doing so, it isimportant to take account of other features of economic data like trends and outliers. 6. Robust methods. Anomalous observations may have a large influence on the results of econometric model building and inference. In this project attention is paid to the behavior of tests to detect nonstationarity in multivariate time series. Also methods are to be developed for the analysis of missing data using robust methods. Further, robust estimation of ARCH en GARCH processes with applications to asset liability management will be pursued. 7. Analysis of longitudinal data. In the period 1993-1997 there were two projects, subsidized by NWO, on longitudinal econometrics. One of them studied to what extent we can find empirical support for theories of capital structure. Models and methods from panel data econometrics and robust statistics have been used to address this question. A robust version of GMM (general method of moments) estimation was developed and applied to firm data. The results of the robust procedures were markedly different from and much more satisfactory than the results of the traditional procedures. The other deals with common trends and common cycles in multivariate time series. The approach is applied in an investigation on the change and mobility of the wealth of nations. It makes use of the so-called Summers-Heston data and an econometric model where the mobility of wealth is modeled as a Markov process. The results indicate a convergence in wealth among the rich countries, a decrease in the variance among poor countries and hardly any mobility. A project on monetary exchange rate models indicated the usefulness of a panel data structure for the estimation of cointegration relations in these models for different countries. Both projects were finished successfully in 1997. |
