Famous Maximum Clique Algorithm Applied in Bond Market Simulation

logo-cropped-11Overbond team on the left, University of Toronto software engineers on the right and University of Waterloo winner Toby Huang on the video link being presented the award

We had a busy week at Overbond, after we announced the closure of our $7.5 million seed financing round earlier last week. This past weekend June 17th-19th, instead of celebrating the close of the deal, we did something a little bit different. At the University of Toronto and the University of Waterloo, weheld an Overbond hackathon event for Canada’s brightest computer engineers. They were applying one of the most advanced AI algorithms in computer science to the bond trading world — a fintech simulation that promised to empirically demonstrate the benefits of the Overbond platform in the marketplace.

Most hackathons are hands-off: the sponsoring company stays out of the action. But this weekend, our approach was different: we worked with the students throughout the weekend — workshopping ideas, holding tutorials, and answering questions — in order to bring the fintech world to them, and help them understand both the complexity and importance of efficiency within primary capital markets. The hard work and commitment paid off.

Students successfully tackled one of the most complex computer science problems ever attempted at a hackathon – the famous maximum clique algorithm. The maximum clique algorithm that students took on is, essentially, the task of finding the largest subset of points (in our challenge, these points represented dealers, issuers, and investors — a simulation of a large trading environment) that were directly connected to one another, forming a complete marketplace. Students ran a simulation of two distinct trading worlds: one where traditional communication technologies (i.e., phone and email) dominate, and a future world where the capabilities of the platform like Overbond make the network both more connected and more efficient. By determining this, students were able to test the simulated network’s connectivity, analyze the base costs of different communication methods, and in effect model real world trading transaction costs.

In the first simulated world, we modeled the network on traditional bond trading markets, where communication is largely done through networks of dealers, issuers and investors. In the second, we introduced to that world the capabilities of the platform like Overbond — namely, more frequent connections between dealers, issuers and investors, and decreased costs associated with all connections. By applying advanced graph theory algorithms to large data sets (each set comprised over a million data points), students were able to statistically demonstrate lower average transaction costs in the second world, in the form of increased efficiency, connection density, and lower communication costs. For our team, this did two things: first, it gave us statistically relevant data demonstrating how our platform improves market efficiency and reduces infrastructure costs; second, it gave us lots of data to analyze. (We are planning to publish the results of the simulation in a white paper later this year.)

It’s hard to overemphasize how challenging this problem was. The maximum clique algorithm problem has been the subject of a number of high-level PhD theses, and is a key area of study in the field of computer sciences. It is a highly relevant and practical problem for software engineers that has applications within the bond trading marketplace. One of the main features of the problem — and what makes it so challenging — is not just finding the clique, but doing so efficiently and quickly (minimizing processing time). Many of the students were able to identify the clique early in the day on Saturday, but (as with any optimization-based challenges) the difficulty comes with trying to do so in the most efficient and streamlined way possible. If not, mere computing time with the average laptop processor would take 100,000 years to execute non-streamlined algorithm.

The winner: Toby Huang, a recent graduate from the University of Waterloo with a dual degree in math and business. His code executed in under two seconds — an extremely fast runtime. “I started with a brute-force solution,” he said. “That was way too slow. And then I found a paper that had a ten-line algorithm that was pretty good.”

For a young company like Overbond, talent is everything, and the right talent has the potential to reap exponential returns. There was a tremendous amount of skill and ingenuity on display this weekend — talents that might soon make for an invaluable addition to the Overbond team.



About Vuk Magdelinic:

Before founding Overbond, Vuk’s career spans over 10 years in capital markets and technology. As PwC Risk and Regulatory consulting manager Vuk led large digital transformation programs at Deutsche Bank and BNY Mellon in New York City. Prior to that he worked at CIBC Fixed Income trading floor in Toronto in structured products origination capacity. Vuk has collaborated on numerous publications addressing key trends in fintech innovation. Vuk holds electrical engineering degree from University of Toronto, MBA from Ivey school of business and is an avid abstract painter.