在本研究當中，先以總體經濟資料與市場價格資料進行特徵工程，形成週頻數據集，接著使用XGBoost演算法進行特徵篩選，挑選出重要變數並進行主成份分析，以建構出精萃數據集。最後再藉由集群分析，希望將經濟情勢相近的週頻樣本凝聚成各個集群。

而在實際資產配置決定時，則取出最近一期樣本之所屬集群，並將該集群內的樣本日期彙整為時間區間，意即歷史上這些時段的市場狀態應與當下相近，故使用該時間區間之日報酬作為資產配置權重估計使用。

歷史回測上，使用此方法建構之投資組合能在2008年轉為保守型配置，避開2009金融海嘯。在回測期間2005年4月1號至2019年4月1號之間，本研究之投資組合在夏普指標、年化報酬、最大回撤皆能優於比較基準。

This study first uses macroeconomics and market price data in feature engineering, creating new features. Then apply XGboost to carry on with feature selection, remaining important variables and using principal component analysis to form the extracted weekly data set. The data set will be used in cluster analysis, aiming to gather weekly samples to clusters that have similar economic climate.

In practice, when deciding asset allocation, the clustering result belonging to the latest sample will be taken and the dates of the other samples in that cluster will be organized to time intervals. It means that the market situation in the time intervals will resemble the situation of the asset allocation decision day. Thus, daily returns in the time intervals are applied to calculate asset allocation weights.

In the backtesting results, the portfolios constructed by the present study's methodology can transform into a more conservative asset allocation in 2008, and avoiding the financial crisis of 2009. Furthermore, the portfolios perform better than the two benchmarks, equal-weighted and historical mean-variance optimization.

論文審定書 i
誌謝 ii
中文摘要 iii
英文摘要 iv
圖目錄 vii
表目錄 viii

第一章 緒論 1
1.1 研究動機 1
1.2 研究目的 1
1.3 研究架構與流程 2

第二章 文獻回顧 3
2.1 資產配置與景氣預測 3
2.2 機器學習於金融之應用 4
2.3 文獻回顧小結 4

第三章 研究方法 5
3.1 實驗架構 5
3.2 資料蒐集與建立 7
3.3 重要特徵篩選 12
3.4 集群分析 17
3.5 資產配置 21

第四章 實證結果 23
4.1 集群結果 23
4.2 資產配置回測 25

第五章 結論與建議 33
5.1 結論 33
5.2 後續建議 34

附錄 34
參考資料 35

游欣慧（1999）。多種情境模式資產配置之研究。國立臺灣大學財務金融學研究所碩士論文，台北市。

張婉蘭（2002）。因應台灣景氣循環的最適資產配置投資組合之研究。國立高雄第一科技大學金融營運所碩士論文，高雄市。

賴曉莉（2009）。主動式V.S被動式管理基金之研究。國立中正大學財務金融所碩士論文，嘉義縣。

Cai, F., Le-Khac, N. A., & Kechadi, T. (2016). Clustering approaches for financial data analysis: a survey. arXiv preprint arXiv:1609.08520.

Chen, S. S. (2009). Predicting the bear stock market: Macroeconomic variables as leading indicators. Journal of Banking & Finance, 33(2), 211-223.

Chen, T., & Guestrin, C. (2016). Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining (pp. 785-794). ACM.

Davig, T., & Hall, A. S. (2019). Recession forecasting using Bayesian classification. International Journal of Forecasting, 35(3), 848-867.

Friedman, J. H. (2001). Greedy function approximation: a gradient boosting machine. Annals of statistics, 1189-1232.

Johnson, S. C. (1967). Hierarchical clustering schemes. Psychometrika, 32(3), 241-254.

Keller, W. J., Butler, A., & Kipnis, I. (2015). Momentum and Markowitz: a golden combination. Available at SSRN 2606884.

MacQueen, J. (1967, June). Some methods for classification and analysis of multivariate observations. In Proceedings of the fifth Berkeley symposium on mathematical statistics and probability (Vol. 1, No. 14, pp. 281-297).

Markowitz, H. (1952). Portfolio selection. The journal of finance, 7(1), 77-91.

Obeidat, S., Shapiro, D., Lemay, M., MacPherson, M. K., & Bolic, M. (2018). Adaptive Portfolio Asset Allocation Optimization with Deep Learning. International Journal on Advances in Intelligent Systems, 11(1), 25-34.

Raffinot, T. (2017). Hierarchical clustering based asset allocation. Available at SSRN 2840729.

Ward Jr, J. H. (1963). Hierarchical grouping to optimize an objective function. Journal of the American statistical association, 58(301), 236-244.