Tim R.L. Fry

The Stochastic Specification of Demand Share Equations Restricting Budget Shares to the Unit Simplex

Abstract The traditional approach to estimating systems of share equations is to append a multivariate normal stochastic component to the deterministic component. The adding up restrictions are then exploited to delete one equation and estimation is carried out by maximum likelihood. Such an approach leads to an implied model which violates economic theory, as there remains a nonzero probability of shares being negative or greater than unity. In this paper we propose the use of a method from the statistics literature, compositional data analysis, and a distribution familiar to statisticians, the additive logistic normal, for the estimation of a system of demand share equations.

A Monte Carlo Study of Tests for the Independence of Irrelevant Alternatives Property

A plethora of tests for the Independence of Irrelevant Alternatives (IIA) property of Logit models of discrete choice behavior has been proposed in the literature. These tests are based upon asymptotic arguments and little is known about their size and power properties in finite samples. This paper uses a Monte Carlo simulation study to investigate the size and power properties of six tests for IIA in the multinomial Logit model. Our results show that the majority of tests based upon partitioning the choice set appear to have very poor size and power properties in small samples. Tests for IIA based upon the DOGIT model, similarly have poor size properties, but in some circumstances do have reasonable power properties.

GARCH modelling of individual stock data: the impact of censoring, firm size and trading volume

Abstract This paper explores the problems of testing and estimating GARCH models with particular emphasis on the impact of data censoring, firm size and trading volume. We conduct our investigation by analysing 1 year of daily returns data on 1014 Australian companies. Generally, our results indicate that GARCH model testing and estimation are impacted by the degree of censoring, firm size and trading volume. Specifically, our analysis produces three major findings. First, we find that low trading volume, small firm size and high censoring tend to be associated with a reduction in the presence of GARCH effects detected in the data by the LM test. Second, according to the estimation of a ‘response surface’ regression, the degree of censoring is found to be the dominant factor and stocks having a level of censoring less than 42.2% are predicted to have significant GARCH errors. Third, we find that low trading volume, small firm size and high censoring lead to a higher persistence of GARCH effects in the estimated models.

ADVERTISING WEAROUT IN THE TRANSPORT ACCIDENT COMMISSION ROAD SAFETY CAMPAIGNS

This paper uses a varying coefficient regression model to investigate whether there is any significant advertising wearout in the Transport Accident Commission (TAC) road safety campaigns on Victorian television. The results suggest that there is some evidence that the effectiveness of the campaigns may be declining with increased exposure.

A Simple Nested Test of the Almost Ideal Demand System

The MPIGLOG specification of an indirect utility function gives rise to Cooper and McLarens (1992) Modified Almost Ideal Demand System (MAIDS) specification, which nests the Almost Ideal Demand System. Following the ‘combined’ approach outlined by Fry, Fry and McLaren (1993), we transform the deterministic equations to logratio form for estimation. This procedure not only restricts the shares implied by the model to the unit simplex, but also provides a transparent representation of the restriction implied by the Almost Ideal Demand System. We estimate MAIDS (with and without the Almost Ideal Demand System restriction imposed) using the ‘combined’ approach and proceed to test the Almost Ideal Demand System restriction.

Alternative Beta Risk Estimators in Emerging Markets: The Latin American Case

In this paper we investigate the empirical performance of an alternative beta risk estimator, which is designed to be superior to its conventional counterparts in situations of extreme thin trading. The estimator used is based on the sample selectivity model. The study compares the resultant selectivity-corrected beta to the OLS beta and Dimson Betas. We demonstrate the empirical behaviour of the selectivity corrected beta estimator using a sample of stocks in seven countries from Latin America. The results indicate that the selectivity-corrected beta does correct the downward bias of the OLS estimates and is likely to better estimate stock risk.

Maximum likelihood estimation in binary data models using panel data under alternative distributional assumptions

Abstract This paper considers an alternative to the commonly used random effects probit model. The likelihood does not require numerical integration and expressions for the score and hessian are provided. The possible diagnostic use of the model is illustrated.