Buy-side risk manager of the year: Vanguard

Risk Awards 2020: Fund giant gave more risk work to machines this year - from duration hedging to op risk

Manish Nagar

Manish Nagar, head of risk at Vanguard, set his team a challenge at the start of 2019: to do less. Since then, the team has automated the equivalent of 10,000 man-hours of previously hands-on work – roughly 5% of its output.

It’s not just the humdrum that has been given over to machines. Vanguard’s risk team constructed a new set of global computerised ‘algo wheels’ that pick the best execution algorithms to allocate the firm’s trades to. It built an optimiser that systematically evens-out unwanted rates exposure. In operational risk, Vanguard used machine learning to scour trading data for signs of mistakes or bad practices. 

“Technology is at the forefront of everything we do,” Nagar says. “The intersection of risk and technology continues to grow. Big datasets are available for us to process, and you need a technical skill set and an investment skill set to get intelligence out of that data.”

With 90 people located across sites in Australia, the UK and USA, Vanguard’s global risk division oversees market risk and operational risk, and includes a quant research team that contributes to the formulation of risk models and relative value tools.

The quant team, which also runs Vanguard’s transaction cost analysis program, spent much of the past year working on the firm’s automated algorithm wheels and a system for generating daily broker-dealer scorecards, Nagar says.

The wheels make their choices with zero human interference, using tick-level market data, on-demand computing power and data storage from the cloud. The best algos get more business. Weaker algos get shunted into the “penalty box”. No single person could do the calculations to make those calls rapidly enough, Nagar says.

“As you can imagine, there are tons of sell-side firms out there. Each will have 10 to 15 different algos they’re trying to push. It’s impossible for us to evaluate the algos unless we measure the performance of trades,” Nagar says. Dealer scorecards are used to track performance versus their peers across the different services they provide.

The project has saved “millions of dollars” in transaction costs for Vanguard’s funds, Nagar says.

In fixed-income trading, Vanguard deployed a new optimiser and automated trader to level out the firm’s exposure to specific points on the yield curve.

Big datasets are available for us to process, and you need a technical skill set and an investment skill set to get intelligence out of that data

Manish Nagar, Vanguard

“We didn’t want portfolio managers to be exposed to duration in corporate portfolios,” Nagar explains, referring to “active” or unwanted duration between funds and their respective benchmarks, because maturities of the bonds in each don’t exactly match, for example. The optimiser and automated trader effectively hedges exposures at different points on the curve without a human trader having to calculate how many bond futures to buy.

The firm also overhauled its methods in its risk-taking framework in fixed income. Previously, when fund managers put on different types of hedging trades – an inflation or steepener trade, for example, or a mortgage trade – the risk team worked to keep exposure within a set of common ‘guardrails’. The risk team realised, though, that these types of universal controls failed to recognise the different levels of risk incurred by different trades.

“We modified the whole framework this year for all rates portfolio managers around the globe,” Nagar says. The team transitioned to a risk-based framework where hedging exposures are driven by past covariances between asset classes. The new curbs are therefore set using a more forward-looking view of the aggregate risk in a portfolio as opposed to managing allocations at the exposure level, Nagar says.

Machine learning

In operational risk, Vanguard applied machine learning to risk-level data. The firm has a forensics team tasked with looking at multiple fields of data and using machine learning tools to find patterns that indicate errors or potential wrongdoing. A portfolio manager might execute buys and sells on the same trade, for example, or input and then cancel orders just before they are placed in the market.

“Those things are very hard to pick up on a day-by-day basis,” Nagar says. “But if you look at longer-term trends of data, using our operational risk framework and the machine learning risk tools the forensic team has put to work, you can figure out any trends of that sort going on.”

One pattern the team observed was that portfolio managers would sometimes take outsized exposures on favoured tickers. That is inconsistent with Vanguard’s philosophy of diversifying sources of alpha, Nagar says, so the firm introduced a new framework this year on issuer concentration.

“We put a policy in place that says we are not willing to have more than ‘x’ position from a single issuer, which reduces this behaviour and encourages more diversification,” he says.

To support its ongoing automation work, Vanguard set in motion a hiring drive – to bring in new people with a knack for technology as well as risk management. “We have made tons of changes in our global organisation, hiring some PhDs, as well as some talent from outside of Vanguard in our three sites,” Nagar adds.

The work will continue, Nagar says. “Technology is something we have gotten much better at in 2019 and something that will definitely be on the agenda in 2020, as well.”