Right: Homeowner renewal premiums for long term, claims free customers double because “theres a storm a comin”. Or, in the current economy, people with good credit history turns slightly negative and huge rate increases follow.
What a pain this has been to manage at the retail level.
If all you do is sell on price,,, then yes, this is difficult to manage on the retail level. However if the agent themselves adds value, and has a “coaching” relationship with their customer, then price changes should not make or break the deal.
Just goes to show you that insurance executives will always believe something they do not understand and computer and data salesmen are always willing to make them believe it. Taking the human element out of it is never a good idea for our business
Seriously…..Brian, if you were running a business that was losing money and someone came to you and told you there was a way to rate risks using a plethera of statistics that will in time turn around your book to be profitable again, you would do it wouldn’t you?
Odds are you were the same guy that complained 2 years ago about how prices were too low.
I think you are not giving enough credence to proven statistical correlations of credit scores and risk (the best example to date). The greater concern is if we can predict risk and charge accordingly, when does it cease to be insurance because there is not longer a spreading of risk, only identyfying and excluding it or making each insured pay for all their risk!
i would like to disagree a bit here.. they are only predicting risk and not the actual event, therefore predictive pricing helps identify customers with risky behaviors to pay for those behaviors..
It is still spreading the risk because when an actual event happens the person is covered and then post facto the risk is spread on those who did not..
If your assessment is correct that “quack” science has driven up price, why not start an insurance company to take advantage of the situation and put those companies using modeling out of business ?
No. If there was modeling and prudent regulation, the melt-down would not have happened. The melt-down was caused by guys likeuys like Alan Greenspan, Larry Summers, Hank Paulson advocating against regulating derivitives, credit default swaps, and CDO’s. Your bank should not be able to sell it’s mortgage. But, that’s what goes on. What incentive is there to underwrite a loan carefully if there are mortgage bankers willing to buy them in bulk and resell them in larger bulk while companies (used to be AIG) sell credit default swaps against their performance? There is still no regulation preventing this practice. I’d love some predictive modeling on that. How about “those who forget the past are condemned to repeat it”?
One of the key points of sound underwriting is taking key stats, and using the data to establish rate. I think that if predicitve modeling is seen as one tool in the tool box, then its useful. If medeling is the only tool in the tool box, then your probably in trouble. So,,, useful tool YES,,, the only useful too, NO. If you are a carrier that is telling your underwriters to use the model as the main UW criteria, then your using it wrong. The model is just one bit of underwriting data.
I’m not sure some of you guys get it. Insurance is, by definition, a pooling of risk. That means consumer rates go up because *other* people have had claims. Insurance is one of the only industries I can think of that has to set a price *before* it knows its costs for the coming year.
If I make widgets I know how much they cost to research, develop, manufacture and distribute, and I can set a price accordingly. If I run an insurance company I have to guess and wonder if there will be storms, fraud, adverse risk selection – any number of problems that will affect the end-result loss results for the year.
Because of commodification of the market, consumers have been taught to look at the lowest price. Some agents make it worse by selling based upon price alone, which justifies that behavior in the consumer’s mind. This focus on price results in the increaing need for carriers to create very sophistocated pricing models that attract their niche customer to yield the best profit.
Insurance prices are driven by fluctuations in soft and hard markets, and those fluctations are driven by carrier profit. The market has hardened, carriers are restricting coverage and raising rates. In a few years, carriers will shore up profits enough that a few will decide to slash prices to buy up market share. Smart carriers resist the temptation to play that game, but some will choose to slash prices to compete…and the cycle starts all over again.
In insurance, it’s grow or die. Multivariate pricing models are aimed at one thing – creating an (all too short-lived) advantage at pricing and attracting the most desirable customers.
All the more reason why you should not sell on price alone. There is a portion of the market agents will never see who are brainwashed by the Flo’s and Geico’s of the world.
Do your job, let Flo screw up. When she does those people will call. Because they will call….eventually.
Here’s the real problem—there is nothing out there and there will NEVER be anything out there that can accurately predict “individual human behavior”. Think about it!!
Agreed… but Statistics is not necessarily about predicting a specific persons behavior. However, I do believe that science has developed where it can accurately predict trends. If an UW is holding a predictive model up as proof positive on how to price an account, then they are using the model wrong.
Insurance is priced across groups, not upon individuals. It’s risk pooling, not individual risk assessment.
Pricing models provide a price based upon the risk characteristics…all consumers with the same risk characteristics and the same rating info should receive the same price. It’s a completely objective way of establishing a price for each type of customer.
Insurance companies aren’t even trying to pretend to price each individual person. To do so would be expensive, slow, and would add in subjective elements – the opposite of what carriers are trying to do.
Your underwriter is the person to talk to when pricing fails at getting to the right price. A good underwriter knows when to follow the pricing model and when to apply subjective decision making.
Maybe you should find another carrier to do business with :) We make exceptions on a daily basis when ot makes sense based upon the account profile. We don’t always say yes, but there is no sense of the pricing model being final.
