We evaluate the reliability of credit gap measures estimated over time samples of different lengths. We augment our empirical analysis (which turned out to be somewhat inconclusive) with Monte Carlo experiments. For this purpose we build an agent-based model that realistically reproduces credit cycles and use it to generate the artificial data set. We found that 12-15 years of available data is sufficient for the estimation of reliable credit gaps (i.e. the reliability of credit gap estimates will not improve substantially as more data are added to the sample).