Ajayi, Olaoluwa (2019) Comparison of alternative inflation forecasting models in OPEC and BRICS countries. (PhD thesis), Kingston University, .
Abstract
We compare the forecasting performance of univariate and multivariate models for BRICS and OPEC economies. For the univariate models, we produce forecasts using ARIMAX models that have a deterministic component to account for structural breaks over the full sample period and different ARIMA specifications over a reduced sample period that avoids the modelling structural breaks. The univariate ARIMA models that we develop over the reduced sample period are, first, a seasonal ARIMA specification identified using the Box-Jenkins method, second, a seasonal ARIMA model identified using EView’s automatic model selection tool and third, a non-seasonal ARIMA model identified using EView’s automatic model selection tool applied to seasonally adjusted data. The other univariate model we considered include the regime shift threshold Autoregressive model (over the full sample and reduced sample) and the naïve model which added as a benchmark. Multivariate models are estimated over the reduced sample period to avoid modelling structural breaks and are based upon Vector Autoregression (VAR) models that utilise differencing and cointegrating restrictions to ensure the stationarity of the data. In particular, we consider the unrestricted VAR model with differenced (stationary) data, the (unrestricted) Vector Error Correction Model (VECM) that assumes cointegration without imposing cointegrating restrictions and the restricted VEC that imposes a single cointegrating equation on the VECM. Our study shows that the benchmark models (naïve) were never favoured over the best selected univariate and multivariate model. The univariate EView’s automatic non-seasonal ARIMA model is generally favoured for the BRICS countries (the exception is South Africa). However, the results are mixed between univariate and multivariate methods for OPEC countries. For OPEC countries that have a history of moderate inflation, for example, Saudi Arabia, the univariate automatic non-seasonal ARIMA model outperforms the multivariate model. In contrast, multivariate models generally outperform univariate automatically selected ARIMA models for countries with high inflation (e.g Angola and Algeria).
Actions (Repository Editors)
 |
Item Control Page |