Development of a macroeconomic forecasting method for post-crisis economies using panel data of structurally similar analogue countries
Abstract
The object of research is the methodology of macroeconomic forecasting for post-crisis economies using panel data of structurally similar analogue countries. The problem addressed is the inability to build reliable macroeconomic forecasts for Ukraine under structural breaks using standard single-country models. A structured forecasting method is proposed. An algorithm for automated selection of analogue countries based on a combined similarity measure is proposed. The measure integrates normalized Euclidean distance and mean Pearson correlation. Selection is performed across 6 macroeconomic indicators. From a pool of 17 candidate countries, six optimal analogues for Ukraine were selected: France, the Czech Republic, Latvia, Austria, Estonia, and Mexico. A panel dataset comprising 1,227 quarterly observations was assembled, based on which quantile gradient boosting models were trained for three scenarios – lower, baseline, and upper quantiles. The model was validated using an eight-quarter out-of-sample holdout applied separately to each country in the panel. Accuracy metrics were calculated on the Ukraine-only holdout sample; for some parameters, on the pooled panel test sample, as the number of Ukrainian observations was insufficient for a reliable benchmark comparison. The mean absolute error for unemployment growth was 0.26 percentage points (pooled panel sample), for the lending rate – 0.33 percentage points, and for the real effective exchange rate – 1.75 index points (Ukraine-only holdout). The twenty-quarter baseline forecast for Ukraine projects moderate consumer price index growth at an average annual growth of 5–6%. Quarterly nominal GDP at current prices is forecast to grow from 2.35 to 3.5 trillion UAH, which is equivalent to 53–80 billion USD. The unemployment rate will gradually decline but will not return to pre-war levels. The methodology is applicable to forecasting other post-crisis economies with structurally interrupted time series.