[논문]익스트림 그라디언트 부스팅을 이용한 지수/주가 이동 방향 예측

김형도

doi:10.5392/jkca.2018.18.09.623

[국내논문] 익스트림 그라디언트 부스팅을 이용한 지수/주가 이동 방향 예측
Prediction of the Movement Directions of Index and Stock Prices Using Extreme Gradient Boosting 원문보기

한국콘텐츠학회논문지 = The Journal of the Korea Contents Association, v.18 no.9, 2018년, pp.623 - 632

초록
AI-Helper

주가 이동 방향의 정확한 예측이 주식 매매에 관한 전략적 의사결정에 중요한 역할을 할 수 있기 때문에 투자자와 연구자 모두의 관심이 높다. 주가 이동 방향에 관한 기존 연구들을 종합해보면, 주식 시장에 따라서 그리고 예측 기간에 따라서 다양한 변수가 고려되고 있음을 알 수 있다. 이 연구에서는 한국 주식 시장을 대표하는 지수와 주식들을 대상으로 이동 방향 예측 기간에 따라서 어떤 데이터마이닝 기법의 성능이 우수한 것인지를 분석하고자 하였다. 특히, 최근 공개경쟁에서 활발히 사용되며 그 우수성이 입증되고 있는 익스트림 그라디언트 부스팅 기법을 주가 이동 방향 예측 문제에 적용하고자 하였으며, SVM, 랜덤 포리스트, 인공 신경망과 같이 기존 연구에서 우수한 것으로 보고된 데이터마이닝 기법들과 비교하여 분석하였다. 12년간 데이터를 사용하여 1일 후에서 5일 후까지의 이동 방향을 예측하는 실험을 통해서, 예측 기간과 종목에 따라서 선택된 변수들에 차이가 있으며, 1-4일 후 예측에서는 익스트림 그라디언트 부스팅이 다른 기법들과 부분적으로 동등함을 가지면서도 가장 우수함을 확인하였다.

Abstract ▼ AI-Helper

Both investors and researchers are attentive to the prediction of stock price movement directions since the accurate prediction plays an important role in strategic decision making on stock trading. According to previous studies, taken together, one can see that different factors are considered depending on stock markets and prediction periods. This paper aims to analyze what data mining techniques show better performance with some representative index and stock price datasets in the Korea stock market. In particular, extreme gradient boosting technique, proving itself to be the fore-runner through recent open competitions, is applied to the prediction problem. Its performance has been analyzed in comparison with other data mining techniques reported good in the prediction of stock price movement directions such as random forests, support vector machines, and artificial neural networks. Through experiments with the index/price datasets of 12 years, it is identified that the gradient boosting technique is the best in predicting the movement directions after 1 to 4 days with a few partial equivalence to the other techniques.

주제어

표/그림 (10)

표 Table 1. Definitions of Input Variables
표 Table 2. The Numbers of Increasing/Decreasing Cases of Index/Stocks by Prediction Periods
표 Table 3. Variable Selection by Prediction Periods for KOSPI
표 Table 4. Experimental Results for KOSPI
표 Table 5. p-Values of Right-tailed t-Tests for KOSPI
표 Table 6. Mean AUC Values for Prediction Period of 1 day
표 Table 7. Mean AUC Values for Prediction Period of 2 days
표 Table 8. Mean AUC Values for Prediction Period of 3 days
표 Table 9. Mean AUC Values for Prediction Period of 4 days
표 Table 10. Mean AUC Values for Prediction Period of 5 days

AI 본문요약
AI-Helper

* AI 자동 식별 결과로 적합하지 않은 문장이 있을 수 있으니, 이용에 유의하시기 바랍니다.

문제 정의

This paper aims to analyze what data mining techniques show better performance with some representative index and stock price data of the Korea stock market. Recently, the extreme gradient boosting (XGBoost)[8] has been demonstrating the best performance through many recent open competitions.
The results of the previous studies on the movement directions of index and stock prices depend on the dataset, the prediction period, and the input variables used. This paper aims to compare the extreme gradient boosting technique with other best ones reported in the previous studies such as SVM, ANN, and Random Forest in the context of the Korean stock market. Some representative index and stock price data of Korea stock market for 12 years are used for the experiments.

