運用計算智慧預測上市首日收盤價與投資組合最適化

Adopting Computation Intelligence to Forecast Initial Listing Closing Price and Investment Portfolio Optimization

周世昊(Shi-hao Chou)

指導教授 ： 林蒼祥 ； 倪衍森

淡江大學/管理學院/管理科學學系博士班/博士(2012年)

https://doi.org/10.6846/TKU.2012.01096

摘要

台灣於2004年3月1日以前，受到每一交易日漲跌幅7%限制，無法研究上市首日收盤價是否已充分反應上市公司真實內涵價值，亦無從判斷承銷價格的合理水準，但俟後實施上市首五日無漲跌幅制度與證券承銷參考價新制度，本文得以取樣新制之上市股票，觀察上市首日收盤價與證券承銷價格差異，並分析此上市首日超常報酬對此項活動參與者間之利益得失。本文運用倒傳遞類神經網路及適應性類神經模糊推論系統，預測上市首日收盤價，從而據以制定出最能平衡各方利益的證券承銷價格，實證結果顯示兩種類神經網路的準確率皆大幅超越新制度之承銷價，而其中又以適應性類神經模糊推論系統表現較為優異。衡量績效亦發現，兩種類神經網路的預測誤差都相當小。
延續對新制度初次上市股票的研究，探討以此類股票納為投資組合，其績效是否出現長期不佳的現象，實證顯示，建構投資組合初期，因受限投資標的數額，當處於市場空頭期間，因無法有效分散風險，致表現略遜於大盤報酬，惟上市家數逐漸增多，可納為投資組合標的亦隨之增多後，實證顯示其績效優於市場大盤表現，本文以Markowitz及CVaR方法建置投資組合，實證評估其報酬率績效，發現CVaR之500交易日模型，在平均及累積報酬均為最佳，且其Sharpe指標也是唯一為正值者。

關鍵字

初次公開募集發行(IPO) ； 倒傳遞類神經網路(BPNN) ； 適應性類神經模糊推論系統(ANFIS) ； 條件風險值(CVaR)

並列摘要

Prior to March 1st, 2004, the Taiwan stock market was subject to the daily 7% up/down limit. Hence, it was not possible to research whether the closing price of the first listing day of initial public offerings (IPOs) had been fully reflected their intrinsic value. After promulgating the new rule which sets the trading of the first five listing days without a price limit, we can observe the gap between an IPO price and the first listing day’s closing price; academia refers to this gap as “IPO under-pricing”. In order to assist involved parties in underwriting activities to find out the best IPO price for their interest, this paper adopts the BPNN and ANFIS model to forecast the first trading closing price of an IPO. By referencing the forecast price, all stakeholders can consider a reasonable price level. The empirical study shows both BPNN and ANFIS possess the superior forecasting power. Both tracking errors are under the projected range, and the ANFIS shows greater performance than BPNN.
In further examining the new rule, this paper investigates another widely discussed topic which is the Post-IPO long-term performance. We adopt the Mean-Variance model and CVaR model to construct the portfolio. The empirical study shows that in the beginning of the sample period, the stock market was in downturn trend and too few stocks could be included in the portfolio to diversify the risk. As a result, the portfolio return underperformed when compared to the benchmark index, TAIEX. Thereafter, as more stocks were included in the portfolio, the return was significantly improved and surpassed the TAIEX by a wide margin. The empirical study shows CVaR with 500 historical trading days performing better than the TAIEX and Mean-Variance model in average and accumulated returns. The CVaR 500 possesses the only positive Sharpe ratio among all returns.

並列關鍵字

Initial Public Offering (IPO) ； Back Propagation Neural Network (BPNN) ； Adaptive Neuro-Fuzzy Inference System (ANFIS) ； Conditional Value at Risk (CVaR)

參考文獻

2. 周宗南、王惠娟(2007)。應用灰色預測與演化式類神經網路建構台灣加權股價指數之預測模式。朝陽學報，第12卷， 89-106頁。

1. Acerbi, C. (2002). “Spectral measures of risk: a coherent representation of subjective risk aversion.” Working Paper.

3. Acerbi, C., and Tasche, D. (2002). “On the coherence of expected shortfall.” Journal of Banking and Finance, Vol. 26, pp. 1491-1507.

4. Alexander G. J. and Baptista A. M. (2002). “Economic implications of using a Mean-VaR model for portfolio selection: a comparison with Mean-Variance analysis.” Journal of Economic Dynamics and Control, Vol.26, pp. 1159–1193.

5. Arnone, S. A. (1993). “Genetic approach to portfolio selection.” NeuralNetwork World, Vol. 6, pp. 579-604.