The Not Unreasonable Podcast

Steve Mildenhall on Insurance History Part 2

May 12, 2021 David Wright
The Not Unreasonable Podcast
Steve Mildenhall on Insurance History Part 2
Show Notes Transcript

Steve Mildenhall returns to talk insurance history! Int his follow up to our earlier episode we dig even deeper into data by different line of business and observe all kinds of interesting things:
- which classes have more premium than claims volatility
- statistical vs narrative analysis
- how well can claims be forecast with prices?
- how did the industry do?
- which lines are more volatile and why?

Check out the youtube video for the slides:
https://youtu.be/HfF53eXFSGQ

Twitter: @davecwright
Surprise, It's Insurance mailing list
Linkedin
Social Science of Insurance Essays

David Wright:

My guest today is Steve Mildenhall principal at convex risk former assistant professor of Actuarial Science at St. John's University and former CEO of analytics at Aon. Today, we are picking up the thread again, and are talking about the history of the macro environment of insurance. As part of our course, Steve is designing on pricing insurance risk. Steve, welcome back again.

Steve Mildenhall:

Thanks for having me.

David Wright:

So we're going to this is like a round two right. And what I think I'd like to do is just recap kind of what I took away, I mean, the two, I still thinking about it and talking to people about it, since we, since we talked, you have this dataset that you've put together, which goes back farther in time than any other publicly available data set that I've seen anyway. And I know you research that very carefully, it was an hour, or a year you add or something like that. And then the the part of that, that I found most profound, shocking, really, in like a good way was the eras of insurance. So I'm wondering if you just recap quickly, the first bunch of slides we got through and then we'll finish off the deck today.

Steve Mildenhall:

Sounds good. Okay. So the data set that is, I think, very interesting in terms of looking at macro insurance cycles is this premium to GDP ratio. This is generally available sort of from 1968, forward, and you see the three cycles, that was the liability cycle in the 70s, there was the LMS cycle in the 80s. And then there was the post World Trade Center cycle. And then we've had more recently very interesting sort of flat period here, where kind of nothing happened through most of the 2010s, relatively low sort of premium level relative to the last 50 years. And a little bit of a spike in the last couple of years. The most 2020 numbers is almost all driven by GDP down rather than premium up.

David Wright:

And so in case, in case actually, one thing I'll add here, Steve, in case, people are only listening on audio, because I got a little feedback that people were doing that the main thing you see here is, like you said flat, and net written premium to GDP ratio through about 10 years or so, which is, if you look back to the prior pieces of the graph is incredibly, I mean, unbelievably unusual. We have as amplitude of cycles, the premium spikes and drops. But then lately, just kind of been a barren desert of excitement.

Steve Mildenhall:

We've traded within sort of about a quarter of a percentage point. whereas previously, we've seen swings of over a percentage point in just a couple of

David Wright:

years of GDP in insurance premium in big numbers, big numbers,

Steve Mildenhall:

big, big secular shifts. And you know, we talked about that a lot of these cycles, you know, we if you have been in the industry, you can put your finger on what caused them lmx spiral being being a big one, most recently, Harvey, Irma, and Maria, so maybe kicking off the hog market, other things that you would have expected to cause a big blip, maybe not so much Katrina, hardly visible. Even Northridge and Andrew really hardly visible, although obviously very significant impacts for different lines of business. We saw how overtime, sort of since the Second World War, kind of through the mid 1990s, you had a secular increase in insurance penetration into the economy. And the sort of clear underwriting cycle that people love to talk about have periods of sort of relatively harder premium, relatively softer premium, but that since sort of, you know, sometime around 1990, really that pattern broke down completely. We've seen sort of a retrenchment in volume of premium as a percentage of the economy for various reasons, tax changes that happened in 1986, pollution exclusion claims made form and lower interest rates, probably most significant there, and really kind of a different pattern that's happened to sort of subsequent to 1990. And then we looked at, I'll just skip over these few slides. But we looked at also the surplus to GDP. And we looked at the relationship between premium to GDP and surplus to GDP to try and predict out okay, what is going to drive a hard market and the regression on the right here is trying to predict kind of next year's premium to GDP ratio, just knowing this year's surplus to GDP ratio. And this is based on that kind of a third phase of the history we've got from 1986 onwards, we see a pretty strong linear relationship here, between surplus to GDP and premium to GDP. So it's clearly it's expressing the idea that if we want to see premium early kind of jump up, that's probably got to be associated with some surplus side shock, taking surplus out of the system. And you know, one of the explanations for the flat premium rates that we've seen more recently, is that actually the industry has got much better at getting surplus into the industry, you can set up a new co you've got all that alternative capital, the bank's cap capacity these days, that can all be done much more quickly and effectively. Maybe then it's been done in the past. And so maybe the days of these jobs, swings are are gone, is certainly an argument that could be made,

David Wright:

if I could just kind of just emphasize something amazing here again, for the if somebody is not looking at this, it's pretty easy to visualize, you see a line from the upper left to lower right hand corner. And then that shows that the higher the premium to GDP ratio is the lower the surplus, direct labor surplus to GDP ratio is the prior year surplus, did you push it? So it shows that like, if you have a decrease in surplus to GDP ratio, the premium goes up, which is pretty self evident, I think, to those of us the insurance industry, but your R squared of point eight six is incredible. And how stable this relationship has been over a long period of time is, I mean, creepy. Like it's unbeliev.

