Financial Markets
Vuk Magdelinic - Overbond

A medical data science technique could help solve a complex bond automation problem should ESMA ban the use of requests for quotes in fixed-income markets. By Vuk Magdelinic, CEO of Overbond

The European Securities and Market Authority (ESMA) Call for Evidence on Pre-hedging spells trouble for fixed-income. ESMA is concerned that a request for quote (RFQ) for a bond trade can be considered inside information if the process entails pre-hedging of the possible transaction. 

Traders could be liable for multimillion-euro fines if RFQs are deemed to be insider information. This would effectively end the use of RFQs for bond trading. Since many major aggregated bond data sources depend on data from RFQs, these data sources would also be heavily affected and of less value, which could see many trading desks flying blind without the full picture of the markets. 

As fixed-income markets stand on the edge of a potentially damaging regulatory change, what can be done to ensure major risk is not brought into the trading of corporate bond markets? This question is crucial now given the market’s significant volatility and illiquidity.

The answer will likely involve automating bond trading. But this is a complex problem. An automated trading algorithm must arrive at the best executable price for a bond while accounting for the uniqueness of the bond issue, market conditions and the preferences of the trader. This includes their approach to margin; in other words, the price or spread difference between where the trade is executed and the next best quote by a competing sell-side bond desk. 

One approach to incorporating margin into automated trading is to train an artificial intelligence (AI) model to the risk tolerance and execution style of the desk by looking back at the margin of all prior trade inquiries processed by the desk, including those that were accepted, rejected and traded away. Once the model is trained, it can be used to optimise the margin for all new trade inquiries as they occur. 

When this approach to margin modelling was first developed, the developers assumed that all market risk was priced into the best executable price and that margin was a second layer related only to the singular trading style of the book managed by the individual trader. Therefore, it was assumed that margin could be added on after determining the best executable price. 

Incorporating complexity

Empirical evidence, however, shows that traders adjust their margin according to the level of risk in the market at the time of trade; therefore, this layer of complexity needs to be incorporated into the margin model. This requires separating the impact of market conditions on historical margin levels, which entails isolating the effects of individual risk categories that were reflected in the price of a bond at the time of the historical trade. 

Bond prices reflect country, sector, issuer and issue-specific risks. Separating these risks and isolating market conditions over time is a complex mathematical problem. To solve this, developers have turned to the data science used to study medical outcomes and drug testing, known as case-mix adjusted cluster (CMAC) analysis. 

Cluster analysis is a technique that groups similar observations into a group, or cluster. For example, all bonds whose prices rise when oil prices fall might be lumped together irrespective of their other characteristics. But there are hundreds of dimensions that affect bonds. Bonds that are sensitive to oil prices, for example, could include bonds from airlines, manufacturing or trucking, with multiple maturities and different ratings. The number of clusters of like attributes can quickly become staggering.

CMAC analysis has provided the modelling breakthrough needed to handle the many dimensions of characteristics and risk affecting bonds. CMAC analysis adjusts for differences in populations and population size before clustering results according to certain common variables. In medicine, this might mean adjusting post-surgery mortality rate data to account for the comparison between a large hospital and a small hospital. In the bond market, this might mean adjusting for call features or bond tenor before clustering results according to spread movements.  

The automated trading AI algorithm uses CMAC analysis to remove pricing movements related to risks at the issuer level and isolate the security-specific risk demonstrated in the pricing movement. The main success of applying CMAC analysis is the ability to isolate those attributable pricing movements versus the broader market purely algorithmically.

When training the model, CMAC is applied to a historical series of post-trade data from the desk to determine if trades were made in a normal, heightened, or low-risk environment, and this allows the model to discern the margin patterns attributable to these markets and adjust according to current market conditions. Margin in the automated trading system becomes dynamic and adjusts to current market conditions as the trader would. And because CMAC analysis clusters attributes from multiple data sources, it will still work without RFQ data, and it will still facilitate automated trading.

A Speaker’s Corner: A Speaker’s Corner is an area where open-air public speaking, debate and discussion are allowed. The original and most noted is in the north-east of Hyde Park in London

Visit for more articles & podcasts on regulatory topics