Predicting missing observations in unbalanced panels: A Kalman filtering-smoothing approach

Rambaldi, Alicia N., Hill, R. Carter and Doran, Howard E. (2002). Predicting missing observations in unbalanced panels: A Kalman filtering-smoothing approach. In: S. Hurn and A. Layton, ESAM02 - Econometric Society Australasian Meeting. Econometric Society Australasian Meeting, Brisbane, Qld, Australia, (1-23). 7-10 July 2002.

Author Rambaldi, Alicia N.
Hill, R. Carter
Doran, Howard E.
Title of paper Predicting missing observations in unbalanced panels: A Kalman filtering-smoothing approach
Conference name Econometric Society Australasian Meeting
Conference location Brisbane, Qld, Australia
Conference dates 7-10 July 2002
Convener School of Economics and Finance, Queensland University of Technology
Proceedings title ESAM02 - Econometric Society Australasian Meeting
Place of Publication Brisbane QLD, Australia
Publisher Queensland University of Technology
Publication Year 2002
Sub-type Fully published paper
Editor S. Hurn
A. Layton
Start page 1
End page 23
Total pages 23
Collection year 2002
Language eng
Formatted Abstract/Summary
Unbalanced panels are often found in empirical settings. In this paper we concentrate on the case where a particular variable of interest is only observed sporadically for the members of the panel, but a set of explanatory variables exists for the complete panel. We propose a methodology that optimally predicts the missing observations and insures that when an observation exists, the prediction equals the actual observation. We start by defining an econometric model for the variable of interest, which is then formulated as a state space model in terms of the observed data at each time period. A contemporaneously correlated error covariance matrix for the predictions is postulated, in order to gain information from other cross-sectional units in the panel. The method produces a prediction (and prediction error) for each missing observation, estimates of the parameters of the original econometric model and one-step-ahead forecasts for each cross-section in the panel. The estimation and prediction are carried out using Kalman filtering-smoothing algorithms, and both stationary and integrated cases can be handled. An empirical application is presented for the prediction of (log) sale prices for houses based on a hedonic price function.
Subjects EX
149999 Economics not elsewhere classified
140304 Panel Data Analysis
Keyword unbalanced panels
Kalman filter
housing prices
Q-Index Code EX

Version Filter Type
Citation counts: Google Scholar Search Google Scholar
Created: Fri, 24 Aug 2007, 00:56:14 EST