Illiquidity at Risk

Battle Count: 50

Highly relevant for quantitative trading, particularly for strategies sensitive to liquidity conditions. The IlliQaR metric provides a principled framework for quantifying tail liquidity risk, which is critical for: (1) execution algorithms that need to anticipate liquidity dry-ups, (2) market-making strategies that must manage inventory risk during stress, (3) portfolio construction where liquidity-adjusted risk measures are essential, (4) risk management systems that need to account for the cost of unwinding positions during crises. The finding that jumps are critical for short-term forecasts and that the S&P 500 acts as a leading indicator for individual stock liquidity stress has direct trading implications. The asymmetric effect (liquidity evaporates more during downturns) is actionable for tactical allocation. However, the paper is primarily academic and does not provide a complete trading system or backtest.

Moderate to high complexity. Requires: (1) High-frequency data processing to compute realized Amihud from 5-minute returns and volume; (2) Implementation of multiple MEM specifications with Gamma-distributed errors; (3) Jump identification via compound Poisson process with time-varying intensity; (4) Computation of conditional CDF as mixture of Gamma and Kappa distributions (requires modified Bessel functions); (5) Maximum likelihood estimation with parameter constraints; (6) Berkowitz test implementation for tail coverage; (7) Model Confidence Set procedure with QLIKE loss; (8) Logit regression for economic determinants. The mathematical derivations (Proposition 1) and the need for numerical integration of the Kappa CDF add implementation difficulty. However, the econometric framework builds on well-established literature (Engle 2002, Corsi 2009, Caporin et al. 2017).

The paper uses publicly available data sources (Oxford-Man Institute for realized volatility, Datastream for trading volume). The econometric models (MEM, HAR, jump specifications) are well-documented in the literature. However, the specific implementation of the realized Amihud measure, the jump identification procedure, and the Berkowitz test implementation details would require careful replication. The Supplementary Document provides additional results but no code is mentioned. The 5-minute sampling frequency and specific data processing steps are described but not fully reproducible without access to the raw high-frequency data.