Stock Price Prediction Using Support Vector Machine Algorithm

Literature Review

Financial time series forecasting is one of the most challenging applications of modern time
series forecasting. According to Yang et al. 2004, financial time series are inherently noisy, non-
stationary, deterministically chaotic and may be influenced by many factors; such as the
economy, business conditions, and political events to name a few. These characteristics suggest
that there is no complete information that could be obtained from the past behavior of financial
markets to fully capture the dependency between the future price and that of the past. It can
easily be seen that there are certain recurring patterns in the history of market prices, and there
are various approaches for classifying them (Nison n.d.). But a much harder task is to recognize
such patterns in the constantly evolving financial markets early and with sufficient reliability.
The challenge of stock forecasting is appealing because a small forecasting improvement can
increase profit significantly. However, the volatile nature of the stock market makes it difficult to
apply linear models, simple time-series or regression techniques.
A number of artificial intelligence and machine learning techniques have been used over the past
decade to predict the stock market. The literature shows that Neural Networks are the most
widely used technique in this area (Egeli, Ozturan, and Badur 2003). However, as pointed out by
Byvatov et al. 2003, a critical issue concerning neural networks is the over-fitting problem. It can
be attributed to the fact that a neural network captures not only useful information contained in
the given data, but also unwanted noise. This usually leads to a poor level of generalization.

In 2003, Kim has therefore proposed an algorithm to predict the stock market direction by using
technical analysis indicators as input to SVMs. It is a relatively new learning algorithm that has
the desirable characteristics of the control of the decision function, the use of the kernel method,
and the sparsely of the solution. The experimental result was able to predict test data outputs
with up to 57% accuracy, significantly above the 50% threshold. Shah n.d. conducted a survey
study on stock prediction using various machine learning models, and found that the best results
were achieved with SVM. His prediction rate of 60% agrees with Kim’s conclusion.

Support Vector Machines (SVMs) as originally proposed by Vapnik 1998 within the area of
statistical learning theory, have demonstrated to work successfully on various classification and
forecasting problems. Many traditional neural network models had implemented the empirical
risk minimization principle; SVM implements the structural risk minimization principle. The
former seeks to minimize the mis-classification error or deviation from correct solution of the
training data but the latter searches to minimize an upper bound of generalization error.
C.~Cortes et al. (1995) explained SVM as a maximum fringe hyper plane that lies in some space
and classifies the data separated by non-linear boundaries which can be constructed by locating a
set of hyper planes that separate two or more classes of data points. After construction of the
hyper planes, the SVM discovers the boundaries between the input classes and the input
elements defining a set of given training samples labeled either positive or negative, a maximum
margin hyper plane splits the positive or negative training sample, as a result the distance
between the margin and the hyper plane is maximized. If there exist no hyper planes that can
split the positive or negative samples, a SVM selects a hyper plane that splits the sample as
austerely as possible, while still maximizing the distance to the nearest austerely split examples
(Lahmiri 2011).

Cao and Tay (2001) and Das (2012) have compared SVM with Back Propagation Neural
Networks (BPN) in context of Chinese & Indian markets respectively. The experimental result
shows that SVM outperformed BPN most often as SVM provides a smaller NMSE (Normalized
Mean Square Error) and MAE (Mean Absolute Error) and larger DS (Directional Symmetry)
than of BPN in most of the cases. These results may be attributable to the fact that the SVM
implements the structural risk minimization principle and this leads to better generalization than
Neural Networks, which implement the empirical risk minimization principle. Another key
property of SVM is that training SVM is equivalent to solving a linearly constrained quadratic
programming problem so that the solution of SVM is always unique and globally optimal, unlike
neural networks training, which requires nonlinear optimization with the danger of getting stuck
at local minima.

L.Yu et al. (2009) and Choudhry and Garg (2008) proposed an evolving least squares SVM
learning paradigm with mixed kernel to explore stock market movement direction, in which
genetic algorithm (GA) is used for feature selection and determination of system parameters.
Similarly, C. J. Huang, Yang, and Chuang (2008) integrated wrapper approach and SVM to
predict the trend of the Korea and Taiwan stock markets. In 2014, Yu, Chen, and Zhang had
proposed SVM and PCA to construct a stock selection model which can do the nonlinear
classification of stocks. First, a stock selection model (PCA-SVM) was established and then
PCA-SVM was applied on test set to forecast the high-return stocks in the next year. All these
studies, by using different approaches, concluded that SVM classification method is accurate and
highly efficient when dealing with complex and highly dimensional data.

