Short-term inflation forecasting models with Bayesian VAR: Evidence from Azerbaijan

In alignment with the Central Bank’s objective to maintain price stability in the economy, this paper is dedicated to forecasting short-term inflation in Azerbaijan with various time series models including Bayesian Vector Autoregressive (BVAR) technique, which, alleviates overparameterization problem. We have also utilized autoregressive (AR) and standard VAR as benchmark models to make a comparison of forecast errors derived from out-of-sample analysis with those of BVAR model. For BVAR estimations, Litterman, Minnesota and Sims-Zha Normal Wishart (NW) prior have been employed. The monthly estimation period covers the date range from 2003M1 to 2019M6 and by using the expanding window strategy, we extended the data window to 24 months and forecasted the subsequent 24 months. We have carried out analysis in two stages: economic category-specific and incremental modelling. In a category-specific analysis, we developed 5 models for VAR and BVAR priors, each focusing on various sectors of the economy. We then applied an incremental approach, where variables from the earlier category-specific models were added step by step, enabling us to evaluate how the forecast performance changed as additional variables were included in the models. The study analyzed different frameworks for the assessment of the final forecast with the objective of improving the forecast accuracy. Overall, the examination of different models and estimation techniques demonstrates that each model and estimation technique have a significant contribution to our suggested four different strategies of forecast assessment.