Monetary policy-maker reaction function in oil exporting countries: STR econometric approach.

Objective: This study is trying to analyze the behavior of monetary policymakers with using Taylor's rule.

Methods: Since it is not expected to have a linear relationship between the variables of the model due to the existence of successive structural changes and changes in the political regime, therefore, the smooth transition threshold regression model (STR) has been used for respect to the variables of oil price changes, official exchange rate changes, inflation gap and production gap in the annual period from 2002 to 2019.

Results: To check the stationarity of the variables, the generalized Dickey-Fuller unit root (ADF) test was used. The obtained results have shown that in all the studied countries, the variables of production gap, official exchange rate changes, inflation gap, oil price changes and nominal interest rate changes (for Iran, money supply growth rate) are at a static level. The next step in estimating an STR model, is to determine the optimal interval for the model variables. For this purpose, according to the annuality of the research period, the optimal interval of the variables has been calculated using the ARMA model and the Schwartz criterion. After determining the interval for research variables, the next step in estimating an STR pattern is to determine the type of pattern in terms of linear or non-linearity, for which the F test statistic is used. According to the probability value of the F test statistic, the hypothesis of non-linearity of the reaction function of the monetary policy maker in oil exporting countries is confirmed. The next step is to choose the appropriate transition variable for the nonlinear model. Based on this, the GDPG variable is determined as a transfer variable according to the presented results. The next step is to select the appropriate model (from ESTAR and LSTAR) for the GDPG transfer variable according to the F2, F3 and F4 statistics. Considering that 4 equations are estimated for each country, the type of STR model based on the obtained results is given in the table below. But for most countries, as in Table (4), the suitable model proposed with the GDPG transition variable is the soft transition regression model with linear transition function (LSTAR), because the probability value of the F2 statistic is 0.00, which is less than 5%. One of the most important results obtained is that specifying the reaction function of monetary policy makers in the studied countries in a linear way is an inappropriate specification and a non-linear model should be used. The findings from the estimation of the specified policy rule based on models (1), (2), (3) and (4) have shown that the targeting of monetary policy makers is the same in the regimes and in the period Recession and prosperity, the same policy should be adopted. It is important to mention that the main difference between the regimes is the intensity of the impact of the variables. Also, the results have shown that in all the investigated countries, by entering the variables of oil price changes and the official exchange rate into Taylor's model (model 1), the targeting of monetary policy makers is stable. Meanwhile, for the country of Iran, with the inclusion of the variables of oil price changes and official exchange rate in the Taylor model, inflation targeting is changed to production targeting. Based on the results obtained from the estimation of models (1), (2), (3) and (4), it can be seen that the variable of oil price changes in the countries of Algeria, Qatar, Kazakhstan, Ecuador, Colombia, Malaysia, Mexico, Belarus and Bulgaria affects the reaction function of monetary policy makers through the production gap channel and in Iran, Russia, Angola, Nigeria, Brazil, Tunisia and Azerbaijan through the official exchange rate gap channel. Based on the obtained results, including the exchange rate in Taylor's law alone does not improve the properties of the equation. Therefore, the claim of Lubik and Echurfide (2005, 2007) regarding the fact that monetary policies do not have a significant reaction to exchange rate fluctuations, is accepted in most of the studied countries. According to the results obtained at threshold levels higher than the threshold of economic growth, it is possible to use monetary policies to stimulate economic growth and real sector activities with low inflationary effects. In the following, after estimating the models, we will evaluate the results in the studied countries. We start this section by examining the possible errors in the estimation stage, which are, respectively: the test of the absence of autocorrelation error, the test of the non-existence of a non-linear relationship in the residuals of the model, and the constancy of the factors in different regimes. In the test of the absence of autocorrelation error, the probability value of the F test for this test (between 0.35 and 0.15) has been obtained, based on which, the null hypothesis (absence of autocorrelation) is accepted at a suitable level of confidence. In the test of the absence of non-linear relationship in the residuals of the model, the probability value of the F test for the transfer function H02 is estimated (between 0.65 and 0.15), which shows that the model has been successful in specifying the non-linear relationship between the variables. In the test of constancy of the factors in different regimes, the probability value obtained for the F statistic is equal to 0.00, based on which, the sameness of the coefficients in the linear and non-linear parts is rejected at the 99% probability level. Among the other tests that evaluate the possible errors in the STR pattern estimation phase are the tests for checking the errors of heterogeneity of variances and normality of the residuals. The results have shown that the null hypothesis of the tests that there is no heterogeneity of variance and the normality of the residuals cannot be rejected at a significant level of 1%. Therefore, it can be said that the model has no heterogeneity of variance and the residuals of the model also have a normal distribution.