Counterparties are increasingly adopting quantitative investing techniques such as systematic alpha and algorithmic trading, merging of fundamental discretionary and quantitative investment styles, consumption of increasing amounts of alternative data, and adoption of new methods of analysis such as AI analytics like Overbond COBI-Issuance algorithm described in this white paper.
Fixed Income Artificial Intelligence
Financial services market is embracing digital processes and artificial intelligence applications to streamline how they do business, and bond origination and bond OTC trading is one of the areas which actively looking into ways to embrace the trend. The current fixed income capital market data flows are inefficient in many respects, limiting precision in assigning proper value to credit risk long term. Markets remain heavily reliant on segregated and manual data operations between counterparties and as a consequence, disparate data sets. These disparate data sets cause the market to suffer from information asymmetry and decentralization. As a result, insight from available data is fragmented and disseminated through manual exchanges between counterparties, which furthers creation of disparate data sets.
Need for centralization of information
There is a great need for a fixed income big-data centralization where advanced analytics such as price discovery, intelligence gathering, pre-trade and post-trade analytics can be performed – to increase the overall efficiency of the fixed income market and understanding of the credit risk valuations. With no centralized hub, issuers and investors operate with partial awareness. There are also limits placed as a result, on applications of AI utilizing deep historical data records of fundamental data elements (audited statements, dealer supplied primary bond price quotations etc.) and secondary market bond trade points. With this, Overbond pioneered to be the first to market with a centralized big-data hub empowered with AI capabilities for fixed income analytics.
Overbond AI Focus Areas:
Price and Issuance Discovery – Predictive price trending analytics and tools and integrated machine-learning modules provide a reduction in credit pricing risk, enabling systematic monitoring of credit pricing tension and alpha-extraction of market opportunities covering large universe of issuer names as well as monitoring of likely new bond issuances.
Opportunity Matching – Buy-side investor canvassing and systematic matching capabilities that are calibrated with Overbond AI models and translate into improved ability to develop and apply custom AI models to precisely determine credit risk valuations, traditional and non-traditional buyer prospects and utilizing proprietary investor preference and market sentiment signals.
Automated Information Systems – Integration and tailored analysis of historical and new indicative pricing data flows empowers trading, portfolio management and deal analytics for optimal decision-making.
Custom AI Solutions
AI Powered Issuance Prediction
COBI-Issuance was created as part of Overbond’s suite of predictive algorithms for the fixed income capital markets. It is an advanced AI algorithm family which makes ongoing measurements of issuer’s propensity to issue bonds. COBI-Issuance assigns a score which estimates the relative likelihood a bond issuer will come to market with bonds in the near future. It analyzes factors from multiple types of data sources including:
AI advantage over statistical methods
COBI-Issuance AI modeling techniques share many similarities with classical statistical modeling techniques starting from the fact that they both deal with data. However, the key difference, between statistical techniques and AI models Overbond applies is in the goal of these approaches. While 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, AI techniques rather aim at finding by themselves the method (with underlying assumptions that are unknown) that best predicts the outcome in consideration.
Clients can use issuance predictions for custom analysis
The predictive time horizon the COBI-Issuance algorithm in standard use cases is optimized to four weeks. A score is assigned for each company in each potential bond issuance tenor. Scores are on a scale of 0-100 and are relative to other issuers and other bond issuance tenors. A higher score in general means that company is more likely to issue in that tenor compared to a company or a tenor that receives a lower score. High scores (~70-80 or higher) across all tenors imply that an issuer is likely to issue a bond in any tenor. High scores in only one tenor implies issuer is more likely to issue in that tenor compared to other tenors. Low scores across all tenors imply issuer is less likely to issue in any tenor. It is important to note that propensity scores are not probabilities. For example, a score of 90 does not mean that issuer is likely to issue new bond in that tenor with 90% probability. It means that issuer is in the 90th percentile in a ranking against all other companies in all other issuance tenor possibilities.
COBI-Issuance Propensity Output
COBI-Issuance Propensity output comes in two formats – historical, and current.
Historical COBI Propensity
Historical propensity is given as a separate time series going back two to five years for each tenor (2, 3, 5, 7, 10, and 30 years) for each issuer. The time series is shown in a graph below with the time across the x-axis, and the COBI-Issuance propensity score on the y-axis labeled as ‘Likelihood To Issue’. The graph also shows black vertical bars at the dates where that issuer actually issued in a given tenor. The following is an example of an output graph for the historical propensity for a single issuer and single tenor prediction:
Current COBI Propensities
Current propensities can be supplied on a weekly basis, although frequency can be scaled according to a client use case need. For example pre-deal analytics applications in investment banking usually require one month or longer time horizon COBI models optimization while use cases in fixed income trading world often entail model optimizations that are as close to real-time as possible.
For the purpose of additional transparency and explain-ability, Overbond exposes underlying factors which would be commonly understood by analysts to contribute to a propensity score at any given time. Breaking down the propensity scores into more detailed categories, factors include: Upcoming Maturity; Average Maturities per Year; Overdue Issuance; Popular Sector for Issuance; Recent Issuance etc.. Note that the COBI-Issuance algorithm is a non-linear, non-parametric algorithm, and the overall propensity scores are not directly proportional to a weighted average of the sub-scores. These sub-scores are intended to give a deeper level of explain-ability to the indicators used to derive the propensity scores.
How COBI-Issuance Algorithm Works
The diagram below and the following paragraphs provide a description of how the Overbond COBI-Issuance Propensity algorithm works.
