Order Imbalance, Crypto Momentum, and Machine Learning in Quantitative Finance
Summary
This research roundup describes several quantitative finance studies. One classifies equity trades by their short-term co-occurrence with other trades and standardizes associated order imbalances into conditional order imbalance measures. These measures correlate with contemporaneous and future returns, and a strategy based on them is reported to have performed well across a multi-year sample of hundreds of stocks. Another study reverses imbalances inferred from quarterly institutional holdings disclosures, on the premise that crowded trades may have already absorbed information and could precede short-term price reversals.
The roundup also covers transfer ranking for cross-sectional momentum when target assets have limited histories. Its encoder architecture shares information across datasets and uses attention to model asset interactions; a demonstration on major cryptocurrencies reportedly beats reference momentum benchmarks, though trading costs remain material. Other papers propose dynamic-graph forecasting models and a configurable reinforcement-learning environment for limit-order-book tasks such as execution and market making. These are summaries rather than full methodological accounts, so details of data construction, robustness, and live tradability are limited.
Key ideas
- Conditional order imbalance built from trade co-occurrence is reported to relate to current and future equity returns.
- Reversing institutional holdings imbalances may exploit crowding and delayed price adjustment.
- Transfer ranking shares information across datasets to address limited training histories in cross-sectional momentum.
- The crypto momentum demonstration reports gains over benchmarks, but trading costs affect the results.
- Dynamic graphs and model-based environments are presented as tools for forecasting and reinforcement-learning research.
Tags
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.