Short-term Crypto Price Prediction Using Volume-Adjusted Mid-Price

This research paper published in December 2022 by Cornell University financial engineering students explores short-term cryptocurrency price prediction using high-frequency limit order book data from Bitstamp for Bitcoin from July 1, 2020, to September 30, 2020. The authors evaluate three mid-price adjustment estimators including Volume-Adjusted Mid-Price, trade imbalance, and quote imbalance to determine a fair price estimator for Bitcoin. The authors demonstrate that the Volume-Adjusted Mid-Price outperforms trade and quote imbalance adjusted mid-prices in predicting both short-term price direction and larger one standard deviation price movements across multiple time windows ranging from 1 second to 60 seconds. The research utilizes preliminary metrics such as volatility, mean-squared error, and trading PnL alongside binary and multiclass classification frameworks. The core finding establishes that Volume-Adjusted Mid-Price with an optimal liquidity cutoff of 50,000 to 60,000 dollars captures liquidity pressure effectively from gaps in the fragile cryptocurrency order book. While trade imbalance tends to overpredict price movements and quote imbalance remains too close to the mid-price, Volume-Adjusted Mid-Price achieves an average binary classification accuracy ranging from nearly 59 percent down to 56 percent. Practical applications of this research include lowering execution costs, developing optimal execution algorithms, and implementing proprietary trading strategies in cryptocurrency markets. Future work suggested by the authors includes expanding data breadth to capture extreme crypto crash conditions and generalizing the Volume-Adjusted Mid-Price model to other asset classes.