In 2006, Ince and Trafalis developed a model for stock selection by using minimax probability
machine (MPM) and support vector machines (SVMs) and the Comparison of the MPM and
SVM technique revealed that there was no significant difference between the testing and training
performance of the algorithms. The best performance of MPM and SVM was 69.62 and 68.35%,
respectively making SVM algorithm slightly better than MPM classification method. Similarly,
to show that the SVM is able to give more accurate predictions than regular ML estimation,
Fernando, Afonso-rodr, and Giner 2003 used SVM to estimate the parameters of a GARCH
model for predicting the conditional volatility of stock market returns. Again the results
concluded that the SVR was always able to predict better than the sample mean, which it was not
possible for the ML technique. These results highlighted the expected relative superiority of
SVM estimation against the ML procedure, developed under more restrictive distributional
assumptions, because the former can be interpreted as a robust estimation procedure. The
research mainly focused on SVM as a linear machine to replace the ML estimation process and
did not obtain a nonlinear estimation using kernels, It lacks the development of nonlinear
machines in which the estimate of future volatilities will depend nonlinearly on the past
volatilities and observations.

In 2013, Lin, Guo, and Hu proposed an SVM-based approach for Taiwan stock market trend
prediction, which hybridizes a correlation-based SVM filter method and a quasi-linear SVM.
Experimental results concluded that the SVM-based stock market trend prediction method
produced better generalization performance over the conventional methods in terms of the hit
ratio. Moreover, the experimental results also showed that the proposed SVM-based stock
market trend prediction system can find out a good subset and evaluate stock indicators which
provide useful information for investors. As there are many factors affecting the stock price the
research needs to test more influencing factors with this prediction system. It requires an
investigation to develop a structured method of selecting an optimal value of the parameters in
the proposed prediction system for the best prediction performance. On the similar grounds,
Madge 2015 aim to calculate price volatility and momentum for individual stocks and for the
overall sector using SVM model and predict its future price. It has used SVM at time t to predict
whether a given stock’s price was higher or lower on day t +m. The study found little predictive
ability in the short-run but definite predictive ability in the long-run. Therefore it concludes that
model’s difficulty in predicting the next day’s stock price supports EMH. The Study only
focused on closing price data. By observing intra-day trends, more robust models that capitalize
on sudden changes in momentum during intra-day trading can be created.

Chen & Chen (2007), Huang, Nakamori, & Wang (2005), and Rao & Hong (2010) trained SVM
and traditional methods like Hidden Markov Models (HMM) on past datasets and used it to predict
future stock movements. The majority of the work review used the MATLAB software for
implementing their models. These studies broadly presents two approaches in helping investors
make better decisions. First, conventional methods such as using the Efficient Market Hypothesis
and technical indicators, for forecasting stock prices and movements were discussed and shown
that these methods are inadequate. Afterwards Support Vector Machines was trained on past datasets
and used to predict future stock movements. The experimental results can help to make decision
whether to choose the target stocks or not. The experimental results show that there is a higher
accuracy rate of prediction using SVM to predict only, because the financial indices as the
experimental parameters are directly selected from the financial statements which are released in
different periods by the stock companies to public, and the experimental results are much more
reliable, nonvolatile and valid. As the ranges of the experimental samples are not large enough,
the overall status of the stock market cannot be described with much accuracy rate. But instead
of training SVM models on predicting stock movements for the next day, if it can be trained to
predict general trends for the next few days, the abnormal volume detection feature can be added
that would help the SVM better predict future stock trends.

