In The Public Eye: Catastrophe Modeling - Uses and Misconceptions

Intro (00:00):
Welcome to the Alliant In The Public Eye podcast, a show dedicated to exploring risk management topics and challenges faced by today's public sector leaders. Here are your hosts Carlene Patterson and Justin Swarbrick.

Carleen Patterson (00:18):
Welcome back everyone to another episode of In The Public Eye. We have spent the last couple of podcasts talking about the weather and how it has impacted the insurance market. One element we push here at Alliant is looking at catastrophe modeling, making sure our clients understand the schedule of values, where your values are, how they're protected and where you have large concentrations. So today, Justin and I have invited one of our in-house CAT modeling analysts to talk about CAT modeling uses and misconceptions. Ali, we're happy to have you join us today. And before we get started, can you tell us a little bit about your background and your role here at Alliant?

Ali Syed (01:00):
All right. So, I am based out of New York VP of CAT modeling. And my background has been in the analysis or analytics arena for the last 10 years. It used to be insurance pricing models and loss models. And then about five years ago, I jumped into CAT modeling. At my previous employer USSA. And then, right around the end of 2019 or actually early 2020, I joined Alliant and for Alliant, I handle all internal CAT modeling needs.

Carleen Patterson (01:39):
All right. So, when we're talking about catastrophe modeling, there are a lot of acronyms that we throw around: AAI, RMS, AIR, thousand-year flood, hundred-year flood, CAT. So, just what exactly is catastrophe modeling?

Ali Syed (01:56):
Catastrophe modeling is not yours, a lot of people think when they hear the word model, they tend to assume that it's something to do or in the analysts or the analytics arena, they tend to think of taking history and just repeating it, running a statistical model and seeing if you can predict the next year, next 10 years. So, CAT modeling is about understanding, a phenomenon, natural catastrophes that are very hard to predict. Nobody knows when the next earthquake will happen. Nobody knows where the next hurricane will strike. Well, we have good ideas and we have probabilistic measures that will tell us, but nobody can know for certain or with a very high degree of confidence let's say. The history of CAT models is that they were developed to fill a role. You know, the insurance industry was blind to the risk of catastrophes until the 1990s.

That's when Hurricane Hugo, Hurricane Andrew, and the Northridge earthquake happened. That's when they realized natural catastrophes, not only pose a very high risk that can wipe out quite a tremendous amount of capital from the insurance industries, but there is a possibility of even worse storms and worse earthquakes acting severely damage. The insurance industry can cause a lot of economic losses too. So, they were developed to fill that gap because your past 10 years can't tell you how the next 10 years of hurricane activity would be. If you think about 2005 to 2010-12, you know, we had a very, or after the Katrina season, and then until 2010-11, we had a very low hurricane activity. And then all of a sudden, boom, for the last two, three years, we are getting cat four cat five every year.

So, if you just look at recent history and try to predict your losses for the next 5, or 10 years, you're going to be severely wrong. So, what CAT models try and do is, simulate tens of hundreds of thousands of years of weather phenomena or seismic phenomena to understand basically what you can expect every year. Like on an average, like if you had what your average loss could be, what's the worst that could happen? So, what's the probability of what could happen every 10 years, what can happen every 20 years, the kind of storms, the kind of events you can expect and then hedge against them because these models were developed for the reinsurance industry. And I'm going to keep coming back to that point of view is that CAT models are developed to be right on average.

They're not deterministic models. They're not supposed to be right every time. You know, if you have a CAT-5 make landfall where you can say, well, this was only supposed to be a 1 in 500-year event. I saw one last year. Why am I seeing, you know, this year too, you're supposed to wait 500 years? That's not how CAT models work. These are annual probabilities. So, what you asked about 1 in 1000, 1 in 100, these are annual probabilities. And there's a lot of confusion about what these numbers mean. So, whenever we advise our clients, whenever you have any questions, do come back and ask us more, you can never ask us enough about CAT modeling. You know, what these numbers are. We're happy to help you understand what AAL means, what 1 in 215 means, rather than making an assumption because that's the worst that you could do about any scientific model is make an assumption and just roll with it.

