Forecasting Bank Credit to Small Business with Short Panels: Partial Pooling Versus Country-Specific Trends, With An Application to Uzbekistan (Academic Journal of Digital Economics and Stability)

Summary

• The study evaluates rolling-origin forecasts across 21 economies using Asian Development Bank MSME loan-stock data for 2019–2024.p.1
• At a one-year horizon, the shrinkage forecast reduces the mean absolute log error by 29% compared to a random walk baseline (0.083 versus 0.117).p.1
• The shrinkage approach cuts the one-year mean absolute log error by 18% compared to country drift (0.083 versus 0.101).p.1
• Empirical 80% intervals constructed from cross-country forecast errors achieve leave-one-country-out coverage of 76–79%.p.1
• Applying the shrinkage model to Uzbekistan projects its MSME loan stock from 93.4 trillion UZS in 2024 to 102.4 trillion UZS in 2025.p.1
• By 2028, Uzbekistan's MSME loan stock is projected to reach 134.7 trillion UZS under the shrinkage model, with an 80% interval of 109.2–183.5 trillion UZS.p.1

Key insights

1. Single-country log-trend models performed worst out of sample among evaluated methods, yielding a one-year mean absolute log error of 0.121.p.5
2. Shrinkage forecasting out-performed the random walk baseline in 17 of 21 economies at the one-year forecast horizon.p.6
3. Thailand recorded the strongest annual loan stock contraction among evaluated countries, averaging -5.4% per year.p.4
4. Sensitivity analysis indicates that setting the shrinkage weight lambda to 0.3 yields the lowest mean absolute log error at both 1-year (0.081) and 2-year (0.141) horizons.p.9
5. Uzbekistan's annual MSME loan growth fluctuated significantly between 2020 and 2024, ranging from -12.7% to +28.0%.p.10