Although most of the studies such as Dai, Zhang, & Wang (2011) and Mager, Paasche, & Sick
(2008) adopted Support Vector Machines method to predict the prices of Stock index futures to
show that predicted results were consistent with the actual one proving the feasibility of the
method none of the previous studies have compared the performance of the economic
information and technical indicators in terms of prediction accuracy. Indeed, it is not known
what type of information leads to better forecasts. Unlike the literature, only predictive variables
that show strong evidence of causal relationship with return series are considered. Moreover,
most of the reported papers refer to forecasting of financial markets in well developed countries.
Very few recent articles shows that return predictability is also possible when dealing with
emerging financial markets (of under development countries). Ferson and Harvey (1999) have
studied eighteen international stock markets of both kinds demonstrating evidence of return
predictability. It is demonstrated that the degree of predictability in emerging stock markets is
higher than that found in well-developed ones. In addition, it is shown that local information
plays a much more important role in predicting returns in emerging stock markets than in well-
developed ones.
These existing stock market trend prediction systems have been shown to be effective and
producing good forecasting performance on stock market trend prediction. But many researchers
have pointed some limitations of SVM on which work must be carried out. First, in the
prediction model, SVM often bears great trouble in over-fitting problem when flexible nonlinear
kernels are used in high-dimensional and high-noise datasets. It may affect the generalization
performance of SVM (Cawley and Talbot 2010). Second, in the feature selection, though these
feature selection methods are able to effectively reduce the dimension and noise of financial
dataset they have their limitations, that cannot exactly select a feature subset contain features that
are highly correlated with the output, yet uncorrelated with each other. (Lee 2009 & Ni, Ni, and
Gao 2011). Third, the existing stock market trend prediction systems usually focus on feature
selection and prediction model to ignore evaluating features. Though, some studies have been
taken care about feature evaluation, but these methods are affected by subjective factor and
multicollinearity (Piotroski 2000) and fourth, the Learning phase of SVM scale with the number
of training data points (Strack et al. 2013). Therefore, even though various techniques have been
used in the literature survey, still there is a need of best techniques to solve these research issues.
BIBLIOGRAPHY:

