header image

overbond logo

How Fixed Income AI Margin Optimization Increases Trading Profit


A Changing Landscape

The structure of fixed income markets has changed dramatically over the past decade. New regulations, government bond purchase programs, new products such as ETFs, the emergence of electronic all-to-all platforms and non-dealer liquidity providers using algorithmic and high frequency trading are among the developments making it more difficult for sell-side fixed income desks to remain profitable.

After the 2008 financial crisis regulators around the world recognized the need to create a safer banking system. Regulations increased capital requirements, reduced the risk banks are allowed to take and increased costs for intermediaries. The unintended consequence of these new regulations has been a reduction in liquidity in the secondary bond market. Dealers are less willing and less able to hold inventories and so they are less willing to act as a principal in bond trading.


new landscape of electronic trading


section divider

The current sell-side trading desk process

Sell-side dealers and buy-side asset managers are rapidly embracing AI applications to price fixed income securities algorithmically in a live trading environment or for end of day reconciliation. Among these is the credit trading desk at Dekabank, a German financial institution, which was introduced to Overbond through Infosys Consulting. Overbond analyzed Dekabank’s credit trading process and determined how it could be improved with the integration of an AI bond pricing model. Under the legacy Dekabank credit trading process traders received RFQs by one of two processes:


electronic vs manual RFQs

section divider

The legacy challenge: low confidence in prices


constraints for traders pricing RFQs
issues with pricing provided by bloomberg or third parties

These factors lead to low confidence in the suggested prices and traders must constantly spend a great deal of time and effort in manually adjusting prices based on prior knowledge and intuition. The major trade-off is thus accuracy versus time, leading to missed deals and direct downward pressure on desk P&L.


section divider

The solution: AI-powered bond pricing

Need for centralization of information


AI modeling techniques share many similarities with classic statistical modeling techniques. Statistics provides the building blocks upon which the machine learning that drives AI is built and both use large amounts of data. But statistics is purely mathematical and largely descriptive with some ability for inference. AI adds additional programming, made possible with modern computing power, to move one step beyond statistics and become predictive.

The goals of the two methods are different. Statisticians start with a set of known assumptions that are given to the model and best explain the expected behavior of the financial outcome in consideration. With AI techniques the underlying assumptions are unknown and the aim of the model is to determine itself the method that best predicts the outcome in consideration.

Overbond has harnessed the AI advantage for pricing bonds


Corporate and Government Bond Intelligence (COBI)-Pricing was created as part of Overbond’s suite of AI algorithms for the fixed income capital markets. It algorithmically finds the optimal indicative new issue bond prices and secondary market bond prices for global investment grade (IG) and high yield (HY) bonds, using machine-learning (ML) algorithms. The ML algorithms analyze millions of data points related to factors such as historical pricing trends between similar bonds and similar issuers, intra-day pricing volatility, trading volume and counterparty composition, company fundamentals, investor sentiment and industry, rating or tenor cluster comparables. Data is aggregated from multiple types of data sources including:


sources of aggregated data

section divider

The solution: Overbond COBI-Pricing Live

Overbond’s COBI-Pricing LIVE is a customizable AI pricing engine that assists traders in automating pricing and trading workflows for global investment-grade bonds. It generates prices and liquidity scores for more than 100,000 fixed income instruments and builds curves for more than 10,000 issuers in various real-time liquidity scenarios.

The full interoperability of COBI-Pricing LIVE allows its AI algorithms to ingest, aggregate and process data from live and historical vendor feeds, internal historical records, OTC settlement layer volume records, and now voice transactions. Overbond AI pricing, liquidity scoring, LIVE trading automation and routing algorithms can now capitalise on all primary and agency trading routes, voice or electronic, across all venues and counterparty types.

COBI-Pricing LIVE has a refresh rate of less than three seconds, enabling sell-side trading desks to fully automate 30% of their RFQs and execute an additional 20% with trader supervision. COBI-Pricing LIVE allows traders to automate trade flow, improve liquidity risk, improve price monitoring and reporting, respond to 80% to 120% more RFQs, maintain an optimal hit ratio and significantly increase desk P&L.


section divider

The solution: the Overbond COBI-Pricing algorithm


COBI algorithm flowchart
Data Intake, Pre Processing, and Model Training

section divider

The solution: COBI-Pricing data intake

Successful data pre-processing is the key stage and a pre-requisite for operation of the COBI-Pricing algorithm. The precision of the algorithm output is critically dependent on the accuracy, timeliness, and relevance of the pre-processed input data. Overbond sources raw data from major data suppliers in the financial sector, including Refinitiv, Ice, The Six Group, EDI, MarketAxess, Tradeweb, Euroclear, Clearstream, DTCC, CDS, S&P Global Market Intelligence, major credit rating agencies, as well as other sources. The data COBI-Pricing algorithms uses includes the following:


COBI pricing algorithm uses

section divider

The P&L enhancement: margin optimization AI model

The case for margin optimization

With restricted liquidity, a reduced ability to take risk and increased speed, fixed income market participants are looking for advantages beyond electronification and automation. Over the past couple of years, we have witnessed increased adoption of quantitative investing, AI-driven liquidity risk monitoring techniques, transaction cost optimization and margin optimization.

