台灣漁會信用部金融預警系統之研究──類神經網路模式之應用

An Application of Neural Network on Early Warning System by Rating for the Credit Department of Fishermen Association in Taiwan

本文探討倒傳遞類神經網路（BPN）應用於基層金融財務危機之預測，本研究比較各預警模式之估計樣本預測能力的實證結果顯示，其正確預測能力依序為原始財務變數倒傳遞網路模式正確預測率（81.1%）為最佳模式，其他依序為因素分析後倒傳遞網路模式（77.85%）及OrderedLogit 模式（75.9%）。事實上，漁會信用部這類基層金融經營若出現問題，將會引起連鎖損害與信用危機問題，故基層金融營運更需要有效的預警系統，基此，本文建議之類神經網路可提供基層金融單位及早發現問題，並採取相關的預防或管理措施。

This paper applies the Back-Propagation Network (BPN) to build thefinancial distress prediction models. Empirical results show that theeffect of BPN on crisis management mechanisms towards communities’financial institutions in Taiwan is doing quite fine. In addition, thepredictability comparison indicates that the highest accuracy is thePrimitive BPN (81.1%) in the surveillance system, followed by theFactory BPN (77.85%) and the Ordered Logit (75.9%). Damages andimpacts to the fishing community and industry are always far moreserious when financial crises occur in the community’s financialinstitutions. Thus, a more accurate financial warning system forgoverning these financial institutions is needed more than ever. Theartificial neural network (ANN) suggested in this study can provide abankruptcy predictor of financial distress among credit unions.