Kinetic approach application in forecasting the stock closing prices

Authors: Saykova E.R
Published in issue: #4(93)/2024
DOI:
Category: Mathematics | Chapter: Computational Mathematics
Keywords: non-stationary time series, distribution function sample density, Liouville equation, time series values forecasting, kinetic approach to stock closing price forecasting Received 07.05.2024
Published: 19.09.2024

The problem of forecasting time series values based on the kinetic approach is considered. The problem of non-closure of the system of equations for modeling the evolution of the sample density of the distribution function is solved by introducing a forecast assumption. An alternative method for obtaining a forecast for more than one time step ahead using a basic model based on the Liouville equation is proposed. A program code for constructing a forecast using the kinetic approach for closing prices of shares of various companies is implemented. The results of its work (in the form of an estimate of the forecast error as the mean absolute percentage error — MAPE) for ten non-stationary time series are presented.

References

[1] Orlov Yu.N., Osminin K.P. Non-stationary time series: Forecasting methods with examples of analysis of financial and raw materials markets. Moscow, LENAND Publ., 2023, 384 p. (In Russ.).

[2] Orlov Yu.N., Osminin K.P. Methodology for determining the optimal sample size for forecasting non-stationary time series. Information technologies and computing systems, 2008, no. 3, pp. 3–13. (In Russ.).

[3] Lerasle M., Magalh?es N., Reynaud-Bouret P. Optimal kernel selection for density estimation. High Dimensional Probability VII. The Cargese. Prog. Probab. Birkhauser, 2016, vol. 71, pp. 425–460. https://doi.org/10.1007/978-3-319-40519-3_19

[4] Murphy K.P. Probabilistic Machine Learning: An Introduction. Moscow, DMK Press, 2022, 990 p. (In Russ.).

[5] Gorlach B.A. Mathematical Analysis. St. Petersburg, Lan Publ., 2022, 608 p. (In Russ.).

[6] Nasdaq Stock Exchange. URL: https://www.nasdaq.com/ (accessed February 25, 2023).

[7] Hapke H., Nelson K. Developing Machine Learning Pipelines. Automating Model Life Cycles with TensorFlow. Moscow, DMK Press, 2021, 346 p. (In Russ.).