Author: Ron Dembo
When faced with radical uncertainty, governments, corporations, and even individuals making difficult decisions must be able to think about the different possible futures that could occur. They must be able to generate scenarios.
A useful set of scenarios will span the whole range of possible future states and contain the best and worst that we may reasonably expect in a given timeframe, including what appear at first glance to be extreme or unlikely. From there, risk thinkers can form their strategy by hedging the worst possible outcomes and optimizing the chances of securing the best possible outcome.
Scenario Generation for Covid-19
We are back in the early part of 2020. COVID has been declared a pandemic by the World Health Organization. It is ravaging Northern Italy and has just begun to appear in Canada with a handful of known cases. The way it is spread, through aerosol transmission, has been verified. To prevent the virus spreading it is also determined that proper masks, handwashing, and social distancing, if properly applied, reduce the spread in the population by over 95%.
Compliance then becomes critical to the effectiveness of preventive measures. But compliance is also highly uncertain. Consequently, the key risk factors are:
- Compliance with mask-wearing
- Compliance with social distancing
- Compliance with regular hand washing.
The government orders the population to voluntarily follow mask wearing with proper masks, hand washing, and social distancing of at least 2 meters in any encounter. The risk is that the population will not comply with these measures and spreading will occur.
Epidemiological models show us just how quickly spreading can occur depending on how well people comply with the combination of these measures.
Obtaining the Forward-Looking Data
So, the first problem is to obtain estimates for how many people will comply with each of these measures and then to develop scenarios on the combinations of compliance with these measures.
To do this we go out to the population at large to estimate how well the population compliance would be. This could be done through Structured Expert Judgment, a fancy way of saying polling done scientifically. The result is a distribution of values for each risk factor, shown below. These come from an actual study of the uncertainty of the risk factors in a poll done on 300,000 Canadians at the end of March and beginning of April 2020.
The original expectation was that 65% of the population would comply with social distancing but the results showed that the percentage compliance was much lower, with a few people not complying at all and most people complying less than expected. But what is clear from the social distancing frequency distribution is that the % compliance could vary wildly.
We then create the same distributions for mask wearing and hand washing and read them in the same way.
Now we can understand various possibilities for how these risk factors interact as a group by drawing a scenario tree, like the one below. Each path in this tree is a different scenario. For example, the red path is the scenario:
Hand washing compliance is lower than we expect and so is social distancing, but mask wearing is better than expected.
To get the values for just how much more or less a given factor is than what we expect we can use values derived from the distributions taken from the polls.
In practice more complicated forms of graphs might be used, but this tree illustrates the idea for how we might generate scenarios on combinations of risk factors.
Using it, risk thinkers can examine the worst-case and best-case outcomes for the combinations of risk factors, and the likelihood of their occurrence. And they can form a cohesive strategy based on the result.