CACEIS brings machine learning to sec lending

CACEIS brings machine learning to sec lending

  • Export:

Asset servicing firm CACEIS is using machine learning to ease the pressure on its securities lending desk and boost returns for clients.

The Crédit Agricole subsidiary, which lends securities on behalf of pension funds, insurers and asset managers, is applying artificial intelligence techniques to price loans of corporate bonds.

By using algorithms that become smarter as they are trained on large amounts of data, the company reckons it’s already seeing higher returns and reduced risks which allow its trading desk to compete with larger players in the market.

“I first started using machine learning technology to predict the outcome of French football league matches,” Aurélien Manson, a securities finance trader at the firm, told Global Investor.

“I made a return of over 200% last season and realised that such techniques could also be applied to our business, especially pricing the lending fee for our clients’ assets.”

Both Aurélien, and Dan Copin, head of equity finance at CACEIS, are convinced that machine learning is the future and will be as important a breakthrough as the internet was in 2000.

Machine learning differs from simple automation in several respects.

In basic terms, it is the science of getting computers to act without being explicitly programmed and constantly improving its own algorithms based on results achieved. Think self-driving cars, practical speech recognition or YouTube’s video recommendation engine.

Despite the scepticism of many fund managers, machine learning is already transforming investment strategies for asset owners.

Many hedge funds, for example, are using vast amounts of data, computing power in combination with machine learning techniques to refine strategies and out-perform the market.

From a securities lending standpoint, Manson believes the technology is improving not only clients’ remuneration opportunities, but also the reliability and robustness of CACEIS’ lending program.

Machine learning-assisted lending also provides a clear growth opportunity at a time when banks face significant regulatory constraints that impact revenue sources.

“We receive thousands of emails every day quoting many underlying holdings. To quote a competitive price, it is necessary to check over 20 features for each asset,” explains Manson. “We aim to think smart rather than just increase manpower for such tasks. Automation and straight-through processing can only take you so far, but true machine learning is providing a revolutionary solution.

“Leveraging innovation within a mature product category is a central part of CACEIS’ strategy. As a custodian bank and asset servicing firm, offering up-to-date technology to our clients is key."

New technologies such as machine learning also enable businesses to allocate human resources in a more efficient way.

“It’s a win/win trade,” Manson explains. “Human resources focus on the added-value tasks, and the more repetitive and robotic tasks can be automated and then incrementally improved via machine learning techniques.”

The trader reckons two thirds of the business volume that is low on the value chain can be automated while the rest needs to be performed by a human.

“Human traders are essential for managing parameters further up in the process, and still out-perform AI in soft data situations such as news analysis, profit warnings, bad/missing data or features not taken into account by the algorithms.”

Another benefit of AI or machine learning concerns risk, which according to CACEIS’ experts, it can help to significantly reduce.

“Our principal role as custodian is to ensure we have visibility on and restitution capabilities for client assets. With our algorithm, we can track each asset to determine how much we can lend safely/or need to borrow to rebuild a buffer,” Manson says.

Lifting the hood on the machine learning algorithms

The algorithms sort through emails received from counterparties, extracting data on asset types and quantities.

Then, cross-referencing CACEIS’ lendable asset database, apply business rules such as minimum tradable size, and in-house rules such as risk limits, liquidity risk etc., and then automatically respond to the counterparty.

“At this stage it is used solely to initiating the negotiation, however our goal is to enable it to handle progressively more, from processing incoming emails all the way to the final booking, with human intervention only when necessary for client relationship matters,” Manson adds.

“In technical terms, we completely modelled a corporate bond within our systems, using more than 36 variables to classify the bond by 3 features: the level, the sensitivity to the utilisation rate, and the convexity. Polynomial regressions and trees are the main analyses used to perfect the model, which was performed on our extensive lendable assets directory to create a vast reference database.”

“To understand how the algorithm operates, imagine our lendable assets are coordinates within a 3D space and the 3 axes, x, y and z correspond to the level, the sensitivity and the convexity.

When a new bond needs to be priced, the technology models the bond, calculates where its coordinates place it, then looks around that point for the nearest neighbours. Using k-NN algorithm, where k is the integer, you can, for example, look at the three “Nearest Neighbours”, and even weight them according to their proximity and provide a very accurate figure the most efficient lending level”.

“It’s certainly a very technical topic," Manson says, “nevertheless, we have the expertise within CACEIS, and are also strengthening our collaboration with Crédit Agricole’s DataLab department which centralises group-wide efforts to leverage FinTech solutions. Our clients will benefit greatly from the increases in efficiency, speed and returns that our investments into technology seek to provide”.

  • Export:

Related Articles