Steve Mildenhall:

That's 35 years, right? It's a generation of underwriters, and actually, so

David Wright:

yeah. So should we pick up the story? Again, Steven, he wants to say on a recap, or we can go to individual lines of business, we can break this down.

Steve Mildenhall:

Right, let's look at individual lines. And really, yo. So why do we have all this surplus? Here? We have it because insurance is a risky business, right? And so we want to get our arms around what is the risk? And how does that vary by by line of business. So the first place that you might think to look at that is let's look at direct loss ratios over time. And what I've got here that the chart if for those just listening is if there's 12 plots, that show the major kind of us statutory lines of business since 1992. And this is where to get I wanted to get the 92, you can get back the 96 reasonably easily. But I wanted 92 because I wanted Andrew and Northridge in there, plus I started working in 92. So attach the 92. So this is it was getting this data and it took me several hours per per year to type in manually of some scans. So this is counting the year numbers. It's from the insurance expense exhibit on a on a direct basis. And I've got lines I've got commercial multi peril commercial auto commercial property, financial guaranty, which has all the financial lines plus mortgage guarantee in their homeowners, inland marine liability med now, other commercial, which is things like sort of aviation, all those sort of smaller commercial line covers, personal auto, welcome, and total. And there's a lot of information on these plugs. So let's just look at one of them. And I'll just talk you through this. So if we look at the top left hand corner, there is CMP. So the solid wiggly line, that is your direct loss ratio over time, so you can see two CMP that's varying between a high of about sort of 95 ish that we hit in in Andrew and that we hit in a lump down to a low in the sort of 50 ish percent range that we saw in 2004 2000.

David Wright:

I find that very surprising. I mean, 95 loss ratio, and Andrew again, in 2001, for very different reasons. Or, you know, the 2001 is a mix of reasons, I suppose. But I would have thought that, you know, even if you're narrowing because of a curse, multi peril is I mean, it's substantially property, I don't know, that's dominated being, you know, some, you probably couldn't maybe you know, what this breakdown is between property and liability, and it's about

Steve Mildenhall:

50. Okay, it's about 5050. Between,

David Wright:

right. So, you know, a lot of cat exposure. It's just surprising to me that even within that very minimal amount of bundling, you have a lot of stability, surprising.

Steve Mildenhall:

Okay, so let's, let's look at this, I want to just sort of draw your attention to the various statistics we got, because I was trying to think, alright, so how do we measure risk? Right? Well, the first place you look to measure risk is you just look at the standard deviation, right? So that's what's reported in the title there, it says SD, in this case, it's 11.1%. And there's two blue lines, you see that sort of go from about sort of 50 to plus 90 to blue, the horizontal lines, they are plus or minus the confidence intervals that you would expect the loss ratio to fall within if the loss ratios were actually just normally distributed. Right. So there, I think there's 22 years of data here. So I'm just doing sort of fee inverse, you know, 23 over 22, which is about 1.7 standard deviations from the mean. And you can actually see if you look across most of the time, that is actually a pretty good estimate of the range of loss ratios. Amazing. Okay. Well, okay, standard deviation CMP for about 11%. And if you look across, you know, commercial auto liability you don't have cats is 8.2%. So obviously much lower commercial property. On the other hand, you can see very volatile 32% right. So that's not a bad measure of risk that you

David Wright:

can I can I just hope I'm saying amazing. I'm having these like on awesome cop, you know, like noises I'm making here. Whereas if you were just, you know, if you're just a statistician walking up and he said, Well, you know, last 25 years are pretty well described by the normal distribution and be like nothing to see here and off you go be like, but you know, I've been trained so much, right, and the actuarial profession and say, like, you know, normal distribution, it doesn't work. And we have all these other traditions, and we learn all this crazy transformations we do, you know, in the in the curriculum, because of the long tail nature of the business. So the fact that you're, and again, we're talking about loss ratios here. So it's not quite same as losses, but in the fact that you're showing that it actually works. Okay, for some subsets of the business is interesting.

