Browsing by Author "Chen, Chiao-Ting"

Now showing 1 - 4 of 4

Cloning Strategies from Trading Records using Agent-based Reinforcement Learning Algorithm
(2018-01-01)
Chen, Chiao-Ting
;
Chen, An-Pin
;
Huang, Szu-Hao
;
資訊管理與財務金融系註：原資管所+財金所
;
Department of Information Management and Finance
Investment decision making is considered as a series of complicated processes, which are difficult to be analyzed and imitated. Given large amounts of trading records with rich expert knowledge in financial domain, extracting its original decision logics and cloning the trading strategies are also quite challenging. In this paper, an agent-based reinforcement learning (RL) system is proposed to mimic professional trading strategies. The concept of continuous Markov decision process (MDP) in RL is similar to the trading decision making in financial time series data. With the specific-designed RL components, including states, actions, and rewards for financial applications, policy gradient method can successfully imitate the expert's strategies. In order to improve the convergence of RL agent in such highly dynamic environment, a pre-trained model based on supervised learning is transferred to the deep policy networks. The experimental results show that the proposed system can reproduce around eighty percent trading decisions both in training and testing stages. With the discussion of the tradeoff between explorations and model updating, this paper tried to fine-tuning the system parameters to get reasonable results. Finally, an advanced strategy is proposed to dynamically adjust the number of explorations in each episode to achieve better results.
Decision Support System for Real-Time Trading based on On-Line Learning and Parallel Computing Techniques
(2016-01-01)
Chen, Chiao-Ting
;
Huang, Shih-Jung
;
Chang, Yang
;
Hsiao, Chih-Yen
;
Lin, Jiun-Yi
A novel intraday algorithmic trading strategy is developed based on various machine learning techniques and paralleled computing architectures in this paper. The proposed binary classification framework can predict the price trends of Taiwan stock index futures after thirty minutes. Traditional learning-based approaches collect all samples during the training period as the learning materials. The major contribution of this paper is to collect a subset of similar historical financial data to train the real-time trading model. This goal can be achieved by an on-line learning technique which is required to calculate an accurate model with training time limitation. In addition, the proposed joint-AdaBoost algorithm is to improve the system performance based on the concept of paired feature learning and planar weak classifier design. The core execution components in this algorithm can be further accelerated with the aid of Open Computing Language (OpenCL) parallel computing platform. The experimental results show that the proposed learning algorithm can improve the prediction accuracy of final classifier from 53.8% to 61.68%. Compared to the pure CPU implementation, the OpenCL version which uses CPU and GPGPU simultaneously can reduce the calculation time around 83.02 times. The efficiency improvement can decrease the delay of investment opportunity which is a critical issue in real-time financial decision support system application. To sum up, this paper proposed a novel learning framework based on joint-AdaBoost algorithm with similar learning samples and OpenCL parallel computation. The extended financial decision support system is also proven to work effectively and efficiently in our simulation experiments to trade the Taiwan stock index futures.
Deep Learning in Model Risk Neutral Distribution for Option Pricing
(2019-01-01)
Chou, Chin
;
Liu, Jhih-Chen
;
Chen, Chiao-Ting
;
Huang, Szu-Hao
;
資訊管理與財務金融系註：原資管所+財金所
Option pricing has been studied extensively in recent years. An important issue in option pricing is the estimation of the risk neutral distribution of an underlying asset. Better estimation of this distribution can lead to a more rational investment, enabling one to earn an equal return with lower risk. To price options precisely and correctly, traditional financial engineering methods make some assumptions for the risk neutral distribution. However, some assumptions of traditional methods have proved inappropriate and insufficient in empirical option pricing analysis. To address these problems in option pricing, this study adopts a data-driven approach. Owing to advances in hardware and software, studies have been using deep learning methods to price options; however, these have not adequately considered the risk neutral distribution. This may cause an uncontrollable risk, thereby preventing the real-world application of the model. To overcome these problems, this study proposes a deep learning method with a mixture distribution model. Further, it generates a rational risk neutral distribution with accurate empirical pricing analysis.
Forecasting Interaction of Exchange Rates Between Fiat Currencies and Cryptocurrencies Based on Deep Relation Networks
(2019-01-01)
Chen, Chiao-Ting
;
Chiang, Lin-Kuan
;
Huang, Yi-Cheng
;
Huang, Szu-Hao
;
資訊管理與財務金融系註：原資管所+財金所
Forecasting exchange rates is difficult because financial time-series data is too complicated to analyze. In traditional financial studies, economic models and statistic approaches were widely used for predicting exchange rates. Recently, machine learning and deep learning techniques have played increasingly important roles in financial technology studies. This study adopts a deep learning technique called relation networks (RNs) to predict the exchange rates of fiat currencies and cryptocurrencies. To discover the relationship among different currencies, the concept of visual question answering (VQA) is applied in RNs. We also propose a specially designed architecture for the feature extraction stage to consider both spatial and temporal relationships simultaneously. The experimental results show that the proposed approach can achieve higher prediction performance for cryptocurrencies with approximately 65% accuracy rate. We aim to improve traditional approaches and construct a model using the concept of VQA based on RNs to optimize the prediction performance between fiat currencies and cryptocurrencies.