Does Collateral Help Mitigate Adverse Selection? A Cross-Country Analysis

It is defined as the weighted average normalized to a scale of 0 to 100 of survey responses to the question, “Generally speaking, would you say that most people can be trusted or that you can’t be too careful in dealing with people?” Where a response of one indicates that most people can be trusted and zero indicates that you can’t be too careful.

Liberti and Mian (2010) also support evidence on the role of financial development in reducing information asymmetries. They show in their analysis that greater financial development reduces the difference in rates of collateralization between high and low risk loans in a country.

Data provided by Djankov et al. (2007) are for each year from 1991 to 2002, and we consider the 2002 value for the following years.

Data for the rule of law indicator are available for 1996, 1998, 2000, 2002, 2003, 2004, 2005, and 2006. Therefore, we extrapolate the value for the index for the missing years, taking into account the fact that law enforcement slowly changes over time.

A term loan is defined in Dealscan as an instalment loan where amounts repaid may not be re-borrowed.

We only comment the thresholds for estimations with the full conditioning information set (odd-numbered estimations), but the findings are similar when the thresholds are considered for estimations with the simple conditioning information set (even-numbered estimations).

We have also carefully checked the cross-correlation matrix for firm, loan, and country-level variables.

We have also performed regressions with firm characteristics for the baseline specifications without interaction terms leading to similar results regarding Collateral as in Table 2. Furthermore, using different specifications regarding firm characteristics variables (for instance with Firm size, Leverage and Liquidity ratio only) lead to similar results.

The DWH test is implemented by including the residuals of each endogenous right-hand side variable (Loan Spread and Collateral respectively) as a function of all exogenous variables (see equation 1), in a regression of the original model. A coefficient of the residual statistically different from zero indicates that the endogenous regressor is correlated with the regression error and that OLS is not consistent.

The Collateral equation is estimated using a Probit regression.

The Hansen-Sargan test is implemented in the following way. We obtain the 2SLS residuals by regressing Loan Spread against the set of explanatory and control variables and the fitted value of Collateral. The latter is obtained with a Probit regression in a first stage (Angrist 2001). Then we regress the residuals on all the explanatory and control variables and instruments to obtain its R². The Hansen-Sargan statistic is computed as this R² times the number of observations less the number of regressors. This statistic is chi-square distributed with degrees of freedom equal to the difference between the number of instruments and the number of endogenous variables. If the Hansen-Sargan statistic is not significant then the instrumental variables are valid and our system of equations is well-identified.