ARIMA vs. Hybrid Models with Machine Learning for Forecasting Ecuador's GDP.

Abstract

The analysis of Gross Domestic Product (GDP) is essential for understanding Ecuador's economic dynamics and guiding strategic decisions in contexts of high macroeconomic volatility. The purpose of the study was to estimate and forecast Ecuador's short-term GDP growth rate using robust and validated statistical models. Historical GDP series (1965–2023) obtained from the Central Bank of Ecuador were used. Stationarity tests (ADF, KPSS), correlograms, and information criteria (AIC, BIC) were applied to select appropriate ARIMA models. The analysis was performed using EViews 12, generating projections for the period 2024–2027 under optimistic, pessimistic and expected scenarios. The results showed that the Ecuadorian GDP series was not stationary at its original level, which required the application of the first difference to stabilise the mean. The identified ARIMA model incorporated autoregressive and moving average components, whose coefficients were statistically significant. The model residuals did not show autocorrelation, confirming its validity. The projections generated for the period 2024–2027 indicated moderate growth under optimistic, pessimistic and expected scenarios. These results were consistent with official estimates, validating the Box-Jenkins methodology as an effective tool for national economic forecasting. The study provides useful empirical evidence for national economic planning, validating the applicability of ARIMA models in GDP analysis. In addition, it promotes interdisciplinary approaches between economics and engineering, strengthening the technical capacity to address macroeconomic problems in contexts of high structural uncertainty.