應用自組織映射神經網路進行公司動態財務行為之體質檢定

Abstract

針對企業體質檢定研究，傳統的股價評價理論及企業體質檢定模型不是評價模式過於簡化，單用線性關係來推導風險與報酬的關係；就是架構在有疑慮的假說，使用上有許多限制條件。但由於實際金融市場系統是動態且高度非線性，不一定符合這些模型的限制條件，使其適用性與準確性可能無法滿足研究的需求。因此本研究提出了「企業動態財務體質檢定投資模型」，期望不受限於傳統模型的諸多條件，利用自組織映射圖神經網路具有非監督式分群的學習能力及視覺化聚類的優點，以靜態與動態兩種構面對企業財務體質進行分群檢定，將體質漸變過程以軌跡方式投射於二維平面，並分析體質移動軌跡的趨勢與能量，區分體質優良與不良的企業以進行投資獲利。研究中以台灣證劵交易所上市公司為實例，應用本模型進行體質檢定，並採用對沖避險交易策略，作多體質優良企業，放空體質不良企業，模擬投資獲利，最後和修正式Buy and Hold模型、無風險利率進行比較。研究結果顯示「企業動態財務體質檢定投資模型」，其模擬投資報酬結果均較前兩者比較模型為佳，其體質檢定成果具有參考價值可提供投資人作為擇股決策之參考。

In research field of corporate constitution judgement, conventional stock price evaluation methodology or corporate financial performance benchmarking model is either too simple, uses linear regression to explain the relationship between stock return and financial statements, or developed under some unreliable assumptions, ex. efficient market hypothesis. However there is considerable evidence that financial market behavior is not fully efficient and is highly nonlinear. This makes above models can’t meet our research desires. Consequently this paper proposes a “corporate financial constitution behavior dynamic judgement model”, which expects to release unrealistic constraints of conventional model as well as takes advantage of unsupervised clustering and visualization capability of Self-Organizing Map to judge corporate financial constitution in static and dynamic point of view. Furthermore, we projects corporate constitution moving trajectory into two dimension grid and calculate energy of trajectory to evaluate corporate financial performance. This study demonstrates proposed model to judge corporate financial constitution behavior in Taiwan TSEC market and adapts hedge investment strategy to buy comparative healthy corporate stocks and sell comparative weakness corporate stocks. The results show that the hedge strategy guided by proposed model generate higher risk-adjusted profit than corrected buy-and hold strategy as well as bit risk-free interest rate. Hence, The “corporate financial constitution behavior dynamic judgement model” can outperform the market index to earn extra-return and provide a valuable reference for making investing decision.