Generating accurate and reliable forecasts for inflation is a necessity for meeting a central bank's primary objective of achieving and maintaining price stability. This issue takes even a greater importance considering that monetary policy effects on the economy are transmitted with time lags depending on the responsiveness of financial markets and real economy to policy interventions. Forecasting inflation is not only an important decision-making tool, but also serves as a crucial communication device. Monetary policy has become considerably more transparent over the recent decades. Open communication is beneficial to the stability and predictability of its transmission into the economy, and essential for the accountability of an independent central bank.
​
In a study conducted through the BCC programme, I analyse the appropriateness of using a Bayesian VAR (BVAR) model to forecast inflation in Albania. Unlike standard autoregressive vector models (VAR), the BVAR allows for the flexible inclusion of a broad range of variables, leading to a more comprehensive explanation of inflation. The BVAR is designed as a medium-sized model, which describes the most important dynamics and interactions between the determinants of inflation in Albania, taking into account developments in the real private sector, the financial sector and the foreign sector. It includes 9 variables (6 domestic and 3 foreign) at a quarterly frequency from 2002Q2 to 2018Q4. A central feature of Bayesian methods, including the BVAR, is to include prior information that the researcher has about the behaviour of the data, these priors are then combined with the data to construct the so-called posterior estimates. The two most common priors found in the literature are the Litterman-Minnesota and Normal-Wishart which are used for the estimation of the BVAR.[1]
 
The results from the BVAR are contrasted against the ones from several simpler econometric models. These include univariate models (the unconditional mean, a random walk, an autoregressive integrated moving average) and an unrestricted VAR. The best performing among these alternatives, namely the random walk and VAR, are used as benchmark against which we evaluate the forecast performance of the BVAR model. The forecasting performance of the various models is measured by the root mean squared error (RMSE) of the out-of-sample forecasts for forecast horizons up to 6 quarters. The forecasts are obtained in two ways (rolling and recursive forecasts) and compared to each-other for different time horizons. In addition, the accuracy of density forecast for both the BVAR and the benchmarks is measured by the average logarithmic scores and they are compared to each other.  
 
The estimated results show that the BVAR approach, which incorporates more economic information, outperforms the benchmark univariate models and the VAR in different time horizons of the forecast sample using both forecast strategies. However, the improvement is moderate as the differences between the various models’ forecast performance are not statistically significant. Of course, this assessment could be revised once we can rely on longer samples of data.
 
A room for future work is to follow a forecast averaging approach, which would be stronger than the BVAR model studied in the paper. In my study, I do not compute forecast averages. This averaging procedure takes different variants of the BVAR that only differ in the prior views on the dynamics of inflation that the researcher inputs in the model (the variants however have the same likelihood function). Such an approach will allow us to consider different a priori views on the inflation process, and contrast their impact on the forecasting of inflation in a well-structured way. As different priors can all be reasonable, averaging over the various variants of the BVAR is likely to give stronger results than selecting only one of them. A forecast combination procedure of the BVAR with other short-term inflation forecasting models could be a successful strategy to improve forecast performance. By combining many misspecified models each incorporating information from different variables, model averaging usually outperforms forecasts from individual models (Aiolfi et al., 2011).


[1] An optimization procedure is implemented to select the best possible combination of the hyperparameters that characterize the priors in such a way that they maximize the marginal likelihood of the model.

​FULL RESEARCH PAPER

Forecasting GDP and inflation is an important part of central bank decision-making, providing inputs for proactive policy formulation. Policy institutions use different forecasting models, going from simple univariate models to more complicated multivariate ones. The proper use of models greatly depends on data availability and data quality. This is an important dimension for many emerging and developing countries that often lack access to high quality data. In contrast, researchers and forecasters in advanced countries benefit from high frequency long time series data in broad area, from micro to macro level.

A standard forecasting framework combining several variables is the vector autoregression (VAR) approach developed by Christopher Sims in the 1980’s. This model consists of a system of equations in which each variable is impacted by its own lags as well as the current values and the lags of other variables. A shortcoming of the approach is that it cannot consider more than a relatively small number of variables. Factor models represent another type of econometric approach and have gained popularity in early 2000s, after works done by Stock and Watson (2002a, 2002b). The central point is to combine a broad range of indicators of the same economic concept (such as different measures of inflation) and extract a factor that is common to them, instead of focusing on one specific variable. Bernanke et al. (2005) have combined the two approaches and developed a factor-augmented VAR (FAVAR) model.

Since then, this method has been widely used in macroeconomics, particularly, in data-rich environment. The FAVAR approach goes around the inability of standard VAR models to include a broad range of data, and helps handling omitted variable bias, as well as providing some robustness in the presence of structural breaks and requires minimal conditions on the errors.

In Rahimov et al. (2019), we build a FAVAR model to forecast inflation and output in Azerbaijan. We contrast the model forecasting ability against other simpler models. The main challenge in this study is the limited quality of data and a short sample (starting from the first quarter of 2003). The variables of interest also exhibit high seasonality and volatility, and include two major structural breaks (in 2007-2008 and in 2015-2016). The number of indicators in some areas, such as labor market and government finance, is also limited. We nonetheless manage to collect 77 quarterly series of indicators on the real sector, prices, monetary variables, government finances and the labor market. We split them into two groups based on their correlations with interest variables (i.e., inflation and GDP growth), and extract three principal components for each group that is included in our FAVAR model.

The forecast ability of the model is assessed using the root mean squared error (RMSE). We show that univariate models outperform the FAVAR in the forecasting of inflation, particularly at a short horizon. Diebold-Mariano (DM) test confirms that the differences in RMSEs between models are significant in short-term horizons. By contrast, FAVAR performs better than univariate models in the forecasting of GDP, even though the DM tests indicate that the differences in RMSE between the models are not significant. The results can partly be attributed to the challenging structure of the variables (short samples, the presence of structural breaks occurred during this period, limited data quality of the indicators used in constructing the factors). The strong inertial component in many
variables – particularly, inflation –could be another reason for superior performance of the univariate model for inflation forecasting.

In summary, our study reveals that FAVAR models do not always produce best forecasts, and researchers should use care in selecting specific models used in forecasting. While having a rich dataset covering a broad range of indicators helps getting better forecasts, other characteristics, such as data quality, length and frequency of data, are more important.