use both, a two-step IV-probit
approach (column (1) and column (3)) and an IV linear probability model (column(2)
and column(4)). The first two columns correspond to the model using the broader distress
definition as dependent variable (Bank Distress) while the third and the fourth
column use the distress definition taking into account only outright bank failures (Bank
Default). We also estimated our benchmark regressions using simple linear probability
models and probit models (results not shown): the results from these robustness checks,
available upon request, leave the main message of the previous results unaltered. For
completeness we also present the results of the IV regressions of the models using the
z-score and the non-performing loans ratio as dependent variables. The results of IVprobit
regression using the broader distress measure (column (1)) tell the same story as
the simple logit approach of the previous sections: Increasing bank-level pricing power
reduces the probability of experiencing a distress event, providing further support for the
competition-fragility hypothesis. Simultaneously, more concentrated banking markets
are, ceteris paribus, characterized by riskier banks. Finally, banks located in states with
a lower competitive conduct, i.e. higher values for the Boone Indicator, have also higher
distress probabilities. Results are slightly different when applying the IV-linear probability
model. The Lerner Index and the Boone Indicator enter significantly and have their
familiar signs (negative and positive, respectively). In contrast, the variable measuring
the market contestability/concentration, although still positive, looses its significant effect.
Concerning the results of the IV-regressions employing the narrowly defined distress
indicator (column (3) and column (4)), we again find that all our previous results remain
valid when applying an IV-probit approach (column(3)).