Justin Swarbrick (05:42):
Yeah. And it's good that you are on the team Ali because it is helpful to have you analyze the models that we get back and use those to our advantage. For us as brokers, having clients all over the country, we have clients that are exposed to earthquakes, wildfires, hurricanes, severe convective storms, and floods. And we know that there are models for all those different perils. Can you explain the difference between the modeling approaches to the different perils that we typically will look at?

Ali Syed (06:20):
Yeah, sure. So, as you said, different perils, different phenomena, earthquakes is a seismic activity. Hurricane is atmospheric activity. In reality, all CAT models have any peril they have three main components. One big component is hazard, the actual damage, you know, how damage occurs. So, let's, work with an example of hurricane. When you think of a hurricane the hazards are the wind, the surge, right? If you think of an earthquake, the hazard is shaking, landslide, LIFA faction, and you have a tsunami. You have a firefalling earthquake. So, that's the hazard component of any peril, every peril, the hazard is developed separately in a different way, then you take that hazard. And then you get into the vulnerability part of where you actually convert that hazard to understand how much damage it will do to a particular location, right? So that's component two its vulnerability. And then the third piece is actually the financial modeling of it. Now I know what my damage ratio would be. Damage ratio is how much of your replacement cost was damaged basically, what percent of loss of your replacement cost, suffer the loss.

Justin Swarbrick (07:51):
So, I just a little follow-up to that. One of, the items we've talked about on this podcast over the past several episodes is what they call the bullseye zone. So, yeah, there's overdevelopment, and it's not necessarily that the weather's getting worse, but we're becoming more overdeveloped, which has caused these losses. So do the models take that into account?

Ali Syed (08:17):
So, the models don't have to take that into account, right? Cause the models are not, I guess we're talking about climate science now, so I'm going to switch a little bit of gears. You are absolutely right. The reason we are seeing a lot of increase in especially severe convective storm losses is because our footprint has expanded and wildfire too. Now people are building houses very close to forests or woods. And that's creating a lot of interaction between human beings and wild land where there was none before, and human beings are the largest source of wildfire triggers. We like to blame a lot of things, but its human activity that causes a lot of the wildfires. So yeah, our footprint has increased, but that's not where I would say the CAT models need are lacking. If you're drop in an exposure in CAT modeling, it will pick up the hazard.

Justin Swarbrick (09:27):
Okay. That makes sense.

Ali Syed (09:29):
Yeah. It will be fine. That's not the blind spot for CAT models.

Justin Swarbrick (09:34):
Got it. All right. Well, I didn't mean to derail us there.

Ali Syed (09:40):
Oh no. So, the only point I would make about your last question about is that  generally that's how all perils are modeled. They have three big components, but within those, there are a lot of differences. So, like the hazard of severe thunderstorms is very tricky, develop race differently, the hazard of wildfire, earthquake earth, so within them, within each barrel, there are a lot of differences, but on the surface there is a general process of how you develop CAT models.

Justin Swarbrick (10:14):
Can you talk a little bit more about that Ali and maybe what information is necessary?

Ali Syed (10:19):
Yeah. From our end when you are, so there are two processes. They were the developers process which I talked about where RMS, AIR, KCC, you know, these companies, they utilize a certain process to develop the models. And we are the consumers of the model, right. We run the models on our client's book. So, for us the most important things are the oldest cliche in the modeling world garbage in garbage out. We need good data, you know, and we can't emphasize enough with our clients, how they need to be invested. And they're spending a lot of money, on insurance and protecting themselves and they should invest a lot of effort in making sure that the data they provide us to run those models is as clean as possible. That you understand where your buildings are, where you understand what the construction types are, what the occupancies are, what all the different values are.

