臺灣博碩士論文加值系統

本研究在個股股價預測方面，嘗試從基本分析(Fundamental
Analysis)觀點，於眾多可能影響股價之指標中挑選出較具代表者，並經
由證券投資專家訪談以及參考相關研究，將最後選定結果作為股價預測之
參考因子。本研究採用倒傳遞類神經網路( Back-Propagation Network ,
BPN)方法，以四家產品結構相近之上市公司的歷史資料為樣本來源，個別
由類神經網路對其各項輸入因子與漲跌趨勢做一訓練與測試 。為避免大
盤整體漲跌與個別產業因素影響而致個股股價之絕對漲跌預測失去意義﹐
本研究亦嘗試預測個股與同一產業群之上市公司平均股價的相對漲跌。結
果顯示，在營收資訊公佈當月之漲跌趨勢預測準確率可達六成，而當月公
佈資訊於次月與第三月後則無明顯預測能力。 在大盤指數預測方面，
本研究以各項總體經濟為預測因子，並分別以二個由類神經網路建構的不
同模式進行預測，發現只要股價預測因子與股價之時間關係能適當調整，
仍能以少數預測因子達成良好預測績效。本研究亦嘗試提出一探討經濟成
長率﹑利率及台灣股市全體上市公司股票益本比(EPR)之交互關係模式，
從歷史資料探討台灣經濟成長率與上市公司獲利狀況之關係，以及利率和
本益比的循環動向，並討論時間延遲效應，使本模式適用於現實環境，如
臺灣證券交易所加權股價指數波段之波峰與谷底的預測。

This study attempts to select the significant ones from
various offinancial ratios and indicators about sale events from
the viewpoint offundamental analysis . After interviewing with
investment experts and referencing the results from related
studies , some indicators withcorrelation were eliminated and a
group of indicators was reserved . Back-Propagation Network
(BPN) approach is used in this study. We collectretrospective
data as BPN''s training samples and testing samples from four
listed companies that are with similar industrial architectures
. To prevent the noise from the effects of whole market and
sector (consists of companies which have similar products) , we
proposed amodule which is used to predict the relative trends
between individual stock price and sector it belongs . Results
indicate that the hit ratiosare about 59% in the individual
stock module and about 67% in the sector-relative module on the
end of the first month in which companiesopen their accounting
information , and there are no obvious predictingcapacities
after the second and the third month . This research also
present a model that discusses the mutualrelations among GNP
growth rate , interest rate and weighted EPR(Earning-Price
Ratio) of whole listed companies on Taiwan Stock Exchange
Company .We adopted neural network approach to identifying the
forecasting - capacities of this model on long-term fluctuation
of TSEWPI . The modelwe have proposed use retrospective date to
discuss the relations betweenGNP growth rate and aggregate
profits of whole listed company , the cyclic phenomenon within
interest rate and PER of listed company , andtime-lagging effect
. We hope this model can be applied on real world like TSEWPI
trend prediction , for example , forecasting the peak and trough
of TSEWPI curve .