 I want to thank the organizers first for including this paper on the program. I'm very excited to present this very interesting symposium. This is joint work with Matthias Crudley, who's in the audience today. And Bridget and Fran were both at the Federal Reserve Board. So in recent years, there's been an intensification of extreme weather events like hurricanes, droughts, wildfires, snowstorms and so forth. And there's some discussion about how climate change will continue to intensify and potentially make such events more frequent. This extensive work about the economic costs and damages associated with such events. But surprisingly, currently little is known about the uncertainty generated for firms by such events. This is surprising because in other contexts, if you think about political uncertainty or macroeconomic uncertainty, so political uncertainty around elections, there have been work that shows that some investments are impacted due to that kind of uncertainty. And so given that the potential for wide-ranging real effects, this is an important concept to understand and see how substantial it is in the data. I want to make clear that by uncertainty, we need expectation of future volatility as opposed to net uncertainty. Now our priority, it's not obvious that extreme weather events will generate substantial uncertainty. On the one hand, you have possible unpredictable impacts to firms, PPE, local labor, demand, supply chain, restrictions and so forth, which can increase uncertainty. On the other hand, particularly vulnerable firms might ensure against such events, they might adapt or relocate away from high-risk areas, which would in turn reduce uncertainty. So how much uncertainty is generated from extreme weather for firms is an empirical question. On top of that, we care about the efficient pricing of climatic risks for financial stability reasons. In 2015, Mark Carney pointed out, for instance, that mispricing could lead to mispricing of climate-related risks can lead to sudden, large price corrections that can be destabilizing. So in this paper, we're going to consider two questions. First, we want to understand whether extreme weather causes uncertainty for firms and if so, how substantial is it, how much uncertainty is generated. And we want to understand whether investors price extreme weather uncertainty efficient. So in particular in the paper, we're going to analyze extreme weather uncertainty at the firm level using financial markets. And we're going to develop a simple framework to formalize our ideas on the sources of extreme weather uncertainty. When you think of the uncertainty generated from an extreme weather event as coming from two sources, the first is incidence uncertainty or uncertainty about whether, when, or where an event will occur. So you can think of this as uncertainty on the extensive margin and impact uncertainty. So this is uncertainty about how an event will impact firms. So even when you know exactly where an event has occurred or whether a firm has hit, there's still associated uncertainty about the ultimate impact for firms. And you can think of this as uncertainty on the intensive margin. Our empirical setting is going to be, we're going to look at single stop option price reactions around US hurricanes. And we'll look at changes to implied volatility as our measure of uncertainty. We're going to use the differences in different setting to identify our effects. And we're going to look at firms located in the forecasted or realized path of a hurricane as a treatment group versus unexposed firms as the control group. Importantly, we're going to determine treatment using firm establishment or facility locations as opposed to purely headquarter locations of firms. And we have multiple hurricane events to chime in our sample, which ensures that we're not sampling just one time period or a particular region or industry. To have an image in mind before we go further in the presentation, so this is showing an image from NOAA of illustration of a hurricane sandy three days before it made landfall. So this is showing the location of the eye of the hurricane. So the current location and this green highlighted area is showing the forecasted region where there's hurricane force winds with different probabilities. And so in our analysis, in our forecast analysis, we'll be using the underlying data that creates such images. So preview of what we find, when we look at the period before landfall, the forecast period, investors pay attention to short-term forecasts and pricing substantial uncertainty. And this uncertainty will reflect both incidence uncertainty and expected impact uncertainty. I'll go into more detail here. After landfall, options of firms in the landfall region reflect large impact uncertainty of up to 20% higher. And these results hold across industries. And then importantly, we find that impact uncertainty resolution is slow. So even when we know that a firm has been hit, implied volatilities remain elevated or uncertainty remains elevated for up to three months after a hurricane landfall, which is a substantial period. While both these results show that investors pay attention to hurricane forecasts and landfalls, both before and after landfall, we find evidence of significant underreaction to the information available to