運用總體經濟變數預測日經指數報酬率

NCU Institutional Repository > 管理學院 > 經濟研究所 > 博碩士論文 > Item 987654321/95387

jsp.display-item.identifier=請使用永久網址來引用或連結此文件: http://ir.lib.ncu.edu.tw/handle/987654321/95387

题名:	運用總體經濟變數預測日經指數報酬率
作者:	劉晉廷;Liu, Chin-Ting
贡献者:	經濟學系
关键词:	日經指數超額報酬;機器學習模型;RMSE及MAE;樣本外預測;Nikkei Index Excess Returns;Machine Learning Models;RMSE and MAE;Out-Of-Sample forecasting
日期:	2024-07-22
上传时间:	2024-10-09 16:45:25 (UTC+8)
出版者:	國立中央大學
摘要:	本研究探討了將總體經濟變數建構資料並運用機器學習方法應用於日經指數超額報酬預測的有效性。近年來，隨著人工智慧與機器學習的迅速發展，這些技術已被廣泛應用於金融領域，尤其是股票報酬的預測，而2023年亞洲股市又以日本股市表現最佳，眾多因素創下了歷史新高點。本文回顧了機器學習在股票市場中的應用歷程，並引入了多種機器學習模型，包括線性及非線性機器學習模型，運用於預測日本股市的超額報酬。研究期間從2000年1月至2023年9月，共285個觀測值，使用30個總體經濟預測因子，並加入了11種不同的機器學習方法以評估這些模型在預測短期、中長期、長期股票報酬方面的表現，計算各期間模型之指數超額報酬預測，比較各自樣本外RMSE及MAE數值大小，且利用樣本外預測(Out-Of-Sample){\ R}_{os}^{2\ }準則來評估是否優於基準模型。研究結果發現在向前一期的預測表現優於其他期間。整體來說，線性機器學習模型表現上皆優於非線性機器學習模型。在短、中長期預測中，線性機器學習模型以LASSO、Elastic net表現最佳，長期預測則為Ridge；而非線性機器學習模型則是不管短中長期預測皆以隨機森林模型表現最佳。進一步與基準模型評比，結果發現短期中所有模型皆優於基準模型；中長期當中，線性機器學習模型幾乎優於基準模型，但非線性機器學習模型皆劣於基準模型；最後在長期當中，只有 Ridge優於基準模型，其餘模型皆劣於基準模型。因此本研究旨在選擇最適合預測日本股市超額報酬的機器學習模型，以幫助投資者提高投資決策的準確性，並實現更穩定的投資報酬。 ;This study examines the effectiveness of applying machine learning methods to predict the excess returns of the Nikkei 225 index using constructed macroeconomic data. With the rapid development of artificial intelligence and machine learning in recent years, these technologies have been widely utilized in the financial domain, particularly in stock return prediction. The Asian stock market, notably the Japanese stock market, performed exceptionally well in 2023, reaching historic highs due to various factors. This paper reviews the application of machine learning in the stock market and introduces multiple machine learning models, including both linear and nonlinear ones, to forecast excess returns in the Japanese stock market. The study period spans from January 2000 to September 2023, totaling 285 observations, utilizing 30 macroeconomic predictor variables. Twelve different machine learning methods are incorporated to evaluate their performance in predicting short-term, medium-term, and long-term stock returns. The index excess return prediction of each model is calculated for different periods, and their out-of-sample root mean square error (RMSE), mean absolute error (MAE), and out-of-sample forecasting{\ \ R}_{os}^{2\ }\ criteria are compared to assess whether they outperform the baseline model. The findings reveal that the predictive performance is better for the one-step-ahead prediction compared to other horizons. Overall, linear machine learning models outperform nonlinear ones. In short and medium-term predictions, linear machine learning models, particularly LASSO and Elastic Net, exhibit superior performance, while Ridge performs best in long-term predictions. Among the nonlinear machine learning models, Random Forest consistently performs best across all prediction horizons. Further comparison with the baseline model indicates that all models outperform the baseline model in short-term predictions. In medium-term predictions, linear machine learning models almost universally outperform the baseline, while nonlinear machine learning models generally underperform. In long-term predictions, only Ridge outperforms the baseline, while other models fall short. Therefore, this study aims to select the most suitable machine learning model for predicting excess returns in the Japanese stock market, aiding investors in making more accurate investment decisions and achieving more stable investment returns.
显示于类别:	[經濟研究所 ] 博碩士論文

文件中的档案:

档案	描述	大小	格式	浏览次数
index.html		0Kb	HTML	25	检视/开启

在NCUIR中所有的数据项都受到原著作权保护.

社群 sharing

数据加载中.....