STAN: Smooth Transition Autoregressive Networks

  • 2025-01-30 19:01:01
  • Hugo Inzirillo, Remi Genet
  • 0

Abstract

Traditional Smooth Transition Autoregressive (STAR) models offer an effectiveway to model these dynamics through smooth regime changes based on specifictransition variables. In this paper, we propose a novel approach by drawing ananalogy between STAR models and a multilayer neural network architecture. Ourproposed neural network architecture mimics the STAR framework, employingmultiple layers to simulate the smooth transition between regimes and capturingcomplex, nonlinear relationships. The network's hidden layers and activationfunctions are structured to replicate the gradual switching behavior typical ofSTAR models, allowing for a more flexible and scalable approach toregime-dependent modeling. This research suggests that neural networks canprovide a powerful alternative to STAR models, with the potential to enhancepredictive accuracy in economic and financial forecasting.

 

Quick Read (beta)

loading the full paper ...