The Credit Trade Paradox and the AI Solution

AI-Powered Trading Alerts, AI and Data Analytics, Credit Trading

The Credit Trading Paradox

Credit market data has long been the “Moby Dick” of quantitative financial analysis, remaining elusive and beyond the grasp of traditional methods of capture.  The reasons for this difficulty are extensive but can largely be broken into two separate categories – structural and data-driven.  

The structural issues tend to get the most focus, with illiquidity, bid/offer cost, and lack of clarity around execution headlining the focus items for most firms (and startup/investment efforts).  

The data-driven issues, however, have seemingly flown under the radar – likely due to the difficulty in modeling the non-traditional price movements of corporate bonds, which often look more like step functions than the smooth time-series curves analysts are used to.  An accurate model also has to account for the fluctuating factors that may have led to the price movement from an issuer perspective…and then try to understand the underlying technical that may have driven further movement within a capital structure.

How is this changed by AI?

Market veterans could be forgiven for thinking they have heard the “technology is going to alter the credit markets forever” narrative before.  It seems to rotate through the financial media annually with one or two large articles profiling the latest tech hotshots poised to “conquer” Wall Street.    You know the story I am talking about – “xyz CEO is taking his success from Silicon Valley to NYC and will drag corporate bond trading out of the stone age”.   The context is seemingly always the same – the machines are smarter and faster so it is only a matter of time…your jobs will disappear in the future.

The team at Deep MM could not disagree more with that conclusion – we built our product “by credit professionals, for credit professionals” and have one goal in mind – to empower the people in the seats by providing the most accurate data and analytics solutions possible.  We want to remove the hours our users spend staring at streaming tickers on computer screens and replace that time with higher value pursuits.  Let our AI “co-pilot” take care of the data…and you can take care of the rest.

It seems obvious to say – but high-quality data is the foundation that a decision making “house’ is built on.  Our CEO spent the last 6 years building our “base” in developing the most accurate corporate bond pricer on the market.  He leveraged the latest in AI technology to solve some of the issues that had plagued traditional quantitative methods.  Here is why our model is different:

  • Removing Time Bucketing of Trades – Each data point in our model is ingested independently, rather than submitting summary statistics for individual periods of time – this better represents the illiquid nature of corporate credit
  • Drastically Increasing the Number of Factors in the Model – Our current model has 80 million parameters and we have tested as high as 3.3 billion.  Traditional financial models have closer to 1,000
  • Incorporating macroeconomic and other data sources – Our competitors only look at corporate bond transactions in reaching a price, we are also incorporating equities, interest rates and credit ETFs.  We ultimately plan to have all relevant data flowing through our “credit data hub” – whether it be messages, financial filings, whatever the potential impact factor.  Our model is updating continuously as the market moves, not waiting until the 15+ minute delayed trace print hits

How Does this Matter Now?

Having accurate real-time and historical pricing on every corporate bond will open a world of opportunities in credit analytics.  However, as anybody who has traded a corporate bond knows, the “mid-price” is only one portion of the equation.   Bid/offer costs and execution probabilities vary dramatically from bond to bond, and even from day to day within the same security.

As can be seen above – our fair market value model is not limited to a single “mid” but shows bid/dealer/offer indications that fluctuate real-time with transaction volume and price discovery.  Which leads us to the final factor in credit trading – the probability of execution at a given level.  Users can adjust the markets displayed on their screen (and on historical price histories) by changing the “percentile” value associated with a given quote.  The percentile can be thought of as a probability distribution – ie, if I put the percentile to 80% on the bid-side, 80% of the “next trades” that occurred would be above my bid.  This helps users to adjust real-time pricing estimates in a way that reflects their role in the marketplace (market maker vs market taker) and provides greater insight into the behavior of the market at any given time.


One implementation that sell side dealers have found particularly helpful so far is the “Runs” screen displayed above.  This allows users to set “percentile” parameters that  define bid/offer width and send out markets on hundreds…or even thousands of bonds at a time.  The days of dealers manually updating and formatting runs in the context of a Bloomberg screen are gone – we have automated the whole process.

Real-Time Market Analysis

Our platform provides accurate real-time pricing and probability assessments for any transaction users are looking to complete – essential information for the execution side of any business.  But we are most excited about the potential uses of real-time data across larger sectors of the market.  


The screenshot above is from our “issuer” screen – which helps users identify dislocations and trends within an individual capital structure in real-time.