| Abstract: | 本篇研究依據李桐豪(1989)統合多種隨機過程的方法,探討台灣股票報酬率的隨機行為。網路式概似比檢定系統被架構起來,以篩撿出解釋台灣股票報酬率較佳的隨機過程。本研究使用樣本共有195家股票,時間則從民國61年1月起至民國83年6月止。研究結果與李所使用美國資料類似,發現含固定成長因子的單純隨機走路過程被資料所拒絕,而複雜如歐費爾德等(1977)的模型亦不如其他較為簡化的隨機過程。本研究並未如李所發現美國股票在跳躍間有輕微的負相關。相反地,如果台灣股票報酬率可以用歐費爾德等的模型解釋的話,則跳躍間有高度正相關的現象。網絡式概似比檢定結果則指出無論是波氏型態(Poisson-type)或是伯努利型態(Bernoulli-type)的跳躍,隨機走路過程配合非零的成長因子與非零的跳躍因子對台灣股票報酬率的隨機行為能有較佳的解釋能力。在此架構下過去相關研究如沛士(1967),貝克士(1981),及伯爾與托絡士(1983)等的模型其對台灣股票報酬率解釋能力都不如上述較完全的模型。本研究亦發現台灣股票報酬率跳躍的幅度與變異程度都較隨機走路本身來得大,這或許是台灣股票市場能迅速反應市場訊息的間接證據。在本樣本期間跳躍的幅度是一正值。相對地,隨機走路過程的成長因子則平均顯得十分的小。
This research project studies the stochastic processes of Taiwan`s stock returns. Lee`s (1989) unified approach to examine the stochastic behavior of Taiwan`s stock returns is adopted. A nested likelihood ratio tests are also performed to identify a better stochastic description of Taiwan`s stock returns. After studying 195 stocks` monthly returns from January 1972 to June 1994, I find results similar to Lee`s finding with the U.S. data. A pure random walk with a drift model is soundly rejected. A complicated Oldfield et al. (1977) model is not only difficult to estimate but also not giving a better result compared to simpler alternative models. Unlike Lee`s negative autocorrelation result, however, the correlation between jumps are highly positive, in any, for Taiwan`s stock returns. The nested likelihood ratio tests indicate that a random walk with a non zero drift in combination with a non zero jump is a better model in describing Taiwan`s stock return behavior. This is true for both the Poisson-type and Bernoulli-type jump. All the previous studies by Press (1967), Beckers (1981), and Ball and Torous (1983) are worse than the corresponding more complete model. The size and volatility of the jump is clearly higher than those of the random walk. It seems that Taiwan`s stock market reacts to economic information sharply, and overtime, such reaction is positive in average. Without inflow of substantial information, which is unlikely in a month time span, Taiwan`s stock returns would follow a random walk with a minimal drift in average. |
