نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه اقتصاد، دانشکده مدیریت و اقتصاد، دانشگاه آزاد اسلامی واحد ابرکوه، یزد، ایران
2 دانشگاه آزاداسلامی واحد یزد
3 گروه اقتصاد، دانشکده مدیریت و اقتصاد، دانشگاه آزاد اسلامی واحد یزد، یزد، ایران
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Purpose:
Providing reliable forecasts of inflation expectations is a constant challenge for policymakers and it is of vital for economic activists and their investment decisions. Accurate forecasting of inflation expectations in a data rich environment when there is correlation between predictor variables is an active research field. The purpose of this paper is to predict inflation expectations by considering economic and monetary variables using machine learning algorithms to help the economic policy makers of the country.
Methodology:
In this study, machine learning algorithms are used to select the most effective variables in predicting inflation expectations because they are suitable for discovering complex patterns and rely on fewer and more restrictive assumptions. For this intention, the data are grouped into two sets of training and testing that the 14 year time period is used for training sets and the models prediction accuracy is evaluated in the 15th year. The dataset of variables includes exchange rate, stock market index, balance of payments, production price index, import price index, wage growth rate, production gap, economic growth rate, money market interest rate, liquidity growth, oil revenues. Forecasting inflation expectations is done using ridge, lasso, Adaptive lasso and elastic net algorithms. Also, to check the machine learning techniques accuracy, the mean square error criterion is used.
Findings and Discussion:
The findings of this paper showed that the, adaptive lasso algorithm has a smaller difference between the observed and estimated values than other techniques in predicting inflation expectations. The mean square error between the real and estimated values in the adaptive Lasso algorithm is 0.0892. The results confirmed that the oracle feature of the adaptive lasso algorithm was able to reduce the penalty by giving less weight to them and leave their effects in the model in real form. Based on this, among economic variables, exchange rate and liquidity growth have the most positive effect on inflation expectations. Their coefficients are 3.144 and 2.904, respectively. Also, the interest rate as a monetary variable has the most negative effect on inflation expectations with a coefficient of -4.383. This result is in line with Fisher's theory because Fisher believed that a higher interest rate leads to a decrease in inflation, and with an increase in the interest rate, the consumption attractiveness in the current period and demand decreases.
Conclusion and Policy Implications:
The difference in the amount of import and export in Iran causes a high dependence on the import of goods, and any change in the exchange rate helps in the consumer expectations formation regarding future inflation. Also, the process of creating liquidity without support due to the lack of coordination with the level of goods production and services causes inflation expectations. The central bank can encourage people to deposit their savings in banks by increasing the deposit interest rate. On the one hand, the demand decreases, and on the other hand, the total consumption and the money circulation speed decrease. Considering the results of this research, policymakers and economic officials can apply directly apply machine learning algorithms to the problem of macroeconomic forecasting in a data-rich environment, using all predictors, without the need for multivariate linear modeling and with minimal size error.
کلیدواژه‌ها [English]
)