Based on a switching-cost model, we examine empirically the hypotheses that bank loan mark-ups are countercycical and asymmetric in their responsiveness to recessionaly and expansionary impulses. The first econometric model treats changes in the mark-up as a continuous variable. The second treats them as an ordered categorical variable due to the discrete nature of prime rate changes. By allowing the variance to switch over time as a Markov process, we present the first conditionally heteroscedastic discrete choice (ordered probit) model for time-series applications. This feature yields a remarkable improvement in the likelihood function. Specifications that do not account for conditional heteroscedasticity find evidence of both countercyclical and asymmetric mark-up behavior. Incontrast, the heteroscedastic ordered probit finds the mark-up to be countercycical but not significantly asymmetric. We explain why controlling for conditional heteroscedasticity may be important when testing for downward stickiness in loan rates.