This paper shows why regressing the realised rates of depreciation within the exchange rate band on a given information set and conditional on (ex-post) actual no-realignment (à la drift adjustment) still encounters a Peso Problem. Such a procedure generally gives inconsistent estimates. The main reason is that the frequency of realignments in the data need no coincide with the frequency of the subjective (even small) probabilities that a realignment may take place. These probabilities cause jumps in the exchange rate even when it is conditional on actual no-realignment. When using an alternative approach that takes care of the peso problem and provides consistent estimates of the expected rate of realignment, we find that our estimates of the expected realignment (devaluation) rates are always greater than the ones obtained using the drift adjustment method.