Econometric Model Building and Forecasting

Summary

This project proposes to tackle a number of important and challenging issues in econometric model building and forecasting under cross sectional dependence, heterogeneity and nonlinearity. This project seeks to establish new and flexible econometric models associated with estimation methods and user-friendly computational techniques to potentially solve some real world problems. The research outcomes of this project are expected to make significant contributions to enrich the academic literature as well as to be practically relevant and useful to empirical researchers in evaluating and improving model building and forecasting from better models in climatology, demography, economics, environment, finance, machine learning and neural networks.

Researchers

Project background and aims

Model building and determination of good models for estimation and prediction is a very important part in much empirical research in econometrics. Since economic and financial data often display various types of characteristics, such as cross–sectional dependence (i.e., interdependency among economic agents, such as firms or individuals, due to externalities, spillovers or common shocks) for cross-sectional and panel data, and nonstationarity for time series data, one important aspect of model building and determination is how to take such characteristics into account. Meanwhile, the increasing availability of big data in recent years has been accompanied by increasing interest in both theoretical research and empirical analysis with big data in many different areas of economics as well as more broadly within the social and medical sciences. In building econometric models for big data analysis, there are various challenges. One challenge is how to do data aggregation when there are many economic transactions. A simple way is to use a tool associated with principal component analysis. Unlike dealing with big data issues in medical sciences for example, we do need to pay attention to many econometric issues, such as multicollinearity when aggregating big data for economic model building. One additional complexity is that aggregated components of a large number of economic variables can become highly dependent even when the original individual economic variables may be mutually independent. Another challenge is the development of new econometric models and estimation techniques that are not only capable of addressing big data issues, but also user-friendly to empirical researchers. Conventional econometric models and techniques often work with many economic and financial data, but there are issues unique to big datasets that may require new tools. For instances, the availability of big datasets facilitates the applicability of nonlinear models and estimation methods, which normally require large sample sizes. When using a nonlinear model and estimation method for a big data problem, however, there are many new issues, such as variable selection and the curse of dimensionality.

The main aims include (i) it introduces a class of semiparametric varying–coefficient models that allow for possible cross-sectional dependence, heterogeneity and nonlinearity; (ii) it proposes flexible model building and estimation methods that are suited to these cross-sectional and panel data models and provide empirical researchers with closed-form estimation methods and user-friendly tools for ease of computation and implementation; and (iii) it is empirically applicable to solve some real world problems.

Methodology

New econometric models and methods are being developed with potential applicability in solving real world problems in climatology, demography, economics, environment, finance, machine learning and networks.

Findings

The main findings include: 1) the project develops a number of new econometric models that are not only theoretically novel, but also empirically relevant and applicable; 2) the proposed models and methods systematically solve some real world problems associated with cross-sectional dependence, heterogeneity, nonlinearity and nonstationarity; 3) the project has been offering some personnel and training benefits in helping to attract international leaders in econometrics and economics to visit and work in Australia as well as in training talented young researchers as an ARC Research Associate as well as PhD and three Masters/Honours candidates.

Output

This research team has been both actively and productively working on this project. Many of the research outcomes have been presented as keynote and invited speeches at leading international conferences and workshops. Some of the research outcomes have already been published by leading international journals in economics, econometrics, finance and statistics.

Funding information

This project has been financially supported by the Australian Research Council Discovery Grants Program under Grant Number: DP170104421.

Research group

This research group is part of the Monash Business School’s International Network of Excellence in High-Dimensional Dynamical Systems.