計畫系統編號PF10907-2081
研究性質應用研究
計畫編號MOST109-2410-H008-014-MY2
研究方式學術補助
主管機關科技部
研究期間10908 ~ 11007
執行機構國立中央大學財務金融學系
年 度109年
研究經費654千元
研究領域財政（含金融，保險）；
研究人員葉錦徽
中文關鍵字序列蒙地卡羅；可預測性；學習；變數選擇；投資組合配置；狀態空間模型；虛無範數；結構改變；LASSO
英文關鍵字Sequential Monte Carlo； predictability； learning； variable selection； portfolio allocation； state space model； zero-norm； structural break； LASSO
中文計畫概述計畫中嘗試以列蒙地卡羅抽樣法作為基礎，推出一套廣泛能適用於適應性監測投資組合管理，同時能夠從海量數據中歸納資訊並在可掌控、容易更新、管理的資產配置中維持良好的績效。在計畫中我們一反過去投資組合建構先估計再最適化的作為，反向思考地先從大量可得的資產中以符合虛無範數條件下的迴歸變數選擇問題為憑藉，優化找出一組相對為數較小的資產個數開展我們的投資組合；再接續為這些為度較低的資產報酬建構較一般性、容許較多建模彈性地的狀態空間動態模型供後續監控、調整。計畫中我們嘗試納入新的基因演算法提升求解的效率並與序列蒙地卡羅法，以及其他相關做法進行比較。我們也將相關的方法應用到指數型基金的績效追蹤、以及時間序列中結構改變點的認定等問題中，並進行相關的比較。最末，我們回到利用所選的資產組合標的配適一般化動態系統的估計、推論、學習與比較，以便日後資訊的更新與調整。我們相信這樣的分析方法工具與架構將有助於許多經濟與財金領域相關模型的估計。
英文計畫概述We propose a comprehensive modelling approach based on sequential Monte Carlo to adaptively monitor portfolio management and enable deduce information from massive data set yet maintain superior performance with an easy-to-update and manageable portfolio allocation and manage system. Set against the classical first-inference-then-optimize procedure, we begin with a reverse engineering by firstly propose to select only a handful of small number of assets and optimal weights with sparsity from a big sky of assets using newly developed variable selection under a penalized zero-norm constraint. The chosen assets are then modeled and estimated with a general state-space model in a concise way. We introduce and examine the performance of a stochastic genetic differential algorithm in our application. The proposed alternative approaches are then applied in the context of portfolio selection/index tracking/identification of structural breaks. We then build a general state-space model with flexible specifications for the chosen assets for dynamic modelling, and adaptive monitoring and updating. We believe this analyzing framework based on SMC may shed lights on future modelling, and effective updating for applications in Economic and Finance.
報告系統編號RW11309-0289
計畫中文名稱以序列蒙地卡羅提升高維計量模型結構式更新/學習效能的新方法及在財務與經濟的應用
計畫英文名稱New Methods for Structural Updating/Learning in High Dimensional Econometric Models via Sequential Monte Carlo with Applications in Economics and Finance
主管機關科技部
計畫編號MOST109-2410-H008-014-MY2
執行機構國立中央大學財務金融學系
研究期間10908 ~ 11007
報告頁數頁
使用語言中文
研究人員葉錦徽 YEH JIN-HUEI
中文關鍵字Sequential Monte Carlo； predictability； learning； variable selection； portfolio allocation； dimension reduction； common factors
英文關鍵字Sequential Monte Carlo； predictability； learning； variable selection； portfolio allocation； dimension reduction； common factors
中文摘要We propose a comprehensive modelling approach based on sequential Monte Carlo to adaptively monitor portfolio management and enable deduce information from massive dataset yet maintain superior performance with an easy-to-update and manageable portfolio allocation and manage system. Set against the classical first-inference-then-optimize procedure, we begin with a reverse engineering by firstly propose to select only a handful of small number of assets and optimal weights with sparsity from a big sky of assets using newly developed variable selection under a penalized zero-norm constraint. The chosen assets are then modelled and estimated with a general state-space model in a concise way. We introduce and examine the performance of a stochastic genetic differential algorithm in our application. The proposed alternative approaches are then applied in the context of portfolio selection/index tracking/identification of structural breaks. We then build a general state-space model with flexible specifications for the chosen assets for dynamic modelling, and adaptive monitoring and updating. We believe this analyzing framework based on SMC may shed lights on future modelling, and effecti
英文摘要We propose a comprehensive modelling approach based on sequential Monte Carlo to adaptively monitor portfolio management and enable deduce information from massive dataset yet maintain superior performance with an easy-to-update and manageable portfolio allocation and manage system. Set against the classical first-inference-then-optimize procedure, we begin with a reverse engineering by firstly propose to select only a handful of small number of assets and optimal weights with sparsity from a big sky of assets using newly developed variable selection under a penalized zero-norm constraint. The chosen assets are then modelled and estimated with a general state-space model in a concise way. We introduce and examine the performance of a stochastic genetic differential algorithm in our application. The proposed alternative approaches are then applied in the context of portfolio selection/index tracking/identification of structural breaks. We then build a general state-space model with flexible specifications for the chosen assets for dynamic modelling, and adaptive monitoring and updating. We believe this analyzing framework based on SMC may shed lights on future modelling, and effective updating for applications in Economic and Finance.
