Abstract: We build on previous work and provide a dynamic model of asset markets with asymmetric information where higher originator screening effort leads to more signaling through delay of sale. We test this theoretical prediction using the mortgage market as a laboratory and processing time as a measure of screening. Our findings are threefold: First, and in line with the theory, mortgage processing time and the delay of sale after origination are strongly positively related in the data. Second, processing time is longer for mortgages with higher ex ante credit risk, i.e., observably riskier loans are processed slower. Finally, both processing time and delay of sale are negatively related to conditional mortgage default, indicating that more screening effort leads to unobservably higher quality loans that are also sold with a longer delay.
Discussant: George Malikov, University of Western Ontario
Abstract: We analyse default risk coefficient estimation when borrowers can inflate, at quadratic cost, a covariate used in credit scoring, with potential heterogeneity in latent sensitivities to interest rates. A qualified version of Goodhart’s law obtains: When the posted model utilizes coefficients from clean historical data, coefficients shift subsequently, provided the coefficient on the true covariate is not zero. As shown, measurement error resulting from manipulation is negatively correlated with the true covariate. This correlation shifts the slope coefficient upward unless noise resulting from heterogeneous interest-rate sensitivities is sufficiently high. We next evaluate internally consistent fixed point (Nash) models. If the clean covariate coefficient is not zero, so Goodhart’s critique applies, intercept and/or slope coefficients of any Nash model must undershoot clean data counterparts, and the Nash slope coefficient cannot be zero. Practically, adaptive estimation converges to a fixed point if manipulation costs are sufficiently high. Finally, an econometrician with commitment power optimally discourages manipulation with marginal increases (decreases) in the posted model intercept (slope).
Discussant: Nicolas A Inostroza, University of Toronto