Steve Mildenhall:

Yeah, and I guess, so, you know, you have discussions with folks, when you start to get into measuring risk. And exactly as you say, there's this feeling that we don't like standard deviation, we don't like normal. And there's a feeling that maybe we actually want to measure some sort of PML like number, but not a one in 100, or one and 250, we want to think about maybe a one in 10, or a one in 20 year kind of statistic. And what I find interesting out of this is that this is saying, so we've got 22 years of data here. So the worst measure is, you know, that's an estimate of you sort of one in 22 result. And we're saying Actually, that's pretty well captured by a normal model, that the standard deviation was 1.7 standard deviations from the mean. So yeah, I agree. I think that the old guys is statistics, they know what they're doing with standard deviations very, very good. It is it's,

David Wright:

I find that if I could just like wind for a second, it is hard to communicate to non practitioner. So it mean, standard deviation, what is it? You know, it would have the units of that something strange. It's like the root of a squared quantity. It's like, it's a bizarre, it's a it's a hard, it's a hard tool for communication. I can attach their definition a little bit here.

Steve Mildenhall:

I agree with that. Yes, there's something more tangible about saying, yo, the worst year in the last 20 has been such and such. So yeah, I agree with that. And one of the big complaints that you hear about standard deviation is that, well, there's a difference between the upside and the downside, right? If I have a big applause, I care about that a lot. But if I have a big down loss, you know, that's actually a great year. And I shouldn't treat those two things symmetrically. So I've addressed that somewhat the the number in parentheses after the standard deviation, so for ccmp, they're the point oh, six, six, that's actually the semi deviation. So that's just the upside, you're the worst loss ratio deviation that you're getting out of this. And if you look through, you see, you know, that number is clearly very highly correlated. deviation. So it is not clear that you are actually getting a whole lot out of, you know, looking at the semi deviation, as opposed to the

David Wright:

interesting.

Steve Mildenhall:

So the next thing we've got here that we could worry about is, well, what's the correlation of this line to the total? So that's shown with the CO off so for CMP that's it's very high, actually, 91 and a half percent is one of the most highly correlated lines, I think, that we've got here. Whereas, you know, commercial Auto 41% commercial property 72% homeowners 70%. So what you're seeing consistently across these lines, is that something I think the whole actuaries are aware of is correlation is real. And it matters. And you're seeing, you know, pretty substantial correlations to the total, across most of these most

David Wright:

that's interesting. So if, if the commercial multi peril is most highly correlated, that tells me Of course, the other ones are less correlated, but it would you say it was it was point eight or something like that for CMP.

Steve Mildenhall:

Point 915 is the COR on them.

David Wright:

Right. Okay. Yep. Point. So that's quite, that's quite big. But the and none of them are I mean, they're all they're all compositions of it is a composition of all the other ones, right? So they're all pieces of it, which one's the least that is it worker's comp at point six,

Steve Mildenhall:

financial guarantee, because it has,

David Wright:

right, so it's Flat, flat, flat, nothing, nothing, nothing kablammo for

Steve Mildenhall:

but then interestingly, other than see if it's the other two that are low correlation, and then the auto lines, right, or no, personal Auto 48%, commercial Auto 41%. And I should see here, the thin gray line that's in each plot, that is the total. So you can sort of eyeball the correlation for that. And, and if you look across, actually CMP is essentially stunningly currently. And I think what's going on is in part, this golf property, and it's got liability, right. So it gets a spike, where you have a cat, which the industry gets the spike, because it has, you know, homeowners and other commercial property, but it also is influenced by the liability side. So I think it probably makes sense that it's got that highest correlation that we're seeing that and you can see it aligns just looking, right, the blue line and the green line in the top left are clearly very highly correlated, right, the spikes there, I think every year moves together in a couple of years, maybe where it doesn't put it.

David Wright:

Yeah, and you know, a few other thoughts. So one, the composition, you know, kind of point there, but also that there's this other idea which Is that capacity is correlated as well. So you're kind of taking from one side of the house to pay another right? capital is shared across these businesses, because they're all insurance companies with a portfolio as well. So, you know, you get to imagine that a property right or to some extent will be constrained in a very hard casualty market. Because all the all the money's in us over there, I think.

Steve Mildenhall:

