Bloomberg Says It Is Using Machine Learning to Deliver Near Real-Time Bond Prices
“You can trust this price as a reference price for that bond at that point in time,” says Tony McManus, global head of enterprise data at Bloomberg.
Near real-time pricing for tens of thousands of bonds is hitting the Bloomberg terminal, dramatically expanding the availability of such information to a broad group of traders and investment managers.
Bloomberg announced on Tuesday that it was launching the Intraday Bloomberg Valuation Service (IBVAL) Front Office, which uses machine learning to analyze “billions” of market data points to predict the prices of approximately 30,000 U.S. high-yield and investment-grade corporate securities that are eligible for TRACE (Trade Reporting and Compliance Engine). The securities are priced as often as every 15 seconds, far more frequently than the 2.5 million bonds and loans tracked by Bloomberg Valuation Service (BVAL) and which are priced at the end of each day based on real-time market observations or, for less liquid securities, comparable relative values.
IBVAL Front Office also captures and displays price information of new bonds issued moments after they are available for secondary trading.
Bloomberg will also provide scores to accompany the new service’s predicted prices to help users understand how liquid that security was at that point in time and the manner in which that price was produced.
The price predictions will also be available on Bloomberg’s real-time streaming market data feed, B-PIPE. “For a credit algo, for an automated ETF create-redeem process, or a portfolio trading strategy, that’s exactly what the market needs,” Tony McManus, global head of Enterprise Data at Bloomberg, told Institutional Investor. In the future, credit markets will only become more electronic, automated and algorithmic, and they need something like IBVAL Front Offfice, he added.
“We can create a pricing source that can give a very high level of accuracy. And it doesn’t matter what the credit rating of the issuer is, it doesn’t really matter what the liquidity profile of the bond is. It doesn’t really matter how volatile the market is,” McManus said. “You can trust this price as a reference price for that bond at that point in time.”
Bloomberg plans to add credit instruments traded in Europe to IBVAL in 2024 and others in Asia and the emerging markets shortly after. The service will also be used to power Bloomberg’s Tradable Trackers, a suite of rules-based indices that consists of 200 liquid bonds measuring the performance of the Bloomberg US Corporate Bond Index and related spreads.
Stocks trade on exchanges and their prices are available in real-time. But fixed income still trades over-the-counter, requiring buyers and sellers to negotiate prices on a screen or over the phone with dealers. That makes it hard to know what a bond — or an entire bond portfolio — is worth at any moment.
Pricing bonds regularly throughout a day is particularly challenging because the data related to them is fragmented or only known by the dealer who traded the securities. Traders and investors have to rely on market observations based on their own aggregation of various sources. “The net effect of that is that customers and market participants are forced to consume as many sources of pricing as they can get their hands on. And then build either quite sophisticated models themselves to figure out what the price in the market is at any point in time, or have very, very experienced trading desks that could interpret that information and figure it out themselves,” McManus said.
Bloomberg says IBVAL can do a lot of that heavy lifting for traders and investors.
Some asset managers tout the ability of their algorithms and traders to price bonds as a differentiator. IBVAL will effectively give all Bloomberg terminal users a tool like that.
“There is no doubt there will be some folks in the market that feel that they’re losing a competitive edge as a consequence of this. But this is completely consistent with Bloomberg’s philosophy since 1981, which is to facilitate transparent markets. It is in everyone’s interest for there to be more transparency in the credit market and across multiple fixed income asset classes,” McManus said. “Our goal is to democratize pricing within the fixed income markets and give the market participants the benefit of the additional transparency.”
How IBVAL is calculating bid and offer predictions is not completely clear. Bloomberg says it is using publicly available TRACE data, including a trove of transaction prices (all FINRA-member broker-dealers are required to report transactions of TRACE-eligible securities) and quotes it receives from broker-dealers with their permission. It is also using “related data that are unique to Bloomberg” and the company’s artificial intelligence and machine learning that “powers a range of analytics throughout the Bloomberg Terminal.” The company didn’t share specifics beyond that.
When asked if any human analysts had a material part in IBVAL’s price predictions, Bloomberg said: “IBVAL Front Office uses a combination of machine learning and more traditional rules-based models. The machine learning approach is being focused at the more illiquid end of the bond universe.”
In January this year, Bloomberg paid $5 million to settle charges brought against it by the Securities and Exchange Commission for failing to disclose to its BVAL customers that the end-of-day valuations for certain fixed-income securities could be based on a single data input, such as a broker quote. That practice did not adhere to methodologies Bloomberg had previously disclosed.
Bloomberg referred to its remedial efforts in the SEC’s order when asked about the data demands of IBVAL’s intraday pricing for similar securities.