Underwriters Compete on Moral Judgment

(note: this is essay 1 of 4 on the social science of insurance)

A great underwriter possesses many skills: communication skills, analytical skills, administrative skills, sales skills, leadership skills. However, the skill that most differentiates great underwriters from the rest is moral judgment. 

Moral codes are our internal rules for how we will conduct ourselves, in good and difficult times. They outlaw certain behaviors and act as a way to credibly signal to others that we can be trusted. Virtue is a value-laden term, it’s an evaluation of one’s moral code against a benchmark of some kind. The virtue that matters for insurance is composed of both the quality of a buyer’s intentions in entering into an insurance contract and also the buyer’s wisdom and experience. It is necessary that the buyer is honest and forthright but it is also necessary that they will not change their character later once they learn some new facts about the world.

The kind of virtue underwriters are evaluating is very specific. There are many forms of virtue and not all relate to the qualities necessary to be a virtuous insurance customer. For example it is virtuous to be a supportive parent, give to charity and be faithful to your spouse. None of these directly matter for insurance. The form of virtue underwriters assess concerns your conduct in dealing with distant financial institutions and whether you will treat them well in the future. This is not a kind of virtue that has existed for all of human history nor one that exists everywhere in the world today. In some ways it’s a very peculiar thing to have a moral code that constrains our actions in dealing with corporations. And we don’t all have it to the same degree. 

This is distinct from the work of risk assessment. Risk assessment as insurers undertake it is to assign a risk to its most appropriate segment. Classifying and aggregating risks to determine their costs is an important function of insurance companies but it is not underwriting.

The underwriting of a person or business starts with an assessment of the quality of someone’s intentions, which in a sense is a forecast of that person’s future actions. The most broad and general guide for those actions is morality. 

Testing morality works to weed out two different kinds of failures of insureds. First is a group that intends to do something bad1 but is trying to conceal it. As an agent, I once sold a hospital indemnity2 policy to a woman that knew she had cancer and would be going to the hospital in a few days. She somehow got through underwriting (possibly by lying) and filed a claim almost immediately after her policy was issued. I also once saw a trucking customer file a total loss ($1m) at 12:01am on the very day their insurance policy came into force. What did they do wrong? I never found out, but it was very suspicious!

The second is a group that might not intend bad actions now but is especially vulnerable to bad behavior in the future. As a reinsurance broker I had an insurer client that was entering a new line of business. The management team scored high on the integrity scale, they were good people. But when the reinsurance actuaries came into their office for an audit of the pricing model it was clear this group had no idea what they were doing and were systematically mispricing risks in a myriad of ways. They simply were ignorant of how to do their jobs. This is a failure of knowledge and wisdom, not of intentions.  

In our daily lives we make judgments of virtue all the time in selecting friends and colleagues and partners of various kinds. In normal life these judgments are heavily influenced by our existing relationships. Moral judgment is a social act. We would have a difficult time accepting or perhaps even understanding the moral evaluations of someone from a very foreign social context or culture. Since our existing relationships share this context they help our evaluation of the virtuous character of another person and offer advice.

Since the evaluation of virtue is so very important, this skill has been the subject of intense competition throughout the history of insurance. And insurers have discovered how to identify and train a talent in certain people so that they don’t need a whole village to pass effective moral judgments: these are the great underwriters. 

In normal life moral judgment is a maximizing exercise. Inasmuch as we make explicit decisions about virtue we approve of those who have a lot of it. Over time we probably select lower levels of virtue in our compatriots through implicit processes hidden to our conscious minds, but we are very unlikely to admit that. Moral judgments by great underwriters are all explicit and must contend with the fact that an insurer can make money by selling to people in a whole range of levels of virtue. 

Virtue vs Profit

You’ll notice that the slope of the line drops off quickly as virtue declines. The least virtuous are nearly certain to be unprofitable.

But how virtuous do you want your customers to be? Well, to anyone who has taken an economics class, you know where to draw the line.

Easy in theory

But what does this mean operationally? Answering this requires knowledge of both an individual’s virtue and the virtue dynamics of the marketplace. Here’s a question: how virtuous are the customers available to you today through your distribution channels? 

Low virtue customers are much easier to acquire than high virtue customers. Partly this is driven by their unprofitability being uncovered by carriers through non-virtuous behavior (and so they get non-renewed from the portfolio). These buyers also work hard at deceiving insurers about their virtue, which doesn’t always work, so they approach many carriers until they slip through.