Please remember that “individuals” make up groups and if “individuals” aren’t placed in the right groups, it’s all a waste of time (garbage in, garbage out). Insurance carriers, like other businesses, make errors all the time and when they report their stat data to ISO, as an example, they also make mistakes when entering data into their systems, etc. The error factors in the above situations usually range from at least 10% to 20%, so the data that comes out is only 80% to 90% accurate, at best. Is this acceptable? The big three credit reporting outfits always make mistakes. I did a recent review on my persoanl info and spotted several situations wherein some showed accounts to be completely paid off, while others still showed balances due and some situations wherein accounts were paid late, as shown by some and not by others. This gives me credit scores all over the board and this is simply not correct. Again, this is part of the error factor referred to above. I am a minor believer of credit scoring as part of the overall underwriting process, however, some carriers place too much weight on this factor and this is when problems occur. NO ONE WILL EVER BE ABLE TO ACCURATELY PREDICT HUMAN BEHAVIOR!!!!
here’s the problem, as those staticians would say —
50% empty or 50% full?
they claim based on statitics that if you have a lower credit score you are more likely to file a claim… well, shucks, i want my car fixed because it was in an accident… but you are willing to charge me more just because of my credit and not just based on my driving record… but if i have a great credit and pay everything off, i even payoff that car i hit – i then don’t file a claim… how unfortunate that i don’t have to worry about that claim with my insurance… i managed to pay it out, just because i was RICH! and can pay anything off… is that truly fair? i like the fact some are trying use predictors of driving behavior.. this makes more sense than if you have bad credit vs good credit… how you drive is more important than how good are you at spending money… if that were the case, what about those CEO’s that put their companies into bankruptcy! they are not good tenders of company and they money! good example — AIG! how many times do we hear them owe another dime or 2 somewhere… yet, i bet that CEO has a great insurance company… so, who’s fair?
This is the same complaint offered by people with lousy credit scores all over the country, and it just doesn’t hold water. As I said above, pricing isn’t about “fair”. It isn’t about establishing rich from poor, or race, or any other socio-political factors.
Insurance companies use credit scoring as one factor, among dozens, to segment their book of new business and their book of policies in force. The first, and most important fact, is that crediting scoring is predictive – all of the studies show it to be predictive. It is even more predictive for auto policies than it is for home policies.
Credit scoring is merely an indicator of overall responsibility. If someone is irresponsible in one part of his life, he is likely to be irresponsible in other parts of his life. Looking at my loss ratios, customers with poor credit and prior claims have terrible loss experience – on average 300-500% higher than customers with clean records and higher credit. If I’m not making money on that slice of customers, I need to raise rates there. Simply and plainly put.
As I noted above, carriers are trying to segment such that they have the best prices for the best customers. They do that by lowering prices on the best consumers (those with lower loss ratios…ie. better credit and clear loss history) and raise rates on less desirable consumers (losses, poor credit).
No one reasonably thinks a consumer isn’t going to file a claim if they’re “rich”. But loss experience tells me someone with a 300 score is more likely (when priced as a class, not as an individual) to have a worse loss experience.
don’t truly think that a person w/a 300 score is more likely to have a worse accident… according to the supposed statistics, it means that person is more likely to file a claim… is it unfair?! i think so… because someone with great credit, could have a bad driving record and will likely be more of a candidate for a loss than someone with a great record and lower credit score… we can twist any set of numbers anyway we feel… credit is one of them… but how do you distinguish the proper credit score because someone is buying a house or getting a credit card and every time one of those hits that score, it deducts from the score… so whose winning? that statisticians are! and so are businesses who think that this model is appropriate…
I agree.. credit score has relationship to financial actions not driving behavior.. we have to find a similar score that makes more sense to Driving behavior and thereby link to insurance premia
You’re still missing the point. Credit is ONE factor, among dozens.
Statistically, someone with no prior claims and a high credit score, is married, has a college education or better, a white-collar job, owns a home, has higher limits (and a host of other variables if we’re talking about home or auto) is the best (i.e. most profitable) risk.
Statistically, the worst risk is someone with prior claims, low credit score, unmarried, rents, didn’t finish high school, is unemployed, has low limits (and a host of other variables if we’re talking about home or auto) is the worst (i.e. least profitable risk.
Most people fall in to the middle. Think about a married small business owner with a Masters degree, own a nice house, has high limits, but a bad year for the business hits his credit score. Is he bad risk? Of course not! But statistically he’s a slightly worse risk than the person I described above. He’s also a slightly worse risk if he doesn’t own a home. Or if he only graduated from high school. Or is single. Or any other factor that deviates from the statistically ideal customer.
If you want to get in to the nuts and bolts of credit modeling, each company has its own proprietary model of how it views credit. Your Beacon score (i.e. what shows up at freecreditreport.com) has absolutely nothing to do with your insurance score.
Carriers spend a lot of time pouring through their policies to find which customers yield the best loss results, and discount those customer’s premiums. Some carriers place a high emphasis on debt to credit ratios. Some put a lot of emphasis on the number of accounts. Some look at the number of accounts open. Each company is different because each company’s loss experience is different.
The whole point is that companies use ALL of this information to create groups of customers and set a price for that group based upon their profit/loss experience. The day low credit score customers yield the best profit results will be the day you see carriers lower their rates for customers with poor credit.
Let’s see now—what good was predictive modeling, credit scoring, algorithms, etc., insofar as the TWO BILLION DOLLAR hail storm that hit Dallas, Texas last week?? Did the hail stones only hit vehicles and other structures that had bad credit? When it comes to predictive modeling and credit scoring, nothing can out wit mother nature. Here’s a question for everyone—-Does credit scoring take natural disasters into consideration when developing a model? Mother nature and human behavior can NEVER be 100% predictive 100% of the time. The longer I’m in the insurance business (40+ years so far), the more I feel that our industry is moving away from good old fashioned “common sense”. God forbid we should asctually use our brains for anything!!
That’s why the models take into account geographic location and historical weather. You are talking about a model like it is only based on one factor and are missing the point because of it.
To assume the models are poorly done without any research into how the models work is a pretty poor intellectual assessment.
Credit score is only one of several factors that go into a good predictive model. So, yes, credit scores could not preditc weather patterns. However, there is other science that can predict weather patters. Its short sited to only focus on credit scores. If your predictive model only uses credit scores, then its no good. The whole point is a comprehensive model that takes many factors in mind.
….and insurance companies break their books of business into territories that are priced based upon group loss experience. So, after the tornado and hail storms in Dallas this year I’d expect their auto and property rates to go up. Whether they have good credit or poor credit.
