Modelling the absolute returns of different stock indices: exploring the forecastability of an alternative measure of risk

Abstract

Conventional measures of the risk of a financial asset make use of the unobserved (conditional) variance or standard deviation of its return. In this paper, we treat the observed absolute return as a measure of risk and explore its forecastability. Two simple models are considered. One is a new AR‐like model which is applied to the absolute return. The other is an ARCH‐like model called Asymmetric Power ARCH. The forecastability is evaluated with the average log‐likelihood of absolute return, instead of that of return itself. While the absolute return is interpreted as ‘volatility’, some quantities of its entire distribution, such as the 95th quantiles, can be interpreted as ‘volatility of volatility’. We apply both models to three stock indices, namely the Hang Seng Index, the Nikkei 225 Index and the Standard and Poors 500 Index. The new model by and large outperforms the ARCH‐like model in both in‐sample goodness of fit and post‐sample forecastability. It performs exceptionally well in the post‐sample period after the outbreak of the Asian financial crisis. Copyright © 2000 John Wiley & Sons, Ltd.

Quantiles, expectiles and splines

“…As such they offer an alternative to methods such as those in Engle and Manganelli (2004) and Granger and Sin (2000), that are based on conditional autoregressive models.…”

The past and future of empirical finance: some personal comments

“…Modeling and estimation is not very difficult and in practice the problem of estimated quantiles crossing appears not to be difficult (see Granger and Sin, 2000). The observed long-memory properties of volatility should be observed in the quantiles due to breaks.…”

A multiple indicators model for volatility using intra-daily data

“…As a matter of fact, next to the traditional volatility modelling from daily returns measured as the log-difference of closing prices, we can consider absolute returns on which considerable modeling effort is present in the literature (Taylor, 1986;Ding et al, 1993;Granger and Sin, 2000) and, with the already mentioned provisos, realized volatility as the standard deviation of intra-daily returns observed at regular intervals. Furthermore, it has long been recognized that the spread between the highest recorded daily price and the lowest recorded daily price is a function of the volatility during the day and, as such, can lead to an improvement of the volatility estimates.…”