The Next Pandemic
The COVID-19 pandemic is pushing insurers and their analytics partners to test the limits of risk modeling, artificial intelligence (AI), and product design to help the insurance industry respond to the crisis and prepare for the next pandemic.
For many employers, losses are mounting, and litigation over coverage of pandemic-related losses under vague or silent contractual language in more general lines, such as business interruption, is spreading.
For insurers and brokers, it is proving difficult to design products that provide appropriate protections for catastrophic events of enormous complexity whose future frequency, severity and duration cannot be predicted. In large part, those events reflect human decisions, both in aggregate and individually, such as whether governments mandate or whether citizens take action to curtail risky business activities. Also, exposure is spreading to different types of insurance policies in unexpected ways, including expanded commercial property losses, and with an extraordinary geographic pervasiveness.
Insurers are trying to respond by better modeling such pandemic risks, including using advanced data analytics and artificial intelligence to help analyze and manage this exposure.
Much has been learned about COVID-19, but the challenges of a worldwide pandemic that persists more than a year after its onset has spurred industry leaders to concede there is simply no way to comprehensively manage exposures this severe and pervasive without national government action to backstop the industry’s response. “The damage resulting from the pandemic by far exceeds the combined balance sheets of the industry,” says Luca Marighetti, a managing director with Swiss Re. “Catastrophes of this size accumulate and as such are not insurable. Yet there may be ways to make them at least partly insurable if governments and insurance join forces to mitigate the risk on the basis of far deeper data wealth.”
Roughly five proposals have been made for various public-private partnerships in the United States, but as of mid-January, it was unclear whether the United States or any other cavalry would be coming to the rescue anytime soon. While the Biden administration has called for some public-private partnerships in fighting COVID-19, most insurance-related efforts appear to be oriented toward enabling more comprehensive and extensive health insurance coverage, particularly through reinvigorating the Affordable Care Act. Biden’s Emergency Relief Package calls for expanding support for the national healthcare law, including by fully funding and expanding authority for the National Disaster Medical System (NDMS) to reimburse healthcare providers for COVID-19 related treatment costs that aren’t directly covered by health insurance. This includes all co-payments and deductibles for the insured as well as uncompensated care burdens incurred by uninsured and underinsured populations. But these proposed solutions do not include a more comprehensive public-private risk solution of the sort envisioned by some in the industry.
“We have reached out to the U.S. and United Kingdom governments to see if they can work with the industry to bear these types of risk that cannot be spread out due to their global nature, but progress has been limited,” says John Merkovsky, head of global risk and analytics at Willis Towers Watson. “It may be that we are planning for how to deal with the next pandemic while they are focused on dealing with the current one.”
The Race for Better Measurement
There have been some data analytics successes in addressing the crisis. BlueDot, a Canadian AI firm, developed an algorithm that on Dec. 31, 2019, processed news reports and ticketing data to source the COVID-19 outbreak in Hubei province and notify its clients. BlueDot also produced one of the first scientific papers on COVID-19, accurately predicting eight of the first 10 cities to import the novel coronavirus.
The quest for better measurement is spurring action by many in the insurance industry. In September 2020, four major insurers (Swiss Re, Aviva, Discovery Ltd., and Legal & General) joined The Trinity Challenge, an international coalition of universities, foundations, and leading technology and health companies that are dedicated to better protecting the world against future health emergencies by harnessing the power of data and analytics. Insurers and analytics firms make data and methodologies available to The Trinity Challenge applicant research teams that aim to better predict, respond to and prevent health emergencies in the future.
For example, Swiss Re has made its data platform, the Risk Resilience Center, available to The Trinity Challenge research teams. Since the beginning of the pandemic, Swiss Re, in collaboration with big data analytics company Palantir Technologies, has been integrating relevant data sources related to COVID-19’s health, economic and social dimensions (such as aggregate mobility) into the Risk Resilience Center, says Vasco Nunes, director of digital and information initiatives at Swiss Re. The platform provides access to a large data set on COVID-19, integrating publicly available global, granular and continuously enriched data from more than 100 sources.
