Logo
Munich Personal RePEc Archive

Nonlinear volatility models in economics: smooth transition and neural network augmented GARCH, APGARCH, FIGARCH and FIAPGARCH models

Bildirici, Melike and Ersin, Özgür (2012): Nonlinear volatility models in economics: smooth transition and neural network augmented GARCH, APGARCH, FIGARCH and FIAPGARCH models.

[thumbnail of MPRA_paper_40330.pdf]
Preview
PDF
MPRA_paper_40330.pdf

Download (328kB) | Preview

Abstract

Recently, Donaldson and Kamstra (1997) proposed a class of NN-GARCH models which are extended to a class of NN-GARCH family by Bildirici and Ersin (2009). The study aims to analyze the nonlinear behavior and leptokurtic distribution in petrol prices by utilizing a newly developed family of econometric models that deal with these concepts by benefiting from both LSTAR type and ANN based nonlinearity. With this purpose, the study proposed several LSTAR-GARCH-NN family models. It is noted that the multilayer perceptron (MLP) neural network and LSTAR models have significant architectural similarities. Accordingly, linear GARCH, fractionally integrated FI-GARCH, asymmetric power APGARCH and fractionally integrated asymmetric power APGARCH models are augmented with a family of Neural Network models. The study has following contributions: i. STAR-GARCH and LSTAR-GARCH are extended to their fractionally integrated asymmetric power versions and STAR-ST-FIGARCH and STAR-ST-APGARCH, STAR-ST-FIAPGARCH models are developed and evaluated. ii. By extending these models with neural networks, LSTAR-LST-GARCH-MLP family models are developed and investigated. These models benefit from LSTAR type nonlinearity and NN based nonlinear NN-GARCH models to capture time varying volatility and nonlinearity in petrol prices. ANN augmented versions of LSTAR-LST-GARCH models are as follows: LSTAR-LST-GARCH-MLP, LSTAR-LST-FIGARCH-MLP, LSTAR-LST-APGARCH-MLP and LSTAR-LST-FIAPGARCH-MLP. Empirical findings are collected as follows. i. To model petrol prices, fractionally integrated and asymmetric power versions provided improvements among the GARCH family models in terms of forecasting. ii. LSTAR-LST-GARCH model family is promising and show significant gains in out-of-sample forecasting. iii. MLP-GARCH family provided similar results with the LSTAR-LST-GARCH family models, except for the MLP-FIGARCH and MLP-FIAPGARCH models. iv. Volatility clustering, asymmetry and nonlinearity characteristics of petrol prices are captured most efficiently with the LSTAR-LST-GARCH-MLP models benefiting from forecasting capabilities of neural network techniques, whereas, among the newly developed models, LSTAR-LST-APGARCH-MLP model provided the best performance overall.

Atom RSS 1.0 RSS 2.0

Contact us: mpra@ub.uni-muenchen.de

This repository has been built using EPrints software.

MPRA is a RePEc service hosted by Logo of the University Library LMU Munich.