VPIN (Volume-Synchronized Probability of Informed Trading), developed by Easley, López de Prado, and O'Hara (2012), estimates the fraction of trading volume that originates from informed (private-information-holding) traders rather than uninformed (liquidity or noise) traders.
Unlike the original PIN (Probability of Informed Trading) model, VPIN uses volume time rather than clock time — computing order flow imbalance over fixed-volume buckets, which adapts naturally to varying trading intensity throughout the day.
Construction
- Divide total trading volume into equal-sized volume buckets
- Classify each trade in the bucket as buy-initiated or sell-initiated (using a bulk classification rule, e.g., Lee-Ready or tick-rule)
- VPIN for a bucket = |V_buy − V_sell| / (V_buy + V_sell)
- Compute the rolling average VPIN over the last N buckets
High VPIN signals high adverse selection risk: order flow is dominated by informed traders and market makers are likely to widen spreads and withdraw liquidity. Empirically, elevated VPIN preceded the May 2010 Flash Crash by several hours, making it a documented leading indicator of sudden liquidity deterioration.