Using the platform, The Trinity Challenge research teams can perform advanced analyses of COVID-19 metrics and their connection with a country’s medical systems, business or travel activity—in short, notes Marighetti, with the country’s economy at large. For example, the platform allows for better planning, such as the assessment of needed medical resources and their allocation.
“It typically takes weeks to have data integrated and curated—our platform updates, checks and transforms data sets automatically and multiple times per hour, with a continuous flow of incoming data, allowing for cross-country comparisons and drilldowns on states and regions,” says Ian Haycock, chief data officer at Swiss Re. “Equipped with these cutting-edge capabilities, The Trinity Challenge participants can analyze the spread of a pandemic, the economic impact of lockdowns, or the success of behavioral changes (like wearing a mask) in limiting further infection.”
Some Swiss Re clients that use the data for COVID-19-related decisions are already using the Risk Resilience Center data, Nunes says. “For example, large companies with global footprints can use the data to make informed decisions on which offices to open and shut given the pandemic,” he says. “The team has a data ‘river’ with different countries and mortalities. We bring in the data from our clients’ office footprints and overlay that with case data to understand which regions are going over COVID-19 thresholds. We also have the opportunity to look at how much the current year compares to the past. For instance, we can analyze whether there are increases in certain types of claims such as mental health related.”
Insurers are using such findings to confirm whether they are sufficiently capitalized in face of the COVID-19 crisis and to adjust capitalization as needed. Swiss Re is also reaching out to governments and multilateral development banks with the aim of providing them with access to the platform on a pro bono basis.
“There are also longer-term goals for our business,” Nunes says. “If we have a better understanding of the claims, then that can help to get a sharper and deeper understanding of the risks and improve the feedback loop to underwriting.”
Still, Swiss Re has been conservative in opining on what such research will allow, though the team hopes socialization of such research will produce a positive feedback loop of knowledge discovery.
“When we started with The Trinity Challenge, it was as an exercise to help address a humanitarian problem. We said, ‘Let’s contribute,’” Marighetti says. “It may also turn out that it has a commercial application. That is the way it should be—first focus on solving the big problem. If you contribute to it, there will be a business application at some time. What we are doing is in the spirit of the ‘Republic of Science,’ where you contribute something and get something back, but the intellectual property remains with whoever develops it, whether it be a university, industry partner or a participant.”
Marighetti says the history of the insurance industry shows that complex problems similar to pandemics can eventually be tamed. “I don’t believe in unsolvable problems,” he says. “Thirty years ago, natural catastrophes couldn’t be covered, and then we figured out a way to do it. Here it is a similar case. There will be application of brainpower to solve it that results in partial solutions. But partial solutions are better than no solutions.”
Catastrophe risk solutions company RMS has also been studying COVID-19, further developing a pandemic business interruption model initiated in 2008 to track the earlier SARS pandemic, an effort that was put on hold after there was little demand for the associated cover. “What has changed is our knowledge of COVID-19, which will be a key part of risk modeling going forward,” says Robert Muir-Wood, chief research officer at California-based RMS. “After SARS, for business interruption, we included the potential for business-directed lockdowns and for that to be the principal part of business consequences. When it comes to COVID-19, we captured some aspects of its impacts well, though there were also some impacts that were underestimated, such as the degree to which governments would shut down their economies to reduce or eliminate the risk.”
Muir-Wood says that RMS and others are performing sector-by-sector analysis of how COVID-19 has affected various industries. “There are huge amounts of data how commercial activity reduced sector by sector,” Muir-Wood says. “The key observation so far is that learning changed through the pandemic, not just that this sector was impacted to this degree. We need to understand how learning shifted over time and improved. In March and April, there was a much more severe business reduction than now, even in the countries with new lockdowns. But then businesses discovered ways to sustain operations during the pandemic. So learning will be very timing-dependent.”
That improved resilience also reflects the current state of home and workplace technology, he notes. “Before this pandemic, it remained to be seen whether people could function out of their homes,” Muir-Wood says. “There have been a variety of technical capabilities that have enabled that. Three years ago, it would have been hard to work at home with all their data and software in the cloud. And if we were to have the same COVID-19 pandemic five years from now, technology will have moved on by then.”
