Sizing up Black Swans
by Richard de Rozario and Tim van Gelder
August 2015
Recently, the Swiss National Bank suddenly lifted its controls on the value of the Swiss franc. As the value of the franc rocketed up, many large banks, hedge funds, and retail brokerages took heavy, unexpected losses.
SNB’s move was what is often called a “Black Swan” - an event that is rare, high impact, and hard to predict. Black Swans are a challenge for risk management in financial institutions and many other business and government contexts.
Given their potentially devastating impact, they should be factored into any thorough risk management plan. Ideally this would mean having good estimates of how likely any Black Swan is, and how severe its consequences might be.
The problem is that the rarity and unpredictability of Black Swans seems to rule this out. In particular, historical data as used in the insurance industry are little use for Black Swans which, by definition, have rarely if ever happened.
As a result, all too often, organisations effectively discount Black Swans, with rationalisations like “Nobody can predict that” and “If it does happen, we’re all in the same boat”.
Unfortunately, when there are many organisations exposed to these risks, or there are many such risks, the likelihood of at least one occurring is substantially increased.
Moreover, optimistic risk strategies are less acceptable these days. One reason is that regulators such as APRA are requiring evidence that financial institutions can deal with the unexpected. Credible risk management plans now need to demonstrate sound approaches to rare catastrophes.
Ironically, once we accept that something needs to be done to prepare for Black Swans, the problem becomes easier. We are now less interested in “what is the probability?” and more interested in “how big can it get?”
The good news is that risk managers are developing better methods for sizing up Black Swans, individually and in combination, even without historical data.
One new development is what we call “scenario analytics.” This combines the use of qualitative scenarios with probabilistic modeling. Both are well-developed and widely-used; the new approach brings them together in a coherent framework.
Scenario thinking involves exploring multiple possible futures, based on discussions with participants about the drivers and consequences of change. Scenarios are qualitative portraits of how things might unfold, rather than predictions of what will happen.
Traditional scenario thinking helps planners expand their understanding of the range of possibilities and develop more resilient strategies, but it typically doesn’t deliver quantitative assessments.
Scenario analytics adds a stage in which various components of the scenario are numerically estimated in workshops, and computer-based modeling is then used to build up a probability distribution for the type of event described in the scenario, given its occurrence.
Starting from a particular scenario technique called "backcasting", participants are asked to assume a future in which a severe event of a particular type has occurred. In the world of financial risk, the focus might be, for example, a very large loss due to a natural disaster. This backcasting step helps to overcome the anchoring to historical and present knowledge. It also aims to deal with possible discontinuities in future trends, by “leaping over” such irregularities.
The participants are then asked for the most plausible way in which such an event might have come about. Experience shows that people typically have difficulty directly estimating probabilities of rare events, but they fare better with finding at least partial explanations of how such events could happen. Scenario analytics does still deal with questions of probability, but they are more about the “relative likelihood” -- meaning, which explanation is more likely than another.
The information about contributing factors is then used to identify key variables for which quantitative ranges of values can be estimated. Possible combinations of outcomes, together with their probabilities, are then generated and quantified using a technique known as Monte Carlo simulation.
A thorough program of scenario analytics for a large organisation might assess dozens of different event types in this manner. In this manner, scenario analytics augments the depth of operational knowledge of an organisation with sound quantitative methods.
This provides more rigour to scenarios, but also adds value in other ways. For example, the scenario models enable us to numerically explore many variations of the scenarios, without necessarily having to re-run the labour intensive discussions and workshops.
There are of course limitations. The technique can only be applied to those Black Swans we can envisage; we may still end up being “side swiped” by the unimagined.
Scenario analysis may also be susceptible to the biases in the way judgements are elicited.
Finally, scenario analytics is not a simple matter. It does require quite a bit of careful effort to thoroughly analyse an appropriate range of event types.
Nevertheless, properly done, the combination of qualitative and quantitative techniques can substantially improve our ability to measure and prepare for Black Swans.
