A Computational Diagnostic Framework for Detecting Structural Break Under Small-Sample Conditions

Abstract Developing countries are often marked by changes in base years, periodic revisions in key macroeconomic indicators and unavailability of consistent back series for many indices. Consequently, researchers are frequently left with a short but homogeneous time series. Applying standard econometric break tests in such environments remains unreliable due to low-statistical power, while splicing together series from heterogeneous sources to satisfy statistical requirements mathematically introduces bias to the analysis and alters the underlying data-generation process. To overcome this limitation, this paper proposes a comprehensive computational diagnostic framework engineered specifically for small-sample settings. The framework consists of a core engine: the Ensemble Break Diagnostic (EBD) which operates as a location extractor by integrating the Zivot-Andrews unit root test with Maximum Entropy Bootstrap to map the frequency distribution of break-dates. The framework relies on a multi-layered validation structure combining an Algorithmic Stability Diagnostic based on Jensen-Shannon Divergence, a Null Monte-Carlo stress test, an Artificial Break Diagnostic, and a simulation-derived Threshold Derivation logic designed to control Type I and Type II errors. Applied to India's macro-fiscal time series (2011-2024), the framework successfully diagnoses a structural break in Real Revenue Expenditure in 2019, while Real GDP was diagnosed with a weak structural break/structural instability in 2019, and no evidence for a structural break was found in Real Total Expenditure and Real Capital Expenditure. A 20-country out-of-sample cross-validation confirms the framework's general applicability, providing the first purpose-built computational toolkit designed for detecting parameter shift in a small-sample time series with serial dependence.