Modeling requires examining and attempting to predict the actions of businesses, governments, the general public and customers in response to a pandemic, Muir-Wood says. In some business sectors, he notes, government-mandated closures had a decisive effect on operations. In other sectors, business-led actions or consumer actions predominated. “The impacts for the cruise liners were interesting because neither governments nor the businesses themselves shut those down,” Muir-Wood says. “Customers shut those down themselves by declining to continue going on cruises.”
Many say the better route for designing pandemic policies is to make them parametrically based, with payouts tied to trigger events, rather than indemnity-based insurance providing reimbursement for actual losses incurred. Muir-Wood says there are two key elements in a good parametric model. “First,” he says, “there must be something that crosses a threshold that indicates it really is a pandemic, such as a government stamp defining it as such or the imposition of lockdowns at some site. The second criterion is impact to the business itself, such as a decline in the occupancy rate for the hospitality industry. You need something that can be measured independently and competently.”
The Fickle Human Factor
Still, others question whether a focus on the challenges of pandemics will be enough to drive sufficient product development and sufficient national and individual human behavioral changes to make the risk profile of future pandemics manageable.
“The human challenge remains the most difficult,” Marighetti says. “I hope we will be able to respond better to a future pandemic. Humans are not renowned for having long memories. After each crisis, we say, ‘It will never happen again.’ If there is a similar pandemic 15 years from now, maybe we will be able to apply lessons from COVID-19. But I am skeptical, if it occurs 50 years from now, the next generation will remember.
“Looking at the spread of COVID-19 today, we had a similar pandemic 100 years ago, the Spanish flu. It is clear what needs to be done to contain such pandemics, including convincing the public to act in certain ways to contain the pandemic. That is really just good old-fashioned infectious disease control. Some places, including many countries in East Asia, have done this during COVID-19 successfully. But Western Europe and the United States have not.”
The degree to which such modeling is helpful in part depends on the efficacy of government response. “Government interventions are slightly less modelable, as the ‘right’ health response is not the same as the ‘right’ economic response and a government may not use the correct response even when overwhelmingly supported by science,” says Colin Dutkiewicz, head of life at Aon’s Reinsurance Solutions business. “So the modeling has to provide scenarios which cover the major moving variables with guesses at government responses. Of course, in some countries there is greater alignment of public behavior to government intervention and of the link between science and government intervention. In these cases, modeling is much better.”
And, of course, there is the question of whether all involved are willing to pay for coverage for an insurance event whose incidence may very well escape generations of business client executives’ successive watches.
“The reality is that pandemic insurance has existed for a long time but is rarely purchased, given its cost and the low likelihood of an event,” noted John Doyle, Marsh’s president and CEO, in congressional testimony. Doyle testified that Marsh clients globally have made more than 11,000 business interruption claims related to the pandemic.
In addition to reinsurers like Swiss Re and other carriers, some of the largest insurance brokerages also have been involved in modeling COVID-19 data. “Aon has been using pandemic modeling for many years on the life and health side,” Dutkiewicz says. “This has involved modeling the potential impact of a pandemic on life and health insurers’ portfolios and extended to the group insurance portfolios of multinational firms with global employee benefits portfolios. This modeling is performed on a stochastic basis, typically running 10,000 simulations of what a pandemic could look like, given the four pandemics over the last 100 years. In COVID-19 times, this has been adapted to be a scenario-based projection of what could happen in the next 12 months—scenarios with effective vaccines, ineffective vaccines, competent health authority interventions, etc.”
Dutkiewicz says the modeling has relevance to at least four lines of insurance:
- Pandemic reinsurance product design: providing reinsurance to protect insurers against pandemic claims
- Bespoke travel insurance: to deal with the changing travel environment
- Airline COVID-19 insurance: to pay out if someone is infected while flying
- Lockdown impact insurance: to pay out if businesses are closed due to government/local authority mandated business closure.
“The modeling for these covers needs to take the modeling of the pandemic itself one step further,” Dutkiewicz says.
