AQFC2015

Machine Learning in Finance

-------------------------------------------------------------------------------------------------------------------
 
 
      Department of Systems Engineering and Engineering Management
                             The Chinese University of Hong Kong
 
-------------------------------------------------------------------------------------------------------------------
Speaker:  Chuan-Hsiang Han 韓傳祥(Department of Quantitative Finance, National Tsing-Hua University, Taiwan)
 
Title: Machine Learning in Finance
 
Abstract:
We review methods of machine learning, and apply pattern recognition, artificial neural network, support vector machine, etc to solve for investment problems in the equity market. Despite that technical analysis remains a disputing discipline between financial practice and academics. We demonstrate that machine learning can be applied for technical analysis, which include chart recognition and technical indicators. Pattern recognition suits for charting. Its primary  difficulty is to identify various geometric shapes from historical price charts in a general and automated way. Query by singing/humming (QBSH in short) is an artificial intelligence technique for audio processing. It has been developed successfully for musical information retrieval in the computer science society during last two decades. We present computational algorithms, statistical inference and empirical implementations for chart recognition by QBSH and show strong evidences on the discrepancies of stock return distributions before and after the presence of chart patterns. GPU parallel computing is further used to accelerate the process of chart recognition. A significant speedup can be obtained under the CUDA C environment.
Artificial neural network and support vector machine(or regression) are developed for trading strategies by combing with a generic approach for time series to detect and characterize breaks for additive seasonal and trend. Portfolio performance of these machine generated trading strategies are compared.
Lastly, we show that a possibility that machine can learn from chart patterns to create another new class of trading strategies. All these techniques can be useful for applications in financial technology.
 
 
Brief Bio:
Dr. Han received his PhD in applied mathematics from North Carolina State University in 2003. After a two-year joint post doctor position in IMA, University of Minnesota and Ford Motor Company in Michigan, he joined the Department of Quantitative Finance, National Tsing-Hua University, Taiwan in 2005. He has been an adjunct associate professor of Department of Mathematics, National Taiwan University since 2012. In 2013, he established Nvidia-NTHU joint lab on computational finance for high performance computing on financial applications by GPU parallel acceleration.
Dr. Han’s research area includes singular and regular perturbation methods, variance reduction methods for Monte Carlo simulations, and stochastic volatility models. In addition to regular academic activities, he also serves as directors for Taiwanese Association of Traders and Financial Engineers and Chinese Society of Financial Technology Management, and involves with several projects with Taiwan futures exchange and financial institutions.
 
 
 
This seminar is hosted by Prof. Chen Nan.
 
 
Venue: Lady Shaw Building(LSB)  C3
 
      The Chinese University of Hong Kong.
 
 
 
Date: 
Thursday, April 14, 2016 - 08:30 to 09:30