Determinants of Inflation Forecast: A Dynamic Model Averaging Approach

First studies in inflation forecasting were mostly based on traditional Philips curve in which the relation between inflation and unemployment is studied. However, after several decades and especially after the Lucas criticism, Philips curve faced great takeovers. The new Philips curve ties real and expected inflation, not to unemployment rate but to a scale of the marginal cost. Since in the original form of Philips curve, marginal cost stimulates inflation, it is difficult to formulate models that are effective in predicting inflation. Therefore, using TVP-DMA model, which has the ability to fix these deficiencies, we try to improve predictability of inflation in Iranian economy. An independent variable in conventional models can be either significant or insignificant while in TVP-DMA model, it may be significant during a period of time and insignificant in rest of the times. Therefore, this approach lets us to determine the periods in which an independent variable is significant and when it is not. In this study, we use seasonal data during the period 1991-2015. The results based on outputs of the TVP, DMS, and DMA models show that, out of 100 time periods under study, the liquidity growth rate in 19, economic growth rate in 7, unemployment in 8, exchange rate growth in 31, changes in the bank deposit rate in 14, oil revenues growth rate in 15, inflation uncertainty in 14 and the budget deficit growth rate in 4 periods have significant effect on inflation. Based on these results, it can be stated that exchange rate growth, liquidity growth and oil revenues growth rate are the most important indicators influencing inflation rate in Iran.