Approach to modelling the yield curve as a multivariate time series

Abstract

Approach to modelling the yield curve as a multivariate time series

Galimnurov A.A.

Incoming article date: 02.04.2025

The study presents an approach to modelling multivariate time series using parameterisation, using yield curve as an example. The effectiveness of adding parameterisation coefficients to predicates is evaluated, and new loss functions are proposed that focus on modelling the shape of the curve. Prediction models including LSTM, Prophet and hybrid combinations were applied. A Python-based system was developed to automate data processing and evaluation. The method improves the accuracy and interpretability of forecasts, offering a promising tool for financial modelling.

Keywords: machine learning, financial engineering, stock market modeling, bond market