DefinePK

Abstract

This paper aims to contribute to the uses of time series econometrics, combining them with some contemporary machine learning techniques for data understanding to enhance the analysis of behaviour patterns, the improvement of the forecasting capability of the models, and the facilitation of policy assessment. The quantitative analysis works through econometric models including ARIMA, VAR, and TVP-SVAR for the selected macroeconomic indicators like GDP, Inflation rate; and machine learning models including LSTM, Random Forest and Gradient Boosting. The data collected was retrieved from different global databases such as the International Monetary Fund, World Bank, and Google Trends with various data points spanning for about 25-30 years. The performance comparison shows that the applied machine learning models, most of all LSTM, are categorically more accurate than the traditional models when forecasting under non linearity and time varying environments. Further, Policy Exercise with the help of TVP-SVAR imply that the fiscal policy is most effective for economic growth compared with a reduction in the interest rate and subsidies. Among variables, interest rates and unemployment have shown the greatest influence in the Random Forest model of GDP. According to the forecasts and simulation of various scenarios proposed by the author, only an integrated policy can bring the highest GDP growth rate. Despite the designs gaining higher performance in the models, the relevant drawbacks in accuracy and interpretability are still challenging, leading to the creation of more hybrids that balance between the two factors. In conclusion, this work highlights the significance of empirical dynamics in the study of increasingly complex economic behavior and contributes to establishing rigorous, flexible, and policy-oriented econometric models.