موضوعات
عنوان مقاله English
نویسندگان English
Measuring unobservable variables such as inflation expectations or potential output is inherently challenging. Inflation expectations, as a key variable, play a central role in many macroeconomic and monetary models. In Iran, despite decades of persistent double-digit inflation, no official survey-based data on inflation expectations has been produced, unlike in many other countries. However, the literature suggests that comparing model-based approaches with survey data can yield valuable insights. This study aims to estimate and present quarterly inflation expectations for Iran during 1996–2021 using the Random Forest regression method within the framework of rational expectations. After training the model, in-sample predictions were generated and evaluated based on rational expectations properties. The analysis also ranked the importance of variables contributing to the formation of inflation expectations. The results indicate that inflation expectations in Iran can be reasonably explained within the rational expectations framework, with no evidence of systematic forecast errors. Among the informational variables considered, inflation lags, exchange rate, and economic sanctions were identified as the most influential factors
Aim and Introduction
Measurement and examination of unobservable variables directly such as inflation expectations or potential output, is really challenging. Inflation expectations have been considered a key variable in many macroeconomic models, particularly in the realm of monetary economics. Macroeconomic models assume that economic agents make consumption, savings, and labor market decisions based on their perception of future inflation levels, and these decisions play a great role in realizing economic variables, including inflation. The role of inflation expectations differs from other inflation-generating factors. While factors such as money supply, budget deficit, exchange rate, and to some extent, economic sanctions can be considered as policy tools. Inflation expectations normally result from the interaction of other factors and may potentially predict future inflation. For example, an increase in the budget deficit, if not addressed independently by the Central Bank, can lead to an increase in money supply, inflation, and intensification of inflation expectations. Thus, inflation expectations can be considered as a variable that evolves within society and changes due to other inflation-generating factors. However, once formed, these expectations themselves become significant factors in inflation and other economic variables. Unlike many countries, in Iran, despite the importance of inflation due to decades of double-digit inflation, no action has been taken to produce and provide survey data related to this variable. However, according to existing literature, comparing the results of alternative methods incorporating inflation expectations with survey data can provide valuable insights. In practice, incorporating inflation expectations can improve the performance of inflation prediction models.
Methodology
Empirical research indicates that methods that consider inflation expectations along with its fluctuations and dynamics outperform models that do not consider these dynamics. Therefore, paying proper attention to how inflation expectations form and fluctuate, as well as avoiding simple methods, is necessary in calculating inflation expectations. In this research, an attempt was made to calculate and present data related to this variable in the framework of rational expectations for the period of 1996 to 2021 using the random forest regression method, considering the strengths and weaknesses of each method of mapping inflation expectations. Subsequently, after learning the random forest-based model, by conducting an in-sample prediction, the data were extracted and the features related to rational expectations regarding these data were examined.
Findings
The coefficient of determination value for the test data was found to be 80%, indicating that, on average, 80% of inflation variations are correctly predicted by economic factors using the model inputs or features. Based on this and by examining the features related to estimation residuals, it was determined that economic factors in predicting inflation do not exhibit systematic errors and, with a sufficiently large time interval and having an adequate information set, can have a proper understanding of inflation behavior. Moreover, the results of comparing inflation expectations based on random forest regression-based predictions show superiority of this approach compared to competing methods such as the Hodrick-Prescott filter. After that, the importance of each of the factors in the basket of information related to inflation expectations was ranked. It should be noted that the selection of features for predicting inflation expectations was not based on the direct attention of households and economic factors to these features. Rather, economic factors and households may find the effect of these features in other evidence. For example, the effect of an increase in the exchange rate on the prices of goods that are somehow related to this variable may be apparent to households, and fundamentally, the prevalent interpretation of rational expectations in the literature of this field is based on this approach. The results of this ranking indicate that among the entire information set, factors such as inflation breaks, exchange rates, and economic sanctions had the highest importance in shaping inflation expectations.
Discussion and Conclusion
It is worth mentioning that inflation breaks have been identified as the most important factor among the entire information set as a manifestation of the adaptive section of inflation expectations. However, this does not mean that expectations are entirely adaptive. Based on the research findings, it is clear that if economic factors rely solely on the adaptive section to predict inflation, zero estimation error, unpredictability of errors, and consequently the formation of rational expectations will not be achieved. Using a combination of three approaches: gradient boosting algorithm, random forest algorithm, and linear regression, a voting regression was also performed, showing a 3% improvement in determination coefficient compared to random forest (83%). Moreover, other results, such as the order and intensity of feature importance, and predicted inflation values, are similar to the random forest method with slight variations which means, estimating rational expectations is reliable
کلیدواژهها English
