Dynamic Panel Models: Arellano--Bond as a Frontier Pointer

Bank lending, fund flows, and balance-sheet risk are persistent: last period's outcome belongs in today's model. In a short panel, adding that lag to a fixed-effects regression makes the transformed lagged outcome correlated with the transformed error. The usual within estimator therefore suffers finite-$T$ bias that does not disappear merely because the number of entities grows. This note is deliberately a frontier pointer: it states the problem, the Arellano--Bond GMM idea, and the diagnostics an applied reader must demand before relying on it.

• write the dynamic panel specification and identify its source of endogeneity;
• explain why within FE is inconsistent when $T$ is short;
• describe first-difference GMM and lagged-level instruments at survey depth;
• state the key serial-correlation and instrument-validity checks;
• recognise when a dynamic-panel result needs specialist review rather than routine panel software.

1. The lagged outcome creates a new problem
2. The Arellano--Bond direction of travel