The industry has changed. Carriers don’t have two prices any more – in fact, most companies have thousands of price points. Why? Because there’s fierce competitive to figure out, and acquire, the best (i.e. most profitable) customers.
Having a poor/unsophistocated/flawed rating model only means that a company will be adversely selected against, because their prices will be wrong (i.e. too low). As an agent, why would you want to do business with a carrier with inaccurate/inadequate pricing? It might feel good to offer a cheap price and win a customer – but you’re jeopardizing your book, and that carrier’s book, because the loss ratio impact will be devastating. That’s exactly why soft markets turn hard.
“Did hail stones only hit vehicles and other structures that had bad credit?”
More than likely, no. It is more likely that insureds of all credit levels (low, middle, and high) suffered losses. A predictive model based on this data would say credit is not meaningful or predictive of wind/hail losses (because all credit levels were hit). This model would validate what you regard as “common sense.” so I’m not sure what you are really arguing here.
Would you be willing to say the same about fire and theft losses? Is a property owner with lower credit more likely to have a fire? A more severe fire? Is a person with lower credit more likely to have theft losses?
Is a person without a garage more likely to have a wind/hail claim? Is a person that lives further from the fire station more likely to have a severe fire loss?
It sounds like we want to charge for these things but we don’t want to use a predictive model to do it. What might not be understood here is that a predictive model takes into all of these variables at once. If you want to do it the “old-fashioned” way and price for these variables independently (not simultaneously), your rates will be excessive because you will be double charging customers.
More than likely,”No.” I think you missed the “no” part. My point is, credit isn’t very predictive of wind/hail losses and actuaries do understand (and adjust for) that. However, credit is predictive of other types of losses like theft. A person who is responsible and pays his/her bills on time is more likely to lock his/her doors and own a home in a neighborhood where crime is less likely to happen. I think there’s sense in that.
They call it predictive modeling however they do not file these plans as “pricing tools” with state regulators. Are they being compliant? They back into their schedule rating / experience rating after determining their predictive price. They say underwriters have say in final pricing however when you adjust the price different from the model it is viewed negative in your performance. Companies want computers pricing to cut out people (salary) and expense (benefits).
Yes, they are being compliant. The model does not alter or change filed rates. So, I fail to see how there is an issue with State Regulators? Also, I do not believe that carriers want to cut underwriters out… instead, I think they want to simply give underwriters another method to review a risk. Just because an underwriter now has a new tool to use, that does not mean the value an underwriter has. I submit to the Insurance Journal Readers that Predictive Modeling enhances underwriting. I do enjoy a creative underwriter,,, but I would rather have it based on Science rather than art.
Wow, I really hope that some of the anti-modeling comments here come from executives for my competitors. I suspect, however, that a lot of the ill will is coming from underwriters who have lost authority in the face of expanding pricing sophistication. The fact of the matter is that the marketplace will judge who is right or wrong. Right now, the guys using more sophisticated models appear to be winning.
As was pointed out before, if you believe that the modeling is hurting companies, then go to a company that does it the old fashioned way. If you’re right, then your old firm should be out of business in no time.
As an aside, I think those comments are from producers who don’t like talking credit scores with their customers and don’t have the relationship to pull off valued-based sales.
I am a producer… and yet I do see the value of Modeling. You know why, because it helps me be more profitable with the companies I represent… hence better profit sharing. The point missed here, is that companies want to retain the good customers,,, and then get the right price for the higher hazard customers. Its my job to discuss the underwriting process with the insured’s. Its my job to point out areas of high hazard,,,, if my customer has bad credit, then its my job to point out why that hurts them from an insurance stand point. I am not passing judgement on them, I am just pointing out the statistics. If my customer is located in a high hail zone, then I talk to them about why their property rates are higher. I find it encouraging that all my peers do is sell on price. My customers love it when I take the time to explain things to them. All the model does is to help understand potential hazards, that gives me talking points with insured’s. And apparently, it allows me to be different from their current agent, who just seems to want to throw the carrier under the bus and blaim some model for all the price increases….
Can you please provide some facts for your position that guys using sophisticated models are winning other than the overhead war? Because when you bring in the geeks that no nothing about an industry and pay them exorbitant salaries for sitting in the bastion of academia their whole lives and have now ventured out with their oh so useful theories based upon some initial guess and that is exactly how these models are started with a guess of what might or might not happen and when proven wrong they guess the other way to appear that they were correct all along…
My dad always said, be nice to the geeks in your class, because one day they will be your boss.
I think your question is a false presumtion… you state that “geeks that no nothing about an industry…” First of all, you are implying that the people who build the model are not fit for the task. That is bogus… the presentations and articles I have read have all been from people who have insurance designations and are clearly experts. So I do not support your claim that the people who designed the model are not hip to the scene.
Finally, the people who made the models are not underwriters… they created a tool that underwriters use.
So they created a tool to be used by people in an industry that that they know nothing about? That is like explaining color to a blind person. If they have not worked in the field in which they are creating and the key word being creating, these tools, how do they they know that what they are providing is what is needed?
Just because some one reads a book about carpentry does not make them a carpenter and a academia type who studied theories their whole life are not in a position to create a tool to be used by underwriters say in the Professional Liability arena.
Underbroker – By once again making the claim that the tool is being created by “people in an industry that that they known nothing about,” I can see you missed the response to your initial statement…
Perhaps a different tact? What makes you so bold as to believe “They” know nothing about the industry?
It seems an interesting claim you’ve made, to be sure. Basically, what I hear you saying is: In my experience Actuaries know nothing about the industry; therefore, as a group Actuaries know anything about the industry. I find this interesting in that (while horribly flawed on a variety of levels) this is the very statistical reasoning you appear to be assailing.
I’ve known individuals in Actuarial who have worked in underwriting, distribution, and claims to flesh out their knowledge of the business. Further, at least in the companies I’ve seen, Actuaries work with others within the organization to ensure that the model is tuned based on “real world experience” (i.e., often annecdotal) of others in business roles.
