Comparing the Forecasting of Iranian Inflation with many predictors

One of the most important economic problems in Iran during the last few decades is the phenomenon of inflation, to an extent that improving the conditions caused by inflation has always been one of the important goals of developmental plans in Iran. Achieving this goal requires the creation of a rigorous and purposeful mechanism for the economic policymaking process, which in its standard form involves forecasting, targeting and policy analysis. Given that inflation affects many macroeconomic indicators, therefore, explaining and forecasting inflation rates based on econometrics models can help improve policy making.

In this study, 108 quarterly variables from 1990-2017 were used. The variables included the consumer price index as a dependent variable and 107 independent variables (predictors) which were included in nine blocks in order to extract the factors. Thus, 1) the price block included different values ​​of the producer price index, GDP deflator and wage indexes (including 13 variables); 2) the demand block included components of GDP on the demand side and some other indexes related to capital stock (including 7 variables); 3) the fiscal block included all types of government revenues and expenditures (including 13 variables); 4) the international block included exports, imports, exchange rates, as well as the inflation and exchange rates of China and Germany (as Iran's largest trading partners) (including 17 variables); 5) the output block, where different types of production variables were used (including 14 variables); 6) the money block included liquidity and monetary base components (including 21 variables); 7) the financial block, including stock market and insurance variable (including 6 variables); 8) the energy block included various variables related to crude oil, electricity and energy (including 7 variables); and 9) the labor force block which included various variables of unemployment, employment and productivity (including 9 Variables). All the variables of this paper were seasonally adjusted using X-13, TRAMO or STL. Also, the statics of the variables were investigated by the use of unit root tests without seasonal integration (Dickey Fuller and KPSS tests) and a unit root test with seasonal and semi-annual integration (HEGY test). In addition, all variables were standardized by differentiating from the mean and dividing by the standard deviation. Principal Component Analysis (PCA) was used to extract the factors using all the variables in each block.The main purpose of this study was to compare DMA and DMS models (9 block factor methodes) with BMA, BVAR, TVP and AR. In addition, the interruptions of each model were determined using BIC. Similar to the study of Koop and Korobilis (2012), the predictions were considered with a short-run horizon (h = 1), a medium-run horizon (h = 4) and a long-run horizon (h = 8).In order to assess the prediction performance, Mean Squared Forecast Error (MSFE), Mean Absolute Forecast Error (MAFE), Mean Absolute Percentage Error (MAPE), the bias of the prediction error, the variance of the prediction error and the sum of log predictive likelihoods (PL) were used. In addition, the Diebold-Mariano (1995) test was used to compare the prediction accuracy.

The results show that the prediction of DMS and DMA is more efficient than other traditional predictive methods to predict the inflation rate in Iran. Findings indicated that in all forecast horizons, money and price blocks were the highest in using the optimal model over time and the fiscal block was the lowest. In general, it can be said that the DMA model is the average of many models and does not only use a single factor block (there is also no irrelevant factor block) and this is the most important feature of DMA / DMS use.

Due to the fact that variables can be changed in DMA / DMS in any time period, so their use for economic policy is recommended.