Typology of macroeconomic disbalances in post-war recovery: justifying the number of clusters and the stability of the classification

Abstract

This article asks whether a cluster typology of post-war recovery trajectories is stable enough to carry the weight placed on it, and separates two claims such typologies conflate: that the episodes are heterogeneous, and that a given episode belongs to a given cluster. Design/methodology/approach. We cluster 50 recovery episodes, 18 of them post-conflict, described by nine disbalance indicators over a five-year stabilisation window, using Ward’s linkage on standardised annual World Bank series. The unit of observation is a country-window pair. Rather than reporting a single partition, we examine the silhouette profile across k from three to seven, the cophenetic correlation, and the agreement with a k-means solution. Findings. The silhouette coefficient rises from 0.134 at k = 3 to 0.196 at k = 6, then falls to 0.192 at k = 7, so six clusters is the last split that still improves separation. The cophenetic correlation is 0.735 and the agreement with k-means is 56 %; all three diagnostics are moderate, supporting the existence of structure but not the assignment of borderline episodes. An external check shows the partition does not reproduce shock type (adjusted Rand index 0.059), and post-conflict episodes differ from post-crisis ones on only two of the nine indicators, the trade balance and the current account. Originality/value. The contribution is a two-level reading of cluster validity, with a decision table linking each quality metric to the class of inference it supports, and a correction rule for transferring the benchmark profile to post-war Ukraine.

JEL Classification: C38, E32, O11, F44.