Also, if you could find me a couple insurance companies that don’t use Actuarial models, then I’d be more than happy to look into whether they do better or worse over the longer (5-10 year min) term. My suspicion is you won’t find any of substantial size and longevity because not using stats as a tool is an antiquated methodology, but I’m perfectly willing to make the effort if you are.
June 19, 2012 at 8:54 am
ComradeAnon says:
Like or Dislike:
0
0
Anyone ever heard of this lowering someone’s premium? This nothing more than a way to charge some people more money.
I have been insured with a small mutual insurance company for a decade. They recently went from an outdated and traditional/basic rating plan to a multivariate plan. Customers were not forced to leave the legacy plan to the new tiers, but my auto premium went from $1200 year to $800 year because I have good credit and few violations and claims. Responsibility matters – and there are many ways to evaluate personal responsiblity. Creditworthiness is but one of them. Those who disagree just don’t want to see the facts.
Yes, I have seen the model lower premium. One thing the model can do is to identify customers that are low risk… all the comments assume that the model only shows Hi risk customers… NOT True. It can be the case that the model shows an underwriter that the risk is not as bad as once thought… so the pricing can be reduced. Trust me, the insurance companies want to retain the customers that the model shows to be low risk.
The question has been raised by others and I would like to chme in: Are there results showing that by predictive modeling, rates have gone done for a particular line of business?
I recall many years ago an underwriter who went out on a limb and agreed to insure a broadcasting tower in Oklahoma. Within a few months of the policy term, a tornado destroyed the tower. The underwriter chalked it up to experience and renewed the policy. In a rare situation a tornado came that second year of the policy and again destroyed the tower. The underwriter swore never again to insure broadcast towers and, so far as I know, never did. Naturally, there has not been a tornado affecting that tower in the past 25 years.
I cannot help but wonder how a predictive modeling algorithm would have handled the policy after the first year and the second year.
There are always anomalies and stories to tell. But that does not invalidate the statistics as a whole.
The model is not designed to predict the future for a specific tower. It is instead designed to predict trends, for similar towers in similar areas. In your specific case, loss history is only one such factor in the model. You presume the model would tell you to non-renew the account as your underwriter friend did. That is just an assumption on your part. Since there are so many factors, and since loss runs are just one factor,,, the model could still tell the underwriter to renew the tower.
Predictive Modeling just means Insurance Companies don’t want to Underwrite anymore! Does anyone remember the saying “that insurance company will not write Pig Iron under water for fire only even with a Rust exclusion’? The companies are removing thier best underwiters…..We the Agents!
How does the existance of a predictive model remove agents? I am confused by your statement Stephen. Please explain how “companies are removing their best underwriters”
Far to often I’ve been told “the model says X and there is nothing I can do about it and if you have to shop it, I understand but I must get what the model says”. PM is eliminating the art of underwriting for total reliance upon the science of underwriting. At some point, some VP will ask “why do we need underwriters?” Think Flo as the underwriter of the future.
I appreciate what you are saying Insurance Fox. I’m not mad about insurance using credit scoring, but I took a hit, and I think a lot of people did when that was used to rate drivers. And it frustrated me. I’ve had credit issues. Got into a little bit of trouble with some credit cards, but got on track and cleaned up my credit. And what bothered me was that while I was cleaning up my credit and paid every single insurance premium on time, I still took the hit. I’m a good driver with a good record, but really felt dupped by my car insurance company.
The GLM is not the answer it reminds me of the Mortgage Industry when they came up with the secret sauce to qualify people for mortgages. Guess what the mortgage industry almost destroyed the Banking Industry where is common sense and good understanding the proper premium for the exposure.
The fringe on the right claims there’s no such thing as climate change. It’s just some tree-hugger’s concocted theory to hurt oil companies. Right? But, aren’t insurance companies supposed to be right-leaning, ecology-bashing conservatives? So, are the insurance people who come up with these predictive formulas pinkos? Wait, I’m confused.
It is quite dangerous to over driven the predictive modeling. My current company is large P&C company. I haven’t heard any successful story to use the predictive modeling on commercial line. It continuely lost money on selling products by pricing based on predictive model. Predictive modeling is an excellent idea and does bring fresh air to insurance industry. But it is just methods and models, I think it is not a good way to 100% to rely them. Statistics used history data to build model. But the insurance industry is heavy regulated. The regulation enviroment changes in recent year and doesn’t reflect insde the data on which the model is built. I saw the fever on predictive modeling so far. It reminds me the IT and website booming in earlier of 2000s. Hard to know when it will end. But it will be told by profitability of company by using the models.
I am not an expert in insurance, but I worked for a long time in finance. The excitement around the predictive modeling rings alarm from my experience in the financial world
Combine predictive skills with underwriting and pricing rules, should provide insurer a more sophisticated results.This, however will largely depend on the modeler’s statistical skills and business acumen as well.
This quack science has been used efefctively to restrict capacity and drive up price nothing more No more corretc than man caused climate change.
Right: Homeowner renewal premiums for long term, claims free customers double because “theres a storm a comin”. Or, in the current economy, people with good credit history turns slightly negative and huge rate increases follow.
What a pain this has been to manage at the retail level.
If all you do is sell on price,,, then yes, this is difficult to manage on the retail level. However if the agent themselves adds value, and has a “coaching” relationship with their customer, then price changes should not make or break the deal.
Just goes to show you that insurance executives will always believe something they do not understand and computer and data salesmen are always willing to make them believe it. Taking the human element out of it is never a good idea for our business
Seriously…..Brian, if you were running a business that was losing money and someone came to you and told you there was a way to rate risks using a plethera of statistics that will in time turn around your book to be profitable again, you would do it wouldn’t you?