Flow of customers

The graph above is very surprising. If most available customers are non-virtuous then how is the insurance industry profitable? The confusion comes from mixing up an analysis of flow of customers (above, measured probably in thousands) with the stock of customers (below, measured probably in millions). 

Stock of Customers

Most insurance customers do not flow, they are parked inside an insurer’s portfolio, paying their premium every year, filing the odd claim when disaster strikes their lives but mostly making their carrier money. They don’t want to think about insurance and everyone is pretty happy about it. Low churn is a signal (but not a component) of virtuous character. I should be clear, the virtuous can shop too! The non-virtuous are simply overrepresented among shoppers. 

Since it is quite hard to get access to a large volume of virtuous customers, underwriters compete on who is better at finding the line and so running a successful business. But as virtue declines deceit increases. There are many insurers that want nothing to do with the minimum virtue zone and focus only on the easiest to underwrite. 

Finding the minimum is hard

But only concentrating on the most obviously virtuous people is no free lunch since the competition for these customers is intense. One can underprice the morally straightforward risk as much as the complex and the end of a bout of underpricing is always the same: years-delayed, spectacular financial detonation once the pricing error is inescapably obvious3

The problem of course comes down to an inability to abstractly measure virtue except after the fact. Carriers that can reliably identify virtuous customers can afford to simultaneously offer better prices, better commission, better service and generate better returns for investors. There are carriers that do just this. 

Many innovators seek to eliminate moral judgment with data and clever technology. And there are indeed ways of improving the efficiency of moral judgment. For example, some distribution channels are so incredibly trustworthy4 that underwriting is superfluous and in other instances claims data is so complete and transparent that a person’s behavior can be modeled with incredible accuracy. In these instances underwriting merges with risk assessment and classification. When moral judgment ceases to be an important dimension of competition underwriting ceases to exist. 

Villains, however, are always looking to profit from infiltrating unprotected networks and from time to time these lightly underwritten portfolios collapse. When that happens, underwriting by people returns in force to expose the bad hiding among the good. Moral judgment is as complex as humans are clever at deceit. 

Having morals as guides for action under uncertainty is a tremendous human achievement and assessing the quality of those morals is an irreducibly human act. For a long time I never really understood what underwriting was. Now I actually think it’s the most deeply human commercial institution we have. 


1Doing something bad means finding a way to reliably profit from your insurance policy. Insurance fraud is an example but anything that deceives an insurer into underpricing its policies counts.

2  Pays out if you are admitted to the hospital.

3   To pick a price, insurers make an assumption about the claims a portfolio will have. If they get this wrong it usually requires an increase in claims estimates for many years at once, concentrating that cost increase in one financial year. This is the most common way insurers go bust. 

4   To pick a price, insurers make an assumption about the claims a portfolio will have. If they get this wrong it usually requires an increase in claims estimates for many years at once, concentrating that cost increase in one financial year. This is the most common way insurers go bust. 

When Data Cannot Do Insurance

Here is David Brooks on what Data can’t do:

Data struggles with the social. Your brain is pretty bad at math (quick, what’s the square root of 437), but it’s excellent at social cognition. People are really good at mirroring each other’s emotional states, at detecting uncooperative behavior and at assigning value to things through emotion.

Computer-driven data analysis, on the other hand, excels at measuring the quantity of social interactions but not the quality. Network scientists can map your interactions with the six co-workers you see during 76 percent of your days, but they can’t capture your devotion to the childhood friends you see twice a year, let alone Dante’s love for Beatrice, whom he met twice.

In insurance we care about scale (the law of large numbers) and not getting f*@#ed over (avoiding moral hazard). Data has definitely helped where moral hazard is somewhat easily guarded against: such as in homeowners or auto liability insurance.

And these are the largest insurance markets on earth. Consumers have no doubt benefited, either through lower insurance premiums or (more likely) through a far more generous tort system and subsidy for people to build their homes in flood planes and on fault lines and  hurricane tracks.

For more complicated lines of insurance, we still need really expensive underwriters to decide who is worthy of trust. Data has a long way to go there.