So, there is no surprise at the end when you see the results. The SOV, the statement of values that we get from our clients. I think it falls on us to our brokers too, as brokers we need to educate our clients because clients don't know right. What the CAT model is or not every client knows. Some are well versed. You know, they've been in the industry long enough, but we as brokers, we have a lot of responsibility on that and do to educate our clients. This is what a CAT model is. This is how it plays a big role in your insurance premium. This is why we need the data that goes into running these models needs to be absolutely as perfect as possible.

Carleen Patterson (12:05):
You raise a really interesting point, in which you talk about how modeling was started for the reinsurance industry. And there's a lot of different firms out there that do modeling. I'm curious why have we, Alliant invested in our own in-house modeling? Like, what is the advantage that you bring to our teams, as opposed to what we could get I guess from a carrier or something.

Ali Syed (12:33):
Yeah. So, we are not developing our own models, right. We license a particular vendor right now, we are using RMS’s, model and RMS is the biggest dog in the industry, as I said. So, the advantage is simple instead of getting third party or some wholesaler running your models we are getting in-house, you're doing you do it in-house now. The amount of effort that goes in, and as I said before, to properly run catastrophe models, it's quite significant, the amount of data cleaning and making sure that all the fields are mapped properly. There's no laziness behind it. And then the constant, the education back, that we can provide to our clients. It's not something that you send an email and you get a PDF back, there's a constant communication with our clients. This is why your results are high. This is what changed from last year. This is where you can improve your model performance. These are locations that cost that have the highest expected losses. We need more information about them. And then we can work with them and some engineering companies to help them understand and get their data in a good order.

Carleen Patterson (13:59):
Okay. So, it sounds like it's a lot when you were talking about just getting the data in like garbage in, garbage out. So, the better we understand what's going into it.

Ali Syed (14:09):
Absolutely. Yes.

Carleen Patterson (14:11):
So, I worked with a client a few years or ago, and we were doing some modeling, convective storm modeling. I understand in the scheme of modeling is a little bit of newer type of model. And we got something back. It was really crazy, something like 0.00016% chance of a tornado hitting a certain area of the country. And six months later that particular tornado hit. How reliable is it? Because when you're looking at something like it's a, you know, 16000th of a chance, I mean, how reliable are CAT models when we're talking to our clients?

Ali Syed (14:54):
Okay. So, this goes back to the fact that CAT models are not supposed to be right every time they're supposed to be, right on average. Your average annual loss. Let's stick with averages because it's easier to understand. And I'll get back to the 1 in 16,000 chance. They say your average annual loss, let's say on a $1 billion book is 20 million, right? Let's assume that, they don't mean that you're going to get a $20 million loss every year. One year you might not suffer annual losses. Well, most of the years, you're not going to suffer any losses. You know, if you're a hurricane pro area, but that one year that you do suffer, that loss is going to be quite significant, or it could be quite significant depending on where that event makes landfall. So, 20 million is just a good number to think about in long term, but it's not how losses are going to pan out. They're going to pan out very violently, there's going to be a lot of variance. So, whenever you start comparing single events, you know, you're not, that's, that's the trap you fall into. Again, you're using a single event to try and validate performance of a model that's supposed to be right over 50,000 years, or hundred thousand years, right, on average.

Carleen Patterson (16:19):
Don't run in and tell our clients, we only need 20 million in coverage.

Ali Syed (16:24):
Yeah, no, exactly not. So, that's where it gives you another metric. One of the most important metric CAT models produces an EP curve or exceedance probability curve, where you say 1 in 250 where those numbers 1 in 250, what 1 in 100 event really means is that you have 1% chance of seeing an event of this size or larger every year. So, it's not that it will only happen 100 years is like, you can see this event every year. There's a 1% chance you see this event or larger every year. So, that's why when you get an EF4 last year or an EF4, again, this year that's where that's the trap people fall in. Well, it was only supposed to happen every hundred years. Why it's happening today or back to back years. That's not, that's not what the metric is saying. It's saying something else.

