Juan A. Parrilla González

Type of the article: Research Article

Abstract
This study examines the predictive relationship between real non-oil economic growth and household bank deposit growth in Azerbaijan using annual data for 2000–2024. Real non-oil GDP and individual bank deposits are measured at constant 2010 prices and transformed into annual logarithmic growth rates. The empirical analysis combines conventional linear Granger causality testing with a nonlinear Kernel Granger causality approach based on radial basis function kernel ridge regression. Nonlinear predictive performance is assessed through expanding-window rolling-origin cross-validation and a circular-shift permutation test. The results indicate a modest improvement in predictive accuracy when nonlinear information is incorporated, with mean squared error declining from 402.41 to 376.16, corresponding to a 3.32% reduction in root mean squared error. Nevertheless, the Kernel Granger statistic is statistically insignificant (KGC = 0.067, p = 0.416), providing no robust evidence of nonlinear predictive causality. Conventional linear Granger testing initially suggests unidirectional causality from non-oil GDP growth to household deposit growth (F = 5.305, p = 0.032), with no evidence of reverse causality. However, introducing a 2015–2016 crisis pulse dummy substantially reduces the lagged GDP growth coefficient from 1.818 to 0.689 and eliminates its statistical significance, while the crisis indicator remains highly significant. These findings suggest that the apparent linear relationship is largely driven by the 2015–2016 macro-financial shock rather than a stable underlying mechanism. Overall, the evidence does not support a robust linear or nonlinear predictive relationship between non-oil economic growth and household deposit growth in Azerbaijan.