Note for Shape-Based Clustering on Stock Prices
Seiji Matsuhashi, Yukari Shirota · 2022 12th International Congress on Advanced Applied Informatics (IIAI-AAI) · 2022
We conducted research with time series data clustering by using stock price data. We investigated global automakers in the early days of COVID-19 by the k-Shape method. Many studies have already shown that stock prices of Chinese companies had been characteristic and not similar to other countries. In this paper, we discuss the effectiveness and problems of the k-shape method on this theme. The k-Shape method requires the standardization of the input data. However, the length of the standardization period and its start date are important to obtain our desired result. We tried two clustering analyses by a month and by four months. By four months, Chinese cluster could be discovered. However, contrary to our expectations, in clustering by a month, both Chinese and other countries’ ones have been found in the same cluster. From these comparative studies, we found that it is significant to determine the appropriate period length for the desirable result. Although this experimental result is specific to this case, it can give in general stock data analyzers a lesson that they should repeatedly evaluate the standardized periods until they can extract their required pattern comparison.