Time Series Clustering을 이용한 한국 기업들의 주가 흐름 분석 : The stock price stream analysis of South Korea corporations using time series clustering

다국어 초록 (Multilingual Abstract)

In this paper, I have looked into corporations that have been listed in KOSPI market for ten years and examined following questions: 1) among stock indicators of these corporations, what kind of fluctuations in stock prices are clustered together 2) whether corporations in similar industry form same cluster.
I have proposed three clustering method in order to answer the questions. First is clustering by calculating rate of change of stock prices. Second is clustering using coefficients calculated from substituting stock price’s rate of change into spectral density function. Last is clustering using second method after smoothing out row-data. To test which method is superior, I have compared results attained from each method.
The results from using hypothetical data show that second method using time series’ rate of change and third method are superior to first method. After eliminating first method and applying second and third method to real life data, I came to conclusion that the third method is the best method for time series clustering.
The significance of this research lies with the fact that I approached a way to cluster seemingly unpredictable time series using row data, which is used in different area of study, to construct a model. Additionally, I have shown to what extent moving average method can fix the flaws of time series.

목차 (Table of Contents)

• 1. 서론 1
• 2. 본론 3
• 2.1 Clustering 3
• 1) Clustering 3
• 2) 계층적 군집화 방법(Hierarchical Cluster Procedures) 3
• 2.2 Time Series Clustering 접근법 4
• 1) Row-data에 기초한 접근법 5
• 2) Feature에 기초한 접근법 5
• 3) Model에 기초한 접근법 5
• 2.3 평활법(smoothing Method) 6
• 1) 이동평균법(Moving Average) 6
• 2) 지수평활법(Exponential smoothing Method) 6
• 2.4. spectral density function 7
• 3. 실증분석 9
• 3.1 생성 data를 이용한 분석방법의 비교 9
• 3.2 실제 주가 data를 이용한 분석방법의 비교 11
• 1) 자료의 설명 11
• 2) Clustering 결과 12
• 4. 결론 15