Stock market volatility simulation with the LSTM neural network

Abstract

Introduction. Stock market volatility simulation and forecast are relevant issues which could contribute into lower risks and higher revenues of the market transactions. These days, AI-based methods, including deep neural networks, are quite promising for volatility simulation.

Purpose. The paper verifies a hypothesis concerning a higher accuracy of LSTM neural network compared to the classical autoregressive models (e.g. ARIMA) and long memory models (e.g. ARFIMA).

Materials and Methods. To check the hypothesis, the authors conducted simulation experiments with S&P 500 index data generally illustrating the dynamics of the US stock market.

Results. The LSTM neural network gave significantly more accurate forecasts compared to the ARIMA- and ARFIMA-based forecasts for learning and test samples; ARFIMA model was more accurate than ARIMA, which supports previous data.

Conclusions. The results of the work prove that the LSTM neural network is a promising method to forecast stock market volatility and could be further examined in this area. Machine learning methods, including the neural networks, could be used to define the future dynamics in the revenues of financial asserts and optimize current algorithms of portfolio imbalances, approximation and simulation of risk metrics, approximation of probabilistic characteristics for financial instruments.

Keywords: stock market, volatility simulation, neural networks, LSTM, ARFIMA

For citation

Patlasov D. A., Garafutdinov R. V. Stock market volatility simulation with the LSTM neural network. Perm University Herald. Economy, 2024, vol. 19, no. 1, pp. 41–51. DOI 10.17072/1994-9960-2024-1-41-51. EDN BHKTOC.

REFERENCES

1. Berzon N. I., Bobrovsky D. I., Vilkul D. E., Dubinsky D. V., Mezentsev V. V. Value-at-Risk and Expected Shortfall approaches for option premiums and the probability of default estimation based on ARMA models. Ekonomika i matematicheskie metody = Economics and Mathematical Methods, 2021, vol. 57, no. 3, pp. 126–139. (In Russ.). DOI 10.31857/S042473880016417-7. EDN ADRRRM
2. Bucci A. Realized volatility forecasting with neural networks. Journal of Financial Econometrics, 2020, vol. 18, iss. 3, pp. 502–531. DOI 10.1093/jjfinec/nbaa008
3. Hu Y., Ni J., Wen L. A hybrid deep learning approach by integrating LSTM-ANN networks with GARCH model for copper price volatility prediction. Physica A: Statistical Mechanics and its Applications, 2020, vol. 557, 124907. DOI 10.1016/j.physa.2020.124907
4. Jiao X., Song Y., Kong Y., Tang X. Volatility forecasting for crude oil based on text information and deep learning PSO‐LSTM model. Journal of Forecasting, 2022, vol. 41, iss. 5, pp. 933–944. DOI 10.1002/for.2839
5. Jung G., Choi S. Y. Forecasting foreign exchange volatility using deep learning autoencoder-LSTM techniques. Complexity, 2021, vol. 2021, 6647534, 16 p. DOI 10.1155/2021/6647534
6. Kakade K., Mishra A. K., Ghate K., Gupta Sh. Forecasting commodity market returns volatility: A hybrid ensemble learning GARCH‐LSTM based approach. Intelligent Systems in Accounting, Finance and Management, 2022, vol. 29, iss. 2, pp. 103–117. DOI 10.1002/isaf.1515
7. Kakade K., Jain I., Mishra A. K. Value-at-Risk forecasting: A hybrid ensemble learning GARCH-LSTM based approach. Resources Policy, 2022, vol. 78, 102903. DOI 10.1016/j.resourpol.2022.102903
8. Lei B., Liu Z., Song Y. On stock volatility forecasting based on text mining and deep learning under high‐frequency data. Journal of Forecasting, 2021, vol. 40, iss. 8, pp. 1596–1610. DOI 10.1002/for.2794
9. Liu Y. Novel volatility forecasting using deep learning–Long Short Term Memory Recurrent Neural networks. Expert Systems with Applications, 2019, vol. 132, pp. 99–109. DOI 10.1016/j.eswa.2019.04.038
10. Verma S. Forecasting volatility of crude oil futures using a GARCH–RNN hybrid approach. Intelligent Systems in Accounting, Finance and Management, 2021, vol. 28, iss. 2, pp. 130–142. DOI 10.1002/isaf.1489
11. Wang T. Stock volatility forecasting: Adopting LSTM deep learning method and comparing the results with GARCH family model. FFIT 2022: Proceedings of the International Conference on Financial Innovation, FinTech and Information Technology (October 28–30, 2022, Shenzhen, China). European Alliance for Innovation, 2023. 12 p. DOI 10.4108/eai.28-10-2022.2328447
12. Balagula Yu. M. Forecasting daily spot prices in the Russian electricity market with the ARFIMA model. Prikladnaya ekonometrika = Applied Econometrics, 2020, no. 1 (57), pp. 89–101. (In Russ.). DOI 10.22394/1993-7601-2020-57-89-101. EDN YJVKGF
13. Berzon N. I., Sulitskii E. A. Primenenie EGARCH modelei dlya analiza spredov Rossiiskikh korporativnykh evroobligatsii. Obligatsionnyi rynok: analiz tendentsii i perspektiv = Bond Market: Analysis of Trends and Prospects. Moscow, 2016, pp. 171–178. (In Russ.). EDN VZPWBZ
14. Garafutdinov R. V. Influence of some ARFIMA model parameters on the accuracy of financial time series forecasting. Prikladnaya ekonometrika = Applied Econometrics, 2021, no. 2 (62), pp. 85–100. (In Russ.). DOI 10.22394/1993-7601-2021-62-85-100. EDN GZHIKL
15. Zagaynov A. I. Investigation of the change in the fractality of chaotic processes in the capital markets. Intellektual'nye tekhnologii na transporte = Intellectual Technologies on Transport, 2017, no. 4 (12), pp. 39–43. (In Russ.). EDN YQKOHT
16. Simonov P. M., Garafutdinov R. V. Modeling and forecasting of financial instruments dynamics using econometric models and fractal analysis. Perm University Herald. Economy, 2019, vol. 14, no. 2, pp. 268–288. (In Russ.). DOI 10.17072/1994-9960-2019-2-268-288. EDN NHKAMR

Show full text