Chronos with ensemble GBT: a hybrid framework for GDP forecasting with zero-shot learning

Abstract

Whether it is global or national, accurate economic forecasting is crucial for a country. It paves the direction of a country in terms of policy making, resource allocation, and risk management etc. There are several economic indicators such as interest rates, inflation rates, gross domestic product (GDP), unemployment rates, etc. to determine economic trends. But among them GDP is one of the main indicators for measuring one country’s economic health. As a result, innumerable time series model and machine learning approaches have been developed to forecast the economic trend of a country. However, accurately predicting the trend of an economy is one of the most difficult tasks due to the highly diverse nature of all the economic indicators. This paper will use Decision Trees Based Ensemble Machine Learning models such as Light GBM, CatBoost and XGBoost, and LLM based model named Chronos to forecast GDP accurately. We have also ensembled Light GBM, CatBoost and XGBoost models to create an Ensemble GBT model. Finally, we create a hybrid model of Chronos and Ensemble GBT. We will be using the Penn World Table Datasets for our model. This dataset contains the Econometric data from 1980 to 2019 from 183 countries of the world. Our Objective is to perform a bench-marking test from our acquired datasets and compare our models. Afterward, this paper will also forecast the global GDP in the upcoming years. The paper has also used some of the traditional Time Series models like ARIMA, VAR and deep learning frameworks such as LSTM from other existing works as benchmarks. The hybrid model (Chronos x Ensemble GBT) generates enhanced predictions as it takes the best from both worlds. Across all calculated values, the model’s performance is superior to all others reflected in MSE of 6.06e+09, RMSE of 7.78601e+4, MAE of 20935.24, R2 of 0.99. The paper has huge potential in the realms of forecasting economic indicators, global GDP growth and downfall.