Google’s Driverless Car

Everyone’s fired up again. This time, however, the debate is moving in a direction that I can relate to. Here’ Megan McArdle (who has obviously been catching up on my blog archive):

Now I’m gloomy again.

Why? Not because of the technology. And not because of the regulation.  But because of the liability.  Self-driving cars represent a massive one–one that I’m not sure companies will take on.

Now, luckily, as many others are observing, a crazy tort system is somewhat unique here in the US and driverless cars need not multiply in the land of their birth.

My guess would be that promising-but-scary technology is more likely to be pioneered in a poorer country, since as people get wealthier they tend to become more risk averse and prioritize safety. But if something proves really useful and basically safe in some subset of countries, the pressure to change the rules elsewhere should become intense.

Good luck to Singapore or wherever but tweak US tort law? It is hard to describe how immense a task that is.

Putting these things onto roads full of human drivers means you probably don’t gain any macro benefits of more orderly roads. Handsfree driving is nice and all but is a few more hours of daily facebook for commuters going to spur Congress to the most fundamental overhaul of the legal system in generations?

For me it’s still filed with tacocopter and segways under ‘cool, technically viable idea: never going mainstream’.

My Sandy Timeline

Mid-April: I move to Hoboken, NJ with my 6-month pregnant wife and Bree and Max, my two 10-pound dogs.

Some time in May: a relatively minor storm floods our parking garage and the nearby street. WTF? Lesson: Hoboken is really bad for flooding and we live in one of the worst parts.

End of July: I sign up for an actuarial exam for the end of October AND my son is born.

October 20: “A strong ridge of high pressure parked itself over Greenland beginning on October 20, creating a “blocking ridge” that prevented the normal west-to-east flow of winds over Eastern North America. Think of the blocking ridge like a big truck parked over Greenland. Storms approaching from the west or from the south were blocked from heading to the northeast.”

Some additional background from the same link:

We expect hurricanes to move from east to west in the tropics, where the prevailing trade winds blow that direction. But the prevailing wind direction reverses at mid-latitudes, flowing predominately west-to-east, due to the spin of the Earth. Hurricanes that penetrate to about Florida’s latitude usually get caught up in these westerly winds, and are whisked northeastwards, out to sea.

Bottom line: normal no longer applies.

October 22th: Tropical Depression 18 forms.

October 24th: Now Hurricane Sandy, the storm hits Cuba hard. The possibility of a US landfall dawns on the NHC for the first time.

October 27th (Saturday): I get a mass email from my building manager saying that the area flooded even during the over-hyped Irene last year and big floods trigger the fire alarm. I reply: as in the building-wide fire alarm? Yep, he fires back, and we can’t turn it off and it’s LOUD.

Well that sews this one up, but where do we go? Here’s our criteria:

  1. Town that has a hotel that wasn’t full
  2. Oh, yeah, and isn’t on a river
  3. That hotel needs to take dogs
  4. Is near a place where I can take my exam (still studying through all this!).

I pull up the intertubes and hit the phones. The answer? Three-hour away Albany.

October 28th (Sunday): no point studying, got to pack up an infant and dogs and supplies and whatnot and hit the road. That takes most of a day. The hotel is great and guest cancellations are rolling in. Papa John’s pizza and a practice exam for me.

October 29th (Monday): Holy Cow this is for real. Glued to CNN. Albany? A brisk wind is the worst we saw. Incredible luck.

October 30th (Tuesday): Hoboken is underwater. Everything is underwater. Uh, oh, when are we going to be able to get back?

October 31st (Wednesday): I write my exam in the morning. I’m the only one sitting it since the CAS let affected folks put it off. Not for me, let’s do this. That’s four hours.

Back at the hotel it’s becoming clear, as I scarf down yet more takeout, that we’re not going home. Looks like it’s back to Canada to my in-laws’. But first someone’s got to go back to Hoboken to get our travel documents. Saddle up!

Driving down the roads I see that about one in ten gas stations in Northern New Jersey is open and each has a gigantic queue of cars at it.

You know what that means: rationing by time instead of price. Far more importantly, however, it means that overall supply is lower. Here’s Yglesias who has been covering this very well:

Chris Christie, also put out a weekend press release warning that “price gouging during a state of emergency is illegal” and that complaints would be investigated by the attorney general. Specifically, Garden State merchants are barred from raising prices more than 10 percent over their normal level during emergency conditions (New York’s anti-gouging law sets a less precise definition, barring “unconscionably extreme” increases).

