基於綜合特徵選擇方法優化投資組合的月交易策略

本論文設計了一個股票市場交易策略。金融市場中存儲的大量數據，如財務報告、年報、招股說明書、財經新聞、分析師報告、交易訊息等，但是這些過多的資料，可能會導致資訊過載的現象。許多前人的研究都顯示出，豐富的市場資訊可以幫助投資者制定高利潤的投資組合。但是，如果沒有有效的分析，數據對利益相關者(即股東、債權人、審計師、財務分析師和經理)就沒有用處，使用這些綜合資訊來預測股票趨勢是非常困難的。本論文幫助依賴機器輔助交易的投資者識別特定範圍(例如國家或交易時段)內的高權重指標，並製定合適的交易策略，以在回測數據和現實世界中取得優異的表現。本研究採用特徵選擇方法檢視台股資訊，從108個技術及基本面指標中找出權重較高的指標，制定月交易策略。分析2018年1月至2020年3月的回測數據，在台灣市場的實驗結果，年化收益率達到56∼132%，夏普比率為0.98∼1.52。在美國市場的實驗結果，年化收益率達到56∼125%，夏普比率為0.92∼1.36。

關鍵字

機器學習；人工智慧；選股機器人；特徵選擇；投資組合

並列摘要

This thesis designs a stock market trading strategy. The overwhelming amount of data stored in financial markets, such as financial reports, annual reports, prospec- tuses, financial news, analysts’ reports, and trading information has led to the phenomenon of information overload. Related works indicate that the most rele- vant stock information can help investors formulate high-profit portfolios. However, without efficient analysis and presentation, the data are not useful to stakeholders, i.e., stockholders, creditors, auditors, financial analysts, and managers. It is very difficult to use these comprehensive features for predicting stock trends. This thesis assists investors who rely on machine-assisted trading in identifying highly weighted indicators within specific scopes (e.g., country or trading period) and developing suitable trading strategies to achieve excellent performance in both the backtesting data and in the real world. By adopting a feature selection method to examine Taiwanese stock data, it identifies the highly weighted indicators from 108 technical and fundamental indicators and devises a monthly trading strategy. Then, this thesis develops a machine learning model to train the data. An analysis of the backtesting data between January 2018 and March 2020 finds that among the Taiwan market’s experimental results, the annualized rate of return reaches 56∼132% and the Sharpe ratio is 0.98∼1.52. Among the US market’s experimental results, the annualized rate of return reaches 56∼125% and the Sharpe ratio is 0.92∼1.36. This thesis also applies the research results to the real world. It began posting the portfolio recommended by the model on Facebook on the 10th of each month in September 2020, and it began monitoring the performance of the recommendations in November to assess the validity and reliability of the proposed model. Between September 2020 and October 2021, the model achieved a return on investment (ROI) of 50.2%. By comparison, the performance of exchange-traded fund (ETF) 0050 and ETF 0056, two of the most popular ETFs in Taiwan, reached 26.37% and 11.2% in the same period. These results show that the proposed model is not merely a theoretical concept but can also be applied to the real world. This thesis also develops a black swan detector to minimize asset volatility caused by black swans (e.g., COVID-19). By pairing the detector with the monthly trading strategy recommended by the proposed model, investors can substantially improve their risk indicators. By testing the proposed model equipped with the black swan detector on the Taiwanese market, a Sharpe ratio of 1.02∼1.524 is achieved, and performance is enhanced by 2∼4%.

並列關鍵字

machine learning ； a.i ； feature selection method ； portfolio

參考文獻

[1] Cloverdx.com. (2021) How much data will the world produce in 2021? [Online]. Available: https://www.cloverdx.com/blog/ how-much-data-will-the-world-produce-in-2021 1

Google Scholar

[2] J. Morgan, M. Kolanovic, and R. T. Krishnamachari, “Big data and ai strategies: Machine learning and alternative data approach to investing,” White Paper, 2017. 1, 2, 3, 23

Google Scholar

[3] I. Bose and R. K. Mahapatra, “Business data mining: A machine learning perspective,” Information Management, vol. 39, no. 3, pp. 211–225, 2001. 2

Google Scholar

[4] Z. McGurk, A. Nowak, and J. C. Hall, “Stock returns and investor sentiment: textual analysis and social media,” Journal of Economics and Finance, vol. 44, no. 3, pp. 458–485, 2020. 2, 7, 41

Google Scholar

[5] Y. Li and L. Liu, “Assessing the impact of retail location on store perfor- mance: A comparison of wal-mart and kmart stores in cincinnati,” Applied Geography, vol. 32, no. 2, pp. 591–600, 2012. 2

Google Scholar

國際替代計量

基於綜合特徵選擇方法優化投資組合的月交易策略

全文下載

主題瀏覽