Heliyon (Jan 2023)
Estimation of the unemployment rate in Turkey: A comparison of the ARIMA and machine learning models including Covid-19 pandemic periods
Abstract
The article focuses on analyzing the robustness of Auto Regressive Integrated Moving Average (ARIMA) and Artificial Neural Networks (ANN) methods in unemployment rate estimation. In this context, a stochastic trend in the unemployment rate was determined by using monthly data in Turkey. The oil price, real exchange rate, interest rate and unemployment rate variables are imported into the ARIMA and ANN models with 176 data samples for the period of 01.01.2008-31.08.2022. The results of the conventional linear ARIMA and nonlinear ANN regressor models are compared. The comparison results show that the ARMA (2,1) model is the most suitable model for the unemployment rate estimation. This conclusion was reached based on ARMA (2,1) and ANN's RMSE, MAE, MAPE and R2 parameters. From the results of the specified criteria, it was found that both models gave results close to the actual unemployment rate however ARMA (2,1) was the more appropriate model for the current data set. The actual unemployment data and the estimated values are also given verifying the better modeling of the developed ARMA (2,1) model. In addition, there are meaningful relationships between month variables and the employment rate. This result supports that the unemployment possesses chronic reasons in Turkey. On the other side, the unemployment rate forecasting error of the ARMA (2,1) is higher than the ANN model for the 2020–2021 period during the intense pandemic. This result is important because it shows that during the times of the economic uncertainty caused by the Covid-19 pandemic, forecasts employing the neural network model is observed to have lower errors than the results of autoregressive moving average model. Therefore, under an economic uncertainty, it is shown that modeling the unemployment rate using artificial neural network provides novel insights for economic forecasting.
