AI is probably the biggest tech that is disrupting so many industries at the moment. It has not just changed the software industries but also industries like finance, agriculture, music, shopping etc. There are many applications of AI in different industries but there are always challenges with getting the quality data to build right AI models. In contrast, capital markets is well matured and have huge amounts of data for AI models to learn from and make accurate predictions.
Many fintech companies and banks are using AI for lots of different applications. Banks like JP Morgan uses it for credit card fraud detection, Bank of America uses chatbots to provide smart assistance to customers, FICCO is using AI to give credit score to customers so that they can get eligibility for loans.
Bonds market has many interesting AI applications. It’s considered less liquid due to various reasons like, over the counter trading, millions of bonds that traders can trade, voice trades, no direct announcement of the current prices of bonds etc.
Liquidity Crunch With Bonds
As far as price transparency is concerned, there has historically been a huge gap between the amount of reference information available to those trading equities versus those trading corporate bonds. Stock exchanges report trades, bids and offers realtime. Free access is available online with a 15 minute delay while traders who demand more information can pay for real-time data. By contrast, bond trades are required to be reported within 15 minutes and only those who pay for the TRACE(USA) & TRAX(Europe) feeds can access this information. No quotes are publicly available and the best way to get a quote is to solicit multiple brokers and wait for a reply. Alternatively, there are data companies that provide end of day prices, published after the market has closed and with no guarantee that the specific information sought will be included. Accurate bond pricing is also hindered by a lack of liquidity. Only a fraction of TRACE & TRAX eligible bonds trade on a given day, so the most recent trade price is often multiple days old. Pricing bonds based on other more liquid bonds that have similar features is common, but again limited by the presence of such bonds.
AI use cases
Below, we have listed a few use cases of AI in Bond Market.
Predicting Bond Prices or Price discovery
Price discovery and price prediction are the next big things around fixed-income trading. With big-data centralization and using machine learning algorithm overall efficiency of the fixed income trading desk could be improved to the next level. The pricing mechanism works on multi-dimensional data points such as trading history and pattern, fundamentals of the issuer and market-related information. As mentioned input data source is multi-dimensional which includes both structural and unstructural data. Hence, using AI based NLP and object detection techniques, we can transform unstructured data into structured format.
Recommending Bonds For Investment
Similar to e-commerce websites product recommendations, AI can be used to recommend which bonds a trader should buy in order to make maximum profit.
Companies like Synechron, ATB Financial uses collaborative filtering, the same algorithm that Netflix uses to recommend the shows to recommends the bonds that trader should invest in. It can work on two sides, meaning that it can recommend which bonds to buy or sell. The algorithms which can be used are,
- Collaborative filtering system
- Hybrid recommendation system
Chatbots are used in so many ways to provide smart assistance to customers. Banks are using them to assist customers in so many different ways from knowing the account balance to loans eligibility.
Companies like AllianceBernstein has chatbot named Abbie which can assist the trader in many ways to make the trade and also recommends him which bonds are good to invest in. It can help in managing the portfolio of the trader and in minimizing the risk involved in trading.
There are different ways to create chatbots. Online platforms like IBM Watson, DialogFlow and many others can be used to create chatbots directly. If the company wants to create from scratch they can use platforms like RASA. Rasa is an open source framework, built for creating chatbots.
The algorithms which are used are,
- Generative based models which used encoder-decoder architecture to create the chatbot.
- Retrieval based models which use intents to classify the query of a user and then provide an answer to the query.
Filling yield curve prices using AI algorithms
There is a research paper in which the idea was to design an unsupervised machine that learns the salient features of corporate bonds yield curves by observing a sufficient number of historical examples in a liquid market and then uses the learned shapes to fill in the missing yields in the illiquid market. Roughly speaking, the machine is a tool for interpolation/extrapolation, however, it also incorporates its memory of typical yield curve shapes.
In this application, the author of the paper used Variational Autoencoders, which is a well known AI model to learn from data in an unsupervised manner. They have collected data from Bloomberg and then used it to train their model. More details on this can be found in the paper.
In order to see, if we can create an AI model which can learn to predict the bond prices, we used this Kaggle Data.
We used LSTM(Long Short Term Memory) network, which is used to learn long term dependencies from the data. It is the most widely used neural network for natural language processing and time-series forecasting.
Below table shows the real trade prices and predicted trade prices using our model.
More details on our experiment and machine learning model can be found here.