Building a cognitive bond with Bondevalue
6 min read
Building a cognitive bond with Bondevalue
Bondevalue has won the pole position in the Asia Pacific region for the Watson Build Contest. They will compete in the final round with 6 other winners from each geography along with one wildcard entry on 2nd November 2017.
“Despite its importance as a source of financing for companies, the corporate-bond market is shockingly archaic. Even basic price data are hard to come by. Whereas stocks can be traded at the click of a button, buying and selling corporate bonds often requires a phone call to a trading desk at an investment bank.” – Economist
Bond market structure & issues
Drawing a comparison against equities, which are light-years ahead in terms of transparency, bond markets continue to be shockingly inefficient on several parameters:
Opaque pricing: Bond investors (usually private banking clients) could pay as much as 2.1% (in terms of bid-ask spread) of the order value when purchasing bonds. At an average trade size of $200K (in Singapore & Asia), the expense amounts to approx. $1K-3K per trade.
Lack of data: Investors have limited access to pre-trade guidance prices and post-trade data/prices. Getting a price quote could take 3-4 hours and investors often have limited options to compare relative prices from 2-3 banks.
Expensive platforms: The price point for receiving bond-specific price information and news through trader terminals is between US$10-20K per year.
Information asymmetry (News): Bond investors have limited access to bond-specific news. It is what can be termed as a classic ‘insider’s market.’ Thus, considerable information asymmetry exists in the bond markets. This situation tilts the odds heavily in favor of institutional players and against individual bond investors.
Need for deep-parsing Bond News
Among the many problems infesting the bond markets, one of the primary concerns of bond investors relates to finding news and information relevant to their portfolio. Fixed income desks at MNC banks have expensive data sources and teams of analysts to maneuver through the considerable noise. In stark contrast, individual investors do not even have a rudimentary access to such information.
Currently, investors need to sieve through an overabundance of financial news and hence finding relevant information is challenging, to say the least.
Crucial information regarding covenant breaches, legal proceedings, new bond issues, price guidance etc. is often relegated to later pages of generic search engines, where it can be rarely found by investors. We found that generic search solutions are unable to fix the issue. Especially considering the sheer volume of news today, it is quite easy for investors and advisors to lose track of the essentials.
For example, a sample Google search based on a keyword of ‘bond news Microsoft’ throws up 3.48 million results. The most prominent news on the front page – multiple republished articles on the same 6-month old bond issue. The search for relevant information will take an investor several minutes and even then finding credible information will remain a challenge.
Investors must rely on word-of-mouth advice from private bankers which results in uninformed investment decisions.
Fact finding and information analysis is a critical part of Fixed Income desks for Institutional Investors & Banks. Hundreds of highly-paid analysts spend several man-hours sifting through financial news through professional software to narrow down relevant information. The process is complex, and banks spend hundreds of millions of dollars on bond research. Using the proposed solution and leveraging the capabilities of IBM Watson, we envision a completely automated ‘news curation’ process. This is a big step towards automating the entire value chain of fixed income due diligence.
The proposed solution is a cognitive search application, designed by bond market specialists and powered by IBM Watson. Leveraging the AI capabilities of the Watson Knowledge Studio, Bondevalue has developed a classifier that not only curate’s news but also sequences it based on relevance. Start here to access Bondevalue-API on the IBM Cloud platform.
Our product builds on well-defined data structures, decades of expertise in fixed income markets and IBM Watson Natural Language Understanding and Watson Natural Language Classifier APIs. The classifier parses structured and unstructured news articles to identify and promote only the most relevant ones. The application generates bond relevant news along with their category, relevance and sentiment score.
List of the Watson APIs used in this solution: Natural Language Classifier & Natural Language Understanding
As a result, the disorganized news-feed is transformed into an intelligent array of relevant information that is much more useful for the investor.
The solution has wide-ranging applicability:
The Watson-based news classifier will save the end customer considerable amount of time and effort. With the power of Watson AI, bond investors can access the most relevant articles on a real-time basis. The impact of this solution will be profound since it fills a much-needed demand in the market.
The application of the classifier can also be extended to other areas of financial markets, such as parsing IPO/Bond prospectus, building intelligent chat-bots, improving finance training, etc.
On the B2B side, private banks and boutique fixed income advisory houses can use the solution to improve their customer engagement levels.
Boutique investment advisory firms in the market, with limited resources to access expensive data solutions, can also utilize this service to improve their debt-market capabilities.
While the solution has strong commercial application, one of the more satisfying benefits of the solution lies in its ability to give a boost to financial inclusion. A distributed, inexpensive and reliable information solution, that we put forth, would enable new retail investors to enter debt markets. New issuers will then find it easier to raise capital as the investor base increases.
Help prevent systemic problems that are caused by vested interests with privileged access to information.
End User Experience
Users will be able to view the output data via the BondEvalue iOS App. Currently the Watson Enabled functionality has been made available in the Bondevalue iOS development environment also known as the BEV DEV App.
Once the user logs into the Bondevalue App, they will be able to access the real-time bond data points across 900 bonds pertaining to different country of risk, sectors and currencies. Each and every bond is displayed across a separate horizontal tile which when clicked takes the user to the Bond Details Screen.
The Issuer or Company Specific news is available on this Bond Details Screen and gets displayed once the user selects the “Latest News” tab as shown below.
As one can observe there is a Watson Enabled switch present towards the top of the news section. The system renders unprocessed news if the switch is in off state. This unprocessed news contains both relevant and irrelevant news and gets sorted on the basis of the latest available news for that particular issuer/company. No category, relevance or sentiment score are made available as it doesn’t involve the processing carried out by Watson Natural Language Classifier and Watson Natural Language Understanding APIs.
Once the switch is turned ‘ON’ to Watson Enabled, the processed news starts flowing and gets rendered in the news section. This processed news contains only the irrelevant news items which the irrelevant ones filtered out using the trained Natural Language Classifier APIs.
We believe that the proposed application is a big step towards opening up bond markets. The solution has a sound commercial potential and wide-ranging applicability in the area of financial information services.
We also strongly believe that the innovative solution powered by Watson and delivered on Bondevalue’s mobile application, effectively bridges the current knowledge gap in bond markets. To learn more, visit the Bondevalue website. And if you have support or technical questions, you can reach us at email@example.com.