RED-2400: Benchmark of Rejected Trading Events

A rare public dataset providing labels for algorithmically rejected trades on a Solana DEX. This benchmark allows researchers to analyze false-positive filter events and the 'cost' of missed opportunities in low-latency trading.

Strategy Decoder Editorial · 2026-05-12 · 3 min read

Key takeaways

RED-2400 provides unique data on algorithmically rejected trades, unlike most datasets focused solely on accepted trades.
The dataset, sourced from a Solana decentralized exchange (DEX), includes detailed outcome labels for these rejections.
It helps quantify the 'cost' of algorithmic filters by classifying rejections as 'saved' (good rejections) or 'missed' (bad rejections).
Quant traders can use RED-2400 to validate and refine their proprietary trade filters, reducing false positives and increasing precision.
The benchmark allows for a deeper analysis of market microstructure events surrounding rejection points and their long-term implications.

Algorithmic trading systems rely heavily on robust decision-making processes, yet the efficacy of their internal filters—mechanisms designed to prevent undesirable trades—often remains opaque concerning their true performance. While datasets of executed trades are plentiful, information on specific rejected trades and their potential outcomes is scarce. A new public benchmark, RED-2400, directly addresses this gap by offering a unique window into algorithmically rejected trading events from a live Solana decentralized exchange (DEX).

Unveiling the Hidden Costs of Algorithmic Filters

Traditional datasets used in quantitative finance typically focus on accepted trades, providing a biased view of filter validation. Filters are designed to prevent bad trades, but they can also mistakenly reject good ones, incurring an opportunity cost. Understanding this 'reject side' is crucial for developing more precise and profitable trading algorithms. The RED-2400 dataset, as reported by Arati U. Kamat, logged continuously over several weeks in early 2026, comprises 6,659 rejection events. Each rejection is richly contextualized with over 169,000 subsequent price and liquidity observations, alongside snapshots of order book depth, providing a comprehensive picture of the market conditions immediately following a rejection.

Detailed Outcome Labels for Deeper Analysis

A standout feature of RED-2400 is its use of a five-tier classification system for rejection outcomes. These outcomes include:

Saved (windowed): The rejection successfully prevented a trade that would have resulted in a loss within a defined time window.
Saved (early-death): Similar to 'saved (windowed)', but the loss materialized very quickly after the hypothetical trade.
Missed: The rejection prevented a trade that, in hindsight, would have been profitable.
Flat: The market remained relatively unchanged after the rejection, meaning neither a significant gain nor loss was avoided or missed.
Unclassifiable: Events where the outcome could not be definitively determined.

This outcome labeling uses specific thresholds based on trough-to-reference and peak-to-reference price ratios within a 24-hour window, providing a quantifiable basis for assessing filter performance. By categorizing rejections this way, researchers and traders can move beyond simply knowing a trade was rejected to understanding why that rejection was beneficial or detrimental.

Bridging the Gap in Filter Validation

The ability to analyze missed opportunities (false positives) is extremely valuable. While a filter might successfully block many potentially losing trades, if it also frequently blocks profitable ones, its net benefit diminishes. RED-2400 enables a direct empirical assessment of filter precision, allowing for the replication and validation of claims about filter effectiveness. Furthermore, the dataset's public availability under a CC-BY-4.0 license, along with companion code, promotes transparent research and collaborative development in this critical area.

Implications for Market Microstructure and Strategy Refinement

The granular detail of market observations linked to each rejection event also opens avenues for advanced market microstructure analysis. By examining price, liquidity, and order book dynamics surrounding these rejections, quants can gain insights into:

The nature of adverse selection: How often do filters successfully protect against trades initiated by more informed participants?
Latency effects: How do microsecond market movements around a rejection point impact its outcome classification?
Liquidity dynamics: Does a rejection event correlate with immediate shifts in order book depth or bid-ask spread?

This kind of detailed data is a cornerstone for refining execution algorithms, enhancing risk management frameworks, and optimizing strategy parameters to minimize opportunity costs while maximizing protection against unfavorable market conditions. The paper notes that RED-2400 is the initial release in a planned series, with future iterations extending the time horizon for even more robust, regime-stratified analysis.

Why it matters for algo traders

For algorithmic and quantitative traders, RED-2400 offers an unprecedented resource. It provides the data needed to move beyond theoretical models of filter performance to empirical validation. Traders can leverage this dataset to:

1. Backtest and optimize rejection logic: Evaluate how different filter parameters would have performed based on historical rejections, not just accepted trades. 2. Quantify opportunity costs: Understand the true financial impact of false positives (missed trades) and adjust risk models accordingly. 3. Improve signal-to-noise ratio in models: By understanding the characteristics of truly 'bad' trades, algorithms can be fine-tuned to react more effectively to genuine market threats while avoiding over-filtering. 4. Develop proactive mitigation strategies: Insights from rejection outcomes can inform the design of algorithms that not only avoid bad trades but also potentially recover from or re-engage after initially rejected opportunities if conditions improve.

This public benchmark is a significant step forward in bringing transparency and empirical rigor to the often-overlooked 'reject' side of algorithmic trading, empowering quant desks to build more intelligent and resilient trading systems.

Frequently asked questions

What is the RED-2400 dataset?

The RED-2400 dataset is a public benchmark containing 6,659 algorithmically rejected trading events from a live Solana decentralized exchange, logged between April and May 2026. It includes detailed post-rejection market observations and outcome labels.

Why is data on rejected trades important?

Data on rejected trades allows researchers and traders to evaluate the true precision of algorithmic filters by analyzing both successful rejections (saving capital) and false positives (missed opportunities), which is critical for optimizing automated trading systems.

How are rejection outcomes classified in RED-2400?

Rejection outcomes are classified into five categories: 'saved (windowed)', 'saved (early-death)', 'missed', 'flat', and 'unclassifiable'. This classification is based on subsequent price movements relative to predefined thresholds over a 24-hour window.

Tags: dataset, dex, solana, algorithmic trading

Based on reporting by arXiv q-fin.TR.