Coronavirus and Modeling Fear

Catastrophe Q&A: Model Maker

Metabiota CEO Nita Madhav

A new coronavirus strain with origins in Wuhan, China is quickly spreading globally and could have the potential to rival previous coronavirus outbreaks such as 2003’s Severe Acute Respiratory Syndrome (SARS-CoV) and 2012’s Middle East Respiratory Syndrome (MERS-CoV) that caused billions in economic and insured losses.

Risk Market News spoke with Nita Madhav, CEO of  disease risk analytics firm Metabiota, about the risk that the current outbreak possesses, the importance that fear plays in loss estimates and the challenges of validating an infectious disease model.

Risk Market News: How would you contrast and compare the current coronavirus outbreak to previous outbreaks involving SARS-CoV and MERS-CoV?

Nita Madhav: This virus is in the same family of viruses causing SARS and MERS , and the symptoms and how people are experiencing the virus are similar to those outbreaks. Current data does suggest the virus is easier to spread, in the sense that we now have close to 8,000 cases within a month. That's a much faster rate of growth than we've seen for the other outbreaks.

But on the other hand, the death rate appears to be lower of those who get infected. About 2.5% of those cases seem to be experiencing a fatal outcome. Of course the data is evolving, but those are a few points on how the current outbreak differs from their previous outbreaks.

RMN: How does a low mortality, high transmissibility virus like that figure that into your model?

Madhav: It suggests that there's going to be quite a bit more spread of the infection.

I'm sure everyone is aware of all of the travel bans and restrictions on movement that are occurring in China. And it remains to be seen whether these measures had been put into place in time. Also, is still needs to be determined what the onward transmission will look like in locations that have imported cases.

We can see from the current data that the incubation period -- from the point of time when someone is exposed to when they might start showing symptoms -- could be as long as 14 days. We do need to keep an eye on this outbreak for some time before we will know for sure what you know it's getting contained or continues progressing.

RMN: What are some of the unique modeling challenges of highly pathogenic strains like coronavirus?

Madhav: As with any model there are going to be challenges and assumptions that will come into play. I think for an outbreak like this, the biggest challenge we see is that there are a high number of unknowns and uncertainties.

As more data come in the picture can change drastically. For example, just about a week ago, there were only limited reports that the virus can spread from person to person. And now it's showing person to person spread is occurring more frequently than we had realized. And it's also showing that people may be able to spread the virus before symptoms show up. That makes it much more difficult to control.

All this new information about the virus, how it spreads and how it infects people, is very important in the modeling and we want to make sure to capture it. In the early stages of an outbreak there's just a lot of information coming in and the model can be sensitive to some of those new bits of information.

RMN: In terms of your model, what are some of the primary drivers of insured and economic losses?

Madhav: In this type of outbreak especially, we do expect to see that health care and response costs are likely to be high. But in actuality, the largest driver of financial losses will often be business interruption and loss of attraction due to fear induced behavioral changes. We've seen from previous outbreaks including SARS.

Some analysis that we've done around hotel occupancy rates shows that there can be a drop in revenue close to 30% to 40%. Even if the levels eventually bounce back, it can take months to recover.

As a result, this outbreak is going to have significant implications for losses in the hospitality industry, especially in areas where people travel or to congregate in specific locations. We're also likely to see some pretty strong economic impacts from travel bans. One of the estimates for SARS, for example, in terms of economic losses was about $54 billion. So that is significant in the global economy.

Given how things are shaping up in the current outbreak, we could see a similar magnitude of loss and the majority of these losses are uninsured. In recent years there have been epidemic related non damage, business interruption policies that have been developed which could help companies with some balance sheet protection against these types of losses.

RMN: How do you measure the potential pandemic mortality tail risk in extreme events?

Madhav: I think it's important to keep in mind there is the potential with pandemics to have a very heavy tail. Although, with this outbreak, is still remains to be seen.

Some past outbreaks had a very fat tail. For example, the 1918 influenza pandemic killed an estimated 2% to 5% of the global population. If something like that were to happen today, that's a pretty big shock and something that really should be taken into account in any sort of shock scenario modeling.

RMN: What has changed in the modeling industry since the previous major outbreaks?

Madhav: One of the major advances is that we've recently been able to quantify the level of fear associated with the outbreak through something that we've developed called the sentiment score.

The sentiment score takes information about the pathogen causing the outbreak, including a number of factors such as; how deadly is it, how is it transmitted, what types of symptoms does it cause, and whether or not there's any vaccines or medicine that would work against it.

We take that information and we apply it into an algorithm that produces a score that provides us a fear rating. We can incorporate that into our modeling to better assess what the economic losses are. That’s because it is often surprising to know that the amount of loss is oftentimes not necessarily correlated to the number of cases and deaths that are seen. In fact, even smaller outbreaks can have very high economic losses due to some of the business interruption and loss of attraction.

RMN: Can you offer an example of an outbreak with a high sentiment score?

