2019-01-012024-05-16https://scholars.lib.ntu.edu.tw/handle/123456789/666751摘要：本子計畫是三年期的研究計畫，計畫目標是應用 AI 技術來輔助金融機構進行優化選股(stock picking)策略、市場擇時(market timing)策略及開發智慧指數型ETF(smart index ETF)自動化交易系統。第一年的計畫主題是優化選股策略，除了納入傳統的結構化金融市場資料(例如財務報表資訊、股票市場、選擇權市場交易資訊等)來進行技術分析和基本分析的選股方式外，還將採用最常被使用的類神經網絡(neural network)、隨機森林(random forest)及支援向量機(support vector machine)等機器學習(machine learning)方法，來挑選出上漲機率或上漲幅度較高的股票。第二年的計畫主題是優化市場擇時策略，建立市場多、空訊號，透過監測即時金融資料的變化，即時對於投資部位做調整等是資產配置的重要工作，本計畫首先將蒐集重要經濟指標、金融壓力指數及其成分變數、指數選擇權及其他市場情緒指標等資訊，然後利用傳統的計量方法(例如:羅吉斯迴歸或馬可夫狀態轉換法(Markov regime switching methods))來建立擇時(多、空訊號)模型，接著再進一步結合類神經網絡、隨機森林及支援向量機等機器學習的方法來提升多、空訊號的準確性。第三年的計畫主題是開發智慧指數型ETF自動化交易系統，近年來被動型指數ETF受到許多投資人的青睞，原因是主動型基金管理費較高且長期投資績效並未優於被動型指數ETF的績效。本計畫首先將研究開發指數型ETF自動化交易系統，按指數權重購買其成分股形成投資組合來形成指數型ETF，由於是自動化交易，可以節省許多人力成本，因此管理費可以相當低廉具有競爭優勢。其次，利用第一年和第二年的研究成果，將選股策略和市場擇時策略納入被動型ETF的投資組合中，找出指數成分股中預期表現較佳(差)的股票群，微幅調高(低)其權重；同時當預期未來市場多(空)頭訊號明顯時，藉由買入(賣出)指數期貨來微幅調高(降)投資組合的beta值；最後再善用自動化交易和AI系統的優勢，有效且快速地累積和處理大量的市場價量資料，找出額外獲利的機會。本研究預期微幅調整後投資組合的報酬率將增加，這樣的投資組合我們稱之為智慧AI指數型ETF。<br> Abstract: This sub-project is a three year project. In the first year, we will focus on the stock picking strategies using machine learning techniques. All the relevant corporate information such as the financial statements, credit rating, stock market data, and stock option data will be collected. We begin with the traditional stock picking strategies such as the usage of fundamental analysis and technical analysis to pick up those stocks with higher probability of larger returns in the future. Comprehensive risk-return analyses will be implemented for each stock picking strategy. We next apply the machine learning methods, such as neural network, random forest, and support vector machine methods, to sort stocks into winner stocks and loser stocks. The inputs of the machine learning methods include all corporate data and some fundamental indicators and technical indicators calculated in the previous step. In the second year project, we will study the market timing strategy and its profits. We first collect the macro economic variables such as economic indicators, financial stress index, index option market data, and market sentiment data and apply the traditional econometric method to predict future states of the economy and stock market. We then apply the machine learning methods to predict future states of the economy and stock market using the above macro variables. In the third year project, we first develop an automatic trading system to replicate the passive index ETF. Based on the findings of the first year project, we will fine tune the index ETF portfolios by increasing the weights of winner stocks and decreasing the weights of loser stocks. Based on the findings of the second year project, we will increase (decrease) the index ETF portfolios by buying (selling) market index futures if the predicted future state of stock market is a booming (busting) state. We expect that the fine-tuned index ETF portfolio can earn a higher risk-adjusted return and thus we term it a smart-AI index ETF.選股策略市場擇時策略智慧指數型ETF基本分析技術分析機器學習類神經網絡隨機森林支援向量機經濟指標金融壓力指數指數選擇權市場情緒羅吉斯迴歸馬可夫狀態轉換法智慧AI指數型ETstock pickingmarket timingsmart index ETFfundamental analysistechnical analysismachine learningneural networkrandom forestsupport vector machinelogistic regressionMarkov regime switching methodseconomic indicator