We all know that the future is not going to look like the past. The mantra that past performance is not indicative of future returns applies to risk as well.
Today we face risks from inflation, climate change and the "irrational exuberance" of record high valuation ratios, that are all stark when compared to the recent past. The volatility of prices over the past few years forms the basis of risk measurement for most all risk applications, but it simply is not reflective of the looming risks. We also know that our greatest risk concerns are richer and more textured than can be put into a smattering of risk statistics.
Because of this knowledge, we rely on scenarios, what-if exercises that project the effect of events that are hypothetical but of real concern. Every risk management application includes scenarios as part of its analysis. And, even if we are not using a particular risk system, positing and mulling over scenarios is a natural part of how we think about risk. There is a critical disconnect, however, between what we do when we are thinking through risk issues and what risk systems actually do. And for the most part, the machines are doing it wrong and we are doing it right.
Risk systems miss five aspects of risk scenarios that we intuitively recognize as we think about scenarios ourselves:
- Scenarios are not one-shot events
We know that events like inflation don't happen all at once. They occur gradually, and recover gradually. Yet, risk systems assert them as a one-shot drop. Look at any risk system and it will depict each scenario as a bar in a chart. Some down 15%, some down 22%, some showing equities dropping by one amount and Treasuries by another. But they miss the point that we all understand: Events are dynamic--a process.
- Its inefficient to consider every possible scenario
You can only constructively worry about so many things happening to the markets. In my experience there are only three or four things that are of material concern; things that could drop the market by, say, 15% or more. Yet, many risk systems stress everything under the sun. If a risk system in the final analysis is simply saying "anything can happen" it is not of much help. No one would keep around a risk manager who comes in every day and simply says, "be careful."
- Scenarios differ in their time frames
When we ignore dynamics of risk, one thing that we miss is the fact that different sorts of events have different time frames. Some, like a leverage-driven downturn can happen quickly. Others, like the effect of a recession are gradual and drawn out. Again, we all know from our own experience, and common market sense, that these are different flavors of risk, but risk models do not capture this important dimension.
- The current market environment matters
Scenario risk has two components: 1) the event itself, and 2) the vulnerability of the market to that event. An event is going to be worse both in the size and the speed of the drop if the market is highly leveraged, more concentrated or less liquid. But risk models do not condition on the current market environment, they measure risk based on the past, usually the volatility of returns over the last one or two years.
- Scenarios are a narrative—a story with twists and turns
I've been the chief risk officer at a number of large financial institutions. When we have been concerned about a scenario, the CEO or principal of the firm will get together with me and the senior traders or portfolio managers and work through the issues. We will be sitting around the table, basically constructing a story, a work of fiction that might turn into a reality. That is, we understand that the large risks posed by scenarios are a narrative. They are not printed out of a computer. There is a story line with twists and turns. I often liken it to driving along a mountain road at night, with headlights only illuminating to the next curve.
Plato is reputed to have said, "Those who tell the stories rule the society." For risk, those who tell the stories rule the market. How do you model this? How do you put it into a risk system?
As things stand right now, for scenario analysis human beats machine. For machines to do better, we need to give them a dose of reality. Inject them with some market sense. There is ready room for a lot of improvement by going after some of the low hanging fruit, like recognizing scenarios are more of a response curve than a bang-bang event. The scenarios in risk applications can improve, but for now we humans are one step ahead.
Rick Bookstaber is Founder and Head of Risk at Fabric. He previously held chief risk officer roles at Morgan Stanley, Salomon Brothers, Bridgewater Associates, and the University of California Regents and served at the U.S. Treasury in the aftermath of the 2008 crisis.