Better modeling of specific coverage depends on better understanding the pandemic-related responses of underlying businesses in different sectors, Dutkiewicz says: “These need to be more transparent and visible, and to the extent that they are now better understood, the models can take into account more risk factors and drivers of severity of loss. But human intervention is required to ensure that these causal links are valid statistical causality, not just correlated statistics. The model/machine can’t tell that.”
Deductions have emerged as to COVID-19’s effect upon different lines of insurance and needed product adjustments, Dutkiewicz says. For example, health insurance needs to adapt rapidly not only to the immediate effects of COVID-19 claims, he says, but to the possible additional claims in the future from the delay in healthcare leading to worse health outcomes.
Modeling can also be useful to understand the impact on incidence of disability due to work-from-home conditions, mental health and long-term COVID developments, Dutkiewicz says.
On the other hand, some aspects of insurance coverage appear to be less affected by COVID-19. Life insurance is largely unaffected by COVID-19 since insured populations have shown a less than 10% additional mortality even with modeling of the second, third and fourth waves of COVID-19, Dutkiewicz says.
Willis Towers Watson’s Merkovsky says his group is working with clients to help them better model both the effects of the pandemic on businesses and the other extreme tail risks that are present for their businesses, such as climate change, political risk or cyber attacks. “The effects of COVID-19 have increased interest in understanding the risk of low-probability events that are of uncertain magnitude, severity and timing but that could destroy significant value in the organization,” Merkovsky says. “The pandemic has driven an understanding of how devastating these risks can be. You may eventually see products to deal with those risks, including products tailored to the needs of a particular client.”
Since May 2018, Marsh has offered PathogenRX, a non-damage business interruption insurance policy tied to pandemic risks developed with Munich Re and epidemiology risk consulting firm Metabiota.
The policy, which has a parametric trigger, was initially designed to respond based on infections or mortalities within a defined coverage area but has since been updated to respond to the underlying cause of economic damages in a pandemic—the lockdowns that civil authorities and public health officials implement to fight community spread of an outbreak, says a Marsh spokesperson. While demand for the product surged in early 2020, COVID-19 was already excluded as an ongoing event.
The danger posed by insufficient care in addressing pandemic risk, whether through pandemic-specific policies or exclusions or other language in more general policies, has become apparent through widespread litigation over whether more general business interruption insurance without standard virus exemptions covers losses related to COVID-19. A University of Pennsylvania Carey Law School Covid Coverage Litigation Tracker found that, through early October, more than 1,300 cases had been filed. Most of that litigation sought coverages that relate to business income, extra expenses and civil authority.
Doug Fullam, director of life and health products at Boston-based risk modeling firm AIR Worldwide, says his company consults carriers on product design and that more pandemic coverage options may be on the way. “Before COVID-19,” Fullam says, “there were few design options and a few specialized players because this was not a top-of-mind issue. Now there are more and more different products. A lot of parametric products may be more useful in a pandemic because they pay out sooner. It may be more important to have some payment during a slowdown than to be made whole after a lengthy process of validating claims under an indemnity policy, at which point the insured may be in bankruptcy.”
Artificial Intelligence and Machine Learning
When it comes to relying on artificial intelligence to gain insights for pandemic modeling and product development, opinions are mixed.
“AI and machine learning have not been helpful in our modeling,” says Aon’s Dutkiewicz. “Unfortunately, most data sources are not reliable, and AI sort of assumes that the data is valid to do its analysis. So human intervention is required to critique data. The pulling of the data sources can be done by machine, as Aon does from its Ireland data center. But then this data needs to be modified to be useful and the models using it adapted to deal with data errors and changing data meaning.”
Some say that a long-term goal of such modeling is recognition of what type of pandemic presents itself and that AI might help there. “The assumption is that data sets from the past can help us understand the future,” Merkovsky says. “But the patterns of past pandemics have varied wildly. For example, based upon COVID-19, we would assume we should be modeling the effects of the pandemic upon an older population, given the disproportionate mortality and morbidity rates for that group, and taking steps to protect them. But in the case of the H1N1 virus, 80% of the deaths were among those aged 65 and younger, so the right containment step might be to keep young people out of schools. Given there is a limited capacity to respond, those are critical distinctions, so the value of the data has to be understood in the context of the nature of the idiosyncrasies of the pandemic.
