MULTIVARIATE FRACTIONAL REGRESSION ESTIMATION OF ECONOMETRIC SHARE MODELS.

Area of Science:

• Econometrics
• Financial Econometrics
• Statistical Modeling

Background:

• Economic share data often presents boundary values (0 or 1) with significant probability.
• Existing econometric methods may not adequately capture the nuances of such data distributions.
• Accurate estimation of conditional mean structures is crucial for understanding economic behaviors.

Purpose of the Study:

• To extend fractional regression methodology for estimating univariate and multivariate share data.
• To address stochastic aspects of share definition and measurement in econometric models.
• To provide robust estimation and inference strategies for economic share outcomes.

Main Methods:

• Extension of Papke and Wooldridge's fractional regression methodology.
• Development of strategies for univariate cross-sectional and panel data.
• Application of full likelihood estimation using Dirichlet-multinomial models.
• Analysis of multivariate share data with considerations for aggregation/disaggregation.

Main Results:

• The proposed econometric strategies effectively estimate regression models for economic share data with boundary values.
• The extended fractional regression approach provides reliable inference for both univariate and multivariate share outcomes.
• Empirical application using U.S. Surveys of Consumer Finances data demonstrates the utility of the methods for financial asset portfolio analysis.

Conclusions:

• The developed econometric strategies offer significant improvements for modeling economic share data.
• The methodology is applicable to various economic contexts, particularly financial portfolio analysis.
• This research enhances the toolkit for analyzing bounded outcome data in econometrics.