Estimation of discrete choice network models with missing outcome data

This paper considers the problem of missing observations on the outcome variable in a discrete choice network model. The research question is motivated by an empirical study of the spillover effect of home mortgage delinquencies, where mortgage repayment decisions can only be observed for a sample of all the borrowers in the study region. We show that the nested pseudo-likelihood (NPL) algorithm can be readily modified to address this missing data problem. Monte Carlo simulations indicate that the proposed estimator works well in finite samples and ignoring this issue leads to a severe downward bias in the estimated spillover effect. We apply the proposed estimation procedure to study single-family residential mortgage delinquency decisions in Clark County of Nevada in 2010, and find strong evidence of the spillover effect. We also conduct some counterfactual experiments to illustrate the importance of consistently estimating the spillover effect in policy evaluation. •NPL estimators can be modified to estimate binary network models with missing data.•Monte Carlo simulations show satisfactory performance of the proposed estimator.•The proposed estimator is applied to study mortgage delinquency spillovers.•Mortgage delinquency decisions in a neighborhood have spatial spillovers.•Counterfactual studies show it is important to address the missing data issue.