How AI can solve your bond liquidity challenge

Equity and forex trading is automated and almost all trades and execution takes place through digital platforms, but bond trading is still performed manually over the phone or instant messaging. We are hearing a lot of AI and ML innovations around equities. When it comes to bond, it is still in its infancy. Indeed, there are few statistics available in Asia, which remains a backwater when it comes to trade-tech. “No one knows what the split between voice and electronic trading is in Asia,” says Jesper Bruun-Olsen, head of Asia Pacific at Algomi, a software company that helps fixed-income fund managers aggregate data.

In terms of digitization and reducing the gap between manual and automated trades, developed markets are consistently innovating. In fact, very few trades have clarity on the gap between voice-to-electronic trades. “Fixed income is so illiquid. You have to talk to each trader at J.P. Morgan or Merrill Lynch until you have spoken with every investment bank,” said Jeremy Son, a fixed-income trader at AllianceBernstein (AB) in Hong Kong. “I have more than a hundred chatrooms with various dealers and a lot of information gets lost.”

How technology is digitizing bond trading

Liquidity scoring

Though liquidity is not a major problem in developed markets like the US, but corporate and high-yield bonds are also mostly voice-traded. In Asia, markets are fragmented and traders have limited access to information which makes it hard to produce liquidity. This in turn makes little information available to feed a trading algorithm. This is

Though liquidity is not a major issue for small traders with average trade below $1 million, But for large trades, especially something over $5 mn is creating a big challenge because traders mostly have to deal with front runs. Due to fragmentation in the bond market, traders have limited access to information which makes it hard to produce liquidity. This in turn makes little information available to feed a trading algorithm.

However, bond liquidity score is a way to counter these challenges. We at Pirimid, have developed an algorithmic model using various data points to predict if the bond is liquid or illiquid. It helps traders to consider in advance that they may have to pay a premium to execute all the quantities if the bond they are considering has less liquidity. The biggest challenge while developing this model was to consider data available from offline channels.

Our liquidity scoring model

We have identified a comprehensive approach to calculate bond’s liquidity than the usual approach of focusing mainly on the bid-ask spread. The liquidity scoring algorithm uses a variety of data points such as bid-ask spread, volume and quantum of trades taking place along with other market depth-related parameters such as the number of dealers and past settlement data.

We also use trade clustering data to understand if it’s a buyers’ or sellers’ market. We aim to calculate one unified liquidity number using all the input variables so that the underlying liquidity related risk could be anticipated and taken care of. Below table highlights the variables used for liquidity calculation along with their relevance on the score.

So, how AI is helping?

Modelling

We can use AI to ensure the liquidity score we get multi-dimensional and capable of predicting future moves. Because we cannot say a bond is not liquid because it was not traded over the last few months, the ultimate goal of calculating liquidity score is to understand its implication in the future. Following are the ways we can use to train and deploy our scoring model:

Clustering of bonds data:- Clustering is one of the most used unsupervised ML algorithms. It has been very helpful when the data is not labelled and we need to find a group of clusters out of it. Using it we can cluster the bonds data into different groups and that can give us the idea of which bonds are less or more liquid than others.

Predictive modelling:- If we can label the data by hands and provide the final output to ML/DL models then they can learn the relation between our inputs and the outputs. This way we can give a score to each of the bond based on the data points that we have for it. This is very challenging as labelling the data is tough and needs expert supervision.

Capturing unstructured data-sets

As mentioned earlier, the majority of bond trades take place over vice or chat. Hence, we used Natural Language Processing and Object Detection models to convert the unstructured data into a structured format so that we can leverage that for our model. Illustrated below how we have transformed quotes received over mail and SMS. We have similar techniques to convert voice quotes too.

This is not all about AI use case in bond trading. Read our detailed coverage on how we are using AI to automate the entire fixed-income trading.

Bottomline

Liquidity score highlights the ability to exit a fixed income position at or near the current value. In case the liquidity score is low, it is very likely all the positions may not get a counterparty and delay the execution. Hence, there is a strong connection between liquidity risk and pricing of bonds in the market. Historically we have seen liquidity having a great impact on yield spreads which widened when the market was volatile. A study conducted by Friewald et al. highlighted that liquidity has 14% impact on bond yield during stable market conditions. However, this number increases to 30% during a volatile market. The study highlighted, this holds true for all bonds with the only exception AAA rated. Another study conducted by Heck et al. confirmed the hypothesis between the yield spread and bond liquidity.

With growing regulations around disclosures and risk management transparency, bond liquidity can play a major role in generating valuable insights with both the investors and regulators can use to derive meaningful insights. Some of the use cases of bond liquidity could be to calculate:

• Days to liquidate: Calculate expected days to liquidate a specific portfolio security
• Stress testing: Simulate how many days it will take to liquidate the portfolio under different stressed conditions
• Portfolio liquidity profiles: Compare portfolio liquidity profiles with comparable benchmarks
• Market impact: Project potential market price impact for a particular trade
• Risk management and compliance: Measure portfolio liquidity trends and demonstrate compliance to regulators
• Investment selection: Use liquidity score as an input in the instrument selection and investment decision-making
• Collateral management: Support eligibility determinations and monitor the liquidity of collateral holdings
• Creating indexes and investable products: Use liquidity score as part of the selection criterion while designing index or portfolios

Drop a comment or write us an email with any feedback about this article, queries info@pirimidtech.com. View our portfolio in building Robo advisory, Large Scale Trading Systems, Algo Trading, Stock Sentiments, Price Trends forecasting, Backtesting frameworks, Credit Model, Open Banking, etc. and services offered on our website. to see our Fintech expertise can help you build cutting-edge solutions powered by AI/ML.