Odds are you were the same guy that complained 2 years ago about how prices were too low.
Watch the movie Moneyball, with Brad Pitt. It shows the power of predictive modeling.
Good movie.
An equation with n+1 variables will always perfectly explain n data points.
…but is not necessarily very predictive.
I think you are not giving enough credence to proven statistical correlations of credit scores and risk (the best example to date). The greater concern is if we can predict risk and charge accordingly, when does it cease to be insurance because there is not longer a spreading of risk, only identyfying and excluding it or making each insured pay for all their risk!
i would like to disagree a bit here.. they are only predicting risk and not the actual event, therefore predictive pricing helps identify customers with risky behaviors to pay for those behaviors..
It is still spreading the risk because when an actual event happens the person is covered and then post facto the risk is spread on those who did not..
If your assessment is correct that “quack” science has driven up price, why not start an insurance company to take advantage of the situation and put those companies using modeling out of business ?
Wasn’t it revolutionary modeling that brought down Wall Street in 2008?
Revolutionary modeling took down wall street b/c such models justified taking greater risk.
Insurance risk modelling is telling insurers to limit their risk.
No. If there was modeling and prudent regulation, the melt-down would not have happened. The melt-down was caused by guys likeuys like Alan Greenspan, Larry Summers, Hank Paulson advocating against regulating derivitives, credit default swaps, and CDO’s. Your bank should not be able to sell it’s mortgage. But, that’s what goes on. What incentive is there to underwrite a loan carefully if there are mortgage bankers willing to buy them in bulk and resell them in larger bulk while companies (used to be AIG) sell credit default swaps against their performance? There is still no regulation preventing this practice. I’d love some predictive modeling on that. How about “those who forget the past are condemned to repeat it”?
One of the key points of sound underwriting is taking key stats, and using the data to establish rate. I think that if predicitve modeling is seen as one tool in the tool box, then its useful. If medeling is the only tool in the tool box, then your probably in trouble. So,,, useful tool YES,,, the only useful too, NO. If you are a carrier that is telling your underwriters to use the model as the main UW criteria, then your using it wrong. The model is just one bit of underwriting data.
you’re = you are……not .
Otherwise, spot on with your comments.
I’m not sure some of you guys get it. Insurance is, by definition, a pooling of risk. That means consumer rates go up because *other* people have had claims. Insurance is one of the only industries I can think of that has to set a price *before* it knows its costs for the coming year.
If I make widgets I know how much they cost to research, develop, manufacture and distribute, and I can set a price accordingly. If I run an insurance company I have to guess and wonder if there will be storms, fraud, adverse risk selection – any number of problems that will affect the end-result loss results for the year.
Because of commodification of the market, consumers have been taught to look at the lowest price. Some agents make it worse by selling based upon price alone, which justifies that behavior in the consumer’s mind. This focus on price results in the increaing need for carriers to create very sophistocated pricing models that attract their niche customer to yield the best profit.
Insurance prices are driven by fluctuations in soft and hard markets, and those fluctations are driven by carrier profit. The market has hardened, carriers are restricting coverage and raising rates. In a few years, carriers will shore up profits enough that a few will decide to slash prices to buy up market share. Smart carriers resist the temptation to play that game, but some will choose to slash prices to compete…and the cycle starts all over again.
In insurance, it’s grow or die. Multivariate pricing models are aimed at one thing – creating an (all too short-lived) advantage at pricing and attracting the most desirable customers.
Underwriter, wish I could “Like” that one 20 times.
“agents make it worse by selling based upon price alone, which justifies that behavior in the consumer’s mind.”
And let’s not forget the GEIGO’s of this world. They spend more money in advertising which reaches more people than agents or brokers could dream of.
All the more reason why you should not sell on price alone. There is a portion of the market agents will never see who are brainwashed by the Flo’s and Geico’s of the world.
Do your job, let Flo screw up. When she does those people will call. Because they will call….eventually.
Here’s the real problem—there is nothing out there and there will NEVER be anything out there that can accurately predict “individual human behavior”. Think about it!!
Agreed… but Statistics is not necessarily about predicting a specific persons behavior. However, I do believe that science has developed where it can accurately predict trends. If an UW is holding a predictive model up as proof positive on how to price an account, then they are using the model wrong.
Insurance is priced across groups, not upon individuals. It’s risk pooling, not individual risk assessment.
Pricing models provide a price based upon the risk characteristics…all consumers with the same risk characteristics and the same rating info should receive the same price. It’s a completely objective way of establishing a price for each type of customer.
Insurance companies aren’t even trying to pretend to price each individual person. To do so would be expensive, slow, and would add in subjective elements – the opposite of what carriers are trying to do.
Your underwriter is the person to talk to when pricing fails at getting to the right price. A good underwriter knows when to follow the pricing model and when to apply subjective decision making.
that’s another problem; the UW has no discretion. The pricing model is final.
Maybe you should find another carrier to do business with :) We make exceptions on a daily basis when ot makes sense based upon the account profile. We don’t always say yes, but there is no sense of the pricing model being final.
If the UW has no discretion, then that is indeed a problem. The model should be one tool amoung many.