Data Intake & Pre-processing
The Overbond platform sources raw trading and fundamental data via automated nightly scripts. Our data sources include Thompson Reuters (primary and secondary bond issuance and trading levels), S&P Global Market Intelligence (company level fundamental data), DBRS (company ratings and macro market data), as well as various other sources.
This raw data is then structured in the Overbond databases. Trading data and fundamental data are structured and mapped to the appropriate issuer ID. The data is systematically scrubbed for anomalies and null values. Finally, a set of key input factors are generated based on the raw input. These include but are not limited to factors that measure recent issuance, issuance frequency, maturity schedule gap, propensity for specific tenors. These factors are divided between sector and company specific and are used as inputs to the machine learning models.
The subsequent stage for the machine learning algorithm is to train and apply several models to calculate the output propensities. An Ensemble Learning strategy is used, meaning multiple models are combined to elevate overall robustness. These models are each trained using a subset of the past data, ranging from one month to a maximum of ten years. Advanced sampling techniques were used to account for class imbalance between positive (will-issue) and negative (will-not-issue) predictions. Finally, the results are back-tested against the entire ten years of data and measured for precision and recall metrics.
COBI-Issuance Data Intake
The successful data pre-processing is the key stage and pre-requisite for the COBI-Issuance algorithm operation. 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 Thomson Reuters, S&P Global Market Intelligence, major credit rating agencies, as well as other sources. The data COBI-Issuance Propensity algorithms use includes the following:
Data Pre-Processing and Model Training
COBI-Issuance is an amalgamation of algorithms which predict if an issuer will come to the market and issue bonds. A variety of pre-processed inputs are fed into COBI-Issuance’s algorithms, to predict issuances. The following is a subset of indicators used:
Within the COBI-Issuance model, multiple supervised machine learning algorithms are trained using past data to predict issuances. The algorithms used include: XGBoost, Neural Network, Random Forest, and Logistic Regression. COBI-Issuance uses a robust ensemble method to combine the results from each algorithm and generate an output score. This score represents the propensity of an issuer to issue a bond in a specific tenor.
Results of COBI-Issuance
The sample back-tested performance of COBI-Issuance algorithm can be seen in the following graph, plotting predictions for a specific issuer to issue bonds in a specific tenor. Propensity values are plotted over time, with black bars representing when actual issuances have occurred. The gray area trailing each issuance on the first graph is indicating the issuance prediction window. The default time horizon for the COBI-Issuance propensity issuance prediction is four weeks. However, this can be adjusted according to client needs.
Overbond data science team ran systematic back-test on 600 issuers, testing 6 issuance tenors for 500 weeks, representing around 1.1 million predictions. For each issuer and every tenor COBI-Issuance algorithm calculated likelihood to issue every week.
To display overall result predictions were categorized in 4 buckets: Highly Likely to Issue, Likely to Issue, Unlikely to issue and Very unlikely to Issue. As you can see in the table, the majority of issuers/tenors that algorithm predicted as being highly likely to issue did in fact issue, above 81% precision. Similarly the vast majority of the issuers/tenors that algorithm predicted as being very unlikely to issue did not issue. This is true for all issuer industry sectors and in general.
Over the past two years, we have witnessed profound changes in the fixed income marketplace with counterparties increasingly adopting quantitative investing techniques. These include systematic alpha and algorithmic trading, merging of fundamental discretionary and quantitative investment styles, consumption of increasing amounts of alternative data, and adoption of new methods of analysis such as AI analytics like COBI-Issuance algorithm.
Overbond client organizations include buy-side institutions with over $2 trillion of assets under management globally, across both passive and active strategies as well as sell-side dealer institutions. Their innovation groups actively explore new technologies that can improve investment performance, deal flow and pre-trade analytics.
Institutions considering AI predictive analytics implementation and big-data transformation projects, can employ acceleration utilizing externally calibrated models and market signals. Below are few key considerations and questions for executives in charge of AI roadmap:
- What is the current state of our fixed income in-house data?
- What are our data science and engineering capabilities?
- Are we building AI capabilities to grow revenue or cut cost?
- How can we redefine the boundaries of our data universe or identify alternative data sources necessary to feed AI engine?
- Given that AI learning curve is steep where do we begin?
- How do we create and execute AI proof of concept use cases rapidly?
- What are key success factors for our AI roadmap?
Custom AI Services
Overbond works with clients to identify and recommend practical AI analytics use cases that are aligned with strategic goals of the financial institution. We help assess current-state AI capabilities, and define roadmap to help clients realise value from AI applications. We manage cross-channel data flows across multiple systems and enable custom font-end visualizations.
With our targeted approach and implementation methodology, we quickly demonstrate value of AI analytics to test use cases, enabling client-side change management approach and stakeholder buy-in.
We help clients build and deploy custom AI solutions to deliver proprietary analytics and tangible business outcomes. Our experience combines calibrated models, design patterns, engineering and data science best practices, that accelerate value and reduce implementation risk.
AI Analytics As-a-Service
Overbond helps customers design and oversee mechanisms to optimize and improve existing fixed income credit valuation, issuance prediction and pre-trade opportunity monitoring using AI. Our team of world-class data scientists and engineers manage an iterative implementation approach from current state assessment to operational handover.
Overbond specializes in custom AI analytics development for clients implementing risk management, portfolio modeling and quantitative finance applications. Overbond supports financial institutions in the AI model development, implementation and validation stages as well as ongoing maintenance.
Chief Executive Officer