Наукастинг и краткосрочное прогнозирование макроэкономических показателей развивающихся стран

Экономика и математические методы. 2026. Т. 62. № 2. С. 58–72.

Makeeva N.

This study analyzes the accuracy of nowcasting and short-term forecasting models for annualized quarterly GDP growth rates in 33 developing countries over the period from Q1 2013 to Q4 2023. The research evaluates the out-of-sample accuracy of various models (MIDAS, MFBVAR, DFM, regularization-based models, a classical pairwise regression and first-order autoregression model) using the last 12 data points (3 years). The results demonstrate that MIDAS models, particularly modifications with Almon exponential lags and constraints based on the Gompertz distribution, achieve the highest accuracy for 70% of the countries due to their flexibility in aggregating monthly data. It was also found that the key factor in accuracy is the consideration of economic structure characteristics: resource-dependent countries achieve minimal errors through commodity and export indicators, while diversified economies rely on financial metrics. Under high volatility, simpler methods (autoregression and regularization) outperform complex models, reducing the Mean Absolute Error (MAE) by up to 58% due to their robustness to noise. Increasing the forecast horizon leads to a 30–50% rise in errors. The study reveals a strong correlation between GDP volatility and the average absolute forecast error, highlighting the challenges of forecasting under unstable conditions. The results provide a foundation for adaptive forecasting systems relevant to central banks and analytical agencies in the context of global economic uncertainty.

Keywords: развивающиеся страны, developing countries, макроэкономические показатели, краткосрочное прогнозирование, nowcasting, macroeconomic indicators, short-term forecasting, наукастинг

Publication based on the results of: Development of a system of models for scenario forecasting of economic growth and key macro-parameters of the Russian economy, including the use of expert survey data (2025)