Alright, so so let's continue to think about the risks, though of this. Right. And you could ask, well, is standard deviation of loss ratio in that way that I've computed there is just, you know, across the years, is that actually a good measure of the risk of the line. And something that springs to mind here is it I remember looking at the schedule p for a very large, personal owner, right. And, in fact, the largest personal owner right away. And, you know, many years ago, they had a fairly substantial increase in their loss ratio, because the market had been, there have been a lot of competition in the market. And I think they'd held the line on rate, and they decided that they were done with losing market share. And they took some fairly aggressive rate actions to sort of address that. And so their loss ratio jumped up. But if you looked at the schedule p, they knew exactly what they were doing, because they nailed the, you know, there was no development after 12 months, right? They booked the number after 12 months, and it went up. So it was a sort of explainable increase. So really, what we'd like to look at here is, we want to measure of the unexpected. loss ratio change, right? If I know I take a lot of right or I lower rates, right, I know what's going to happen to the loss ratio. The question really is after I've adjusted for all of that, what's that residual volatility? Now, that's obviously a hard thing at this macro level, when we're just looking at industry trends, that's a hard thing to pick up. I don't you know, I could try and do some fancy Yo, on leveling and trending process and what have you, but I don't really have the information available to me to run analysis like that. So instead, what I what I did was I just looked at, let's just fit an autoregressive model for the loss ratio, right, that probably says like the weather, the best estimate of loss ratio, a high level for next year is going to just be I'll take my loss ratio for this year. And yeah, if ideally, we would adjust the rate and trend but you know, I don't know that. So let's just do an AR one model. So on each of the plots, you see there's a color dotted line as well. And that's the AR one fit, okay. And so you look at commercial auto, for example, the second one or the orange one, you can see there's a quite a lot of volatility in that loss ratio, you've got a big cycle effect, right? 8.2% standard deviation, but it's sort of predictable in the sense it goes up and then it goes down, and then it goes up again, you're a one fit is actually pretty good. And the residual error of that auto regressive fit is only 3.8%. So that's the RS is zero, that's shown in the second line of the title there shows you sort of the unpredictable volatility for the line, just based on the simple AR one model. And then also on the second line, I report the R squared. So you can see, for commercial auto, it's a it's 80% r squared. So that's actually a pretty good fit. So that would argue that your commercial model maybe is not really as volatile, as the standard deviation would lead you to believe, you know, once you sort of take out what's known, you get a much lower volatility. Now compare that to the next one over commercial property. Okay, it's very interesting. So that there is a dotted line fit there for commercial property, but you can see it lies almost entirely underneath the line, right? That the R squared is basically zero because the coefficient that got applies, the last year's loss ratio is basically zero. Okay. So in that case, you're saying there's sort of unpredictable uncertainty of commercial property is 32.9. It's actually slightly higher than the Unreal, unconditional standard deviation, that there's a degrees of freedom adjustment there, but it's accounting for that you would have thought it couldn't be higher, but because the degrees of freedom, it can come out slightly higher. So this is saying, this is a completely different line of business, right? This is a line of business where, as we know, it's cat driven. So it's all about sort of innovation. The innovation in the sense of the unknown thing that happens during the year is driving the volatility and very different picture, then you've got with interesting

David Wright:

and so like, I think of it kind of endogenous and exogenous, right? So like endogenous means something within the system that affects the system. So for commercial auto, and we're going to come to this I think in a moment see, so I don't want to steal too much thunder, but probably mostly a pricing problem, right? So it's like it's within the insurance system. And looking forward to talking about that. Whereas with commercial property, it's like you're getting hit by Meteor strikes, which are hurricanes and like, and you don't see them coming, and they hit you. And maybe there's some predictability there. Maybe someday, Steve, we'll do a deep dive on to do it. How on earth? do you predict hurricanes? I don't know, you can. But this data says that the insurance industry should camp? Because we don't, you know, it's just, there's just no, well, at least, the loss ratio doesn't predict the subsequent loss ratio. And so,

Steve Mildenhall:

right. And I think it's very interesting, you know, it's, we picked commercial auto and commercial property, because they're next to one another, but look at homeowners on the second row there, inland Marine, commercial property, they're all exactly the same, right? The AR, one model does nothing, the residual volatility is about the same or slightly higher than the underlying standard deviation. So we've got those cat lines, which I like, love the language, right? exogenous shocks drive, the volatility. And then you've got, you know, the other the remaining lines, where it may be a combination of loss volatility, and, but also internal kind of pricing cycle volatility as well, driving the overall risk, in the results.

David Wright:

big difference. Now, worker's comp, okay, that one's pretty good, too, isn't it there? So that when it follows the line, as well, right?

Steve Mildenhall:

Well, workers comp where you've got. So actually, your substantial so I can three cycles? Yeah. How many? One and a half, one and a half full cycles, I guess. Overall standard deviation 10.3%, residual standard deviation 6%. So that's suggesting that yield is a much lower sort of unexplained volatility than we see in the unconditional and

David Wright:

that one has the visit the lowest residual.

Steve Mildenhall:

Now, first of all, 3% 3.6% personal auto's where it's going to be

David Wright:

sure, yeah, and that one's me lowest volatility and all kinds of ways cuz even look at the, I mean, it's you gotta watch out for this, like, the scale of the y axis and all of your graphs is different in zoom Britain.

Steve Mildenhall:

Great, great, great points up. So we do have a little bit of light. Okay, next slide. Because on the next slide, I actually show this is everything on the same scale. Okay. So this brings out the Yeah, we were talking about volatility and commercial auto, but it's a fraction of the volatility and commercial interesting that you're seeing here. You know, liability, pretty sporty med now, you know, but there's some pretty serious volatility going on their personal auto here, you know, really tight ranges. And you can see the plus or minus kind of normal bands very clearly.

