Analysis and Forecasting of Azerbaijan’s Balance of Payments Based on VECM and Comparison with the ARDL Model

Abstract

The aim of this study. To empirically establish long-run cointegrating relationships between the main determinants of the current account using the Engle–Granger two-step Vector Error Correction Model (VECM), to quantify the speed of the short-run error correction mechanism, to evaluate the diagnostic reliability of the model, and to construct a medium-term forecasting framework encompassing oil price scenarios. Materials and methods. The study is based on a dataset of 96 quarterly observations covering 2001Q1–2024Q4, drawn from the official macroeconomic statistics of the Central Bank of Azerbaijan [1]. The Engle–Granger (1987) two-step procedure [2] was applied: in the first step, a long-run OLS regression was estimated; in the second step, an ADF test on the residuals [3] independently confirmed cointegration (t = −4.784, p = 0.000). Short-run dynamics were then estimated through the VECM equation. Diagnostic validity was assessed using the Breusch–Godfrey LM, Jarque–Bera, White, and Durbin–Watson tests. All computations were carried out using the EViews 12 software package. Results. The ECT coefficient of the VECM (−0.2929, t = −3.119, p = 0.003) indicates that approximately 29% of each quarter's equilibrium deviation is automatically corrected, with full restoration requiring roughly 3.4 quarters (~10 months). The long-run regression confirmed the dominant role of oil and gas exports (+1.167, p < 0.001), the structural burden imposed by FDI through the repatriation channel (−0.880, p < 0.001), and the adverse structural effect of foreign trade via import growth (−379.47, p < 0.001). The model demonstrated high explanatory power: R² = 0.897, F = 92.61***. The dynamic forecast outperforms the naïve benchmark (Theil U = 0.159); under the baseline scenario (Brent ≈ USD 75–80/barrel), a positive current account balance is projected to be sustained throughout 2025–2026. Conclusion. This study completes the methodological link that complements the preceding ARDL-based work [4]: the ECT coefficients of both models are negative and statistically significant (ARDL: −0.396; VECM: −0.293), and the sign structure of the long-run coefficients is identical — demonstrating that two independent methodological approaches reflect the same underlying economic reality. The scenario-based forecasting framework provides an empirical foundation for the diversification targets embedded in Azerbaijan's 2030 Strategic Roadmap.