自1970 年以來，世界各國銀行危機此起彼落，對全球經濟造成嚴重之傷害，至今業已有多篇研究致力於對銀行危機之形成原因與預警模型之探討，希望能夠及早獲知警訊並防患於未來。本研究承續先前的研究脈絡嘗試利用訊號法的雜訊比來篩選重要變數後，進一步結合Panel Logit 模型對panel data 的解釋能力，來建構銀行危機預警模型。並將資料集分成樣本內資料集作為模型建構，以及樣本外資料集作為建構模型之測試。實證結果顯示，實質匯率波動、出口量、股價指數、M2 貨幣乘數及商業銀行存款量對於銀行危機有顯著影響，證實了過去假設的推論。而經訊號法篩選之變數所建構之Panel Logit 模型，其預警能力在樣本外資料集確實優於訊號法所建構的四種綜合指標，此實證結果可提供實務界防患於未來之參考。

Banking crises have been showing a great damage to the global economic since 1970. Since then there have been many papers trying to find out the main causes and formulate the early warning model to avoid its recurrence. Following the same purpose, this research is trying to construct an early warning system for banking crisis by combining signal approach and panel logit method. The data set is divided into two sets, in-sample and out-sample data sets. In-sample data set is used for model construction, and out-sample data set is used for the model validation. The empirical results show that real exchange rate change, export, stock index, M2 multiplier, and commercial bank deposit have significant effects on the occurrence of banking crisis, which corroborate the past results. The panel logit model with variables filtered through signal approach has better prediction power than the four indexes constructed from the signal approach for the out-sample data. The proposed model can be of great help to avoid the occurrence of banking crisis for the authorities in the practical sense.