On April 20, 2020, the impossible happened: prompt WTI futures price settled below zero. Traders were willing to pay counterparts to take delivered oil off their hands at Cushing, Oklahoma, because there were no remaining places to store it.
Managing a portfolio of non-linear derivatives, such as options and more exotic products, through volatile markets like we’ve seen in the COVID-19 pandemic is challenging even when systems are up and running. When core assumptions like “oil prices are always positive” are violated, however, risk management analytics break right when we need them the most.
Exacerbating the pain of a broken risk system is the fact that getting it back up and showing you accurate risks takes a long time - unless you’ve got a system designed to be both agile and controlled.
I’ve lived through several periods of extreme volatility in my career as a quant and market maker. I’ve seen desks profit handsomely by managing their positions effectively while continuing to service their clients because they had robust and nimble risk systems, and I’ve seen desks drop years of accumulated profit in weeks because they did not.
The main lessons I’ve learned: know your book; understand your models; be able to react nimbly and with confidence; and partner with your developers.
Know Your Book
Portfolios of options and exotics can have thousands to hundreds of thousands of open positions at any time, and understanding what happens to your book under large market shocks is impossible without flexible real-time position and scenario reporting.
Position reporting lets you zero in on the makeup of your portfolio where it matters. For example, knowing what strikes are nearby for options expiring in the short term helps you understand your “pin risk”: how delta changes quickly as the underlying asset price crosses the strike level because of the concentrated gamma there. This is especially interesting for options that settle into more complex structures, like swaptions.
Similarly useful is being able to view how risk metrics change under different market scenarios. Traders are used to “ladder” reports that show their headline risks under, for example, different parallel or shaped shocks to the futures curve. These are useful for understanding non-linear risks beyond instantaneous gammas and cross gammas, especially for larger moves when the local “Taylor series” second-order approximation breaks down.
In stressed markets, traders often want to see risk metrics under a custom set of specific scenarios which they feel represents plausible moves for the particular crisis period they’re experiencing. Your risk system should allow for this as well, either through custom scenario reporting, or more likely, through new custom reports written by financial engineers working on the desk who can translate the qualitative instructions from traders into quantitative scenario definitions.
That ability to create new fully custom reports quickly, generally through programmatic extensions to the reporting tool, is a hallmark of the best risk systems. As Federal Reserve Board Governor Kevin Warsh said, “Once you’ve seen one financial market crisis… you’ve seen one financial market crisis.” It’s critical to give traders the flexibility to understand their book in the ways that make sense for the specific environment in which they find themselves.
Being able to report on theoretical risk moves is useful, but even more important is seeing your risk calculated live from real-time market data inputs. Some risk systems do this through approximation techniques such as a Taylor series based on the previous night’s closing risks, but that approach is extremely fragile in extreme market conditions - like when oil prices drop $50 in a day. Full real-time risk recalculation is generally the most robust approach, and that often requires considerable compute resources to effect with a cycle time that’s small enough to matter in fast-moving markets. A cycle time of 15-30 seconds is generally sufficient for complex portfolios. Elastic cloud computing is an efficient way to scale up computing resources when you need them, and avoids paying for compute when you don’t.
Understand Your Models
The worst thing a risk system can do is pretend to be right when it’s wrong, rather than display an error. Traders trust the risk metrics that they’re shown, and if those metrics are not accurate, they’ll put on the wrong trades.
A common problem with risk systems is testing pricing and risk models under normal market conditions but not under stressed conditions. To be fair, it’s hard to get this right: there are a lot of complicated ways that markets can be stressed, and testing a significant fraction of them takes a lot of disciplined effort in model review when conditions are normal.
One example of this is the approximate intraday risk described above, but it also happens in more subtle ways. A precious metals trading desk had traded self-quanto options and used the pricing models from their foreign exchange desk; but both at-the-money volatilities and skews in metals were much higher than in FX at that point, and the FX model had not been tested in those conditions. A limit on a numerical integration was inappropriate for that case, and the self-quanto options showed model prices materially below fair value. The precious metals desk discovered this error after selling a significant notional of the product at the wrong price to some lucky counterparties, and paid $9M to learn the lesson that they should make sure that models are tested for their case before using them.
Be Able to React Nimbly and With Confidence
When oil prices went negative in April 2020, many risk systems stopped working: the assumption that oil prices are always positive was baked into the low-level analytics for derivatives pricing.
Quants were able to quickly identify the places in their analytics that needed to be upgraded. But for most shops, risk systems require days or weeks to be deployed. Traders were pressed into using questionable approximations in hand-crafted spreadsheets to cover the gap, and ended up running very conservative positions.
At shops with the best risk systems, however, quants could deploy their fixes intraday, while still satisfying standard controls on technology deployment. Their traders navigated that market turbulence with all their instruments working and were able to continue to serve their clients and investors.
Partner With Your Developers
The flexibility to quickly fix a system is similar to the example of custom scenario reporting described above, and lets quants and financial engineers be true partners to the traders in the business.
Giving your developers - whether they’re quants, data scientists, or financial engineers - the tools to be able to partner with you is a hallmark of the most successful shops. They need to be able to join in high bandwidth conversations about how to improve the business, then translate those ideas into code - and then deploy that code into production so traders can use it to make more money.
An important challenge in that model is balancing nimble and iterative system changes with enterprise control policies designed to protect against accidental errors. Setting controls too tight leads to inflexible and slow-moving systems that can’t respond in a crisis and cuts developers off from the business, but setting them too loose leads to traders hedging off the wrong numbers.
It’s tricky to strike that balance, and getting that right is what we at Beacon have spent our careers refining.