近年來,因發生許多重大金融危機事件,使得各類創新的金融商品相繼上市。但也由於投資工具愈趨於複雜,相關人員得承受更加複雜的風險型態,風險管理之學問便成為炙手可熱之議題。 在一般的理論架構之下,其報酬率往往假設為常態分配,但許多研究指出,金融風險資產大多有厚尾的情況。因此,對於金融資產的報酬建模上,使用常態分佈假設會失去其準確性,且與實際情形不符。在本研究中,我們直接藉由歷史資料來模擬未來的情境,因為使用其方法即可以不需知道報酬率的分配,便可以利用較為符合真實情況的模式架構,來分析其投資之最適組合。 本研究使用類神經網路模擬原始歷史資料,其模擬神經元的特性可以記憶可用的經驗與知識,並以Safety-First投資組合模式為基礎求解投資組合最佳化問題。 在本研究中,我們選取摩根台灣股價指數(MSCI Taiwan Index)中的前20支股票作為投資標的,並且與大盤指數作績效比較。最後結果證實,藉由Safety-First投資組合模式所選取的投資組合會優於大盤。
In recent years, there are more and more kinds of innovative financial merchandises emerging. These merchandises also make the environment of investment more complicated and let investors having more chance to expose under the risk. Due to those reasons, risk management on investment has become a serious issue and investors may to find their portfolios with downside risk controlled. Under general theoretical frameworks, the distributions of the returns are supposed to be normal distribution. But many researches indicate that most of financial assetshaving fat tail. In the study, we directly use historical data to simulate scenarios of future. It can have more flexibility to fit real conditions. Using this framework we can analyze optimal portfolios under unknown return distributions. In this study, we use Artificial Neural Networks to simulate historical data and solve safety-first optimal portfolios under the simulated scenarios. We choose top 20 stocks of MSCI Taiwan Index as investment targets and compare performances with the Market index. The results show that ours are better than Market in the sense of risk averse.