Margin optimization AI modeling measures the distance to cover on all prior executed transactions and RFQs and minimizes it with respect to trade information at point of execution. Sell-side market making desks can double or triple the volume of RFQs they can respond to without incurring negative margin on those trades and increase profitability in an environment where it is becoming increasingly difficult to do so.

The Overbond approach to margin optimization AI model

The Overbond margin optimization model incorporates a variety of factors as inputs to AI ensemble to arrive at a distance to cover price for each transaction. The model aims to capture and convert various margin optimization measures to one unified, optimized distance to cover price. These factors include but are not limited to:


some factors to the AI model

section divider

The implementation: Dekabank project structure

Overbond structured the Dekabank project into two phases. Overbond first deployed and tested end of day data on a smaller universe of ISINs to which ML algorithms were applied to find the best executable secondary market bond price for each bond.

Intra-day pricing was approached as a Phase 2 deliverable of the project because Overbond ML algorithms analyze millions of data points aggregated from multiple types of data sources and the models are computationally intensive.

phase 1 deliverables phase 2 deliverables

section divider

The implementation: back-testing simulation

Overbond conducted an intensive back-test of its COBI suite of AI algorithms on a sample of 153,587 unique RFQ prices submitted by the Dekabank credit trading desk during trading sessions between August 2017 and October 2019.

The goal of the back test was to determine the logic to be used to recommend a mean distance to cover with a high and low bound that would be used and submitted in a trader’s RFQ, without requiring adjustment by the trader.


backtest graphs

Seven per cent of the sample data was used as the testing data on which the algorithm was run.

The Overbond AI engine iterated through all “accepted” and “covered” RQF results in the trade book on both bid and ask sides and performed a margin optimization analysis.

Three values were generated: the predicted mean of distance to cover, the lower bound of distance to cover and the upper bound of distance to cover.

The actual distance to cover as a percentage of the interval between the lower bound and upper bound of the distance to cover was calculated. There are some outliers for the bid side in terms of distance to cover, but about 93% of the actual distance to cover is captured by the estimated confidence interval. Similarly, on the ask side, about 90% of the actual distance to cover is captured by the estimated confidence interval.


section divider

The implementation: back-testing results

Of the total universe of 153 587 sample RFQs, the COBI engine with the margin optimization model was able to provide an estimation of confidence intervals capturing about 93% of the actual distance to cover on the bid side and 90% on the ask side. This meets the criteria for both distance to cover accuracy and profit optimization required to be suitable for RFQ margin optimization operations without manual trader intervention.

The figure below shows how the results would appear to a trader user of the Overbond UI front-end platform visualization.



The columns “Margin Bid (M)” and “Margin Ask (M)” are composed of the “distance to Bid (M)” and “distance to Ask (M)” respectively, on top of the COBI priced “Bid” and “Ask” quotes for each bond.


section divider

The business impact

Overbond has created positive business impact for clients around the globe, including sell-side institutions with significant trading volumes (200-500 RFQs+ a day per trader). We work with clients’ innovation groups to actively explore the application of new technologies that can serve as the catalyst for trading automation and improved risk management, trade flow, pre-trade and post-trade analytics. These technologies have direct business benefits.


business impact from automation

section divider

Implementation considerations

key considerations for firms in charge of AI roadmap Overbond's custom ai services, proven methodology, operational acceleration and AI analytics as a service

section divider

About Overbond

Overbond is a developer of process-redefining, AI-driven data and analytics and trade automation solutions for the global fixed income markets. Overbond performs market surveillance, data aggregation and normalization, and deep AI quantitative observation on more than 100,000 corporate bonds and fixed income ETFs. Applying proprietary artificial intelligence to pricing, curve visualization, market liquidity, issuance propensity, new issuance spreads, default risk and automated reporting, Overbond enables trade automation and enhances trade performance and portfolio returns. Clients of Toronto-based Overbond include global investment banks, broker dealers, institutional investors, corporations and governments across the Americas, Europe and Asia.


Contact:
Vuk Magdelinic | CEO
+1 (416) 559-7101
vuk.magdelinic@overbond.com

Andrew Zippan | Global Director of Sales
+1 (647) 405-0895
andrew.zippan@overbond.com