Trade co-occurrence, trade flow decomposition and conditional order imbalance in equity markets

Yutong Lu, Gesine Reinert, Mihai Cucuringu

TL;DR

This paper introduces a novel method to classify individual stock trades based on their temporal proximity to other trades in the market. By decomposing trade flows into five distinct types and analyzing their "conditional order imbalance" (COI), the authors demonstrate that these decomposed signals have significant predictive power for future stock returns and can be used to build profitable trading strategies, outperforming traditional methods.

What Problem Does This Paper Address?

The modern equity market, characterized by high-frequency trading (HFT) and nanosecond-level latency, presents a complex microstructure where understanding the subtle interactions between trades is crucial for price discovery and profitable trading. While order imbalance has long been studied as a predictor of price movements, existing methods often treat all trades as homogeneous. This paper addresses the challenge of extracting more granular and informative signals from trade flows by recognizing that trades don't occur in isolation but are influenced by their temporal context within the broader market. The authors aim to uncover how the "co-occurrence" of trades—trades happening close in time—shapes their impact on prices and predictability.

Methods

The core innovation is the concept of trade co-occurrence, defined by a neighborhood size (δ), which groups trades that happen within a specified time window. This allows for a trade flow decomposition into five categories:

  • Isolated (iso): Trades with no other trades in their δ-neighborhood.
  • Non-isolated (nis): Trades with at least one other trade in their δ-neighborhood. This category is further broken down into:
  • Non-self-isolated (nis-s): Only trades of the same stock in the neighborhood.
  • Non-cross-isolated (nis-c): Only trades of different stocks in the neighborhood.
  • Non-both-isolated (nis-b): Trades of both the same and different stocks in the neighborhood.

The paper then defines Conditional Order Imbalance (COI) for each of these five trade types. This allows for a more nuanced analysis of price impact and return predictability compared to using a single, aggregated order imbalance measure. The authors empirically select the optimal neighborhood size (δ = 1 ms) by maximizing the difference between empirical and null (random arrival) co-occurrence probabilities.

Datasets

The study utilizes limit order book data from the LOBSTER database for 457 US stocks listed in the S&P 500 index, covering the period from 2017-01-03 to 2020-12-31. This data provides detailed information on limit order submissions, cancellations, and executions at nanosecond precision. Daily price data was obtained from the Center for Research in Security Prices (CRSP) database. Factor data (market, size, value, profitability, investment, momentum) was sourced from Kenneth R. French's online Data Library.

Key Insights and Ideas

  • Temporal Proximity Matters: Trades are not isolated events. Their timing relative to other trades, both within the same stock and across the market, carries significant information.
  • Decomposition is Key: Simply looking at overall order imbalance masks important distinctions. Breaking down trades into types based on co-occurrence reveals different price impact and predictive properties.
  • "Isolated" Trades are Special: COIs from isolated trades (iso) show a strong positive correlation with contemporaneous returns and are particularly predictive of future returns, suggesting they might represent informed trading activity.
  • Interactions Reveal Predictability: While aggregated order imbalance loses predictive power for future returns over time, decomposed COIs, especially "iso" and "nis-s" (non-self-isolated), retain significant predictive signals.
  • Trading Strategies Profit: Trading strategies built on these decomposed COIs, particularly long-short portfolios, generate substantial profits and Sharpe ratios, demonstrating the economic value of this granular approach.

Key Results

  • Contemporaneous Impact: All types of COIs show a positive and significant relationship with same-day returns. The 'iso' COI yields the highest adjusted R² (5.91%) in contemporaneous regressions, comparable to the unconditional order imbalance despite representing fewer trades.
  • Predictive Power: 'Iso' and 'nis-s' COIs are positively related to future returns, while 'nis', 'nis-c', and 'nis-b' COIs are negatively related. The adjusted R² for predictive regressions using decomposed COIs is higher than using only control variables or the unconditional order imbalance.
  • Portfolio Performance: Long-short portfolios based on COIs achieve significant annualized returns and Sharpe ratios. The 'iso' single-sort portfolio yields an annualized return of 6.76% with a Sharpe ratio of 1.20. Double-sorted portfolios, like 'iso/nis-c', achieve even higher returns (34.87%) and Sharpe ratios (1.79), significantly outperforming benchmarks like the SPY ETF and momentum strategies.
  • Robustness: The findings are robust across different neighborhood sizes (δ) and market indices (S&P 100, Dow 30), although performance can vary.

Limitations and Future Work

  • Trader Identification: The paper acknowledges the difficulty in identifying who initiates trades, which limits the direct interpretation of the generating mechanisms behind each COI type. Future work could leverage private data to distinguish trader types (e.g., informed vs. HFT).
  • Market Index Choice: The classification of trades can be influenced by the choice of the market index (M). While robust across S&P 500 and S&P 100, using a smaller index like the Dow 30 led to deteriorated forecasting power.
  • Threshold Hyperparameter: The current classification of "non-isolated" is binary. Future research could explore a threshold hyperparameter to better distinguish noise trades.
  • Other Market Events: The analysis primarily focuses on trades (executions of limit orders). Future work could extend the co-occurrence analysis to other limit order book events like submissions and cancellations.
  • Data Granularity: Computational constraints limited the analysis to daily COIs. Future work could explore intraday COIs for even finer-grained insights.

Summary updated

Abstract

The time proximity of high-frequency trades can contain a salient signal. In this paper, we propose a method to classify every trade, based on its proximity with other trades in the market within a short period of time, into five types. By means of a suitably defined normalized order imbalance associated to each type of trade, which we denote as conditional order imbalance (COI), we investigate the price impact of the decomposed trade flows. Our empirical findings indicate strong positive correlations between contemporaneous returns and COIs. In terms of predictability, we document that associations with future returns are positive for COIs of trades which are isolated from trades of stocks other than themselves, and negative otherwise. Furthermore, trading strategies which we develop using COIs achieve conspicuous returns and Sharpe ratios, in an extensive experimental setup on a universe of 457 stocks using daily data for a period of 4 years.
BibTeX
@inproceedings{lu2022trade,
  author = {Yutong Lu and Gesine Reinert and Mihai Cucuringu},
  title = {{Trade co-occurrence, trade flow decomposition and conditional order imbalance in equity markets}},
  booktitle = {Quantitative},
  year = {2022},
}
Figure