IMPROVING FORECASTING OF AGRIBUSINESS DEVELOPMENT BASED ON GARCH MODELS AND INTERNET OF THINGS (IOT) TECHNOLOGIES

Abstract

Based on the analyzed aspects, forecasting the development of agribusiness using IoT is an essential and necessary process in the modern agricultural sector. It is argued that the application of IoT technologies significantly enhances agribusiness efficiency through resource optimization, process automation, and improved risk management. It has been established that GARCH models are powerful tools for modeling and forecasting conditional volatility in time series, particularly regarding agricultural product prices and yields. However, it is demonstrated that GARCH models have several limitations, including complexity, stationarity requirements, limitations in predicting extreme events, linearity, assumptions of normally distributed residuals, high data requirements, limited interpretability, lack of external factor integration, and delayed response to changes. To address these limitations, an improved model incorporating a comprehensive forecasting approach is proposed. It is substantiated that using AIC and BIC criteria to select the optimal number of lags simplifies the model and reduces overfitting risks. Additionally, performing the Dickey-Fuller test ensures the time series meets stationarity conditions. Extended models such as EGARCH are employed to account for asymmetric effects and shocks, improving extreme event prediction. Nonlinearity in data is addressed by integrating machine learning models, specifically LSTM, capable of capturing complex relationships. Alternative residual distributions, such as t-distribution or GED, overcome the assumption of normality. Methods of transfer learning mitigate the requirement for large data volumes, enabling improved modeling on smaller datasets. Limited interpretability is addressed through result visualization and simpler models for initial analysis. Additionally, external factors like weather conditions are integrated using GARCH-X models, ensuring a comprehensive forecasting approach. The delayed response to changes is resolved by applying adaptive GARCH model versions that update parameters in real-time. Thus, it is justified that employing a comprehensive modeling approach combining traditional GARCH models, modern machine learning methods, and external factors provides more accurate and reliable forecasts for IoT-based agribusiness development.