제안 방법

To the best of our knowledge, there is no research about the application of the technique to the prediction issues of stock prices and their movement directions. This paper intends to compare the prediction techniques for the movement directions of stock prices such as SVM, RF, and ANN with the XGBoost using the Korean stock market data. Prediction periods of 1 to 5 days are experimented with the 12-years data for analyzing its effects on the quality of prediction results.
(2011) estimated the National 100 index of Istanbul stock market in Turkey by applying ANN and SVM for the daily movement directions from 1997 to 2007[4]. They used 10 technical indicators of 10-days Simple Moving Average(SMA), 10-days Weighted Moving Average(WMA), Momentum, Stochastic %K, Stochastic %D, RSI, Moving Average Convergence Divergence(MACD), WPR, ADO, and CCI. Experimental results showed that the average performance of ANN model was found significantly better than that of SVM model.
Manojlovic and Stajduhar(2015) conducted a comparative experiment on predicting the movement directions after 5 and 10 days for the CROBEX index of the Croatian Zagreb stock market and several stocks of various sectors[6]. They selected 12 technical indicators of 5/10-days WMA, Stochastic %K, Stochastic %D, MACD, CCI, 5/10-days Disparity, Percentage Price Oscillator(PPO), 10-days ROC, 10-dyas Momentum, RSI, 5-days Standard Deviation(SD) for testing SVM, ANN, and RF models. RF was reported as the best in accuracy.
In this study, XGBoost will be compared with the existing methods in predicting movement directions of stock prices. As comparison targets, SVM, ANN, and RF, reported to be superior in the previous studies, were adopted.
As comparison targets, SVM, ANN, and RF, reported to be superior in the previous studies, were adopted. As potential input (predictor) variables, a number of technical indicators suitable for daily forecasting were selected: 5/10/20-days SMA, 5/10/20-days WMA, Stochastic K%, Stochastic D%, MACD, MACD Signal, WPR, Accumulation / Distribution Line(ADL), CCI, 5/10-days Disparity, PPO, 5/10/20-days Momentum, 5/10/20-days ROC, RSI, and 5/10/20-days Psychology Line(PSY). That is, 30 input variables are used in total.
Most previous studies have been predicting the movement directions of stock prices after one day later, but it is not uncommon to predict those after 5 days(1 week), 2 weeks, 1 month, 3 months, and even 1 year. In this study, all the experiments on prediction techniques are designed to examine the effect of prediction periods of 1 to 5 days. The movement direction is represented as 1 if the stock price after n days is higher than the current stock price, otherwise 0.
A T-tests is applied to the 10 AUC values obtained from the 10-fold cross validation for each combination of techniques, stock, and prediction periods. The composite AUC and accuracy values, obtained from the total predictions combining all the partial predictions in the 10-fold cross validation, are also referred to as a comparison measure.
For the selection of variables, an XGBoost model was constructed by using 80% of the total data as training data and the remaining 20% as validation data to predict the movement direction of stock prices in each prediction period. All the XGBoost parameters are set with default values and the number of decision trees is optimized by stopping additional decision tree generation if the AUC of the validation data does not improve over 100 times.
Some representative index and stock price data of Korea stock market for 12 years are used for the experiments. It is also tested in the experiments about the effect of prediction periods of 1 to 5 days, compared with most other studies of single prediction period, and the effect of variable selection by prediction periods. The difference among the best sets of input variables by prediction periods identifies the importance of variable selection in predicting the movement directions of index and stock prices.
More extensive research will be needed for making the comparison of the techniques more general. This study used 30 potential input variables, commonly used in the literature, for variable selection. In this respect, the set need to be extended for better quality of predictions and strategic decisions suitable for the Korea stock market.

