Predicting Merger and Acquisition Targets：An Empirical Study of Electronic Industry in Taiwan

公司可能因為各種不同理由而成為購併活動的標的。這些理由可以被量化而以會計、財務或市場變數來代表。過去有許多國外學者以這些理由為基礎形成假說，進而以對應的代理變數建構出購併活動中目標公司的預測模型。這篇研究即是將此一概念應用於台灣的上市櫃電子業，試圖建立出一套可以預測出目標公司的模型，並觀察是否可利用此模型獲取超額報酬。此研究中樣本區分為估計性和驗證性兩類，前者用於模型建構而後者則用於檢定超額報酬。在研究方法上，使用二元向前逐步羅吉斯迴歸建構預測模型，分類及預估正確性則運用三種門檻率進行檢測，並利用最小化錯誤成本之門檻率決定潛在被併標的，最後形成投資組合以檢驗模型賺取超額報酬的可能性。 實證結果顯示，在預測模型所使用的六個假說變數中，公司規模、經理人不效率及股利發放假說的代理變數是顯著的。在三種不同門檻率檢測下，模型皆具有高且顯著的整體分類正確性及預估正確性。然而無論在何種門檻率之下，模型對於目標公司的發掘能力卻非常差，以致於預測出之潛在目標公司所形成之投資組合未明顯擊敗大盤指數，而無法利用此預測模型賺取超額報酬。此一結論也與過去國外學者所得到的結論相同。Firms may be merged or acquired for various reasons, some of these reasons are hypothesized and quantified into accounting, financial, or market variables. In previous studies, these variables were used to develop a target prediction model for foreign firms. The same concept is applied in this study to the Taiwan electronic industry. A forward stepwise binary logit regression is used to construct the model. Three different types of cutoff are used to test the classifying and predictive accuracy for the estimation sample and the holdout sample respectively. A portfolio of predicted targets is constructed using a cost-minimizing cutoff to test the possibility of earning excess returns from the model. The empirical results show that among six variables which are hypothesized to be important factors in predicting M&A targets, only three of them: net sales, return on equity, and retention rate, are found to be significant. The overall classifying and predicting accuracies of the model perform significantly better than pure chance, regardless of which cutoff is used. However, the ability of identifying M&A targets is quiet poor, indicating that it is difficult to earn excess returns from the model predictions. This conclusion is consistent with previous studies.TABLE OF CONTENTS 1. Introduction 1 2. Statement of hypothesis 10 3. Methodology and Data 16 3.1 The Sample 16 3.2 The Variable 19 3.3 The Statistical Technique of Model 21 3.4 The Determination of Cutoff Point 23 4. Empirical Result and Discussion 27 4.1 The Selected Variables 27 4.2 The Estimated Model 30 4.3 The Cutoffs 33 4.4 The Classifying and Predictive Accuracy 35 4.5 The Excess Return of Model Predictions 38 4.6 Discussion 40 5. Summary and Conclusion 42 Reference 4