The bipartisan indignation is heartening, but there’s one problem. These laws are hideously misguided. Stopping price hikes during disasters may sound like a way to help people, but all it does is exacerbate shortages and complicate preparedness

And more:

But when it comes to things like gasoline and bottled water, neither the short-term nor the long-term supplies are genuinely fixed. Transportation routes into the area have been severely disrupted and many gas stations’ supplies are hard to access due to power outages, but it’s not impossible to transport this fuel from where it is into people’s cars and generators. It’s just much more annoying and difficult than usual. But the possibility of windfall profits is exactly the lure we need to get people to start making extraordinary efforts to get more fuel to the people who need it. There are things people will do to sell gasoline for $10 a gallon that they won’t do to sell gasoline for $3.40 a gallon (note that in Norway this is what gas costs all the time) and that’s what we need.

Power lines were down everywhere and electrical crews were working away. Roads were closed, though, and it took forever to get back to Hoboken. Eventually I had to park about a mile away and, now under the cover of darkness, run into town with my rubber boots, flashlight and backpack.

Very post-apocalyptic.

The phone networks were overloaded so there were definitely people around. You could see refugees sitting in cars charging their devices before going back up to, I dunno, play angry birds by candlelight, I suppose. The gold standard of disaster certifications is of course an on-location broadcast by Anderson Cooper, which happened while I was there! I didn’t see AC360 himself, though.

Anyway, got my stuff and booted it back to the car. Back to Albany by 11pm. Phew. what a day.

November 1st: quick check of the newswires. Still flooded. Ok, back in the car for 8 hours to Canada!

Post Scripts:

The insurance loss is getting picked at 10-20bn, which should put this after Katrina and Andrew as the third most costly hurricane in US history. That’s probably about right. There’s also a debate about whether hurricane deductibles (higher than normal storm deductibles) are going to apply to this “Post-Tropical Cyclone”. See here for example.

There is also a debate about the role of FEMA in these kinds of disasters. Here is an interesting point (via MR):

We’ve nationalized so many of the events over the last few decades that the federal government is involved in virtually every disaster that happens. And that’s not the way it’s supposed to be. It stresses FEMA unnecessarily. And it allows states to shift costs from themselves to other states, while defunding their own emergency management because Uncle Sam is going to pay. That’s not good for anyone.

When FEMA’s operational tempo is 100-plus disasters a year, it’s always having to do stuff. There’s not enough time to truly prepare for a catastrophic event. Time is a finite quantity. And when you’re spending time and money on 100-plus declarations, or over 200 last year, that taxes the system. It takes away time you could be spending getting ready for the big stuff.

…I think another issue is some people see the failed response to Hurricane Andrew as the reason George H.W. Bush lost Florida to Clinton. So now, you have presidents who are very concerned about the potential impact, from an election standpoint, of disasters. That created an incentive to nationalize things.

Finally, here’s a statistical wrap-up (great image at the link):

Death toll: 160 (88 in the U.S., 54 in Haiti, 11 in Cuba)

Damage estimates: $10 – $55 billion

Power outages: 8.5 million U.S. customers, 2nd most for a natural disaster behind the 1993 blizzard (10 million)

Maximum U.S. sustained winds: 69 mph at Westerly, RI

Peak U.S. wind gusts: 90 mph at Islip, NY and Tompkinsville, NJ

Maximum U.S. storm surge: 9.45′, Bergen Point, NJ 9:24 pm EDT October 29, 2012

Maximum U.S. Storm Tide: 14.60′, Bergen Point, NJ, 9:24 pm EDT October 29, 2012

Maximum wave height: 33.1′ at the buoy east of Cape Hatteras, NC (2nd highest: 32.5′ at the Entrance to New York Harbor)

Maximum U.S. rainfall: 12.55″, Easton, MD

Maximum snowfall: 36″, Richwood, WV

Minimum pressure: 945.5 mb, Atlantic City, NJ at 7:24 pm EST, October 29, 2012. This is the lowest pressure measured in the U.S., at any location north of Cape Hatteras, NC (previous record: 946 mb in the 1938 hurricane on Long Island, NY)

Destructive potential of storm surge: 5.8 on a scale of 0 to 6, highest of any hurricane observed since 1969. Previous record: 5.6 on a scale of 0 to 6, set during Hurricane Isabel of 2003.

