VWAP Participation Algorithm for Cryptocurrency Markets

Overview

This project implements a simple VWAP (Volume-Weighted Average Price) participation algorithm for cryptocurrency markets. The algorithm simulates position accumulation while attempting to minimize trading costs in electronic markets, using high-frequency cryptocurrency trading data from 2024. Nanosecond level data was pulled from our professor at the University of Chicago. If you have any questions please email me at kunjs@uchicago.edu

Data Structure

The project utilizes high-frequency trade and order book data with the following formats:

Trade Data

Each record contains:

• Received UTC nanoseconds
• Timestamp UTC nanoseconds
• Price (in millionths)
• Size (in billionths)
• Side (-1 or +1, representing the trade direction)

Order Book Data

Each record contains:

• Ask/Bid prices at multiple levels (in millionths)
• Ask/Bid sizes at multiple levels (in billionths)
• Received UTC nanoseconds
• Timestamp UTC nanoseconds
• Mid price

Algorithm Specifications

The VWAP participation algorithm:

• Inputs:

  • Target quantity (Q, positive for buying, negative for selling)
  • Start time (τs)
  • Target participation rate (p)

• Key Parameters:

  • Minimum size threshold (g) - representing unlikelihood of being first in queue
  • Quoting participation rate k(p) - necessarily larger than p (max 5%)
  • Pause duration (P) - ranges from 0.05 to 5 seconds

• Conservative Assumptions:

  • Only trades with opposite Side are available for participation
  • For each price level, accumulation size = max(0, (total_level_quantity - g) × k)
  • Algorithm pauses for P seconds after participating in a flurry of trades

Transaction Fees

• 50 basis points (0.5%) for trades between crypto-tokens and traditional currencies
• 10 basis points (0.1%) for trades between crypto-tokens

Implementation Notes

• The algorithm is post-only (passive)
• Target quantity (Q) selection can be based on quantiles of 5-minute volumes
• Minimum size threshold (g) can be derived from the 5th percentile of trade sizes
• The algorithm accommodates both buying and selling scenarios