Madhav: In the grand scheme of things, the SARS event would have been considered a smaller outbreak but with large loss. Another example is the Zika virus that struck a couple of years ago and had a significant economic impact in Latin America. That was mainly due to the presence of symptoms in newborn children, specifically microcephaly. The presence of potential birth defects was a very important driver for the loss in the tourism industry.  However, we saw very few deaths recorded from Zika even though there were a large number of infections in the general population. That’s something in terms of economic and insured loss that would be not necessarily captured if we were looking at the number of deaths that were caused.

RMN: What are some of the best practices you’ve seen in infectious disease model validation?

Madhav: Like all modeling, we always want to keep in mind is ‘what is the objective of the model that we're building.’  We want to make sure that we can make use of the model, but that it may not always give us exact to answer. There’s a famous quote by George Box [British statistician] that “All models are wrong, but some are useful.”

We do try to make the models as useful as possible and validation is a very important component. We validate both with a bottom up approach and a top down approach.

As we're building these models there are many components and parameters that we incorporate. For example, we're using computational epidemiological approaches which allow us to follow a disease as it starts in one location and then use a state transition model to follow where the disease spreads on a daily basis from person to person, place to place over the global population.

Within that model, we're trying to estimate things such as how it spreads and then incorporating parameters such as the incubation period, the infectious period and transmissibility. We compare the scientific literature to the distributions, parametric or non-parametric, and make sure that the science is supporting what we come up with.

Then we run the models and we assess how well they perform against historical outbreaks, both with the magnitude of infections and hospitalizations and deaths that they're estimating along with how the temporal progression of the outbreak looks as well as the geographic footprint. We're comparing all of these different outcomes to see whether our model is producing plausible results. That allows us to estimate the severity component and we also look at the frequency, including things such as the timing distribution between how often the outbreaks occur.

When we put that all together we're looking from the top down. How does this exceedance probability curve compare to what we've observed historically? What indicates if it scientifically plausible? We also understand that there is a lot of subjectivity that goes into the modeling and we run an extensive model review process with independent experts in the field.

We're very open and transparent about the assumptions and the results and work collaboratively to  ensure that we're incorporating the latest methods and the best science into creating these models.

RMN: What sort of validation questions do you receive from users?

Madhav: To some extent depends on who is the user. From time to time we get questions that are really in the nitty gritty and from a bottom up view.

I would say for most of the users it's more of a top down view. Looking at how does this compare to the historical data? I think there are always pitfalls to simply comparing to historical data since things change over time, but we make sure that we're accounting for the latest transportation patterns, the latest types of therapeutics and vaccines available.

There's also strong evidence that the number and frequency of these events has been increasing over time and so we need to account for that.

Below is a list of what RMN paid subscribers read this week…

• Underwriting "Three-Peat"
• Disasters Are Pressuring The Mortgage Market

Consider becoming a full subscriber at the special offer rate to get access to daily emails and exclusive features, along with discussion access.

Subscribe now

Risk Reads: What they are saying about coronavirus

AXIS Capital Holdings CEO Albert Benchimol, January 30 earnings call

I know that, there are a lot of questions around the coronavirus. And let me address that right now. We certainly in our A&H book write, a range of business and within that range of business, most of it would not be exposed to the coronavirus. There are two places where that might be.

One is, we do write a book across the world of excess mortality. Those are generally for life companies. Those are generally based on the entire population and require a meaningful increase above the annual average expectations and those attachment points are relatively remote.

So we do not believe that certainly at this point in time with the data that we're seeing the – this would be an effective portfolio. We do have one exposure, which we acquired as part of our corporate citizenship. We own a $10 million World Health Organization Pandemic Response Bond and that's $10 million. We bought it in 2017. It starts to pay out at a 250 fatalities and runs out at 2,500 fatalities. So that's a bond that we've identified. There are currently about 130 fatalities. So, clearly, you're getting close to starting to have a loss on that bond. But at the end of the day, it's $10 million and it's a limited amount in the context of our overall portfolio.

Starbucks President and CEO Kevin Johnson, January 30 earnings call

…due to the dynamic situation unfolding with the coronavirus, we are not revising guidance at this time. But as we get more clarity on the situation, we will transparently communicate with investors.

Our immediate focus is on two key priorities in China: first, caring for the health and well-being of our partners and customers in our stores; second, playing a constructive role in supporting local health officials and government leaders as they work to contain the coronavirus. That said we remain optimistic and committed to the long-term growth potential in China, a market we have been in for more than 20 years.

Tim Cook, President and CEO of Apple, January 30 earnings call

With respect to the supply chain, we do have some suppliers in the Wuhan area. All of these suppliers, they’re our ultimate sources, and we’re obviously working on mitigation plans to make up any expected production loss. We've factored best thinking and the guidance that we've provided you.

With respect to supply sources that are outside the Wuhan area, the impact is less clear at this time. The reopening of those factories after Chinese New Year has been moved from the end of this month to February 10th, depending upon the supplier location, and we've attempted to account for this delayed start up through our larger range of outcomes that Luca mentioned earlier.

With respect to customer demand and sales, we've currently closed one of our retail stores and a number of channel partners have also closed their store fronts. Many of the stores that remain opened have also reduced operating hours. We're taking additional precautions and frequently deep-cleaning our stores as well as conducting temperature checks for employees. While our sales within the Wuhan area itself are small, retail traffic has also been impacted outside of this area, across the country in the last few days.