investors. We find that exposed realizable utility is differentially larger for exposed firms than the ex ante expected volatility. And then interestingly, we find that this underreaction is attenuated following hurricane Sandy, which was a particularly salient destructive hurricane in our sample. And then finally, consistent with Merton 87, the theory of under diversification impacting associating implied volatility with positive expected returns, we find that extreme weather uncertainty is associated with higher cost of capital firms. Just quickly on the related literature, we, we contribute to three different strands of the literature. So to start with the pricing of climatic risks. This is a small but growing area of the literature. So papers like Hong and co authors and a look in Kumar and co authors, they look at stock price reactions and find mixed evidence for both underreaction and overreaction in the part of investors for climatic risks. And there's a separate literature that looks at transition risks. That is, when there's a change in the expectation of climate change regulation, how that's reflected in financial markets. So Engel and co authors and Ilhan and co authors, they look at stock and option markets, respectively. In this paper, we test the efficient pricing of climatic risks with both forecast and realizations using multiple exogenous clearly identified saving climatic risks. We analyze the second moment, and we show the user forecast and underreaction and subsequent learning about uncertainty. There's a big literature on uncertainty and we contributed to it by adding a comprehensive analysis of extreme weather uncertainty and developing a conceptual framework to think about uncertainty on the extensive and intensive margins. And then finally, on the literature on idiosyncratic volatility and how it's associated with expected returns. So theories like those by levy and merton show that market segmentation or the diversification can lead to your syncretic volatility or uncertainty to firms being associated with positive expected returns. And the empirical evidence for these theories are remain mixed. And so what we can do is exploit our unique empirical setting to test these theories using identified exogenous increases to the synthetic volatility. Alright, so we're going to quickly present the simple framework we have in our paper to think about how extreme weather uncertainty impacts firms. So suppose there's a extreme weather event, so time t does an extreme weather event that's expected to hit it in the next period t plus one, we have a firm whose stock return is is as follows. So you have two components. The first is the is a random shock that is independent of the extreme weather event. So you can think of this as the baseline process for the stock. The second component is a random component that's due to the extreme weather event. So this G is normally distributed with a mu G and a variance sigma G squared. And this is the impact of the extreme weather on the event on firm I. And theta is a one draw of a binomial or Bernoulli distributed variable that is one if the firm is hit by the extreme weather event or zero otherwise. So fee would be the probability of the event occurring or the probability of incidents. And so the data is one with probability. So we can think of the variance of the firm conditional on whether it is hit or not hit by the extreme weather event. So if theta is zero, so the firm is spared and not hit by the extreme weather event, you have the variance of the firm as sigma squared, the baseline variance of the firm, the stock process of the firm. If the event occurs or the firm is hit by the extreme weather event, you have the baseline variance plus the sigma G squared, which represents the impact uncertainty associated with extreme weather event. What we can do is combine the expected conditional variance and the variance of the conditional expectation and obtain the total variance with expected variance at time T of the stock return process. And that is depicted by these three terms. So again, the first is just the baseline variance of the firm. The second captures the expected impact uncertainty. So that's the impact uncertainty discounted by the probability of the event occurring. And then the third captures what's what we call incidents uncertainty. And so for a given new G, this is highest when the probability of incidence is 0.5, as you have with the Bernoulli variable. And it is zero when phi is zero or one. There is total uncertainty about whether or not the event will occur. It's also zero when mu is zero as well. This figure is illustrating the total variance prior to an extreme weather event, as you vary the probability of the incident occurring. So on the y-axis, you have the total variance. On the x-axis, you have the probability of incidence. So the red line here is what you would get if there is no landfall uncertainty, no incidence uncertainty. So this straight line. And as you keep everything constant, so for a given level of phi and you vary mu G, what you can see is the distance of the curve from the red line depicts the amount of incidence uncertainty associated with those set of parameters. So what I want to point out is, so even when you have a relatively low probability of the event occurring, you can have cases