“We need to better understand the data to be able to react more quickly. Machine learning may offer the hope of using larger streams of data to use the patterns of different viruses to determine whether, for example, the next virus is more similar to H1N1 or to COVID-19. The sooner you understand the pattern, the sooner you can place the right protective steps in place.”
AI and machine learning have their place in modeling some risks, Muir-Wood agrees. “You can’t throw AI at all aspects of the pandemic, but where it is viable is to look at the impact on different sectors and how it changed week by week,” Muir-Wood says. “We do have that data week by week for the first wave and second wave. We will find elements of this story, and what we need for modeling will come out of insight into credit card data and the probability of impacts upon businesses week by week, month by month. That will help with coming up with pandemic business insurance. You would like to have losses covered through a period of a pandemic, and insurers can’t diversify risks they do not understand.”
COVID-19 Coverage in Small Bites
For one brokerage that is actually offering COVID-19 pandemic insurance to clients, the second half of 2020 presented a rapid learning curve. In July, Elite Risk Insurance, a specialty brokerage based in Newport Beach, California, began offering Pandemic Outbreak with COVID-19 Relapse coverage with limits from $2 million to $25 million.
Elite Risk has painstakingly mapped exposures and insured perils to its deep knowledge of certain industries, including the risks involved, which risks should be insured, which exposures should be assumed by policyholders, and what payouts are necessary to allow their policyholders to survive rather than be made whole.
“We are using the data and creating our own parametric indemnity policy for moments in time,” says Elite Risk owner Jeff Kleid. “We took the film industry and pandemic coverage, looked at results from March, April and May, and looked at it manually to figure out what we could do to make sense to allow productions to be completed.”
“You can intelligently handle pandemic charges for this type of risk and charge for it and take small bites,” Kleid says, adding that coverage is provided by a group of privately controlled captive insurance companies. “The client has to accept the pricing and get used to not getting what they used to receive from a policy. For example, with business interruption, previously, if they were shut down, they got paid all interruption costs, often receiving a payout of $20 million, and the client believed they were bulletproof. Now, we only give them $3 million for a COVID-19-related payout. I could go to more than one carrier and get more than $3 million for them, but the others are often hard to get. The smaller insurance payout actually works by providing an incentive for the production company to keep moving forward even though they were insured for COVID-19. I have given them a band-aid with coverage to give them, we hope, what they need to continue filming.”
Elite Risk has sold about two dozen policies since beginning in July, including many in the film and television industry, Kleid says. The premiums are high, at between 8% and 15% of coverage, meaning it is likely to remain a niche program aimed at particular industries and the discrete exposures they face, such as ensuring COVID-19 does not prevent completion of a movie or television show.
Kleid says there already have been a few claims, which have been handed off to his entertainment claims partner. “I’m not sure how they will play out,” he says. “There is a lot of learning, and a lot is by trial and error here. For example, we realized very quickly that we had to clarify further what is and is not covered. Some people sent COVID-19 tests to labs that took too long to produce results, delaying production, and asked, ‘Are we covered because we had to shut down while we waited for results?’ The answer was, ‘No, you are not covered.’ Because that is not the risk we are trying to cover. An example of what we do cover is a cast member who is positive, who can’t fit into a production schedule when they have COVID-19, but who can make it back to a production at a later date, or costs associated with the production reshooting the scenes with someone else.”
Kleid thinks it might be possible to dramatically expand the availability of this type of coverage by working with associations, say a restaurant association, with tens of thousands of members to provide very narrow categories of standardized coverage, such as $500 in premium to cover a $10,000 shutdown and cleaning fee related to incidences of COVID-19.
Some might characterize this type of approach as medieval—in the good sense of the word. Muir-Wood notes that in medieval times European guild members got together and raised money to provide various financial and other services to members, including support during periods of sickness. “These days, one thing some industries could do is set up a policy for their own members so they could be taken care of in the event of an interruption like a pandemic,” Muir-Wood says. “It would have to be a reasonably wealthy sector. I think, for example, you could imagine some profession like dentists, where a national dental society did quite well covering their members and possibly getting reinsurance.”