August 2015
Recently, the Swiss National Bank suddenly lifted its controls on the value of the Swiss franc. As the value of the franc rocketed up, many large banks, hedge funds, and retail brokerages took heavy, unexpected losses.
SNB’s move was what is often called a “Black Swan” - an event that is rare, high impact, and hard to predict. Black Swans are a challenge for risk management in financial institutions and many other business and government contexts.
Given their potentially devastating impact, they should be factored into any thorough risk management plan. Ideally this would mean having good estimates of how likely any Black Swan is, and how severe its consequences might be.
The problem is that the rarity and unpredictability of Black Swans seems to rule this out. In particular, historical data as used in the insurance industry are little use for Black Swans which, by definition, have rarely if ever happened.
As a result, all too often, organisations effectively discount Black Swans, with rationalisations like “Nobody can predict that” and “If it does happen, we’re all in the same boat”.
Unfortunately, when there are many organisations exposed to these risks, or there are many such risks, the likelihood of at least one occurring is substantially increased.
Moreover, optimistic risk strategies are less acceptable these days. One reason is that regulators such as APRA are requiring evidence that financial institutions can deal with the unexpected. Credible risk management plans now need to demonstrate sound approaches to rare catastrophes.
Ironically, once we accept that something needs to be done to prepare for Black Swans, the problem becomes easier. We are now less interested in “what is the probability?” and more interested in “how big can it get?”
The good news is that risk managers are developing better methods for sizing up Black Swans, individually and in combination, even without historical data.
One new development is what we call “scenario analytics.” This combines the use of qualitative scenarios with probabilistic modeling. Both are well-developed and widely-used; the new approach brings them together in a coherent framework.
Scenario thinking involves exploring multiple possible futures, based on discussions with participants about the drivers and consequences of change. Scenarios are qualitative portraits of how things might unfold, rather than predictions of what will happen.
Traditional scenario thinking helps planners expand their understanding of the range of possibilities and develop more resilient strategies, but it typically doesn’t deliver quantitative assessments.
Scenario analytics adds a stage in which various components of the scenario are numerically estimated in workshops, and computer-based modeling is then used to build up a probability distribution for the type of event described in the scenario, given its occurrence.
Starting from a particular scenario technique called "backcasting", participants are asked to assume a future in which a severe event of a particular type has occurred. In the world of financial risk, the focus might be, for example, a very large loss due to a natural disaster. This backcasting step helps to overcome the anchoring to historical and present knowledge. It also aims to deal with possible discontinuities in future trends, by “leaping over” such irregularities.
The participants are then asked for the most plausible way in which such an event might have come about. Experience shows that people typically have difficulty directly estimating probabilities of rare events, but they fare better with finding at least partial explanations of how such events could happen. Scenario analytics does still deal with questions of probability, but they are more about the “relative likelihood” -- meaning, which explanation is more likely than another.
The information about contributing factors is then used to identify key variables for which quantitative ranges of values can be estimated. Possible combinations of outcomes, together with their probabilities, are then generated and quantified using a technique known as Monte Carlo simulation.
A thorough program of scenario analytics for a large organisation might assess dozens of different event types in this manner. In this manner, scenario analytics augments the depth of operational knowledge of an organisation with sound quantitative methods.
This provides more rigour to scenarios, but also adds value in other ways. For example, the scenario models enable us to numerically explore many variations of the scenarios, without necessarily having to re-run the labour intensive discussions and workshops.
There are of course limitations. The technique can only be applied to those Black Swans we can envisage; we may still end up being “side swiped” by the unimagined.
Scenario analysis may also be susceptible to the biases in the way judgements are elicited.
Finally, scenario analytics is not a simple matter. It does require quite a bit of careful effort to thoroughly analyse an appropriate range of event types.
Nevertheless, properly done, the combination of qualitative and quantitative techniques can substantially improve our ability to measure and prepare for Black Swans.