Using SVMs and Order-Book Features for High-Frequency Futures Trading

Summary

This research summary describes a machine-learning approach to predicting very short-term price changes from futures order-book data. It proposes using support vector machines with features such as best bid and ask prices and sizes, order-book depth, slope and relative spread, alongside open interest, volume, basis, and conventional indicators. The stated target is the next-tick price movement, with potentially actionable moves identified at a two-tick observation interval.

The reported example uses the IF1311 index futures contract. The summary says there were roughly 4,000 candidate moves on one October session and reports prediction accuracy of about 70% for moves meeting its stated price-change condition. A simulated session is described with fees, assumed slippage, 605 trades, a 56% win rate, and positive net profit. These are limited historical examples, not evidence of robust live performance: the underlying report is not reproduced, and the summary gives no broader sample, validation design, or treatment of market impact and changing conditions.

Key ideas

• The proposed model uses an SVM to estimate the next-tick price change from order-book and market features.
• Features include top-of-book prices and sizes, depth, spread measures, volume, open interest, and basis.
• The summary reports roughly 70% prediction accuracy for a specified subset of price moves.
• A single simulated IF1311 session is reported with assumed fees and slippage.
• The reported results are limited historical evidence and do not establish performance across markets or regimes.

Tags

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.