請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91216
標題: | 以風險規避之深度值分布強化學習進行市場摩擦下之選擇權避險 Risk-averse Deep Distributional Reinforcement Learning for Option Hedging under Market Frictions |
作者: | 林鼎鈞 Ding-Jun Lin |
指導教授: | 呂育道 Yuh-Dauh Lyuu |
關鍵字: | 風險規避強化學習,值分布強化學習,選擇權避險,市場摩擦,時間一致性,期望短缺,演員–評論家演算法, risk-averse reinforcement learning,distributional reinforcement learning,option hedging,market friction,time consistency,expected shortfall,actor-critic method, |
出版年 : | 2023 |
學位: | 碩士 |
摘要: | 本篇論文結合風險規避強化學習以及深度值分布強化學習之技術,應用於可能存在市場摩擦之離散時間選擇權避險。具體而言,我們提出一個使用深度強化學習之選擇權避險架構,透過時序差分學習獲得損益分布函數之神經網路表徵,並從偽樣本中即時性地估計風險。此架構具多用性,因為其風險估計量以及分布之表徵形式皆為模組化。相較於其他以直接估計風險為目標的深度強化學習模型,此架構亦具有較高的獨立性與穩健性,因為它能在一個不依賴選擇權定價模型來建構報酬的選擇權避險情境中習得更好的避險策略與更準確的風險估計。另外,透過損益分布之表徵,此架構可進一步延伸為一種具有時間一致性的期望短缺最佳化方案之深度學習實現。最後,我們提出一個概念性驗證來展現此架構可從隨機的避險情境中習得可泛化的避險模型。 The thesis applies the combined techniques in risk-averse reinforcement learning (RL) and deep distributional RL to discrete-time option hedging with possible presence of friction. Specifically, we lay out a deep RL option hedging framework in which a neural network representation of the profit and loss distribution function is obtained through temporal difference learning and the risk is estimated from a pseudo-sample on the fly. The framework is versatile because the risk estimator and the distribution representation are both modular. It is also more independent and robust than several deep RL models that aim to directly estimate risk, in that it learns better hedging policies and more accurate risk predictions in a hedging setting where the reward formulation does not depend on an option pricing model. Moreover, the access of a representation of the profit and loss distribution allows extension of the framework to a novel deep learning implementation of a time-consistent optimization scheme of expected shortfall. Finally, we demonstrate a proof of concept that this framework can learn a generalizable hedging model from randomized hedging instances. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/91216 |
DOI: | 10.6342/NTU202304116 |
全文授權: | 同意授權(全球公開) |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-112-1.pdf | 2.3 MB | Adobe PDF | 檢視/開啟 |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。