David Wright:

Yeah, that's pretty, pretty neat. I wonder, and this is not something for this for this presentation is deep, but I wonder what the variability is within the groups. So like, for personal auto writers like the most and you know, the distribution of performance, right? I wonder if you know, that is as volatile or to what degree that's correlated to overall volatility.

Steve Mildenhall:

that's a that's a whole nother topic, I'd love to discuss it with

David Wright:

all right now, a future episode.

Steve Mildenhall:

Let's Let's stay on topic. There's an elephant in the room here that we haven't discussed. And that is the graph and the top right, which is financial guarantee line. And to the spirit of lying with statistics, I just want to go to the next slide. That was very pleased, I could figure out how to make this that shows what it's like, if you don't cap the results, you know, at the at the same point. So when we talk about volatility here, and you talk about a sharp loss, right, this one is literally off the

David Wright:

breaks through the top of the graph.

Steve Mildenhall:

100%, right, it explodes out of the top of the chart. And so your this is your ultimate example, in a sort of Nassim Taleb thick tail distribution. And be super careful about Yeah, because you've picked up those pennies in front of the steamroller for the last 30 years, doesn't mean to say that you're necessarily going to continue to be able to do I was particularly you know, if the environment changes and credit quality of mortgages, genuine, we all know what happened in 2008. But I think that's it is a very nice sort of graphical illustration of just how bad that got. Plus, this isn't just mortgage on its own just mortgage on its own would look even worse. This is blended with some of the

David Wright:

Yeah, that's incredible. I mean, you know, you're just describe it, your red line of the loss ratio is up to the top of the graph and then out again, and entire chart height above the graph. That's just pretty cool. Nicely done. Anything else to say about this slide? I wonder like, you know, it seems like

Steve Mildenhall:

well, you can you can see very clearly the bands the plus minus bands for the normal approximate You know, that does track for certainly all the liability lines there there, you know, this thing kind of nicely within that. But I think, at this point where I think we said everything that is easy to say about loss ratio, and we I'm sure you're already thinking this, hey, Alice ratio has got two components, right, it's got a premium component, and it's got a loss component. And if we really want to understand the volatility, what we should probably do is analyze them separately rather than pulling them together. So let's look at how that looks here on on the next slide. So what we've got here, let me just talk through this. It's the same 12 lines of business. And then for each chart, we've got kind of two groups of lines, the blue shows the premium over time, and then the green shows the losses, okay. And then I've got it, going back to our idea of, we were really interested in sort of the unexpected component of next year's premium and next year's loss. And if you've got a line, like, you know, look at homeowners here, right, the premium for owners, it's just been a straight line, I mean, the, you know, very, very tight, the, the R squared on that regression is 98.9%, you know, straight line increase in premium. So there's really been very little uncertainty about what's happened to the premium for homeowners. So I'm fitting a, a, I fit both a linear regression and an auto regressive model to the premium and the loss for each of these charts. And then what I'm showing on the chart is, if the R squared is greater than I think I used 85%. For the linear regression, I show the linear fare. Otherwise, they show the batter of the auto regressive and under linear fit, right, so you can see here for commercial auto, for example, we're picking up the auto regressive model for both of them because they're, they're not straight lines, commercial property is linear for both of them, etc, etc. Mostly, it's, it's, it's turns out to be linear for premium and an auto regressive model. Interesting.

David Wright:

An exception is personal auto, which diverges from commercial auto on both dimensions, commercial auto is auto regressive, and both and personal auto is linear on both. Any comments there?

Steve Mildenhall:

And, yeah, so let's look at personal auto is a really interesting example. Um, so remember, from a loss ratio perspective, personal was the tightest, right, I think we showed a 3% residual volatility for personal during the loss ratio. Where is the uncertainty live for personal auto, while for the premium side that we're showing the RSC. So it's, it's the residual squared error for the regression. And then I just normalize it by dividing by the average premium, right, it's not perfect measurement, you know, the reasonable measure, so scale, that for the premium, the uncertainty is 7%. For the losses, it's six into three quarters percent. So you could make a firm case that there is a greater degree of the uncertainty in the results or person lotto comes about because of sort of the premium cycle. And the premium movements, then comes about because of the losses, and you can see that pretty clearly from the lines, right? I mean, it's personal, we wouldn't, we would expect premium to pretty much grow at a straight line, the loss is much more costly to grow as a straight line than the premium does. Right? So is that is a sort of ultimate endogenous risk coming,