Diameter of tropical storm-force winds at landfall: 945 miles

Diameter of ocean with 12′ seas at landfall: 1500 miles

One More Step And The Barbarians Are In

I’m immersed in the history of insurance regulation these days. What was that? Oh, yeah, it’s hilarious and stuff. Good one.

Anyway, know what’s striking about the history of insurance regulation? Let’s quote Bastiat:

The State is the great fiction through which everyone endeavours to live at the expense of everyone else.

Insurance is brutal: everyone thinks that they can screw someone else over indefinitely. Consumers do it, insurance companies do it, the government does it.

Take the National Flood Insurance Program. Here’s a report from the Government Accountability Office:

Why GAO Did This Study

The National Flood Insurance Program (NFIP) has been on GAO’s high-risk list since 2006, when the program had to borrow from the U.S. Treasury to cover losses from the 2005 hurricanes. The outstanding debt is $17.8 billion as of June 2011.

17.8 billion with a ‘b’. All because nobody will bother to do either of these things:

1. Move out of flood zones.

2. Pay enough for insurance to cover the cost of repairing flood damage.

Realistically this is a risk that is too costly to insure for most. If your house costs $100,000 and you annual insurance policy is $10,000, what’s the point?

But how did the government get involved? Get this: FANNIE And FREDDIE! A timeline from my notes:

1973: Flood disaster protection act is passed for owners of properties who had mortgages from federally regulated lenders.
1994: National Flood Insurance Reform Act strengthened mandatory purchase requirement for owners of properties in flood zones and with mortgages from federally regulated lenders.
2004: Bunning-Bereuter-Blumenauer Flood Insurance Reform Act authorized grant programs to mitigate properties that experienced repetitive flood losses. Owners that did not mitigate *could* face higher premiums.

They vacillate between statist coercion and open market reforms. But we’re at 17.8bn now and private capital is the only way out of the mess.

It’s a dizzying spiral of interlocking regulations and market distortions that press down on our economy. If only, like before, it were strong enough that we didn’t care.

Will Google Write Catastrophe Insurance?

Catastrophe insurance is the sexy part of my industry: lots of data and “analytics” and in tune with the information age. It’s also alternated between the most and least profitable line of business in the business.

Here’s what you need to write the stuff:

  1. A really good map of where buildings are.
  2. Some knowledge of what those buildings are made of and, just as importantly, what they’re worth.
  3. An idea of the susceptibility of each region to natural catastrophes.

In my experience, people in the insurance business put a bit too much emphasis on #3, which a cursory understanding of is easy to get but a deep understanding of is currently beyond any intelligence yet discovered. The reality is that all of the science in the underwriting is in #s 1 and 2: where are the buildings and what are they worth?

What if Google just suddenly realizes it can probably do this better than anyone else?

“We already have what we call ‘view codes’ for 6 million businesses and 20 million addresses, where we know exactly what we’re looking at,” McClendon continued. “We’re able to use logo matching and find out where are the Kentucky Fried Chicken signs … We’re able to identify and make a semantic understanding of all the pixels we’ve acquired. That’s fundamental to what we do.”

More here.

I like imagining an even more tantalizing project: open source cat underwriting. Open Street Maps does most of what Google does except for free.

Will some actuary use this public data to check an industry-changer into Github one day? Might Index Funds (capitalizing this automated underwriting platform) and governments (subsidizing coastal homeowners) one day split all catastrophe insurance between them?

Thoughts On Mathematical Finance (3F/MFE Exam)

I moved over to the actuarial side of the business a few years ago from our capital markets team. I’m not an actuary, but I took an interest in modeling and followed the advice of my first boss in the business: “want to get ahead? Be more useful to us”. We were starting an actuarial department and I joined in.

I didn’t start taking the exams seriously until my wife got pregnant. It’s been a breakneck pace since and I just wrote my fourth exam, the 3F/MFE, subject of which is the math of derivative securities. I’m fairly familiar with the area from the CFA exams, but what does this have to do with valuing P&C liabilities, the job I’m training to do? And anyway, isn’t all this math exactly what is supposed to be wrong with finance?

