Applying Behavioural Analysis to Assess Trading Activities
Background
An audit team was preparing for an assessment of the risk and operational controls of a large and active trading desk. The rates desk under examination included several traders, each with a handful of books where they recorded their transactions. As well, the desk was using three different trading systems, each for a different class of products. Inputs for valuation were provided by traders, while a specialized group in the back office independently verified and corrected them. As in other institutions, an independent risk management group measured risk and produced a detailed set of market, counterparty, and liquidity risk indicators.
The Challenge
The main challenge for the audit team in assessing the rates desk was the massive amount of data composed of data points coming from different sources. As well, data objects (transactions, for example) were not static over time. A transaction (“trade” or “deal”) may be amended after its first inception in the database and pricing intpus (swap rates, implied volatilities, and exchange rates, for instance) may also be corrected. As a consequence, the valuation of the transactions changed not only because of market movements but also because transactions and pricing inputs changed.
Our Approach
Based on discussions with the audit team, we performed analysis at different levels of aggregation: rates desk, individual traders, and individual books. The desk view provided the team with information about its overall risk and profitability, trends in pricing, liquidity risk profiles and concentration on certain counterparts. We also built behavioural profiles for traders, which included not only risk and profitability but also trade cancellations and amendments, among other operational risk indicators. Traders would use different books for different strategies; therefore, our book-level analysis gave auditors a profile for each strategy.
These analyses allowed the audit team to describe the activities at different levels of detail, but we were also able to provide audit sample of atypical transactions (as compared to other trades at desk-level, and also at book-level) and atypical behaviour (at trader level).
The Benefit
The audit team was able to use a complex data set that included transaction information, pricing, and risk indicators (including, operational risk). In addition, the analysis allowed auditors to observe the joint evolution of these observations over time. Using anomaly detection methods, including breakout detection, we provided the audit team of samples of atypical transactions, atypical behaviour, and sustained changes in behaviour (sustained increase in a particular type of risk measure).
Contact us
