Conventional trade surveillance solutions employ static rules or conditions. They apply these rules to each trade and flag those which meet every condition as an alert. Although this ‘rules-based’ approach can support an automated surveillance system, it is fundamentally ill-suited to the scale and complexity of financial markets. In response, TradingHub developed MAST to overcome the many pitfalls of this legacy technology.
High false positive rate
Arguably the most significant drawback is accuracy. The vast majority of alerts generated by rules-based systems are false positives. This is because binary rules are incapable of covering all possible market behaviours. When combined with the colossal scale of financial markets, the results can be hundreds of thousands of alerts.
To overcome this problem, providers of legacy systems have integrated AI and machine learning. These models ‘learn’ from previous resolutions to determine which new alerts are not likely to be genuine. Although these innovations can reduce false positives, they do not necessarily increase the accuracy of their solutions. This is because each alert – occurring in different instruments with its own dynamics – is fundamentally unique. These models mistakenly assume similarities between alerts when in fact underlying market conditions have changed.
MAST adopts a fundamentally different approach to trade surveillance. It models the market impact of each individual trade by considering the dynamics of each instrument, including volatility and liquidity. This ensures that its alerts only capture instances where a trader could realistically have driven the price.
Inability to detect cross-product abuse
Rules-based approaches apply their rules to individual trades. As they do not have any intrinsic understanding of correlation between different instruments, they are unable to detect cross-product abuse. To overcome this deficiency, with some rules-based providers firms ask their customers to maintain pairwise mappings of correlated instruments. However, this is almost impossible to perform for all instruments, especially in the OTC derivative space, and does not establish the degree of correlation. The result is fragmented and inaccurate cross-product coverage.
MAST employs a risk-based approach which models what risk position a trader takes across all their trades. For example, for fixed income instruments it will model how a trader positions themselves along a yield curve across all of their trades. This enables it to natively detect abuse involving different instruments, irrespective of venue.
No understanding of abuse severity
Legacy systems employ binary rules. Where a trade meets all the rules, the system generates an alert. They do not measure the degree to which a trade meets or exceeds a rule. This means there is no capacity for quantifying an alert’s severity and that ultimately each alert is equal. This drawback compounds the issue of false positives in that a compliance team can potentially face hundreds of thousands of alerts, without knowing which alerts are the most serious.
MAST models the market impact of each trade whilst considering the dynamics of each instrument. This means it ultimately produces a USD Value score for each alert, which is an estimate of the trader’s intended gain. This score is standardised across all asset classes and abuse typologies. This lets analysts prioritise the key alerts across the entire business for prompt investigation.
Constant recalibration
Legacy approaches use fixed rules which reflect market conditions at the time that they are applied. However, markets can be both highly volatile and cyclical. Inevitably therefore, these rules become ill-suited to prevailing market conditions. One outcome is that they become too ‘attuned’ and overwhelm trade surveillance teams with even more alerts, such as during the beginning of the Covid-19 pandemic in March 2020. The other is that they become less sensitive and let genuine abuse go undetected. As a result, these systems require regular recalibration. Given the number of rules, this involves significant and continuous resource on the part of trade surveillance teams.
MAST avoids the need for rules by modelling the market impact of each trade whilst considering the asset’s dynamics, such as liquidity and volatility. This lets it automatically tailor its parameters to each individual instrument. Crucially, MAST uses its own market data sources in these calculations. This enables it to automatically adjust its thresholds every day without requiring intervention. As well as increasing the accuracy of MAST’s analysis, this significantly reduces maintenance costs.
Conclusion
Rules-based systems inherently struggle to deal with the complexity of financial markets. This is manifest both in terms of system maintenance, which involves regular intensive recalibration, and the quality of its analysis. These systems have high rates of false positives, no understanding of the severity of each alert and no comprehensive way of detecting cross-product abuse. Though some vendors have bolted on newer features such as machine learning, these cannot fully compensate
for their fundamental drawbacks. Instead of rules, MAST uses models. This enables it to self-calibrate its thresholds every day to reduce false positives and quantify abuse, whilst minimising maintenance costs. Additionally, it also permits a risk-based approach where MAST is able to understand a trader’s position across different instruments.