Justin Swarbrick (17:20):
Okay. So are there any other misconceptions Ali that you can think of that insureds will latch onto when looking at modeling?

Ali Syed (17:32):
It's generally around these terms, you know, I guess average, annual loss. I explained it's not supposed to be that number every year. 1 in 100 events are not supposed to be only every 100 years, there's a chance that, or worse can happen every year, but the general idea of CAT models being right, every on deterministic events, like, for example, let's say you have a hurricane that makes landfall in Miami and a CAT modeling firm public out, a few sample storms and says, Hey, these, this storm is similar to the one that's making land for right now. Why don't you run it against your portfolio and see, what your loss is going to be. They don't expect it to be right. You know, they just want to give you an idea.

Ali Syed (18:25):
You don't think of a CAT model, as a point estimate it's always a estimate with some variance surrounded. When you're modeling year over year, you should always use get model numbers as a watermark. I had this number last year, I had this number this year. What changed? That's what's the key thing to understand is, okay, what changed from year to year? Is it my PIV that's growing. If I grow too fast. I had this client, I don't know if I remember their name, but it was for Alliant and they actually reduced a couple of hundred million of their PIV, and they added some other properties, but overall the PIV came down, but their average and loss almost doubled. And I got asked that question. This does not make any sense. You know, like the client was like, this does not make any sense, what are these models? So when I dug again they removed a significant amount in properties, but they added Florida. Right. So these are the kind of things you need to understand, never look at numbers on the surface, always dig through what change, what does it mean? You know, never expected to be right year over year.

Carleen Patterson (19:49):
So, Justin talked a little bit about how from the bullseye effect that as we grow and develop it's in certain areas where there is catastrophic exposure, but do the models actually take into account climate change and weather patterns? I mean, I know you've got that piece over there, which is where we are developing, but that's just the basic question. Does it actually take that into account that not economic development, but climate change itself?

Ali Syed (20:24):
There is very clear evidence that the climate is changing. The climate is changing and is changing abruptly. It is trickier evidence, harder evidence to relate that to actual weather phenomena. You know, I'm not saying there's not any evidence, but it's a harder step forward to actually say this hurricane activity was caused directly due to result in climate change. Well, there is direct evidence that the last few years we have seen, were basically supercharged because the sea surface temperatures were so high. There's evidence of that, but to directly pin it on climate change, it's a harder step still. For example, there's some evidence for hurricanes and the tracks are shifting eastwards more and the storms might be, and not as frequent, but might be getting stronger. So, what CAT modeling firms do is that they have a catalog of events, usually. They assess that catalog of events and make sure that that catalog of events is still, it reflects, you know, the current climate conditions and the conditions they expect over the next five years.

So like RMS, has something called the medium-term rates, which is their best estimate of the hurricane activity for the next five years. It's an ensemble of different models. I don't want to get into details, but it's an ensemble of different models that competes to see which model can basically predict the best, hurricane activity for the next five years, and then AIR has something that they call the sea warm sea surface temperature models. In simple terms, if you jack up the sea surface temperature, what's your hurricane activity going to be like? So, for climate change, the best thing is not to wait for models or not to know depend just on climate modeling. What you should do is basically do a lot of sensitivity testing. If you see more hurricanes you can work with most vendors that can help you understand, talking about CAT modeling vendors, they can help you design these studies. What if hurricane connectivity increases, what if severe storm activity increases?

Carleen Patterson (22:58):
You know, that really does make a lot of sense. And we really appreciate your input as to how the CAT modeling is working and how it works. And then how we as brokers can use it to help our clients. So Justin and I really thank Ali for joining us today. We recognize this is a challenging time to be in public entity risk management, and we will continue to focus our podcasts on information and resources as we navigate 2020 and beyond.