Understanding Extreme Price Movements in Large-Cap NASDAQ Equities: A Microstructure and Liquidity-Focused High-Frequency Analysis

Abstract

This master thesis examines intraday Extreme Price Movements (EPMs) as liquidity-driven phenomena and asks two questions: which signals most reliably flag an imminent 10-second EPM, and whether liquidity-based information adds incremental predictive power beyond standard market-state controls. Using LOBSTER limit orderbook data for the ten largest NASDAQ-listed equities, we reconstruct prices on a 10-second grid, label EPMs at the 99.9th percentile of the distribution of absolute mid-quote returns, and estimate parsimonious logistic models that layer liquidity signals on top of market-state predictors. Elastic-Net and unpenalized logistic regression specifications are evaluated both in-sample and out-of-sample, with January to October 2020 used for model estimation and November to December 2020 for model evaluation. We find that selectively incorporating a compact set of liquidity signals consistently improves short-horizon EPM detection and probability calibration relative to the market-state benchmark. Across assets and specifications, the most reliable precursors to EPMs are: recent absolute returns as a volatility measure, sustained trading intensity, multi-level spread deterioration (loss of market tightness), and execution pressure captured by fill rates. Together, these results indicate that EPM risk is not mere noise but the joint outcome of market-state conditions and liquidity dynamics. These indicators enable better, real-time detection of windows of heightened EPM probability, supporting better-informed risk management for practitioners, regulators, and market participants.