Tomoya Suzuki, Ph.D.

Profile

Dr. Tomoya Suzuki received his B.S., M.S., and Ph.D. degrees in physics from the Tokyo University of Science in 2000, 2002, and 2005, respectively. Then, he joined Tokyo Denki University as an assistant in 2005 to teach electric circuits. From 2006 to 2009, he was a lecturer at Doshisha University, teaching computer languages and computer engineering. Since 2009, Dr. Suzuki has been an associate professor at Ibaraki University, teaching mathematics, statistics, and computer science.
His research interest is the physics of complex systems, such as financial markets, and his research methods are time series analysis, prediction, machine learning, and data mining with computers.
In particular, his recent research is involving the relationship between technical analysis and complex science. From this viewpoint, nonlinear prediction models based on neural networks can be considered very useful for developing new technical analysis methods.
He is a member of the Nippon Technical Analysts Association (NTAA); the Institute of Electronics, Information and Communication Engineers Information and Systems Society (IEICE); the Information Processing Society of Japan (IPSJ); and the Physical Society of Japan (JPS).

Presentation Abstract

Ensemble Neural Networks for Identifying Market Patterns and their Confidence

Along with the recent development of computer power and its price reduction, machine learning tools like neural network has been familiar to us. In particular, by releasing free computer languages such as R and Python, we can easily use the latest machine learning tools even if we do not have advanced programming skills.

The largest advantage of machine learning is that it can automatically identify some regular patterns hidden in enormous historical price data. Because it uses historical price data for analysis, it could be categorized as one of technical analysis, but is sometimes called modern technical analysis in terms of using computer techniques.
However, the machine learning itself is not enough for a trading algorithm because the identified patterns are the past things. Similarly, it has some problems such as the overfitting problem and the local-minimum problem. Moreover, the more popular the machine learning becomes, the less its advantage would be.
In my presentation, I would like to point out these problems and introduce some tips to apply machine learning algorithms more efficiently to trading stocks. Especially, I introduce my own ideas[1,2] based on the ensemble learning method used with neural networks to quantify the confidence of machine learning as a new technical indicator. By using this indicator, we can focus only on more predictable stocks having higher confidence. This is how we can reduce the number of trick sings that we often encounter in technical analysis.

[Reference]
[1] T. Suzuki, T. Hayashi: Financial Technical Analysis Based on Deterministic Nonlinear Prediction, IEIEC Journal, Vol.J98-A, No.2, pp.237-246, 2015.
[2] T. Suzuki, Y. Ohkura: Modern Technical Analysis based on Nonlinear Bagging Predictors, Physica A, under review.