本篇論文主要係研究我國於民國 105 年起初次上市 (櫃) 的競價拍賣新制度，並探討競拍底價與競價拍賣投標結束日前 10 日的興櫃市場平均成交價格對投標者是否具有參考性及是否可作為預測競拍加權平均得標價格的因子。本文以民國 105 年 1 月 1 日至 110 年 12 月 30 日的 180 家初次上市 (櫃) 公司，其中 150 家公司為樣本內資料，而其餘 30 家公司作為預測的樣本外資料。我們透過迴歸分析 (regression analysis)、隨機森林 (random forest)、類神經網路 (artificial neural network) 三種模型預測競拍加權平均得標價格結果與競拍公告實際結果進行實證分析。實證結果發現以迴歸分析模型預測之結果與競拍公告加權平均得標價格有最小的均方誤差 MSE (mean-square error)，且相較一般實務僅採用競拍投標結束日前三日的日成交均價折價 13%~15% 作為投標價有更佳的效果。本研究依據上述結果對實務界提出建議，作為未來研究之參考。

This thesis mainly explores the new auction policy for initial public offerings (IPO) implemented in Taiwan since 2016. Furthermore, it discusses whether the upset price and the average transaction price of the emerging stock market, 10 days prior to the end date of the auction, can be used as references for the bidder, as well as whether they can be used as factors to predict the weighted average bid prices. A hundred and eighty firms that completed their IPOs between January 1, 2016, and December 30, 2021, were included in this study wherein 150 firms are in-sample data, and 30 firms are out-of-sample data used for the prediction. Three models were adopted in this study, namely, regression analysis, random forest, and artificial neural network, to predict the weighted average bid prices of the auctions, which along with the actual results in auction announcements are further subject to an empirical analysis. According to the empirical findings, the results predicted by the regression analysis model have the smallest mean-square error (MSE) with the weighted average bid prices in auction announcements. Moreover, compared with the general practice, which simply uses the average transaction price 3 days prior to the end date of the auction with 13–15% taken off as the bid price, the regression analysis model achieves better results. Based on the above findings, this thesis puts forth recommendations for practitioners, hoping to provide references for future studies.

中文參考文獻
1.江淑貞與吳桂燕 (2013)，「為何台灣的競價拍賣逐漸式微於初次上市市場」，《管理論叢刊》，23，29-64。
2.李孟霖 (2016)，「競價拍賣新制政策效果分析」，碩士論文，台北： 國立政治大學財務管理研究所。
3.姜堯民與戴維芯 (2007)，「興櫃交易對於初次上櫃績效的影響」，《證券櫃檯月刊》，129，17-32。
4.姜堯民與戴維芯 (2016)，「台灣股票初次上市 (櫃) 相關文獻研究回顧」，《經濟論文叢刊》，44，77-125。
5.徐燕山與徐政義 (2004)「IPO 競價拍賣中投資人標單的資訊內涵：以臺灣為例」，《政大財管叢刊》，12，27-54。
6.郭峻豪 (2009)，「競價拍賣折價程度與承銷商市佔率之關係」，碩士論文，台北：國立政治大學財務管理研究所。
7.陳香吟 (2014)，「興櫃股票市場發行面之興利與改革措施」，《證券暨期貨月刊》，32，第9期6-16。
8.陳湘琴 (2015)，「推動初次上市 (櫃) 案件優先採用競價拍賣方式辦理承銷」，《證券暨期貨月刊》，33，第12期24-41。
9.陳香吟與吳怡瑩 (2018) ，「淺談初次上市 (櫃) 承銷制度及價格訂定」，《證券暨期貨月刊》，36，第4期21-38。
10.黃茂欣 (2011)，「新股初上市 (櫃) 報酬分析與興櫃市場價格發現機能」，碩士論文，台北：國立政治大學財務管理研究所。

英文參考文獻
1.Breiman, L. (2001), “Random Forests,” Machine Learning, 45, 5-32.
2.Caudill, M. (1987), “Neural Networks Primer,”. AI Expert, 2, 46-52.
3.Chiang, Y. M., Hirshleifer D., Qian, Y. and Sherman, A. E. (2011), “ Do Investors Learn from Experience? Evidence from Frequent IPO Invertors,” Journal of Financial Economics, 78, 3-29
4.Derrien, F., and Womack, K. L. (2003), “Auction vs. Book-building and the Control of Underpricing in Hot IPO Markets,” Review of Financial Studies, 16, 31-61.
5.Ibbotson, R. G. (1975), “Price Performance of Common Stock New Issues,” Journal of Financial Economics, 2, 232-272.
6.Rock, K. (1986), “Why New Issues Are Underpriced,” Journal of Financial Economics, 15, 187-212.
7.Welch, I. (1992), “Sequential Sales, Learning and Cascades,” Journal of Finance, 47, 695-732.