Volume Impact

Quantitative trading algorithm that investigates how visible (public) order-book volume translates into realized price impact, and when the displayed book cannot be trusted. Using the broker’s L2 feed as the observation surface, the work identifies integrity failures and “leakage” patterns where published depth/queue state can be incorrect, stale, or systematically bypassed. It characterizes the conditions under which apparent priority can be jumped, and how these anomalies distort impact estimates and fill expectations. The output is meant to be operational, not only descriptive: the project converts LOB/infrastructure pathologies into detectable conditions (awake/asleep states, anomaly flags, reliability scores) that can drive event-based logic-filtering signals, throttling aggression, or switching execution modes. Learning is framed as an online/local regression problem to adapt impact parameters under partial observability, and the framework is used to trade announcement-driven moves with heavy-tail-aware sizing.