대상 데이터

All the experiments in this paper were conducted using the R environment on a 64-bit Windows PC with an i5-4460 processor and 32G RAM. RF was implemented using the randomForest package, XGBoost the xgboost package, SVM the kernlab package, and ANN the keras/tensorflow package.
In spite of the successful application of the gradient boosting technique to the prediction problem of the movement directions of index and stock prices, there are some limitations to be noted. This study tested only 4 index and stocks in Korea stock market although they represents the stock market. More extensive research will be needed for making the comparison of the techniques more general.

이론/모형

In order to predict the 1-day later movement directions of the daily Korea composite stock price index(KOSPI), Kim(2003) used 12 technical indicators of Stochastic %K, Stochastic %D, Stochastic Slow %D, Momentum, Rate of Change(ROC), Williams'%R(WPR), Accumulation/Distribution Oscillator(ADO), Disparity 5, Disparity 10, Price Oscillator(OSCP), Commodity Channel Index(CCI), and Relative Strength Index(RSI)[13].
These include the SPY return for the current day and three previous days, the relative difference in percentage of the SPY return, exponential moving averages of the SPY return, Treasury bill(T-bill) rates, certificate of deposit rates, financial and economic indicators, the term and default spreads, exchange rates between the USD and four other currencies, the return of seven world major indices(other than the S&P 500), SPY trading volume, and the return of eight large capitalization companies within the S&P 500. Applying three dimensional reduction techniques such as Principal Component Analysis(PCA) to these potential variables, they used the most important and influential factors as inputs to ANN. Experimental results showed that combining ANN with PCA gives slightly higher classification accuracy than the other two combinations.
There are various measures for evaluating the performance of classifiers such as Accuracy, TPR, TNR, G-mean, Precision, Recall, F-measure and so on. Based on the report that AUC can appropriately and accurately measure the performance of binary classification[16], AUC, which is the area under the ROC(Receiver Operating Characteristic) curve, is adopted as the performance evaluation measure. As the AUC value of a classification system approaches 1, the better it can be evaluated.

성능/효과

They used 10 technical indicators of 10-days Simple Moving Average(SMA), 10-days Weighted Moving Average(WMA), Momentum, Stochastic %K, Stochastic %D, RSI, Moving Average Convergence Divergence(MACD), WPR, ADO, and CCI. Experimental results showed that the average performance of ANN model was found significantly better than that of SVM model.
(2011). The experimental results suggested that RF outperforms other three prediction models on overall performance with ten technical parameters are represented as continuous values.

참고문헌 (18)

M. Ballings, D. V. Poel, N. Hespeels, and R. Gryp, "Evaluating Multiple Classifiers for Stock Price Direction Prediction," Expert Systems with Applications, Vol.42, pp.7046-7056, 2015.

상세보기
B. G. Malkiel and E. F. Fama, "Efficient Capital Markets: A Review of Theory and Empirical Work," The Journal of Finance, Vol.25, No.2, pp.383-417, 1970.
S. B. Imandoust and M. Bolandraftar, "Forecasting the Direction of Stock Market Index Movement Using Three Data Mining Techniques: the Case of Tehran Stock Exchange," Int'l Journal of Engineering Research and Applications, Vol.4, No.6, pp.106-117, June. 2014.
Y. Kara, M. A. Boyacioglu, and O. K. Baykan, "Predicting Direction of Stock Price Index Movement Using Artificial Neural Networks and Support Vector Machines," Expert Systems with Applications, Vol.38, pp.5311-5319, May 2011.

상세보기
W. Huang, Y. Nakamori, and S. Y. Wang, "Forecasting Stock Market Movement Direction with Support Vector Machine," Computers and Operations Research, Vol.32, pp.2513-2522, 2005.

상세보기
T. Manojlovic and I. Stajduhar, "Predicting Stock Market Trends Using Random Forests: A Sample of the Zagreb Stock Exchange," Proc. of MIPRO 2015, Opatija, Croatia, pp.1189-1193, 2015.
X. Zhong and D. Enke, "Forecasting Daily Stock Market Return Using Dimensionality Reduction," Expert Systems with Applications, Vol.67, pp.126-139, 2017.

