Детекція структурних зламів при прогнозуванні фінансових часових рядів

Abstract

У роботі представлено вдосконалений метод прогнозування фінансових рядів, що є поєднанням методу Elastic

This paper presents an improved method for forecasting financial time series, which combines the Elastic Net method with Gaussian weight decay and a mechanism for automatically detecting structural changes in time series. We propose training the model on a recent segment of data that reflects the current market conditions, rather than using the entire available history. This method reduces the gap between the latest observations and the first forecast value compared with baseline weighting schemes. Our research confirms that accounting for shifts in market regimes helps improve the overall accuracy of forecasting models.