Welcome to the web page of Nvidia-NTHU Joint Lab on Computational Finance (Nvidia-NTHU 計算金融聯合實驗室).
Contact us via chhan@mx.nthu.edu.tw
Click here to enter its Chinese version.
This laboratory focuses on parallel computing by GPU for solving problems of computational finance. Numerical methods include
(1) Numerical PDE
(2) Fast Fourier Transform Method
(3) Monte Carlo Simulations
These numerical methods can be benefited from parallelization and are implemented by Matlab GPU and CUDA C. Our current research reveals positive results.
A lecture note on "Introduction to Matlab GPU Acceleration for Computational Finance"
can be
downloaded here
or via Software Development.
Recently we have applied techniques of machine learning on investment problems. Those techniques include neural network and support vector machine.
Financial applications contain:
- Topic 1: Volatility Estimation by Fourier Transform Method
Analysis of volatility information content: the instantaneous volatility estimated from Fourier transform method is a leading index compared to the historical volatility.
Reference: C.H. Han. Instantaneous Volatility Estimation by Fourier Transform Methods. Handbook of Financial Econometrics and Statistics (C.F. Lee eds.), Springer-Verlag, New York. 2015.
a. Volatility-based algorithmic trading strategies for high frequency data
Develop volatility-based algorithmic trading strategies for high-frequency data. The test example is Taiwan index futures data with backtest.
Strategy 1: A back-testing result from trading Taiwan index futures. (trading frequency: per sec.)
Strategy 2:A back-testing result from trading Taiwan index futures. (trading frequency: per 15 sec's. Data period 1/1/2010 ~ 8/6/2013)
b. Risk Management for VaR/CVaR estimation with empirical backtest
Traditional VaR estimation methods include historical simulation, RiskMetrics (EWMA), GARCH(1,1) model, etc. Given TAIEX(Taiwan Capitalization Weighted Stock Index), SPX(S&P 500 index), and some other daily currency rates, we develop a new approach for VaR/CVaR estimation under stochastic volatility models.
Backtesting results for VaR indicate a better significant level of this approach.
Example 1:99% VaR/CVaR of S&P 500 Index under Stochastic Volatility Model
Example 2:99% VaR/CVaR of JPY/USD Exchange Rate under Stochastic Volatility Model
Reference: C.H. Han, W.H. Liu, and T.Y. Chen. VaR/CVaR Estimation under Stochastic Volatility Models. International Journal of Theoretical & Applied Finance. Volume 17, Issue 02, March 2014.
- Topic 2: Model Calibration to Implied Volatility Surface
a. GPU-based Monte Carlo calibration
Combine variance reduction methods with GPU computing to reduce standard errors. Current results reveal over 100 times acceleration and provide a better fit to the implied volatility surface. These show outstanding of GPU parallel computing.
Current research projects focus on multi time-scale and hybrid volatility models.
Example:Model calibration to implied volatility surface of SPX options under multi-scale stochastic volatility model
b. GPU-based FFT calibration
Under investigation.
- Topic 3: Rare Event Simulations
a. GPU-based importance sampling
Study high-dimensional joint default probability estimation with application to credit risk and systemic risk. Compared with some existed commercial codes, our new algorithms have advantages of (1) high performance, (2) dimension free, and (3) parallelization.
Example: Simple instructions of Matlab GPU can speed up computation easily.
b. Entropy-based (a.k.a. CE) importance sampling
Study default probability estimation of a portfolio and portfolio optimization problems.
Example: Efficient frontier of 100 stocks
Reference: C.H. Han and Y.-T. Lin. Accelerated Variance Reduction Methods on GPU. Proceedings of the 20th IEEE International Conference on Parallel and Distributed Systems, 2014.
Latest News:
Invited booth demo at Matlab Computational Finance seminar. Taipei. June 23, 2015.
Invited booth demo at Matlab Computational Finance seminar. Taipei. June 4, 2013.
產學合作課程:「期貨計量系統性交易技術概論」(見授課大綱) 將於2013年春季開設,本課程與元大寶來期貨自營部合作教學。修課優良者可獲得由賀氏基金會所提供之『計量財務金融獎學金』
(見詳細辦法),進行實習合作。
Jan. 25, 2013. Congratulations to the student team (Christie Chen 陳靜、 Lichia Yeh 葉力嘉、 I-Chien Lai 賴怡誠、Chien-Liang Kuo 郭建良) for winning the first place of 2012 Taiwan CUDA Contest with the annual theme BIG DATA.
2012 台灣 CUDA 程式設計大賽(主題巨量資料(big data)處理)第一名.
新聞聯結: 國家高速網路與計算中心
或 NVIDIA .
Project: GPU-Based Monte Carlo Calibration to Implied Volatility Surfaces under Multi-Factor Stochastic Volatility Model.
First brown bag presentation and the seminar of Dr. Simon See (NVIDIA Chief Solution Architect. Director for Solution Architect). Dec. 7, 2012. NTHU, Taiwan.
held at QF, NTHU.
Second brown bag presentation. Time: 12 noon- 2 PM, Feb. 7, 2013. Place: Room 644, Delta BLD, NTHU, Taiwan.