An automatic stock market categorization system would be invaluable to individual investors and financial experts, providing them with the opportunity to predict the stock price changes of a company with respect to other companies. In recent years, clustering all companies in the stock markets based on their similarities in the shape of the stock market has increasingly become a common scheme. However, existing approaches are impractical because the stock price data are high-dimensional data and the changes in the stock price usually occur with shift, which makes the categorization more complex. Moreover, no stock market categorization method that can cluster companies down to the sub-cluster level, which are very meaningful to end users, has been developed. Therefore, in this paper, a novel three-phase clustering model is proposed to categorize companies based on the similarity in the shape of their stock markets. First, low-resolution time series data are used to approximately categorize companies. Then, in the second phase, pre-clustered companies are split into some pure sub-clusters. Finally, sub-clusters are merged in the third phase. The accuracy of the proposed method is evaluated using various published data sets in different domains. We show that this approach has good performance in efficiency and effectiveness compared to existing conventional clustering algorithms.