Herding and Liquidity in Order-Book Markets. III. Leverage and the Onset of Endogenous Liquidity Crises under Weak Anchoring

Battle Count: 75

The paper provides important theoretical insights into when and how endogenous liquidity crises emerge in order books, which is directly relevant to market-making strategies, execution algorithms, and risk management. The finding that the margin buffer (jointly set by leverage and maintenance margin) governs crisis onset has practical implications for position sizing and leverage limits. However, the model is highly stylized (single asset, simple ABM), uses short-memory clustering, and is not calibrated to specific market data, limiting direct quantitative trading application. The phase-diagram approach is more useful for understanding regime boundaries than for generating trading signals.

The agent-based model is deliberately simple (single CDA, four agent populations, ~10 market makers, ~30 holders) and the code is publicly available. However, reproducing the full experimental suite (multiple sweeps across anchor, leverage, herding, and maintenance margin with 8-20 seeds per cell, long runs of 32,000 events) requires significant computational resources. The statistical analysis (Hill estimator, ACF computation, Fano factor, Wilson intervals, z-scores) adds moderate complexity. The conceptual framework (matched-herding increment, margin buffer collapse) requires careful implementation to avoid confounding channels.

Code and data reproducing all results, figures, and sweeps are available in an accompanying GitHub repository. Companion studies (Parts I and II) also have public repositories. However, the model relies on specific calibration points (fundamental-value process, market-maker and holder capital, population sizes) whose robustness is not yet mapped. Seed counts of 8-20 per cell may limit statistical power near onset. The buffer collapse is established at a single onset cell and not re-verified across the full parameter plane.