Hierarchical time series analysis requires probabilistic forecasting techniques to account for inherent uncertainties. A probabilistic forecast proposes a range of potential outcomes. In domains like retail and electricity, where time series data exhibit significant cross-correlations and multiple hierarchical levels, existing research has not emphasized the development of models that consider these dependencies. This lack of attention is mainly due to the recently reported good performance of the simpler independent models. In response to this challenge, we introduce HRA (Heuristic Reordering Approach), a novel approach designed to enhance predictive accuracy and preserve the dependencies. Notably, HRA does post-processing using a heuristic recording technique on forecasted values and is adaptable to samples of any size. Our detailed experiments demonstrate the effectiveness of HRA by improving accuracy by up to 7% compared to existing state-of-the-art (SoTA) methods on simulated and well-established benchmark datasets. These results underscore HRA’s ability to significantly improve forecasting accuracy, preserve the correlation and address the unique complexities associated with hierarchical time series data.