After reaching the limits of MATLAB, a group of quantitative analysts at Bloomberg set out to develop a new system for creating and exploring machine learning models. That effort ultimately grew into BQuant, a Python-based data science environment designed specifically to help finance professionals analyze data more effectively. Yesterday, he unveiled BQuant Enterprise, which takes this data science notebook environment to new heights.
Data science is nothing new for Bloomberg, which has for some time been integrating advanced statistical capabilities into its popular Bloomberg terminal. But the story behind BQuant’s development is slightly different.
The story begins in 2013 with Bloomberg’s Quant Research, a team tasked with incubating new ideas for customer products, solving customer problems, and generally pushing the boundaries of quantitative analysis. The company was using a popular but proprietary data science language, and it was time to move on.
“We have reached the limits of what we can do with MATLAB,” Bruno Dupire, head of quantitative research, said in a Bloomberg blog post. âWe realized that we had the tools and knowledge to create our own rich IT environment that could be used to simultaneously explore financial analyzes in multiple dimensions. “
Quant Research engaged members of Bloomberg’s Desktop Build Group, which creates custom machine learning models for Bloomberg clients using the Bloomberg Desktop API. This group was looking for tools to speed up development and avoid rewriting boilerplate code. They eventually came up with a series of new tools, including a data query language called Bloomberg Query Language (BQL), a Bloomberg-specific extension of the D3.js visualization library called bqplot, finance-specific libraries, and links to Python and .NET.
Researchers saw the writing on the Python wall in 2014, and hooked BQuant to the open source group iPython, which would eventually become Jupyter. In addition to being able to write and test Python code in this data science notebook, the BQuant offering gave users access to Bloomberg’s comprehensive financial data.
The desktop version of BQuant has been a hit over the years among traders, especially when it comes to quantitative exercises such as factor model assessment, backtesting strategies and portfolio analysis across all classes. of assets. This popularity has grown thanks to the ability to tap into Bloomberg’s growing universe of financial data, including alternative data and geospatial data.
With the launch this week of BQuant Enterprise, Bloomberg is scaling up the quantitative tool. Bloomberg runs BQuant Enterprise in a Kubernetes container, which allows it to scale as needed with available cloud resources. It’s also integrated with Apache Spark, giving the offering the parallel processing capability needed to handle very large data sets.
âBy using distributed computing, users can run tasks overnight on a secure cluster to solve problems that are far beyond the capabilities of a desktop computer,â Bloomberg explains. âIn addition, BQuant Enterprise can ingest data from customers’ own datasets or from third parties. “
In addition to bundling these frameworks and libraries into a cloud-based offering, BQuant Enterprise also helps customers monitor user activity, enabling better monitoring of data flows that become critical to operations.
âUsing the power of the cloud, we provide clients with a turnkey environment where they can connect to their existing systems, bring in their own data, mix it with Bloomberg’s full data sets, and improve collaboration for their business professionals. investing when testing and deploying new quantitative investing strategies, âsaid Shawn Edwards, CTO of Bloomberg, in the Bloomberg press release.
The hosted offering is already paying dividends to Thornburg Investment Management, an independent investment firm with $ 49 billion under management. According to Igor Kuznetsov, head of portfolio analysis at Thornburg, the company created its own web-based data science application, but it was unable to ingest Bloomberg data.
âAfter selecting BQuant Enterprise, we were able to bypass these limitations Kuznetsov,â Bloomberg said in the press release. âMy team was able to immediately access a wide range of standardized data; spend less time handling it; apply additional backtesting capabilities; and generate more ideas faster than before.
How geospatial data drives insight for Bloomberg users
Data science inside the Bloomberg terminal