應用關聯性規則探勘於股市時間序列分析─以TFT-LCD產業股價行為為例

Abstract

TFT-LCD是目前台灣政府積極推動的「兩兆雙星」產業之一，本研究希望能夠透過資料探勘，找出存在於TFT-LCD產業上下游間股價之特殊現象並改進Apriori algorithm使其適用於跨交易日離散型非連續性之資料探勘。 本文之研究採二階段實驗，研究方法係先利用R. Agrawal et(1996)提出之方法進行資料分類及前處理，再利用Apriori algorithm計算support and confidence,找出第一階段之strong association rules；第二階段則是改善Apriori，突破傳統利用圖形比對找尋股市行為模式時之連續性比對的限制，針對時間序列資料進行跨日探勘，並得到統計檢定上顯著之成果。本研究提出一有效之跨交易日離散型非連續性資料探勘方法論，應用於TFT-LCD產業供應鏈上下游間股市時間序列資料分析之模型。研究結果發現產業上下游並非完全存在著正向相關。 研究結果顯示利用此規則，當上游供應商Sintek單日漲幅超過5%時，於之後第七日放空下游廠商Client LiteON Co.，並於第十日回補，在統計檢定95%的信賴區間下，有非常顯著之年報酬率。換言之，應用此結果將有助於投資人在未來某特定時間區段，進行樍桿性之套利投資策略，並可將本研究所建構之跨交易日離散型非連續性資料探勘方法論模型，應用於其他產業或時間序列資料的分析上。

TFT-LCD is one of industries currently promoted by the “Two Trillion and Twin Star Industries Development Plan” in Taiwan. This study endeavors to find out the stock price associations between the suppliers and manufacturers in the value chain of the TFT-LCD industry by means of data mining techniques, and meanwhile, to improve the Apriori algorithm so that it can facilitate association mining of discrete data points in a time series. An efficient data mining method which consists of two phases is proposed. In the first phase, data are classified and preprocessed using the algorithm proposed by R. Agrawal et al. (1996), then Apriori algorithm is applied to extract the strong association rules. The second phase further improves the Apriori algorithm by breaking down the traditional limitation of relying on pattern matching of continuous data for disclosing stock market behavior. By mining the association rules from the discrete data points in a time series and testing the corresponding hypotheses, statistically significant outcomes can be obtained. The proposed data mining method was applied to some real time-series of the stock prices of companies in the supply chain of TFT-LCD industry in Taiwan. It is suggested that a positive correlation does not necessarily exist between the companies’ stock prices in the supply chain of TFT-LCD industry. For instance the result shows that, if the stock price of Sintek Phonrotic Corp., a company in the up stream of the value chain, soars for more than 5% in a day, the stock price of LiteON, a company in the down stream of the same value chain, may not respond positively accordingly. If an investor can short the stock of LiteON on the 7th day and long it back on the 10th day after Sintek’s stock price soaring for more than 5%, the annual return on investment is striking with 95% confidence interval. In conclusion, the results may reveal helpful information for the investors to make leveraged arbitrage profit investing decisions, and it might be interesting to apply this proposed data mining method to the time series in other industries or problems and investigate the results further.