Please remember that “individuals” make up groups and if “individuals” aren’t placed in the right groups, it’s all a waste of time (garbage in, garbage out). Insurance carriers, like other businesses, make errors all the time and when they report their stat data to ISO, as an example, they also make mistakes when entering data into their systems, etc. The error factors in the above situations usually range from at least 10% to 20%, so the data that comes out is only 80% to 90% accurate, at best. Is this acceptable? The big three credit reporting outfits always make mistakes. I did a recent review on my persoanl info and spotted several situations wherein some showed accounts to be completely paid off, while others still showed balances due and some situations wherein accounts were paid late, as shown by some and not by others. This gives me credit scores all over the board and this is simply not correct. Again, this is part of the error factor referred to above. I am a minor believer of credit scoring as part of the overall underwriting process, however, some carriers place too much weight on this factor and this is when problems occur. NO ONE WILL EVER BE ABLE TO ACCURATELY PREDICT HUMAN BEHAVIOR!!!!
is absolute accuracy the goal or incrementally getting close the goal?
here’s the problem, as those staticians would say —
50% empty or 50% full?
they claim based on statitics that if you have a lower credit score you are more likely to file a claim… well, shucks, i want my car fixed because it was in an accident… but you are willing to charge me more just because of my credit and not just based on my driving record… but if i have a great credit and pay everything off, i even payoff that car i hit – i then don’t file a claim… how unfortunate that i don’t have to worry about that claim with my insurance… i managed to pay it out, just because i was RICH! and can pay anything off… is that truly fair? i like the fact some are trying use predictors of driving behavior.. this makes more sense than if you have bad credit vs good credit… how you drive is more important than how good are you at spending money… if that were the case, what about those CEO’s that put their companies into bankruptcy! they are not good tenders of company and they money! good example — AIG! how many times do we hear them owe another dime or 2 somewhere… yet, i bet that CEO has a great insurance company… so, who’s fair?
This is the same complaint offered by people with lousy credit scores all over the country, and it just doesn’t hold water. As I said above, pricing isn’t about “fair”. It isn’t about establishing rich from poor, or race, or any other socio-political factors.
Insurance companies use credit scoring as one factor, among dozens, to segment their book of new business and their book of policies in force. The first, and most important fact, is that crediting scoring is predictive – all of the studies show it to be predictive. It is even more predictive for auto policies than it is for home policies.
Credit scoring is merely an indicator of overall responsibility. If someone is irresponsible in one part of his life, he is likely to be irresponsible in other parts of his life. Looking at my loss ratios, customers with poor credit and prior claims have terrible loss experience – on average 300-500% higher than customers with clean records and higher credit. If I’m not making money on that slice of customers, I need to raise rates there. Simply and plainly put.
As I noted above, carriers are trying to segment such that they have the best prices for the best customers. They do that by lowering prices on the best consumers (those with lower loss ratios…ie. better credit and clear loss history) and raise rates on less desirable consumers (losses, poor credit).
No one reasonably thinks a consumer isn’t going to file a claim if they’re “rich”. But loss experience tells me someone with a 300 score is more likely (when priced as a class, not as an individual) to have a worse loss experience.
don’t truly think that a person w/a 300 score is more likely to have a worse accident… according to the supposed statistics, it means that person is more likely to file a claim… is it unfair?! i think so… because someone with great credit, could have a bad driving record and will likely be more of a candidate for a loss than someone with a great record and lower credit score… we can twist any set of numbers anyway we feel… credit is one of them… but how do you distinguish the proper credit score because someone is buying a house or getting a credit card and every time one of those hits that score, it deducts from the score… so whose winning? that statisticians are! and so are businesses who think that this model is appropriate…
I agree.. credit score has relationship to financial actions not driving behavior.. we have to find a similar score that makes more sense to Driving behavior and thereby link to insurance premia
You’re still missing the point. Credit is ONE factor, among dozens.
Statistically, someone with no prior claims and a high credit score, is married, has a college education or better, a white-collar job, owns a home, has higher limits (and a host of other variables if we’re talking about home or auto) is the best (i.e. most profitable) risk.
Statistically, the worst risk is someone with prior claims, low credit score, unmarried, rents, didn’t finish high school, is unemployed, has low limits (and a host of other variables if we’re talking about home or auto) is the worst (i.e. least profitable risk.
Most people fall in to the middle. Think about a married small business owner with a Masters degree, own a nice house, has high limits, but a bad year for the business hits his credit score. Is he bad risk? Of course not! But statistically he’s a slightly worse risk than the person I described above. He’s also a slightly worse risk if he doesn’t own a home. Or if he only graduated from high school. Or is single. Or any other factor that deviates from the statistically ideal customer.
If you want to get in to the nuts and bolts of credit modeling, each company has its own proprietary model of how it views credit. Your Beacon score (i.e. what shows up at freecreditreport.com) has absolutely nothing to do with your insurance score.
Carriers spend a lot of time pouring through their policies to find which customers yield the best loss results, and discount those customer’s premiums. Some carriers place a high emphasis on debt to credit ratios. Some put a lot of emphasis on the number of accounts. Some look at the number of accounts open. Each company is different because each company’s loss experience is different.
The whole point is that companies use ALL of this information to create groups of customers and set a price for that group based upon their profit/loss experience. The day low credit score customers yield the best profit results will be the day you see carriers lower their rates for customers with poor credit.
Let’s see now—what good was predictive modeling, credit scoring, algorithms, etc., insofar as the TWO BILLION DOLLAR hail storm that hit Dallas, Texas last week?? Did the hail stones only hit vehicles and other structures that had bad credit? When it comes to predictive modeling and credit scoring, nothing can out wit mother nature. Here’s a question for everyone—-Does credit scoring take natural disasters into consideration when developing a model? Mother nature and human behavior can NEVER be 100% predictive 100% of the time. The longer I’m in the insurance business (40+ years so far), the more I feel that our industry is moving away from good old fashioned “common sense”. God forbid we should asctually use our brains for anything!!
amen!
That’s why the models take into account geographic location and historical weather. You are talking about a model like it is only based on one factor and are missing the point because of it.
To assume the models are poorly done without any research into how the models work is a pretty poor intellectual assessment.
Credit score is only one of several factors that go into a good predictive model. So, yes, credit scores could not preditc weather patterns. However, there is other science that can predict weather patters. Its short sited to only focus on credit scores. If your predictive model only uses credit scores, then its no good. The whole point is a comprehensive model that takes many factors in mind.
WHERE WAS THIS SCIENCE FOR THE TWO BILLION DOLLAR HAIL STORM LAST WEEK?