There are two big ideas in the syllabus that are worth discussing. The first has some notoriety, which is that we assume asset returns are distributed normally and so prices are distributed lognormally. The immensity of this assumption cannot be overstated. First because just about everything in mathematical finance requires it to be true; second, it’s just about complete garbage. In fact, there is absolutely no attempt to justrify this assumption in any of the required readings.

The other idea is a bit more esoteric and is called risk neutrality. Risk neutrality is a way of dealing with the problem that people are risk averse, in that we’re unlikely risk $1 of loss for $1 of gain on a 50/50 bet. So we’d only take a 50/50 bet if we could win $1.5 and lose $1.

In a strange twist, the math skips over the ‘utility’ of a dollar of profit vs dollar of loss. Instead, we reweight the probabilities so that the return is the risk free rate. In a risk neutral world, it’s not a 50/50 bet.

Don’t worry, I don’t really understand it, either. One of my complaints is that, once I’ve passed the test (8 weeks ’till they tell me!), it doesn’t matter whether I understand it or not, I’m never going to use this knowledge. The understanding is literally worthless.

So why did I have to learn about it? In the the released sample problems there’s this really interesting remark (number iii after the solution to problem 71 in this pdf document) in respect of risk-neutral pricing:

Arguably, the most important result in the entire MFE/3F syllabus is that securities can be priced by the method of risk neutral pricing… Some authors call the following result the fundamental theorem of asset pricing: in a frictionless market, the absence of arbitrage is “essentially equivalent” to the existence of a risk-neutral probability measure with respect to which the price of a payoff is the expected discounted value.

I didn’t understand what that meant until I really got stuck into the Black Scholes derivation. The story of the that derivation goes like this: borrow some money and buy a derivative. Now hedge away the risk with some shares. All that’s left is the risk free rate. The big trick is figuring out how many shares you need to buy/sell to hedge he risk of the derivative. That’s where normally distributed returns come in and that’s where everything falls apart.

The CAS/SOA don’t make similar comments about any other part of syllabus, so it’s interesting that they decided to play Hitchcock and jump in for a cameo here.*

Why do they think this here is such a big deal? And, more importantly, why do they think actuaries should know this stuff?

There’s a bit of history to this exam. Originally it was paired with a more statistics-heavy exam for the complete Exam 3 (now they’re 3F/MFE and 3L). The split exams were half-exams, too, but the MFE has crept back up to a full-on 3-hour test. Why?

And think of the social context. During the 90s and early 00s, financial mathematics was a pretty big deal. Physicists were pouring out of science departments and onto trading floors, probably amazed that all this abstract math was a valuable skill. It’s no surprise that the CAS/SOA tried to hop onto that bandwagon.

But ultimately the ideas aren’t terribly compelling. I’ve studied these models intimately now (the fixed income ones are laughably useless and WAY more complicated), and I’d never give someone my money to trade with them.

*Get ready for a tangent:

There’s an essay at the beginning of my copy of Moby Dick that for no good reason pops into my head all the time. Melville is describing the most essential characteristic of a whale, the spout, and the essayist claims he has sheds the character of Ishmael and starts writing as himself:

the spout is nothing but mist. And besides other reasons, to this conclusion I am impelled, by considerations touching the great inherent dignity and sublimity of the sperm whale… He is both ponderous and profound. And I am convinced that from the heads of all ponderous profound beings, such as Plato, Pyrrho, the Devil, Jupiter, Dante and so on, there alwasy goes up a certain semi-visible steam while in the act of thinking deep thoughts. While composing a little treatise on Eternity, I had the curiosity to place a mirror before me and ere long saw reflected there a curious involved worming an undulation in the atmosphere over my head. The invariable moisture of my hair, while plunged in deep thought, after six cups of hot tea in my thin shingled attic of an August afternoon, this seems an additional argument for the above supposition.

Incidentally, I don’t buy it; I think he’s still Ishmael here (treatise on Eternity? Puh-leeze). But I think the sentiment remains. Melville is clearly in awe of the whale and pays it a kind of honor by comparing it to his greatest heroes.

Maybe the SOA/CAS is just so impressed with the elegance of the math that they HAD to break in for a chat.

Self-Driving Cars Approach? Doubt It.

Geekdom is a-flutter over Google’s self-driving car project.