상세보기
T. Chen and C. Guestrin, "XGBoost: A Scalable Tree Boosting System," Proc. of the KDD '16 of the 22nd ACM SIGKDD Int'l Conf. on Knowledge Discovery and Data Mining, San Francisco, USA, pp.785-794, August 13-17, 2016.
B. E. Boser, I. M. Guyon, and V. N. Vapnik, "A Taining Agorithm for Optimal Margin Classifiers," Proc. of the Fifth Annual Workshop on Computational Learning Theory," Pittsburgh, USA, pp.144-152, July 27-29, 1992.
J. Schmidhuber, "Deep Learning in Neural Networks: An Overview," Neural Networks, Vol.61, pp.85-117, 2015.

상세보기
L. Breiman, "Random forests," Machine Learning, Vol.45, No.1, pp.5-32, 2001.
J. H. Friedman, "Greedy Function Approximation: A Gradient Boosting Machine," Vol.29, No.5 pp.1189-1232, 2001.

상세보기
K. Kim, "Financial Time Series Forecasting Using Support Vector Machines, Neurocomputing," Vol.55, pp.307-319, 2003.

상세보기
K. Manish and M. Thenmozhi, "Forecasting Stock Index Movement: A Comparison of Support Vector Machines and Random Forest," Proceedings of Ninth Indian Institute of Capital Markets Conference, Mumbai, India, http://ssrn.com/abstract876544, 2005.
J. Patel, S. Shah, P. Thakkar, and K. Kotecha, "Predicting Stock and Stock Price Index Movement Using Trend Deterministic Data Preparation and Machine Learning Techniques," Expert Systems with Applications, Vol.42, No.1, pp.259-268, 2015.

상세보기
F. Provost, T. Fawcett, and R. Kohavi, "The Case against Accuracy Estimation for Comparing Induction Algorithms," Proc. of the Fifteenth Int'l Conf. on Machine Learning, Madison, USA, pp.445-453, July 24-27, 1998.
http://finance.naver.com/
B. Shahrirai, "Taking the Human Out of the Loop: A Review of Bayesian Optimization," Proceedings of the IEEE, Vol.104, No.1, pp.148-175, 2016.

상세보기

저자의 다른 논문 :

표제어: PCR

동의어: Packet Collision Rate

용어 설명 출처 목록 (6)

용어 설명: PCR은 세균 특이성이 있는 primer를 이용하여 적은 수의 세균이 있을지라도 쉽게 검출할 수 있는 유용한 방법이며, 이를 이용하여 구강 내 치면세균막이나 타액에서 직접 세균을 검출할 수 있게 되었다[8].

내보내기 구분	파일저장 인쇄 메일전송
구성항목	기본정보 상세정보 관리번호, 논문명, 저널/프로시딩명, 저자 , 발행년, 권, 호, 시작페이지, 끝페이지, 발행기관 관리번호, 논문명, 대등논문명, 저자 , 저널/프로시딩명, 발행기관, 발행년, 발행언어, 권, 호, 시작페이지, 끝페이지, ISBN, ISSN, 주제분야, 키워드, 초록(한글), 초록(영문), 저자(소속기관)
저장형식	Text(ASCII format) Excel format RefWorks Direct Export RIS format (for Reference Manager, ProCite, EndNote), Scholar's Aids, Mendeley
메일정보	받는사람 (필수) @ 보내는사람 (선택) @ 제목 내용 KISTI 검색결과 이메일 서비스
안내	총 건의 자료가 검색되었습니다. 다운받으실 자료의 인덱스를 입력하세요. (1-10,000) 검색결과의 순서대로 최대 10,000건 까지 다운로드가 가능합니다. 데이타가 많을 경우 속도가 느려질 수 있습니다.(최대 2~3분 소요) 다운로드 파일은 UTF-8 형태로 저장됩니다. 파일의 내용이 제대로 보이지 않을실 때는 웹브라우저 상단의 보기 -> 인코딩 -> 자동선택 여부를 확인하십시오. ~ Text(ASCII format) Excel format

연합인증