We present a novel empirical approach based on categorizing systematic risk, the beta of a stock, for evaluating the performance of two recently reported interval estimation methods, the asymptotic and the wild bootstrap, suitable for estimation from high-frequency data. In a dual-beta context, the robustness of the estimation methods is assessed using three different lengths of the sampling interval that fall within the range deemed reasonable for achieving a balance between bias and precision of estimates derived from intra-day data. We apply clustering of variables to categorized betas to assess similarity of market risk experienced by various stocks in up and down market conditions when such risk is estimated using different methods and data sampled at differing sampling intervals. Our study suggests that regardless of the length of the sampling interval, the estimation procedure and market conditions are the major influencing factors. The effect of the length of the sampling interval in producing dissimilar estimates is more in up market conditions compared with down market conditions for both estimation methods. The study also suggests that categorization based on the wild bootstrap method provides more robust results than the asymptotic results, to the choice of different intra-day sampling intervals.