• Corpus ID: 15125109

Addressing endogeneity in residential location models

@inproceedings{GuevaraCue2005AddressingEI,
  title={Addressing endogeneity in residential location models},
  author={Cristi{\'a}n Angelo Guevara-Cue},
  year={2005}
}

• Cristián Angelo Guevara-Cue
• Published 2005
• Economics

Some empirical residential location choice models have reported dwelling-unit price estimated parameters that are small, not statistically significant, or even positive. This would imply that households are non-sensitive to changes in dwelling unit prices or location taxes, which is not only against intuition, but also makes the models useless for policy analysis. One explanation for this result is price endogeneity, which means that the price is correlated with the error term in the econometric model. This problem is caused either by the simultaneous determination of the supply and the demand for dwelling units in aggregated models, or by omitted attributes that are correlated with the price, in the disaggregated ones. The treatment of endogeneity in discrete choice models is an area of ongoing research in econometrics. Therefore, methods to treat this problem began to be proposed only in the last decade, and have not been thoroughly analyzed for residential location models. This thesis evaluated the available methods to treat endogeneity in discrete choice models. Each method was tested in terms of its applicability and robustness in a residential location choice framework, using a set of Monte Carlo experiments. The results showed that the control-function method (Petrin and Train, 2004) is the most promising one to address endogeneity in this framework because it is the best to handle individual level endogeneity and it is tractable with available estimation software. Finally, the application of the control-function method to an example based on real data from Santiago de Chile showed not only that the problem of price endogeneity does exist in residential location choice models, but also that the control-function method gives a satisfactory answer to the problem. Further venues of research are discussed at the end of the thesis, in particular, the usage of non-parametric methods to improve the estimation results of the control-function method. Thesis Supervisor: Moshe E. Ben-Akiva Title: Edmund K. Turner Professor of Civil and Environmental Engineering