where the total variance is higher than what the variance would be if the event would occur, which is this level, just because of how much incidence uncertainty or uncertainty about whether or not an event will occur. That's very high. So moving on to the empirical design and data, we put together several novel data sets in the paper. So prior to landfall, we used county level probabilities of hurricane level wind speeds from NOAA forecasts. So these are five-day forecasts that are available from 2007, and they cover 41 storms. Importantly, we include storms that both make landfall and also dissipate without making landfall as hurricanes. So we have both hits and misses in the forecast analysis. After landfall, we use the location and distance from the eye of a hurricane to determine the landfall region. And so we have 33 hurricane landfalls starting from 1996 in our analysis. We identify firms exposed or unexposed to a hurricane using establishments, and these data come from nets. And we measure the change in IED relative to just before inception as our measure of how much uncertainty is generated from a particular event. And so for this, we obtain single stock options data from option metrics, and we have a constructed daily average implied volatility measure for each firm, and we get depicted with IV. So in terms of the identification strategy, I want to be as clear as possible at this point because it will make the later results presentation much simpler. So the first dimension of our identification is the pre-post timeline. So if you think about a hurricane, a storm appears out in sea, there's no anticipation of a storm before it appears. So the inception is at the point where the storm appears in the ocean. So the pre-period there is a very clean control period. There's no anticipation of the event because the possibility of the event did not exist before inception. So this is our control period. For the forecast analysis, we're going to use the number of days before landfall as points in time to identify a particular post period. For the landfall analysis, correspondingly, we'll be looking at how much time it has been since landfall to understand how uncertainty falls. So that would be child days after landfall. On the spatial dimension, our identification, I illustrate with this figure. So this grid is, the squares within this grid, you can think of as counties. And we have three illustrated firms here, A, B, and C. So A has four establishments shown by the yellow circles. I think someone's unmuted. Firm B has four red triangles showing four establishments. And firm C has the three establishments. So firm establishments are spatially distributed across different counties. And so from NOAA forecast, so when we're looking at a firm's forecast exposure, we can obtain from NOAA the counties that have hurricane level wind speed potential. And these would be the exposed counties. And so how do we think about the forecast exposure per firm? So for firm A in this example, two out of its four establishments are in the exposed region. So it would have a forecast exposure of 0.5. For firm B, three out of its four establishments are in the exposed region. And for firm C, none of its establishments. So the blue squares are all in the unexposed counties. So it has a forecast exposure of zero. So forecast exposure is going to be a continuous variable that ranges from zero to one, reflecting treatment intensity. We have a corresponding measure for a landfall exposure, which we use the landfall region. So the distance from a hurricane eye when it makes landfall will determine the landfall region. And so we use a corresponding methodology to figure out the landfall region exposure. And it's again a continuous variable that ranges from zero to one, reflecting treatment intensity. So what does the real data look like? So here I'm showing a hurricane Sandy four days before landfall. And this is part of the US map. So you can see Florida here and here and so on. And the highlighted region are showing the counties that NOAA is forecasting. From NOAA forecast, we obtained a set of counties that are likely to have hurricane four swings with greater than 1% probability. Now, as you get closer and closer to landfall, you can see the shifting around and NOAA forecasting with higher and higher probability about which counties will be impacted. So if you go from four days, three days, two days, one day, and so on. So in our analysis, we'll separately consider both the number of days before landfall and also the probability level. And then landfall. So this is illustrating hurricane Sandy's landfall region. So the darkest area is 50 miles around the hurricane eye and the subsequent colors that go from red to yellow are showing 100 miles, 150 miles, 100 mile radii around the landfall region or the eye of the hurricane. So this will determine whether a hurricane affirm is within the landfall region or not. So as I mentioned, we have 33 hurricanes like this. And so this is showing another landfall region to Hurricane Harvey, which hit Texas in 2017. And we have hurricanes that hit all throughout this region. This is illustrating the firm establishment data, so as of 2010. And so at the county level, this is showing the density of firm facilities. And what you can see is that in the region where you