Kimiaki Shinkai

Fuzzy Clustering Level Analysis Using AIC Method for Large Size Samples

This paper investigates the fuzzy clustering level analysis using AIC (Akaikes information criterion) method for small size samples. Since AIC is obtained by the asymptotic normality for the maximal likelihood estimator, it is difficult to apply it to small size samples. Therefore, in the paper, we would show that the AIC method can be applied to large size samples which are constructed by a simulation with pseudo random numbers obeying several distributions.

Fuzzy Cluster Analysis and Its Application on International Stock Prices

This paper applies fuzzy cluster analysis to investigate co movement of Asian and U.S. stock prices from the viewpoints of both region and industry. Specifically, we analyze daily stock price data of Chinese, Indian, Japanese, South Korean, and U.S. firms from 2005 through 2011. The past literature has never used daily data because of non-synchronous trading times and holidays, but we resolve this problem by analyzing American depositary receipts traded in the New York Stock Exchange instead of underlying shares traded all over the world. Partition trees computed each year provide overwhelming evidence that the country effect always surpasses the industry effect (i.e., shares from the same country tend to move together but shares within the same industry do not). This finding is particularly informative for portfolio managers, choosing a country and then many kinds of industry therein is a riskier strategy than choosing an industry and then many countries. Besides this practical implication, the dominant country effect highlights a slow process of globalization. Nationality of shares should not matter in a globalized world, but there still exist barriers segmenting countries. All these results and implications are robust to different clustering methods, the frequency of data, and foreign exchange rates.

Statistical scheme via AIC for evaluating the optimal cut off level in fuzzy clustering

In this paper we show a new statistical scheme to find the optimal cut off level in fuzzy clustering which is an improvement of Uesu and Shinkai et. al [4]∼[7]. Deterministic algorithms which seek a certain equilibrium cluster level have essential disadvantage in principle. We focus in it and propose a statistical scheme via AIC.