1: \begin{abstract}
2: Here we present a theoretical study on the main properties of
3: Fractionally Integrated Exponential Generalized Autoregressive Conditional
4: Heteroskedastic (FIEGARCH) processes. We analyze the conditions for
5: the existence, the invertibility, the stationarity and the ergodicity of
6: these processes. We prove that, if $\{X_t\}_{t \in \mathds{Z}}$ is a
7: FIEGARCH$(p,d,q)$ process then, under mild conditions,
8: $\{\ln(X_t^2)\}_{t\in\mathds{Z}}$ is an ARFIMA$(q,d,0)$, that is, an
9: autoregressive fractionally integrated moving average process. The
10: convergence order for the polynomial coefficients that describes the
11: volatility is presented and results related to the spectral representation
12: and to the covariance structure of both processes
13: $\{\ln(X_t^2)\}_{t\in\mathds{Z}}$ and $\{\ln(\sigma_t^2)\}_{t\in\mathds{Z}}$
14: are also discussed. Expressions for the kurtosis and the asymmetry measures
15: for any stationary FIEGARCH$(p,d,q)$ process are also derived. The $h$-step
16: ahead forecast for the processes $\{X_t\}_{t \in \mathds{Z}}$,
17: $\{\ln(\sigma_t^2)\}_{t\in\mathds{Z}}$ and $\{\ln(X_t^2)\}_{t\in\mathds{Z}}$
18: are given with their respective mean square error forecast. The work also
19: presents a Monte Carlo simulation study showing how to generate, estimate
20: and forecast based on six different FIEGARCH models. The forecasting
21: performance of six models belonging to the class of autoregressive
22: conditional heteroskedastic models (namely, ARCH-type models) and radial basis models
23: is compared through an empirical application to Brazilian stock market exchange index. \\
24:
25: \noindent {\bf Keywords:} Long-Range Dependence, Volatility, Stationarity, Ergodicity, FIEGARCH Processes.
26:
27: \noindent {\bf MSC (2000)}: 60G10, 62G05, 62G35, 62M10, 62M15, 62M20
28:
29: \end{abstract}
30: