Do Insurers Need To Be In Predictions Markets?

Yes, insurers should participate in prediction markets. But maybe not in the way you think?

To state the obvious, I’m not suggesting insurers should allocate some of their investment portfolios to whether the Yankees will win tonight.

Rather, I think there is a role of insurers to offer proprietary products where customers can “wager” to protect themselves from insured losses.

What would this look like? It would be somewhat similar to the parametric market. Customers would buy a contract that would pay out (or not) based on a trigger event.

The main difference between what I am proposing and parametric offerings is the price of the trigger would fluctuate based on buyer assessment of their risk.

The other big difference is this product could be structured more like ILS where non-insurance experts provide the capital (i.e. they sell the “no” outcome) and insurers would act like the ILS manager. The manager would be responsible for operating the platform and acquiring customers who would buy the “yes” contract.

This accomplishes a lot of things the parametric market can’t as well as many things traditional insurance struggles with.

But I fear I jumped to the answer too fast without proper explanation first, so let me back up a bit and review some of the basics first for anyone who is struggling to keep up.

How Do Prediction Markets Work?

Prediction markets offer a way to bet on the occurrence of an event. If the event happens, you get paid a $1. If it doesn’t happen, you get $0.

If this reminds you of an insurance contract, it should. I even used the word “occurrence”! Think of an insurance contract with a sublimit that caps payment. The $1 outcome is like a full loss at the sublimit. The $0 outcome is obviously when there is no loss.

What prediction markets don’t have is the middle scenario – the partial loss. However, that also means they don’t have disputes over reimbursement. It’s either a full loss or no loss.

This simplifies the claims process materially. There is no need for defense costs or adjusters or even fraud prevention.

How Does This Differ From Parametric?

There are a lot of similarities to parametric insurance. Both are contracts that pay out based on a pre-defined trigger that is (hopefully) widely observable so there is no dispute about whether a loss occurred and payment is due.

Both carry basis risk (see below) that a buyer might not get paid if they have a personal loss that did not correspond to a trigger event (or they may get a windfall payment if the trigger is hit but they suffered no damage).

However, prices for parametric covers are set by the insurer. As a customer, I have no negotiating power, nor do I have a great way to gauge if the price is fair or if the insurer is overcharging me relative to the risk.

I also take credit risk that the seller of the cover will be around to pay me. Paradoxically, the more people who buy the same trigger as I do, the more credit risk I bear (because the insurer would have to pay all those customers at once if a loss is triggered).

While a parametric insurer can manage this by selling multiple products with aggregation limits, I, as the customer, have no insight into whether they do this responsibly or not.

However, in prediction markets, there is no credit risk. If I buy “yes”, then if the event happens I collect the premium from those who bet “no”. It’s a zero sum game.

Also, the price is set by the market participants, not the insurer. Thus, there is no excess profit margin built into the price (I’m handwaving away the fees to trade and the bid-ask spread for now).

If the market thinks there is a 10% chance of a hurricane making landfall, the contract will be priced closer to 10c than 15c as there are no administrative or other costs to cover. (There would still need to be a small premium to expected loss to compensate for volatility given the asymmetric payout profiles.)

That is not to say there is no risk of overpaying. In popular prediction markets, retail participants tend to overpay and lose money to “professionals” who are better at assessing fair value.

That said, there is some ability to free ride the smart money and a lot of the overpaying is due to participants chasing a gambling high. If you are looking to protect your home from a flood, you are likely looking for the cheapest price, not to watch the Weather Channel rooting for extreme weather.

How Are Prices Set?

As noted, prices are set by market consensus of buyers and sellers. If done responsibly (i.e. not as gambling), there would be enough smart money who would “price enforce” by buying or selling if market value strayed too far away from fair value.

Prices would respond to new information and change in real time rather than be fixed in place for a year as dictated by a rate filing.

Perhaps a good analogy here is NFL point spreads. NFL betting is a very efficient market (gamblers win about 50% of the time). If Vegas sets a price that is too favorable to one team, professional investors will bet heavily on it and force the line to move to a more balanced level.

I would expect the same thing to happen with insurance contracts. Thus, the average retail consumer can have some reasonable level of confidence that they’re paying a fair price for protection.

There would certainly be times psychology would make prices more expensive or cheap than normal. My guess is, if prediction markets existed today for marine insurance, they would be priced above where they should be given what’s going on in the Persian Gulf.

But, that’s OK, because actual insurers are likely spooked by recent events and pricing coverage above where it should be. The key here is prediction markets wouldn’t be systemically biased in their pricing. They would be priced rich during uncertainty and priced cheaply during complacency.

What Would Be Covered?

OK, so now that we’ve addressed the basics, let’s get to the key issues here. What kinds of insurance events could be covered by prediction markets?

I would say any kind of cat event would be fair game since triggers can be designed based on wind speed, rainfall, quake magnitude, etc. Other low frequency property events like aviation, marine, political violence, etc. should also be possible.

It gets much harder to create markets for fire losses or for most liability classes. However, I could even see markets for workers’ comp that are tied to national measurements of workplace accidents or personal auto based on highway accident data.

Those latter examples would not be a consumer product though. They would be more akin to portfolio hedging bought by an insurer as an alternative to reinsurance and would take more time to develop.

That said, I would envision, over time, a robust market for buying on the part of insurers. There would be good reason to buy contracts based on aggregate tornado or hail events.

Similarly, governments may want to buy contracts to protect their exposure. Imagine if Venezuela had a contract that paid tied to a 7+ quake. They could use the proceeds to help rebuild and it would diversify the basis risk (an individual may or may not have damage but we know there would be damage somewhere in Caracas).

However, the most common contracts would likely be consumer focused with protection for cat risk – think hurricane or tornado triggers tied to wind speed within a zip code.

The most natural place to start, in my view, would be rainfall based triggers to proxy flood insurance. Given how little residential flood damage is covered by traditional insurance, offering payouts based on local rainfall would make a lot of sense.

Today, the consumer is likely to receive zero regardless of whether the parametric trigger hits, so any payout from a predictions contract would be all upside.

The side benefit is it would create an incentive for new approaches to improve pricing of flood damage which would bring more capacity and innovation over time.

So, my first product would be residential flood contracts that pay a set dollar amount if rainfall exceeds X inches in Y zip code. I will use the flood example for the remainder of the discussion.

Are Insureds Better Off?

I think so, especially if we are talking about a product like flood where few good options exist today. Coverage will be cheaper and, more importantly, available.

One reason it will be cheaper is underwriters have certainty over their exposed limits. If the buyer purchases a $50K payout if the trigger is met, the underwriter doesn’t have to worry about the actual loss being $120K or a lawyer getting involved and escalating it to $250K.

While there is obviously basis risk where a rain event doesn’t meet the contract trigger but still floods your home and you get nothing, there is also basis risk that exists in traditional coverage since water losses are often excluded (even sometimes when legit but lacking proof), so it goes both directions.

The other basis risk lies in the insured having to guess their severity, meaning if they only buy a $50K contract but actual damage is $80K, they are underinsured by $30K. On the other hand, they don’t have the challenge of believing they have a $80K loss and the insurance adjuster only offering them $50K.

There is no perfect system to ensure 100% matching of incurred loss to paid loss, especially for a product like flood. One can imagine though, as the market matures, that the ILS manager will invest in developing tools to help insureds better estimate their insurance needs as this would grow the market.

Can I Day Trade My Coverage?

The other interesting aspect is whether there will, or should be, a holding period requirement. Can someone not buy coverage most of the year, then see a week out that the radar looks bad, and buy coverage expiring in a week just before the price goes up?

Or should coverage not kick in until say a month after the contract is purchased to prevent adverse selection (which would lead to higher market prices over time)?

Also, what if I buy coverage when the rate on line is 3% and some news comes along that moves it up to 5%, can I sell and profit? Or can only “underwriters” sell?

I think anyone should be able to sell at any time. If I feel my coverage is now overpriced, why shouldn’t I be able to profit and then try to buy it back cheaper later? This is not how insurers are used to thinking obviously but if I’m the underwriter who sold the limit, why do I care who gets the payout if the event occurs?

If we’re trying to provide fair prices to coverage buyers, then I think you have to let anyone sell, not just “experts”. If homeowners are willing to sell their in force for cheaper than the underwriters will provide new cover, then the underwriters can keep their powder dry.

Similarly, I think you probably have to let anyone buy, even if they are a speculator. I know that violates the principle of insurance interest but if speculators lead to more efficient prices that is a benefit to the buyer with insurable interest.

Regulators will want to be careful not to create incentives for people to be day trading their insurance coverage but making it completely illiquid disadvantages the consumer and makes it an unbalanced market.

How Will Underwriters Participate?

One big difference between insurance prediction markets and sports or political ones is that there would be professionals, i.e. underwriters, taking the risk of loss if the event occurs.

As I suggested above, that doesn’t mean individuals couldn’t sell the “no” contract but they are unlikely to have the capital needed to participate meaningfully. Thus, most of the time the buyer would be a retail consumer paying a small rate on line and the seller would be an underwriter offering a large limit.

However, as I noted earlier, that doesn’t mean the “underwriter” needs to be a traditional insurance company. I suspect there would be some “syndicates” that would dedicate insurance company capital to bespoke contracts that may have slightly different contract language than the “average” contract, but I also think there would a lot of “third party” capital.

How might this work? There are a few varieties I could see emerge but I think they would all be a flavor of the ILS market or, if you prefer, names at Lloyd’s. Prediction market managers would organize syndicates to create contracts that would specify triggers and payment terms. They would then handle marketing to steer activity to their contract.

On the “yes” side, they would market to consumers to buy protection. On the “no” side, they would market to institutional investors who want “diversification” just like with cat bonds or insurance sidecars. They would also provide actuarial support for “expected” losses to help investors understand what price would make sense to transact at to meet their return hurdle.

They would make a small fee on each trade and, possibly, be able to act as a market maker buying and selling contracts when liquidity is thin and profiting off the bid/ask.

So why would institutional capital prefer prediction markets to ILS or sidecars? For a lot of the reasons I mentioned earlier, but also they would have better liquidity (daily), their would be no trapped capital (the loss amount is known with no risk of future development), and the expected return should be higher due to the lower cost structure.

As long as investors believe the actuarial pricing is accurate, prediction markets are a far more efficient way to commit capital than other alternative insurance options.

Other Considerations

So hopefully I’ve established that prediction markets for insurance are a viable market with the potential to gain traction where parametric has struggled and the ability to grow even faster than ILS has. There are some other loose thoughts I’d offer before wrapping up.

The information produced by these products about consumer buying preferences, how different contract language affects basis risk, how much cheaper settling these contracts would be relative to traditional insurance, etc. is incredibly valuable to insurers. Incumbent insurers should launch these contracts just for the learnings they can apply to their core business.

There are also potential risk segmentation benefits from watching how buyers participate in these markets. Is there a difference in risk between someone who buys once a year and forgets about it vs. someone who trades often or who delays purchasing until they perceive a potential loss event as imminent?

Are people who use prediction markets only for buying insurance better risks than those who also use it for sports gambling and betting on news headlines or elections? If so, is there a way to charge them different prices? Or do you not care because the loss isn’t correlated to their behavior? It’s only dependent on the external risk trigger. But what if they come to you for a traditional insurance quote?

There are tons of possible insights here for some enterprising actuaries to discover and apply.

Would insurers or reinsurers look to these markets for live cat retro or for hedging any “bets” in their underlying cat exposure? Yes, insurers could turn out to be big “yes” buyers rather than only offering “no” capacity. In terms of dollar volume, it may turn out the corporate market turns out to be much larger than the retail market.

In fact, I could see a future where smart insurers are buying underpriced contracts from dumb insurers. In other words, if reinsurer X prices wind too high but quake too low, reinsurer Y might but some quake contracts from them and net out their exposure by writing that risk with one of their other clients.

You can think of this like how some reinsurers both buy and sell retro to “optimize” their portfolio. But, in this case, they could buy the cheap contract and then, if the market price goes up, sell it at a gain rather than have a portfolio that is theoretically generating excess returns but can only be monetized if loss events occur.

Would any insurers “do a Progressive” (circa 2000) and lean hard into this model of running a prediction market syndicate along with their core insurance business and find it becomes their growth engine?

I’m going to guess none of them are that adventurous but someone probably should. It actually solves the inherent MGA flaw of not having permanent capital so, arguably, managing an insurance prediction platform deserves a higher valuation than being a MGA.

I think I’ll end it there. There are so many fun things one can speculate about on this topic. I hopefully have shown though that it would make sense for a market for tradable binary insurance events to exist.

I have no idea if someone is trying to do this already or if there will be any appetite for the product if someone launches it. If anyone is pursuing something like this, feel free to reach out if you want a sounding board.