Application of Neural Networks to Short Time Series Composite Indexes: Evidence from the Nonlinear Autoregressive with Exogenous Inputs (NARX) Model

Roman Matkovskyy (1) , Taoufik Bouraoui (1)

1 ESC [Rennes] - ESC Rennes School of Business

The aim of this paper is to extend the index of financial safety (IFS) approach with improving its predictive performance and to show the applicability of artificial neural networks to economic and financial short time series. To this end, prediction is performed by means of the nonlinear autoregressive with exogenous inputs (NARX) model that represents the neural networks and can emulate any nonlinear dynamic state space model. Thus, a NARX model, trained by means of Levenberg–Marquardt algorithm, was chosen since it gave the best performance. Results reveal that the NARX models are suitable for performing short time series composite indexes prediction.

Mots clés: Index of financial safety (IFS); Forecasting; Nonlinear autoregressive with exogenous input (NARX) model; Neural networks

Domaines: Économie et finance quantitative [q-fin]

Roman Matkovskyy, Taoufik Bouraoui. Application of Neural Networks to Short Time Series Composite Indexes: Evidence from the Nonlinear Autoregressive with Exogenous Inputs (NARX) Model. Journal of Quantitative Economics, 2019, 17 (2), pp.433-446. ⟨10.1007/s40953-018-0133-8⟩. ⟨hal-02155402⟩