….and insurance companies break their books of business into territories that are priced based upon group loss experience. So, after the tornado and hail storms in Dallas this year I’d expect their auto and property rates to go up. Whether they have good credit or poor credit.
The industry has changed. Carriers don’t have two prices any more – in fact, most companies have thousands of price points. Why? Because there’s fierce competitive to figure out, and acquire, the best (i.e. most profitable) customers.
Having a poor/unsophistocated/flawed rating model only means that a company will be adversely selected against, because their prices will be wrong (i.e. too low). As an agent, why would you want to do business with a carrier with inaccurate/inadequate pricing? It might feel good to offer a cheap price and win a customer – but you’re jeopardizing your book, and that carrier’s book, because the loss ratio impact will be devastating. That’s exactly why soft markets turn hard.
“Did hail stones only hit vehicles and other structures that had bad credit?”
More than likely, no. It is more likely that insureds of all credit levels (low, middle, and high) suffered losses. A predictive model based on this data would say credit is not meaningful or predictive of wind/hail losses (because all credit levels were hit). This model would validate what you regard as “common sense.” so I’m not sure what you are really arguing here.
Would you be willing to say the same about fire and theft losses? Is a property owner with lower credit more likely to have a fire? A more severe fire? Is a person with lower credit more likely to have theft losses?
Is a person without a garage more likely to have a wind/hail claim? Is a person that lives further from the fire station more likely to have a severe fire loss?
It sounds like we want to charge for these things but we don’t want to use a predictive model to do it. What might not be understood here is that a predictive model takes into all of these variables at once. If you want to do it the “old-fashioned” way and price for these variables independently (not simultaneously), your rates will be excessive because you will be double charging customers.
More than likely? PLEASE!!!!
More than likely,”No.” I think you missed the “no” part. My point is, credit isn’t very predictive of wind/hail losses and actuaries do understand (and adjust for) that. However, credit is predictive of other types of losses like theft. A person who is responsible and pays his/her bills on time is more likely to lock his/her doors and own a home in a neighborhood where crime is less likely to happen. I think there’s sense in that.
They call it predictive modeling however they do not file these plans as “pricing tools” with state regulators. Are they being compliant? They back into their schedule rating / experience rating after determining their predictive price. They say underwriters have say in final pricing however when you adjust the price different from the model it is viewed negative in your performance. Companies want computers pricing to cut out people (salary) and expense (benefits).
Yes, they are being compliant. The model does not alter or change filed rates. So, I fail to see how there is an issue with State Regulators? Also, I do not believe that carriers want to cut underwriters out… instead, I think they want to simply give underwriters another method to review a risk. Just because an underwriter now has a new tool to use, that does not mean the value an underwriter has. I submit to the Insurance Journal Readers that Predictive Modeling enhances underwriting. I do enjoy a creative underwriter,,, but I would rather have it based on Science rather than art.
Brilliant. Would you marry me.
Wow, I really hope that some of the anti-modeling comments here come from executives for my competitors. I suspect, however, that a lot of the ill will is coming from underwriters who have lost authority in the face of expanding pricing sophistication. The fact of the matter is that the marketplace will judge who is right or wrong. Right now, the guys using more sophisticated models appear to be winning.
As was pointed out before, if you believe that the modeling is hurting companies, then go to a company that does it the old fashioned way. If you’re right, then your old firm should be out of business in no time.
As an aside, I think those comments are from producers who don’t like talking credit scores with their customers and don’t have the relationship to pull off valued-based sales.
…and people with lousy credit.
I am a producer… and yet I do see the value of Modeling. You know why, because it helps me be more profitable with the companies I represent… hence better profit sharing. The point missed here, is that companies want to retain the good customers,,, and then get the right price for the higher hazard customers. Its my job to discuss the underwriting process with the insured’s. Its my job to point out areas of high hazard,,,, if my customer has bad credit, then its my job to point out why that hurts them from an insurance stand point. I am not passing judgement on them, I am just pointing out the statistics. If my customer is located in a high hail zone, then I talk to them about why their property rates are higher. I find it encouraging that all my peers do is sell on price. My customers love it when I take the time to explain things to them. All the model does is to help understand potential hazards, that gives me talking points with insured’s. And apparently, it allows me to be different from their current agent, who just seems to want to throw the carrier under the bus and blaim some model for all the price increases….
I hope you represent my company!
Can you please provide some facts for your position that guys using sophisticated models are winning other than the overhead war? Because when you bring in the geeks that no nothing about an industry and pay them exorbitant salaries for sitting in the bastion of academia their whole lives and have now ventured out with their oh so useful theories based upon some initial guess and that is exactly how these models are started with a guess of what might or might not happen and when proven wrong they guess the other way to appear that they were correct all along…
My dad always said, be nice to the geeks in your class, because one day they will be your boss.
I think your question is a false presumtion… you state that “geeks that no nothing about an industry…” First of all, you are implying that the people who build the model are not fit for the task. That is bogus… the presentations and articles I have read have all been from people who have insurance designations and are clearly experts. So I do not support your claim that the people who designed the model are not hip to the scene.
Finally, the people who made the models are not underwriters… they created a tool that underwriters use.
So they created a tool to be used by people in an industry that that they know nothing about? That is like explaining color to a blind person. If they have not worked in the field in which they are creating and the key word being creating, these tools, how do they they know that what they are providing is what is needed?
Just because some one reads a book about carpentry does not make them a carpenter and a academia type who studied theories their whole life are not in a position to create a tool to be used by underwriters say in the Professional Liability arena.
Underbroker – By once again making the claim that the tool is being created by “people in an industry that that they known nothing about,” I can see you missed the response to your initial statement…
Perhaps a different tact? What makes you so bold as to believe “They” know nothing about the industry?
