實證分析的結果顯示，將高頻資料使用以周為調整資產配置時間、持股公司數為25、使用一般化柏拉圖分配做為證券報酬之邊際分配，並且使用常態關聯結構模型，代入最佳化投資組合進行資產配置，所得到的投資組合最能夠快速的反映市場狀況，並且有效的降低投資風險，本研究建議積極型的投資人使用此投資策略來提升其投資績效。 How to get the information and analyze is one of the most important issue in financial market. This research devotes to providing the investors the proper investment strategies through the optimal asset allocation. There are several aspects need to be considered before one could approach the optimal strategies.
First of all, the inputs of the objective function are the key ingredients to reach the minimum risk. Secondly, the marginal densities fitness of the assets and the correlation between the assets could fine tune the input estimation if the statistical methods are used properly. At last, the data frequency is also another ingredient of information. Different data frequency provides different microstructure information, therefore, lead to a different strategy.
To achieve the goal, this research adopts the best fitted asset return marginal distribution out of six marginal densities, Generalized Pareto distribution. Two copula models (normal copula and T copula) are incorporated to catch the correlation between the assets. Different portfolio sizes and rolling agenda settings are investigated.
Using intra-day 5 minutes high frequency data, it is found empirically that the optimal portfolio return can be reached at portfolio size 25, adopting Generalized Pareto distribution and normal copula model, rolling out and reinvesting weekly.