have hurricanes, there's a lot of economic activity. And so you have a lot of firms that will be exposed to hurricanes or will be hit by, have the potential to be hit by hurricanes. All right. So this is some summary statistics just to show that. So we have over 25,000 unique firms in our sample. And the number of unique hit firms are substantial. So and I just wanted to point out that firms that I hit and those that are never hit be, when you look at different dimensions, they are on average, not significantly different from the total sample. All right. So to onto the results. So the first set of results I'm going to show you is how the measurement of uncertainty before landfall or dissipation. So to remind you, this is our timeline. So inception is before. So this period control period is when the storm didn't exist. So we're going to look at, we're going to estimate this panel regression. Gamma days before landfall or dissipation, the dependent variable is the change in IVs since just before hurricane inception. So from this point to this point, we're going to look at the implied volatility change. And Lambda will capture the uncertainty increased due to exposure to hurricane forecast. And it's positive if uncertainty increases with forecast exposure. So what do we find? So this is showing, as I mentioned, we looked separately at the number within the number of days before landfall and different probability buckets. And so there's up to 21% increase in IV for a firm that's fully exposed to a hurricane. We find that Lambda is positive and significant. And so we knew that there is uncertainty being priced in due to forecast exposure to firms. What I can do is illustrate this graphically. So when I can, when I look at the regression coefficient, so five days before landfall at different probability minimum property thresholds, implied volatility is elevated as much as five days before landfall. And as I get more kind of observations at different levels of probability, I can estimate this uncertainty increase as we go along. And implied volatility increases with the minimum probability of hurricane force winds. And what you get, again, this is a day before landfall when there is significant, you know, high probability observations, we get a nice graph like this. So ultimately, what this is showing is that investors are paying attention to hurricane forecasts, and that, you know, even forecast exposure is generating substantial uncertainty for firms. Moving on to uncertainty after landfall. So, you know, once a hurricane landfall has occurred, we have incidence uncertainty or uncertainty associated with a hurricane event will occur, is fully resolved and only impact uncertainty remains. And so what we can do is look at the implied volatility difference between the post landfall period and pre-inception. So the dependent variable here is a change in ID since just before hurricane inception, and lambda here captures the uncertainty increase due to exposure to hurricane landfall. And it's positive if the uncertainty increases with landfall exposure. So again, this is showing the impact uncertainty that we estimate a week after landfall for different landfall regions. So the 50 mile radius, 100 mile, 150, 200 mile radius. And what you can see is that there's significant uncertainty being priced in. We can run this regression separately for, you know, different days after landfall, which is what this graph is showing. So I showed the table for five days after landfall, but we repeat this regression for up to 100 days after landfall. And this is showing how the estimated impact uncertainty is changing. The coefficient peaks at around 20% and reverses to two hurricane levels after three months. Similarly, I can look at the 200 mile radius and rerun the regressions. And while there's a lower peak for firms further away from the hurricane eye, as you would expect, because they are less intensely treated, IV remains elevated for up to three months. All right. So we want to understand whether these, so while these results show that investors are paying attention to hurricane events, we want to understand whether the expectations of this future volatility are correctly priced, if it's efficiently priced. As I mentioned, inefficient pricing of climatic risks to impose financial stability risks. And so how we're going to do this is we're going to define the difference between the option implied volatility, which can be thought of as a measure of expected volatility, and the subsequent realized volatility over the same period. So over the remaining life of the option, we're going to define this as a volatility risk agreement. We're going to analyze differences in VRP between firms exposed to hurricane forecasts and landfalls versus control firms. So again, it's a differences in different specifications similar to the previous regressions. And so a negative lambda would imply that the ex ante expected volatility is systematically lower than the exposed realized volatility for the set of exposed firms compared to the set of control firms. So this would be evidence of systematic underreaction. And so what do we find? So this is the VRP difference we estimate prior to landfall. And it's, we find that it's negative and significant. So VRP is systematically lower for exposed firms compared to control