David Wright:

you know, it occurs to me on this is like, this. This is almost it's an interesting example of, um, distinct distinguishing between cost statistical analyses and narrative analyses. Because I look at that, and I see three, clearly three different worlds, right, you kind of have the world before 2000, which is, which is like a straight line, I mean, that I don't know what the R squared would be on that. But it's something like 99 point something percent. And then you have an episode where something happened, right, where the losses pumped up a little bit, and then the market overreacted with premium. Right. And then the losses didn't actually it seems like they didn't the market didn't, sort of the losses didn't materialize the way you say like the premium increases were anticipating as my interpretation here. And so you had this like flat Suddenly, the premium stops increasing, while the losses keep coming and they catch up and you have a loss ratio compression, or rather like the lines compressive loss ratio goes up. And then you have this another overreaction to premium. So like, it seems to me that there's like a pretty clear cycle here. We're zoomed in I suspect because it's personal auto so everything's, you know, everything's tighter and personal auto. But to me, this is like a fairly classic cycle of overreach actually, amongst personal auto writers and like the latter half of the 2000s. And then, you know, all the hand wringing in the early 20 teens, you know, so I want you to react to that. And I will add one more thing, which is to say like, you know, I remember the early 20 teens, and it was all about, oh my god distracted driving, it's the smartphones. Whereas I would say this analysis says actually, the loss ratios are entirely because of the softening market.

Steve Mildenhall:

So, I love the statistical analysis versus narrative analysis. And, obviously, we do need to be careful with narrative analysis, because you can, you know, you have confirmation bias, and you just sort of fit the facts. But I think your analysis was was was spot on. And I think it's interesting to compare personal auto with commercial auto, because commercial auto sort of does the same, you know, you can see the same trends, they're just much more exaggerated, in commercial auto. So you see, you know, going, let's look at let's divide into the three periods you divided into right, you've got prior to 2000. Well, I don't know what the trend is there on the losses. But it's clearly a heck of a lot more than the trend on the premium. So loss ratio, compression, dam breaks. And interestingly, the dam broke for commercial auto before 911, it broke in late 2000, early 2000, when people were starting to get substantial on commercial auto. So you see inflection point in the premium right premium jumps up. And then you see losses kind of flatline. Now, that's an interesting phenomenon, because what's going on there is, that's the impact of underwriting is a couple of things. One is some of the losses in the early very early 2000s, were actually development from prior years, because this remember is IWA information can be a basis. But then you have the ability in commercial auto to you know, you can raise deductibles lower limits, and you try to cap out the lawsuit. So all of a sudden, premium starts growing sort of in the way it had been suggested from surviving on 98 to 2001. got very steep slope on it. But at the same time, you're taking underwriting actions, you cut the loss out system, and you get this sort of massive spread. And then, you know, this, I think we could probably do a whole thing on on auto and sort of what's driving, because you see on the personal line side, right? Where you wouldn't really have had those coverage differences happening nearly the same way. And people buy 100 302 5500, they are they're not buying to 510 million dollar limits. But you still see that sort of softening out, you still, you know, the lines that track together quite closely. But we booth a much smaller effect. But you see, for commercial auto, you've still got volatility on the premium is more than volatility on the last right. So in a very real sense. It's sort of self inflicted volatility for the industry, from how we price it, and how we how we think about it, and then you also see a much bigger effect. So eight 910, that the dip there for the global financial crisis and the economic impact, which makes sense, right, commercial is probably going to be more impacted by that. And then subsequently, we've got, as you say, the will the distracted driver or what have you. But we've also got the beginning of, you know, Uber and delivery and a culture maybe driving more commercial

David Wright:

here, Amazon. Right, right. And other delivery services, the technology, boom, a couple other thoughts here. So one of them is we have to of course, acknowledge that the early 2000s was a general hard market. And so maybe there's a capital constraint that's kind of creeping in here, which is going to enhance your premium requirements. So the required return just goes up, perhaps, and probably that's gonna be directionally what's going to happen there. So you're sort of in retrospect, anticipate expecting, I would be surprised if we didn't kind of see a bit of an overshoot on these because the cost of capital would have gone up. Another thought here is on the commercial auto one. If you were to fit like your, if you were to fit a linear regression to the 92 through 2001. It'll be pretty good fit and losses, right? And you can almost think like that the commercial auto actuaries were probably thinking the same thing, thinking, Well, hey, if we just fitted less than the linear regression here, I mean, that we can just extrapolate and see 2003, four and five. And you could imagine that the the premium line, there would have been a pretty good bet. If it was in fact, a linear extrapolation of the late 90s. Last trends, right? Good. Yeah.

Steve Mildenhall:

It looks like the pricing actually is talking to the underwriters, right? They were working on exactly what you said 98 to 2001, I draw a line is still enough to get to decent loss ratio, I get the premium that's shown there. But the underwriters, were taking all these other underwriting actions that improve the book and cut the cut the losses, and you see that exact same pattern liability here, you know, pretty much the same thing. It's shifted over a bit. I think in liability, you've got a lot more development flowing through than you had on commercial property. But you see the same pattern going on there. Same effect, premium sort of tracking what you'd have expected from losses, underwriting actions, taking loss out of the system, and you get that sort of second dip. Yeah, wonderful financial crisis. And welcome to Friday, you get the same

