Variable Misclassification and the Ecology of Financial Crises: A Framework for Early Warning Signal Validation Across Crisis Types

The ecological early warning signal (EWS) literature and the financial stability literature have produced contradictory findings for fifteen years. Ecological theory predicts that systems approaching critical transitions exhibit rising variance and rising autocorrelation — critical slowing down (CSD) — yet empirical studies of financial crashes find rising variance without rising autocorrelation, leading to the conclusion that the ecological framework fails in financial applications. This paper resolves the contradiction. The apparent failure is a variable misclassification error. The ecological framework distinguishes two transition types: bifurcations, in which a control parameter approaches a threshold and both variance and autocorrelation rise; and noise-amplification transitions, in which stochastic forcing increases and variance rises without autocorrelation increasing. Prior financial EWS applications applied bifurcation diagnostics to state variables — equity prices and credit growth — that theory identifies as noise-amplification candidates. The correct variable for bifurcation diagnosis is the 1 control parameter: credit-to-GDP level. Applied correctly, the framework predicts the observed empirical pattern exactly. We validate the reframed framework on three datasets: daily equity prices across three US markets and three historical crashes (Guttal et al., 2016); annual macroeconomic data for 17 countries and 25 post-1970 financial crises (Jordà-Schularick-Taylor dataset); and quarterly US credit-to-GDP from 1947 to 2026 (BIS data). Across all datasets, the variable-classification prediction holds: credit-to-GDP level exhibits bifurcation signatures before credit-driven crises and no signal before non-credit crises; asset price growth exhibits noise-amplification signatures. A formal 2×2 test — level variables versus growth variables crossed with crisis type — yields p = 0.0001 for the level-versus-growth signal-rate difference. Simulation analysis under four adversarial conditions identifies a structural break confound and derives a shape-based filter that reduces false positive rates 2.9-fold. A direct comparison against GARCH(1,1) on the same 25 crisis events shows the ecological diagnostic achieves a signal-to-noise ratio of 1.63× against GARCH's 0.99×. A multi-variable composite score requiring three of four correctly classified variables to fire simultaneously achieves a 3% false positive rate at 7.1× signal-to-noise. An out-of-sample prospective test, parameterised on pre-2000 data, produces correct true negatives through the 2012–2019 deleveraging period and the 2020 COVID shock, and currently signals bifurcation in US credit-to-GDP as of early 2026.

Keywords: early warning signals, financial crises, critical slowing down, credit cycles, systemic risk, ecological transition theory