It seems an interesting claim you’ve made, to be sure. Basically, what I hear you saying is: In my experience Actuaries know nothing about the industry; therefore, as a group Actuaries know anything about the industry. I find this interesting in that (while horribly flawed on a variety of levels) this is the very statistical reasoning you appear to be assailing.
I’ve known individuals in Actuarial who have worked in underwriting, distribution, and claims to flesh out their knowledge of the business. Further, at least in the companies I’ve seen, Actuaries work with others within the organization to ensure that the model is tuned based on “real world experience” (i.e., often annecdotal) of others in business roles.
Also, if you could find me a couple insurance companies that don’t use Actuarial models, then I’d be more than happy to look into whether they do better or worse over the longer (5-10 year min) term. My suspicion is you won’t find any of substantial size and longevity because not using stats as a tool is an antiquated methodology, but I’m perfectly willing to make the effort if you are.
Anyone ever heard of this lowering someone’s premium? This nothing more than a way to charge some people more money.
Plenty of people have lower premium because they are no longer being lumped with bad drivers, you guys just focus on those whose premiums go up.
I have been insured with a small mutual insurance company for a decade. They recently went from an outdated and traditional/basic rating plan to a multivariate plan. Customers were not forced to leave the legacy plan to the new tiers, but my auto premium went from $1200 year to $800 year because I have good credit and few violations and claims. Responsibility matters – and there are many ways to evaluate personal responsiblity. Creditworthiness is but one of them. Those who disagree just don’t want to see the facts.
Yes, I have seen the model lower premium. One thing the model can do is to identify customers that are low risk… all the comments assume that the model only shows Hi risk customers… NOT True. It can be the case that the model shows an underwriter that the risk is not as bad as once thought… so the pricing can be reduced. Trust me, the insurance companies want to retain the customers that the model shows to be low risk.
Yes but they cast such a narrow line for the ideal risk that it doesn’t take much to turn them into a gutter ball.
The question has been raised by others and I would like to chme in: Are there results showing that by predictive modeling, rates have gone done for a particular line of business?
I recall many years ago an underwriter who went out on a limb and agreed to insure a broadcasting tower in Oklahoma. Within a few months of the policy term, a tornado destroyed the tower. The underwriter chalked it up to experience and renewed the policy. In a rare situation a tornado came that second year of the policy and again destroyed the tower. The underwriter swore never again to insure broadcast towers and, so far as I know, never did. Naturally, there has not been a tornado affecting that tower in the past 25 years.
I cannot help but wonder how a predictive modeling algorithm would have handled the policy after the first year and the second year.
There are always anomalies and stories to tell. But that does not invalidate the statistics as a whole.
The model is not designed to predict the future for a specific tower. It is instead designed to predict trends, for similar towers in similar areas. In your specific case, loss history is only one such factor in the model. You presume the model would tell you to non-renew the account as your underwriter friend did. That is just an assumption on your part. Since there are so many factors, and since loss runs are just one factor,,, the model could still tell the underwriter to renew the tower.
Good article
Watch the movie Moneyball, and read this article.
http://www.towerswatson.com/assets/pdf/6703/Towers-Watson-Emphasis-The-Future-of-Predictive-Modeling.pdf
Yeah “Actuary Fellow” – he should just borrow some money from his dad and start an insurance company, great idea.
Predictive Modeling just means Insurance Companies don’t want to Underwrite anymore! Does anyone remember the saying “that insurance company will not write Pig Iron under water for fire only even with a Rust exclusion’? The companies are removing thier best underwiters…..We the Agents!
How does the existance of a predictive model remove agents? I am confused by your statement Stephen. Please explain how “companies are removing their best underwriters”
Yea right!!!!
Far to often I’ve been told “the model says X and there is nothing I can do about it and if you have to shop it, I understand but I must get what the model says”. PM is eliminating the art of underwriting for total reliance upon the science of underwriting. At some point, some VP will ask “why do we need underwriters?” Think Flo as the underwriter of the future.
I appreciate what you are saying Insurance Fox. I’m not mad about insurance using credit scoring, but I took a hit, and I think a lot of people did when that was used to rate drivers. And it frustrated me. I’ve had credit issues. Got into a little bit of trouble with some credit cards, but got on track and cleaned up my credit. And what bothered me was that while I was cleaning up my credit and paid every single insurance premium on time, I still took the hit. I’m a good driver with a good record, but really felt dupped by my car insurance company.
The GLM is not the answer it reminds me of the Mortgage Industry when they came up with the secret sauce to qualify people for mortgages. Guess what the mortgage industry almost destroyed the Banking Industry where is common sense and good understanding the proper premium for the exposure.
The fringe on the right claims there’s no such thing as climate change. It’s just some tree-hugger’s concocted theory to hurt oil companies. Right? But, aren’t insurance companies supposed to be right-leaning, ecology-bashing conservatives? So, are the insurance people who come up with these predictive formulas pinkos? Wait, I’m confused.
It is quite dangerous to over driven the predictive modeling. My current company is large P&C company. I haven’t heard any successful story to use the predictive modeling on commercial line. It continuely lost money on selling products by pricing based on predictive model. Predictive modeling is an excellent idea and does bring fresh air to insurance industry. But it is just methods and models, I think it is not a good way to 100% to rely them. Statistics used history data to build model. But the insurance industry is heavy regulated. The regulation enviroment changes in recent year and doesn’t reflect insde the data on which the model is built. I saw the fever on predictive modeling so far. It reminds me the IT and website booming in earlier of 2000s. Hard to know when it will end. But it will be told by profitability of company by using the models.
I am not an expert in insurance, but I worked for a long time in finance. The excitement around the predictive modeling rings alarm from my experience in the financial world
Combine predictive skills with underwriting and pricing rules, should provide insurer a more sophisticated results.This, however will largely depend on the modeler’s statistical skills and business acumen as well.