David Wright:

one of the thought that I just got to wonder about a rhetorical question is To what degree do we can we predict what the impact of our actions will be as a as an underwriting team and pricing team, right. So I almost feel like a lot of, you know, a lot of underwriting strategy is, is, let's get kind of directionally correct, just kind of hope it's gonna work out pretty well, because it's very hard to be precise about, you know, he raised the deductibles by a certain amount, and I'm sure there's gonna be someone else's, you can do some things to this, but though the behavioral change that's going to manifest after you do that, you just don't know what the customer is gonna do. Right. So, and that's kind of this sort of this thing that I think about a lot, which is, insurance is a product that is kind of CO created between the insurance company and the customer, because you can have good and bad customers, and they will have very different claims activity, given all the other you know, everything else is the same two people have a different because they just have different. Some people want to file insurance claims, and some people don't. Right. And for a certain certain category of percentage of claims, you're going to have people who are good insurance, and in the sense that they won't, you know, and you'll charge them commensurately. Right. And so some distribution systems exist, if only to select people within a certain kind of behavioral profile, then you can set the rate, you know, to be exactly what they were, you can make money on doing that. So all that to say, like, I wonder how much you can know about, you know, we got to make some big changes to our portfolio that's disruptive. I, you know, I would think that I'm not sure what the hell we're gonna have in our hands two years from now, after we change all of our underwriting parameters. And you know, you probably fired some agents and all this kind of stuff, like, once you mixing stuff up, the prior data set doesn't work anymore, and you just gonna have to hope it works out. You know, I even do it. As long as me Steve, what do you think about that question?

Steve Mildenhall:

So, I think a lot of people This quote is attributed to a lot of people, including Yogi Berra, and somebody once said that, you know, it's difficult, especially about the future. And and I, you know, I think, when you look at this prediction is especially difficult when you have a turning point, right? Pretty much all predictions that people do. Yeah, it's just draw that line, right? You've got a few dots, you've got two dots, you got a line, you draw it through extrapolating. And I would say, you know, if you look at this, right, we're not good at you look at when the dips come in, where the turning point comes, the premium change always happens after the death rate, or it doesn't seem to be sort of leading indicator. That said, in the more recent years, we do seem to be doing for commercial auto for personal liability, doing a reasonable job, they're sort of keeping those two lines moving in the same direction and keeping the distance between them sort of reasonably constant. Yeah, the one that draws your eye is work comp, where you know, any sort of different than the other, the other guys on the on the chart right there. I'm not, not an expert on comps, I'm not quite sure what's going on there. But that that one seems to be. And we saw this on the premium, I don't know if you remember the premium volume, right that the premium volume for comp is, is not nearly kept up with GDP in the same way it has for other lines, it tracks much more closely with actually employment,

David Wright:

I mean, come in to just put it into words like the premium is ridiculously more volatile, in the sense of like, on the upside, you know, increasing and then overshooting this gap. And I remember this old chart that I first saw from guy carpenters group, you know, the one where it shows like the progression of loss ratios in the in the workers comp cycle, phenomenal, wonderful piece of visualization. And I've used that myself to explain to people here's what this unrecycled means. It's, you know, it's very long tail comp, you know, it's just very hard to know what's you know, where it's going to go? I suppose I've heard some people who do a long time worker's comp reinsurance underwriter saying, What the hell happened in the early 20 teens? Because like the losses decision show up something. It was like a blonde, like something weird happened and it is a broke all they're kind of like, you know, intuitive models. Super strange.

Steve Mildenhall:

Yeah, the I mean, the quantification of that is that the residual volatility on premium is 11%. And the residual volatility on losses is 8%. So, you know, almost 50% higher volatility from the premium side, then you've got on the low side, and that's the I think, the biggest, the biggest swing between these and just, you know, let's just go back and sort of connect the thoughts on the premium side. What we see there look at homeowners look at commercial property to some extent look at CMP mn ream, you see much more predictable premium your varamaan is yo ha the residual volatility on previous only 5% but on the loss is 21% right so so clearly that's all about care. And I think over that entire time it's basically been people chasing rate because you got behind yourself with the cat states and you got to get the hurricane rates up up to snuff and you know yet 2004 2005 lower rate rate rate And then what happened after that was, well, the Midwest fell apart and you had, you know, hellacious severe convective storm and what have you. So that pushed a lot of rate. And then recently, obviously, we've had the issues on the West Coast with wildfires and what have you. And we've also had more hurricane activities, those that have really been pushing homeowners. commercial property, you see had a more of a cycle and had more of a dip in 2008 sort of overshot maybe a little more. So it's got a bit of volatility around his premium line. 12% Yeah, the last volatility Yo, for 39%. So that's still very much a

David Wright:

in what an incredible years of, of opulence, and no man, was it raining gold, and most commercial property underwriters in 2006. And 20. You know, 2016, I suppose? Holy cow. Yeah. What a gap.

Steve Mildenhall:

Although if they were international writers, you know, there was other stuff.