Google announced a new phase of its self-driving car project Tuesday. The test vehicles, of which there are “about a dozen on the road at any given time,” have so far logged 300,000 miles of road testing without a single accident under computer control. In the next phase of testing, team members will start commuting to work solo, with the robot at the wheel.

Google also showed off a new vehicle type added to the program, the Lexus RX450h SUV. Now that the self-driving car software is comfortable in a variety of traffic conditions, the next phase will test snowy roads, temporary construction signals and other unusual terrain.

More here. Via MR under a very optimistic headline. An optimism I don’t share, unfortunately, but not because of the technology.

What I’m worried about is whether our society is genuinely capable of putting the most lethal weapon on earth in the hands of AI.

Remember that the auto liability insurance market is the largest in the world by an order of magnitude. This is so because everyone who can drive has the power to maim and destroy a lot of property and life around him/her. Auto insurance works because agents have control over their actions and are responsible for those consequences. Each person pays premium.

Who pays when Google’s driver hits a schoolbus full of children and sends it rolling down a cliff? What if Google’s driving algorithm isn’t at fault but a court pins the blame anyway? Remember Google’s car need never cause an accident for people to scream “Skynet!” and pull the plug.

Like with Kickstarter, Google’s car will only truly be tested when someone gets effed over. You tell me how long Kicktarter will last when someone commits genuine fraud and everyone’s confdience evaporates. Caveat Emptor? Yeah right.

It is our liability system (which mostly reflects an underlying extreme risk aversion) that will probably kill these technologies.

The Twilight of Catastrophe Modelers

One interesting idea in Kevin Kelly’s *What Technology Wants* is that technologies undergo a life cycle where they are at first specialized and poorly designed (they just don’t effing work right) and progress to the point where they are ubiquitous.

I am reminded of that by this article on cat models (via Jim Lynch):

Speakers at several recent insurance conferences stressed the need for property insurers and reinsurers to develop their own independent views of catastrophe risk, rather than outsourcing their risk views to third-party vendors. But they differed on exactly how to get to there.

While experts observed that reinsurers and insurers are increasingly using multiple models to inform their views of cat risks (with the smallest insurers enlisting the help of reinsurance brokers to accomplish this), Peter Nakada, managing director of RMS, a Newark, Calif.-based firm, suggested that a multi-parameter view is preferable to a multi-model view…

“Pick one of the giant simulation things and then force the modeling firms to give you the secret sauce from inside the models,” he advised, suggesting that users can then select “multiple points of view on the parameters that run the model” to develop a range of estimates.

Nakada is fighting a serious rearguard action here. RMS overreached with their last update and modeled claims costs skyrocketed. Instead of recanting on their update (unthinkable), they instead downplay the importance of their technical view of the risk. And they’re right. But I wonder if they realize how much pushing the commoditization of their black box will fundamentally change their business.

Kelly would phrase it like this: what happens when open sourced cat models are ubiquitous? How does that affect the industry?

The day may well come when the ‘secret sauce’ of the cat models goes open source and the state of the art is free to all. In that world RMS goes from R&D shop to industry consultant. They’ll provide outsourced analysis and data cleaning services.

They’ll fight like rabid dogs to avoid the billable hour revenue model, since nobody gets rich in businesses that don’t scale, so they’ll need products. Maybe they’ll look to compete with ISO and offer some kind of master database of property values in the US, who knows.

Their heyday, though, is perhaps ending.

Not From An Actuarial Textbook

One of the mainstays of actuarial education is thinking about the ways one might underwrite auto liability insurance. The most accurate predictor of risk should be miles driven; after all, the more you’re on the road, the more likely you’re going to get into an accident.

The problem with calculating the number of miles driven is that it’s simply impractical. You’d need an insurance company to install a mileage monitor in every car, scoffs the textbook, and that’s just too costly to do. Some day, perhaps…

Well well well, the day has arrived!

Telematics insurance relies on a databox the size of a mobile phone which is installed by the insurance company into your car. The box does not damage the car and will not affect the warranty; it uses less energy than a car radio so should not drain your car battery.

What Data Is Collected?

Data from this box is collected by GPS, enabling insurers to monitor:

  • The distance it travels at those times
  • Where the car is located
  • On what type of roads the car travels
  • Speed of travel
  • Braking behaviour of the driver
  • Direction and speed of travel before and after a collision
  • Force of impact in a collision
  • At what times the car is used