Abstract: There are a variety of indices (attributes) to evaluate the investment value of risky assets. It is vital for investors to assess their investment values of risky assets based on less indices. Taking stock market as an exmple, this paper proposes a stock-choosing approach that is based on attribute reduction and dynamic time warping distances. First an improved algorithm based on overlap ratio of attributes is proposed to pick out less attributes to evaluate stocks’ values. Then single-index dynamic time warping algorithm is extended to multiple indices and a set of dynamic time warping distances between stocks is computed based on reduced attributes. These distances are clustered and partitioned to match individual stocks into categories, and then the categoroies with better return-risk performance are chosen to construct the asset pool. Numerical experiment validates that the cumulative returns of both random-weight and equal-weight portfolio have better return than market benchmark, and the optimal allocation of assets improves the characteristics of returns and risks. This approach reduces the difficulties for investors to evaluate the stocks’ values via reduced attributes and measures the similarities between the trends of times series related to these indices by multi-index dynamic time warping distances. The procedure to choose risky assets is based on actual data, which avoids the influences of subjective factors and obtains robust assets pool. The proposed approach is suitable for the selection of general risky assets and is helpful for diversified investment if different markets and industries are taken into consideration.

Key words: attribute reduction, overlap ratio of attribute, dynamic time warping distance, similarity of tracks, cluster