Volatility forecasting model selection with exponentially weighted information criteria.
In this paper, we consider a recently proposed information criteria (IC) for selecting among forecasting models. This IC involves the use of exponential weighting within the measure of fit in the standard IC, such as the Akaike’s IC or Schwarz’s Bayesian IC. The effect of this is that greater weight...
| Main Author: | |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | English English |
| Published: |
2009
|
| Subjects: | |
| Online Access: | http://psasir.upm.edu.my/id/eprint/18844/1/ID%2018844.pdf |
| Summary: | In this paper, we consider a recently proposed information criteria (IC) for selecting among forecasting models. This IC involves the use of exponential weighting within the measure of fit in the standard IC, such as the Akaike’s IC or Schwarz’s Bayesian IC. The effect of this is that greater weight is placed on more recent observations in order to reflect more recent accuracy. This new exponentially weighted IC has previously been used to select among models for point forecasting. In this paper, we consider its use for selecting between a variety of volatility forecasting models, including GARCH, exponential smoothing, and smooth transition exponential smoothing. In the likelihood functions for all the models, we allow conditionally non-Gaussian distributed errors in order to capture more fully the leptokurtosis in financial returns data. We demonstrate the use of model selection based on the new IC using stock indices data obtained from four major stock markets. |
|---|