GAN 모형을 이용한 주가 시계열의 확장
Expansion of Stock Price Time Series using GAN
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
최근 시계열 데이터를 통해서 예측하는 기술과 방법론은 급속도로 발전하고 있으며, 이는 국가적, 사회적, 경제적 등 인류의 다양한 의사결정 문제를 효과적으로 해결해 주고 있다. 이처럼 과학적인 분석 기법을 적용하기 위해서는 데이터의 질과 양이 중요한데 데이터의 양이 부족한 경우에는 기계학습과 빅데이터 분석과 같은 과학적 방식에 적용하기가 매우 어렵다. 최근 신규 상장된 기업들의주가 데이터는 그 양이 제한적이고, 이러한 데이터의 부족은 기계학습 및 인공신경망 모델의 학습에있어 과소적합과 과적합 같은 문제를 일으킬 수 있다. 이 논문에서는 신규 상장 기업의 주가 데이터를 확장하는 방법으로 Time-series Generative Adversarial Network(TimeGAN)을 적용하였다. TimeGAN은 시계열 데이터의 내재된 시간적 동적 특성을 모델링하여 실제와 유사한 합성 데이터를생성할 수 있는 효과적인 방법을 제공한다. 우리는 이 모델을 사용하여 2019년부터 2023년까지 한국주식 시장에 신규 상장 기업의 부족한 주가 데이터를 확장하고, 확장된 데이터에 대한 통계적 실효성을 PCA, t-SEN 기법을 통해서 분석한 결과, TimeGAN을 통해 생성된 데이터는 원래 데이터의 통계적 특성을 잘 보존하는 것으로 나타났다. 본 논문은 시계열 데이터가 부족한 상황에서의 데이터 확장 기법에 대한 유용한 접근 방법을 제시함으로써, 금융 시장 분석 및 예측에 있어 새로운 가능성을열어주길 기대한다.
Abstract
Recently, technologies and methodologies for predicting through time series data are rapidly developing, and they are effectively solving various decision-making problems of humanity such as national, social, and economic. In order to apply such a scientific analysis technique, the quality and quantity of data are important, but when the amount of data is insufficient, it is very difficult to apply it to scientific methods such as machine learning and big data analysis. The amount of stock price data of newly listed companies is limited, and the lack of such data can cause problems such as underfitting and overfitting in learning machine learning and artificial neural network models. In this paper, Time-series Generative Adversarial Network (TimeGAN) was applied as a method to expand the stock price data of newly listed companies. TimeGAN provides an effective way to generate synthetic data similar to the real world by modeling the inherent temporal and dynamic characteristics of time series data. We used this model to expand the stock price data of newly listed companies in the Korean stock market from 2019 to 2023, and analyze the statistical effectiveness of the expanded data through PCA and t-SEN techniques, and found that the data generated through TimeGAN well preserves the statistical characteristics of the original data. This paper is expected to open new possibilities in financial market analysis and prediction by presenting a useful approach to data expansion techniques in the context of lack of time series data.
참고문헌 (17)
- J. Patel, "Predicting stock and stock price index movement using Trend Deterministic Data Preparation and machine learning techniques", Expert Systems with Applications, 42(1), pp.259~268 (2015)
- Y. Ma, "Research on public opinion effecting on stock price during crises based on model checking", Expert Systems with Applications, 249 (2024)
- Z. Geng, "Coercing machine learning to output physically accurate results", Journal of Computational Physics, 406 (2020)
- C. Sujung, "Herding among local individual investors : Evidence from online and offline trading", Economics Letters, 144, pp.4~6 (2016)
- E. Halilaj, "Machine learning in human movement biomechanics: Best practices, common pitfalls, and new opportunities", J. Biomech, 81, pp.1~11 (2018)
- A. Esteva, "Dermatologistlevel classification of skin cancer with deep neural networks", Nature, 542(7639), pp.115~118 (2017)
- K. M. Rashid, "Times-series data augmentation and deep learning for construction equipment activity recognition", Advanced Engineering Informatics, 42 (2019)
- K. Kamycki, "Data augmentation with suboptimal warping for timeseries classification", 20(1), pp.959~994 (2020)
- K. Mikalsen, "Time series cluster kernel for learning similarities between multivariate time series with missing data", Pattern Recognition, 76, pp.569~581 (2018)
- G. H. Rosa, "A survey on text generation using generative adversarial networks", Pattern Recognition, 119 (2021)
- K. Wang, "Automatic generation of sentimental texts via mixture adversarial networks", Artificial Intelligence, 275, pp.540~558 (2019)
- J. H. Yoon, "A Study on GAN Algorithm for Restoration of Cultural Property", Journal of The Korea Society of Computer and Information, 26(1), pp.77~84 (2021)
- U. Sivarajah, "Critical analysis of Big Data challenges and analytical methods", Journal of Business Research, 70, pp.263~286 (2017)
- M. K. Back, "Improving Fidelity of Synthesized Voices Generated by Using GANs", KIPS Trans. Softw. and Data Eng, 10(1), pp.9~18 (2020)
- A. Odena, "Deconvolution and checkerboard artifacts", Distill, 1(10), pp.1~3 (2016)
- J. S. Yoon, "Time-series generative adversarial networks", 33rd Conference on Neural Information Processing System (2019)
- I. Goodfellow, "Generative adversarial nets", Advances in neural information processing systems, pp.2672~2680 (2014)