1. Byvatov, Evgeny, Uli Fechner, Jens Sadowski, and Gisbert Schneider. 2003.
“Comparison of Support Vector Machine and Artificial Neural Network Systems for
Drug/Nondrug Classification.” Journal of Chemical Information and Computer Sciences
43(6): 1882–89.
2. C.~Cortes, C.~Cortes, V.~Vapnik, and V.~Vapnik. 1995. “Support Vector Networks.”
Machine Learning 20(3): 273~-~297. http://link.springer.com/10.1007/BF00994018.
3. Cao, Lijuan, and Francis E. H. Tay. 2001. “Financial Forecasting Using Support Vector
Machines.” Neurocomputing 1(2): 1–36.
4. Cawley, Gavin C., and Nicola L. C. Talbot. 2010. “On Over-Fitting in Model Selection
and Subsequent Selection Bias in Performance Evaluation.” Journal of Machine
Learning Research 11: 2079−2107.
http://jmlr.csail.mit.edu/papers/v11/cawley10a.html%5Cnhttp://www.jmlr.org/papers/vol
ume11/cawley10a/cawley10a.pdf.
5. Chen, Rung-ching, and Rung-ching Chen. 2007. “Using SVM with Financial Statement
Analysis for Prediction of Stocks.” 7(4).
6. Choudhry, Rohit, and Kumkum Garg. 2008. “A Hybrid Machine Learning System for
Stock Market Forecasting.” World Academy of Science, Engineering and Technology
2(15): 315–18.
7. Dai, Hezhong, Yichi Zhang, and Dapeng Wang. 2011. “Price Prediction of Stock Index
Futures Based on SVM.” 2011 Fourth International Conference on Business Intelligence
and Financial Engineering (1): 54–57. http://ieeexplore.ieee.org/document/6121087/.
8. Das, Shom Prasad. 2012. “Support Vector Machines for Prediction of Futures Prices in
Indian Stock Market.” 41(3): 22–26.
9. Egeli, B., M. Ozturan, and B. Badur. 2003. “Stock Market Prediction Using Artificial
Neural Networks.” Proceedings of the 3rd Hawaii International Conference on Business:
1–8.
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.134.1114&rep=rep1&am
p;type=pdf.
10. Fernando, P, Julio A Afonso-rodr, and Javier Giner. 2003. “Estimating GARCH Models
Using Support Vector Machines *.” 3: 1–10.
11. Ferson, Wayne E., and Campbell R. Harvey. 1999. “Economic, Financial and
Fundamental Global Risk In and Out of the EMU.” SSRN Electronic Journal.
http://www.ssrn.com/abstract=152889.
12. Huang, Chenn Jung, Dian Xiu Yang, and Yi Ta Chuang. 2008. “Application of Wrapper
Approach and Composite Classifier to the Stock Trend Prediction.” Expert Systems with
Applications 34(4): 2870–78.
13. Huang, Wei, Yoshiteru Nakamori, and Shou-yang Wang. 2005. “Forecasting Stock
Market Movement Direction with Support Vector Machine.” 32: 2513–22.
14. Ince, Huseyin, and Theodore B Trafalis. 2006. “Kernel Methods for Short-Term Portfolio
Management.” 30: 535–42.
15. Kim, Kyoung Jae. 2003. “Financial Time Series Forecasting Using Support Vector
Machines.” Neurocomputing 55(1–2): 307–19.
16. Lahmiri, Salim. 2011. “A Comparison of PNN and SVM for Stock Market Trend
Prediction Using Economic and Technical Information.” 29(3): 24–30.
17. Lee, Ming Chi. 2009. “Using Support Vector Machine with a Hybrid Feature Selection
Method to the Stock Trend Prediction.” Expert Systems with Applications 36(8): 10896–
904. http://dx.doi.org/10.1016/j.eswa.2009.02.038.
18. Lin, Yuling, Haixiang Guo, and Jinglu Hu. 2013. “An SVM-Based Approach for Stock
Market Trend Prediction.”
19. Madge, Saahil. 2015. “Predicting Stock Price Direction Using Support Vector
Machines.”
20. Mager, Johannes, Ulrich Paasche, and Bernhard Sick. 2008. “Forecasting Financial Time
Series with Support Vector Machines Based on Dynamic Kernels.” 2008 IEEE
Conference on Soft Computing in Industrial Applications: 252–57.
http://ieeexplore.ieee.org/document/5045969/.
21. Ni, Li Ping, Zhi Wei Ni, and Ya Zhuo Gao. 2011. “Stock Trend Prediction Based on
Fractal Feature Selection and Support Vector Machine.” Expert Systems with
Applications 38(5): 5569–76. http://dx.doi.org/10.1016/j.eswa.2010.10.079.
22. Nison, Steve. JAPANESE CANDLESTICK CHARTING TECHNIQUES.
23. Piotroski, Joseph D. 2000. “Value Investing: The Use of Historical Financial Statement
Information to Separate Winners from Losers.” Journal of Accounting Research 38: 1.
http://www.jstor.org/stable/2672906?origin=crossref.
24. Rao, Satish, and Jerry Hong. 2010. “Analysis of Hidden Markov Models and Support
Vector Machines in Financial Applications.”
25. Shah, Vatsal H. “Machine Learning Techniques for Stock Prediction.” : 1–19.
26. Strack, Robert, Vojislav Kecman, Beata Strack, and Qi Li. 2013. “Sphere Support Vector
Machines for Large Classification Tasks.” Neurocomputing 101: 59–67.
27. Vapnik, Vladimir N. 1998. “Statistical Learning Theory.” Adaptive and learning Systems
for Signal Processing, Communications and Control 2: 1–740.
http://eu.wiley.com/WileyCDA/WileyTitle/productCd-0471030031.html.
28. Yang, Haiqin, Irwin King, Laiwan Chan, and Kaizhu Huang. 2004. “Financial Time
Series Prediction Using Non-Fixed and Asymmetrical Margin Setting with Momentum in
Support Vector Regression.” Analysis 152: 334–350.
http://books.google.com/books?hl=en&lr=&id=rsJ6VXKvsMQC&oi=fnd
&pg=PA334&dq=Financial+Time+Series+Prediction+Using+Non-
fixed+and+Asymmetrical+Margin+Setting+with+Momentum+in+Support+Vector+Regr
ession&ots=GiqQ0n8Pt8&sig=nPJO9BFE9.
29. Yu, Huanhuan, Rongda Chen, and Guoping Zhang. 2014. “A SVM Stock Selection
Model within PCA.” Procedia - Procedia Computer Science 31: 406–12.
http://dx.doi.org/10.1016/j.procs.2014.05.284.
30. Yu, Lean, Huanhuan Chen, Shouyang Wang, and Kin Keung Lai. 2009. “For Stock
Market Trend Mining.” 13(1): 87–102.