David Wright:

Yeah. diversification, I mean, you want to just be putting all your money on Florida for a period. In retrospect, of course, I'm sure Steve, you have lots of friends and I that made tons and tons of money, they have nice boats, as a result, good for them.

Steve Mildenhall:

And then I don't, I'm not qualified to comment about it. But then we have med mal is also living a little world of its own that got itself know, incredibly upside down nearly 2000s. And I know a lot of sort of tort reform and legal reform went on. And I think y'all, I would say that that picture suggests a lot of a lot of caution. In underwriters in terms of Yo, do, we really believe these remediations are going to have the effects that are desired. And, you know, we're seeing through several years of both declining losses and sort of shrinking gap shrinking sort of profitability.

David Wright:

So funny about this, too, as I kind of like cast my over these over these graphs. And we're gonna make sure we put a link to this presentation, of course, in the in the episode. But we're seeing almost like, if I can use fancy words, that exposition of upside volatility, right, so it's like, it's almost as though what we see certainly in the history of the last 20 years are examples of insurance industry, getting it wrong in a way that's benefited them, you know, wherever he has these loss ratio gaps, just widening. Of course, there's survivorship bias, there's all sorts of things and not every company makes it. And so maybe it's just the smart ones who remain that are able to kind of reap the rewards. And that's kind of how the heart market works. But, you know, I see a lot of like, you know, surprisingly large profit margins are emerging at when when we think about one side of volatility to the point you made earlier about standard deviation, right? We tend to think, Oh, it's all downside risk. You know, we're obsessed with the downside, it's a business that's explicitly exists to risk manage. But, you know, there's there's also upside volatility once in a while, at least on an intraday basis.

Steve Mildenhall:

That is true. We're all taught that a rate is a, you know, prospective estimate. But in reality, the Yeah, right, john retro rating plan, right. That is, I think, the surest indicator that rates are going to harden and the people are going to grow, is the management stops talking about growing, and they start talking about the need to rebuild the balance sheet. And at that point, they will grow if you go back and you listen, yeah, what were people saying, you know, one or two or three, that time a period. That's what the conversation was about, whereas, you know, late 90s, and maybe sort of, you know, back into the sort of teams, it was all about grow, grow, grow. The industry is sort of, it's just going to grow with the economy, right? That's kind of what we saw when we are starting slides on premium to GDP, you can't magically make people aren't going to demand more insurance than they need. And so it tends to be cannibalization, growth through cannibalization rather than sort of new finding new new markets. And so everybody talks about growing How do they do that? Well, they still business

David Wright:

and I think that too, there's there's a cultural analysis of insurance there's always going to fascinated me where the think the very first podcast episode I did was with Rob Johnson, who was a mentor of mine actuary, and he had a story that he would tell that I've heard from him share too many times tell it and I still think about it, where he was actually an underwriting manager running a portfolio of personal lines business or I think small commercial in Australia and is wholly traded company and every year they go to the conference, and then they would the fruit he said the first two days or the first day let's call it day and a half of the conference was the underwriting management and the CEO berating the team's for about not performing well enough or loss ratio is too high and and we need to be more conservative. We need to make you know, we need to like you know, hunker down and you know, We need to, I don't know, for moral fortitude of restraint, you know, all this kind of interesting that you know what I mean, right? Like you can see in truth, he will tell it talking to each other, they're always talking about how we know we need to have more, more restraint and raise rates, and if only what you're talking to mark it up. And he said in the last guy, they came up and presented was the finance person who would show a very steady, neat progression of net assets increasing every year 9% 10% 8%. And thinking like, the finance, so one interpretation is these hypocrites, their interpretation is it's working click, we kind of have to have this sort of culture of restraint. And, you know, self flagellation, or something like that. And artists hold the discipline to be able to get that. So if you get a little too happy about it, it's gonna get taken away.

Steve Mildenhall:

So that reminds me of somebody pointed out to me once that if you look at the sort of industry history, how many years has the industry actually lost money in total surplus decline? Very few. And when it does decline here, it doesn't decline very much. Let's compare that to banking. Right. And when you went through 2008, what was the mortality rate amongst major investment banks? It was 30%. Wasn't it three or five blue whale? What? one blow up to go purchase that have? You know, so? Yeah, we have we I think the industry is so aware of risk, and it's a debate I've had with various people that you know, is, is cap really risky, because it doesn't, you know, you look at things like Ms puts out that survey of what causes companies to become impaired. And cap doesn't really sort of figure on the list, and it's easy to think, Oh, well, maybe we're over overplaying this. And in fact, cat isn't really, I don't think that's correct. I think everybody is so aware that cat could kill you. Absolutely. That they manage it extremely carefully. And as a result, we get, you know, a reasonably good sort of set of set of amazing,

David Wright:

so I think that probably it's your last slides at the time, any closing comments on this? We're going to do more of these. And we haven't covered No, you'd like to say great