firms. And so this is evidence of underreaction. And we see similar results for the VRP difference after landfall that it is systematically lower for hit firms compared to control firms. And again, so underreaction. You won't understand whether investors learn over time because these are events that, you know, the multiple events through time might think a particularly damaging hurricane could increase the saliency of hurricane strikes for investors. And it could lead to investors pricing hurricanes more efficiently in option markets. So we test if the underreaction result changes after hurricane Sandy, which was a particularly destructive hurricane where a large share of institutional investors in the West resided in the landfall region. And it, and it really changed the narrative around extreme weather events and the potential for climate change to intensify this, it really changed the narrative just after the event, unlike with prior destructive hurricanes. So what do we find? So we basically run the same regression we did before, but with an interaction for the time period off the Sandy. And the evidence is in line with so what you should do to think about what the uncertain the VRP difference is following Sandy is to add up these two coefficients. So what we find is that the inefficiency in pricing stream weather uncertainty diminishes post hurricane Sandy, as you can see the sign flips following hurricane Sandy. So the VRP difference is less negative. The final result that I'm going to present, I actually lost track of the times. I'm not quite sure how much time I have left. But we want to understand whether it would be good if you could wrap up. Okay, sorry. Thank you. All right. So the last thing we want to think about is whether the heightened extreme weather uncertainty leads to higher cost of capital. And this is important because it we have a unique setting that will allow us to exploit the identified exogenous shocks to volatility that we have. And we're again using a difference in different specification, similar to previous regressions. And so what we find is that the relationship between excess returns and uncertainty is as predicted by Martin and Levy after hurricane Sandy. So greater exposure to extreme weather uncertainty leads to higher cost of capital. We have other results in the paper I'm not going to go into. Let me conclude. So we show that extreme weather events cause substantial uncertainty that investors pay attention to these events, but they underreact for a long period of time. It takes a long while to resolve the uncertainty associated with extreme weather events. And just one point I want to make is that given these evidence of significant pricing efficiencies, even for repeated events like hurricanes and the fact that learning required at a particularly destructive salient event like Sandy raises concerns for the efficient pricing of novel risks caused by climate change. And finally, it's important that we are able to show that extreme weather uncertainty increases the cost of capital for exposed fronts. And thank you for your attention. Sorry for going over time. I'm looking forward to the discussion and audience questions. Thanks so much for a minute. The discussion is Zacharias Zautner from the Frankfurt School. Thank you very much. So I'm delighted to discuss this paper and we have an exciting panel after my discussion. So I'm trying to talk very fast to make up the time. It's a great paper. It's very topical. I think it could become an anchor paper in the literature on climate finance. And it's not just my personal view that this is a great paper. I mean, it made it on the program. That's one thing. Plus it's R&R and the JF, which is another indication that this has great potential. So a huge amount of work went into the data collection, analysis, robustness checks. It's a very deep and dense paper, which is also jealous, because I think the audience could see it. It was very difficult to follow all the amazing work that has been done. My comments are mostly about understanding and interpreting the results. And given the five minutes, I'll focus on three comments on the economic channel, most important on the uncertainty measures and the hurricane measures. The paper is currently quote agnostic about how firms are affected by the hurricanes. I think it's important that the authors try to shed some light on the economic channels from which these volatility and return effects are originating from. There's large heterogeneity across firms that they can exploit and how different firms are affected by the hurricane and how their business activity is hampering or benefiting from hurricanes. So let me make concrete suggestions. I think you should explore somewhat more the different dimensions that are available to understand in a better way the channels behind these results. You could look at real effects. So what are the effects of the hurricanes on changes in investor demand, on product demand or the demand for services as well as the supply of goods and services? You could look, and that is similar to a comment by Sönke in the chat, at the role of adaptation measures insurance across firms. So two firms are identical, but one may have insurance, so the effect should be different if that information is available. I think it would be important