A0678
Title: Combining classification tree analysis with network grouping of Japanese stock returns
Authors: Takashi Isogai - Bank of Japan (Japan) [presenting]
Abstract: A set of classification trees is built that provide sorting rules to reproduce the stock groups identified by correlation clustering of the Japanese stock returns. The clustering is achieved by hierarchical network division by the modularity maximization algorithm of complex networks theory. We try to link the clustering results that are based on the stock price data with non-price external data in order to explore how the hierarchical division process can be explained by other categorical and numerical variables. Various non-price data including price performance data and sector classification are examined as effective variables to explain the splits of the stock groups. Variables with a high level of relative importance scores are identified; specifically, the market capitalization and price book-value ratio are included as significantly important variables. The selected variables seem to be consistent with variables included in the standard stock price model such as Fama-French factor model. Some other variables are also identified as ones that clarify the properties of the Japanese stock market. Further, variables that represent local features of the Japanese stock market are also detected. The classification tree analysis method can also be applied to find the closest group of stocks even for stocks that have limited price data due to low liquidity.
