Incapacity to pay or moral hazard? Public mortgage delinquency rates in Chile

Abstract

High delinquency rate of publicly provided mortgages in social housing programmes are often interpreted to be due to moral hazard. In this article we show that the typically used parametric approaches give misleading results because of overlooked confounding and selection biases. We show that using the more appropriate impact or treatment nonparametric approach the problem of high delinquency rate in publicly provided mortgages is due to the incapacity to pay and not due to moral hazard. The results caution against public policies to encourage private mortgage providers to move down market, and suggest eliminating mortgages and correspondingly increasing the grant component of the programmes.

Notes

¹For a detailed description of the programs and a meta-impact evaluation of them, see Ruprah and Marcano (2007).

²See Pardo (2000).

³See Morandé and García (2004).

⁴The full list of control variables includes age, age squared, gender (dummy), years of schooling, years of schooling squared, income of household head, household head with labour contract (dummy); household beneficiary of public housing program, head in public employment, self-employed as an entrepreneur, household poor (dummy); household income, household income decile, residential area urban or rural (dummy); location in which region with respect to the capital, the number of household members. The coefficient is statistically significant with different set of control variables ranging from a maximum of 0.51 and a minimum of 0.46. The results are available from the authors on request.

⁵Let us assume that t is treatment (private versus public), d the outcome indicator (delinquency rate) and y the confounding variable (income). If t is independent of y, then only, it could contribute to a nonsystematic variation of d. The impact of source of the loan on the delinquency rate would be . If the source of the loan were independent from income, would be zero and it would be easy to estimate the impact of the source of the loan on the delinquency rate. However, because is not zero, it would be necessary to separate the impact of each other. This involves the problem of knowing or identifying the real structural form of the relationship. In such a case, it would be necessary, for example, to use instrumental variables.

⁶See Duflo and Kremer (2003) for the argument in favour of randomization.

⁷See Blundell and Costa (2002) for a comprehensive review of this approach.

⁸The National Survey of Socioeconomic Characteristics (CASEN) for its Spanish abbreviation) of 2003 includes 4903 households with open loans of which 3633 were SERVIU and 1270 were Private. After dropping households with missing information, the sample was reduced to 3261 consisting of 2514 SERVIU and 747 Private. Using the 3261 households, the propensity of having a private loan was estimated using 28 explicative variables. The list of variables includes imputed rent as a percentage of household income, age of head, gender of head, years of schooling of head, income of head, head in public employment, household is poor, household income decile, total household income, region located and distance from regional capital. Depending of the balancing process, the support group depends on the parameters and methodology used and possibly changes every time the iteration of creating a support group and balancing is performed. The preferred estimations reported in the text consisted of a support group formed with 312 SERVIU and 310 private mortgage holders.

⁹The full list of control variables includes age, age squared, gender (dummy), years of schooling, years of schooling squared, income of household head and household head with labour contract (dummy); household beneficiary of public housing program, head in public employment, self-employed as an entrepreneur, household poor (dummy); household income, household income decile, residential area urban or rural (dummy); location in which region with respect to the capital, the number of household members. The coefficient is statistically significant with different set of control variables ranging from a maximum of 0.51 and a minimum of 0.46. The results are available from the authors on request.

¹⁰ See MINVU (1998).