to shed more light on whether the effects are coming from a cash flow or discount rate channel. So can you show some evidence on whether earnings, cash flows or cost of capital risks, discount rates are affected? And I think you could also kind of decompose the implied volatility and look at a little bit more at where and the return distribution side effects are coming from. Do they only affect the left tail or are they also right tail effects? Is both affected? The implied volatility is a composite measure that captures both sides, so it would be interesting to see that. Some firms may actually benefit as well if in a local region a hurricane comes. Second, I think you should do a little bit more on the uncertainty measures. Your model argues that uncertainty before landfall is all about forecast on where the hurricane will hit. After landfall, it's all about how much damage the hurricane will produce. So there's an assumption here that there's a clicker difference between incidents and impact uncertainty before versus after landfall. Now my concern here is that there remains a lot of incidents uncertainty even after landfall because the movement of the hurricane is usually rather unpredictable. So I think the two phases may be much more intertwined in practice than what the model assumes and what also the test as a result of that implies. So here again my suggestions. I think I would address a little bit more the concern that it's difficult to distinguish these two phases, these two uncertainties. I would consider merging the two uncertainties and constructing a single measure from the inception of the hurricane until its dissipation or if you're unable to do that think of designing a shorter test to distinguish between these two types of uncertainty. Third comment is about the hurricane measures. It's clear that there's a challenge to decide whether and how much a county therefore a firm is affected by a hurricane. So the authors use a radius of 75 miles around the center for some of the main tests. You've also seen they're using robustness jack using different Rs. Now it's plausible even that the results are a kind of robust that this will not affect the results a lot but still you know given that NOAA reports that a typical hurricane has a width of 300 miles or 250 and that NOAA at the same time says that the effect of a hurricane depends a lot on the size of the hurricane and can range between 250 and 300 miles. But there's a lot of heterogeneity still that you may want to capture. For example, Hurricane Katrina had an hour of 150 miles and 70 of 450 miles. So rather than taking one specific numbers and looking at for robustness checks at different Rs across all hurricanes, why not simply considering the actual data? Consider the actual parameter or for each hurricane the data is available. It's a small number of hurricanes in the sample so it should be possible to collect this use this data and give an R to every single hurricane in the data. I was very fast in interest of time. Let me summarize by saying I enjoyed the paper a lot. It's a fantastic piece. Some work needs to be done to consider the economic channels and there are some additional aspects to explore. Thanks a lot for the opportunity to present the discussion. Thanks, Zacharias, especially for going through your discussion at the speed of a hurricane. I pick up one question from the floor. Question is whether the larger firms are also more geographically diversified and whether this could explain why they face smaller risks. And a related question is whether the exposure should be measured based on the size of the establishment. So indeed, we don't have a lot of time but I do want to give you the opportunity to respond either to the floor or to Secretary Yes's questions. I'll be very quick. So on the next, Zacharias, thanks for a great discussion. All points very well taken. I can just address one point about the uncertainty before and after landfall. I might not have been clear when we think about landfall. We're actually looking at the 24-hour period. What we take as the landfall region is after you first make landfall, we take a 24-hour period around that to construct the landfall region. And hurricanes generally dissipate very quickly once it makes landfall. So within 24-hour period, you have the hurricane force winds dissipating and going away. And the results I showed started from five days to subsequent. So that is definitely without any landfall incidents uncertainty. But the points well taken can be that much clearer. And on the economic channels and maybe even more granular about hurricane measures that we can do more on that point well taken. And on Sonke's question very quickly on the size of establishments. We have alternate measures that we use for instance by sales weighting on the establishments. And the results are for robustness we do this the results are qualitatively similar. But the sales data at the establishment level is not as reliable as the location data. So don't feel comfortable using that as the baseline measure. Even with like you know if you went to say another data set it's similar like at the facility level you can get location data but everything else is a bit more noisy. So it's not the cleanest way to think about exposure. But thanks for the questions and great discussion.