Modelling the density of Egyptian quarterly CPI inflation

Abstract EN

This paper aims at modelling the density of quarterly inflation based on time-varying conditional variance, skewness and kurtosis model developed by Leon, Rubio, and Serna (2005).

They model higher-order moments as GARCH-type processes by applying a Gram- Charlier series expansion of the normal density function.

We estimated seven univariate models, including GARCH-M and TARCH-M models, assuming three different distributions for the error term, namely: normal, student t, and GED distributions.

Additionally, the model that allows for non-constant higher order moments, GARCHSK-M, has been estimated.

Moreover, the paper utilizes two multivariate models, Dynamic Conditional Correlation (DCC) and Diagonal VECH models to isolate the time-varying conditional correlations between inflation and two financial variables, including growth in domestic credit and real exchange rate.

Results revealed the significant persistence in conditional variance, skewness and kurtosis, which indicate high asymmetry of inflation.

Diagnostic tests indicated that models with invariant volatility, skewness and kurtosis are inferior to the models that permit them to vary over time.

Moreover, depending on models of static historic correlation between inflation and the highly financial variables in order to evaluate inflation dynamic behavior is misleading and is a poor informative.

Comparing the predictive power of different models showed that basic models are more accurate in forecasting out-of-sample inflation according to some criterions and GARCHSK-M is better for other criterions.

By applying Diebold and Mariano’s (1995) encompassing test, it was found that all models could be combined together to form a more accurate forecast.

We have done the combination of forecasts using equal weights, Bayesian Model Averaging (BMA), and Dynamic Model Averaging (DMA).

Results of forecast combination showed that the combined forecasts outperform the projection of best single model.