financial modeling

[danhammer]

We will ultimately estimate the value of a house under conditions of uncertain sea level rise. To do so, here are the resources we have access to:

1. Estimated value of the house, according to Zillow
2. Estimated rental rate of the house, according to Zillow, which presumably equates to the value of housing services each month.
3. The amount of sea level rise (according to the NOAA bathtub model) that would put the house under water at high tide.
4. The uncertainty associated with the value in Item 3. Just two categories: low or high uncertainty.
5. A probability distribution of sea level rise rates, collected from the best available science to date. This is rarely updated.

We need to translate all of these items into an expected value. One suggestion is to calculate the time at which the value of housing at the given location goes to zero -- due to rising seas. There will be a distribution of end dates, with associated likelihoods. We should be able to convert this information into an expected present value of the house, based on the current valuation (presumably without taking into account sea level rise).

I am opening this issue to start the conversation about properly modeling the financials, given the environmental conditions we get from other data sources.

[glenearthgenome]

This is a great discussion. Here are my initial thoughts, trying to balance end-user orientation and something that is defensible/easy to understand.

1) Timing/date. First and foremost, the end-user (likely a home owner or prospective home owner) wants to know the best scientific estimate on the year in which the house will be affected by sea-level rise. I think we should report a date, perhaps at varying levels of damage, even if not tied to a specific financial calculation for the home. e.g. Zillow website should say "90% confidence that this home will be affected by sea level rise by year 2038" 2) Home value. Complementing the time estimate should be an estimate of financial damage to the home. Importantly, I think a current/prospective owner wants to be able to see the home value versus sea-level rise curve. Y axis is home value (measured in $ with the axis intercept being current value with existing sea levels). X axis is increasing sea-level rise (measured in inches). Importantly, I'm not sure this graph has a time component. In this sense, the timing/data in #1 above complements the #2 home value.

The key question here is what does this curve look like? We would expect it to be non-linear and likely with potential discontinuities (e.g. at some point, the value of the home will likely drop precipitously once sea level rise appears inevitable or imminent). I think one way to potentially address this is to use the same methodology that FEMA or insurance actuaries use to currently estimate flood damage to a particular home. After all, sea-level rise is just another type of flooding (albeit more long-lasting and permanent). In this case, what we would do is take current home value zestimate and then subtract out the flood damage calculation due to the rising inches of sea-level rise.

I agree that another way to do it would be to try and estimate the exact date where the Future Value (FV) of the home = $0 and then you calculate back the adjusted Present Value (PV) given sea-level rise and then report the "discount rate" at which this home value will decline (e.g. Zillow would state "this home has a -3.5% yearly decline in home value, given 90% confidence level of sea-level rise damage"). My problem with this is the great unknown of when is a home FV really equal $0? Hard to tell when that really happens.

It is probably worth also noting that we can't factor in the potential mitigants that may be enacted to save homes (cost likely born by other actors, not just the home owner). That would make the FV = $0 calculation difficult as well. Also, there is an infrastructure consideration here too...e.g. even if you own a home on top of a tall hill...if you can't access the roads (or other basic services) at the lower level, that may also dramatically affect home values.

Happy to discuss this further...we'll need to iterate together. Note: I have no specific details on how FEMA calculates flooding damage, but this feels knowable.

[danhammer]

These are my takeaways from @glenearthgenome's comments:

1. After basic data on SLR and coastal properties have been collected, there are many useful ways to package the information to the consumers. The date when the property would be under water elicits the greatest response from a potential homebuyer -- even greater than the discounted value of the house (which requires a bunch of financial modeling and assumptions).
2. There is value in operationalizing regulations or federal protocols, not just the science. Picking through all of the FEMA regulations, for example, is a non-trivial exercise. We can add value by interpreting the regulations for a specific, local property. In addition, this accepted protocol is a useful reference for our own estimates.

The objective of this project, @glenearthgenome reminds us, is to install environmental information within applications that already command attention from end-users -- rather than trying to build a new platform that people need to adopt. How can we deepen exiting interactions or, better yet, transactions along environmental dimensions? How can we intensify the intensive margin, rather than create one more app or platform on the extensive margin?

Now, the question is: How do we model the time at which the housing value goes to zero? Or, better yet, how do we trace out the value curve of a coastal property under conditions of SLR?

[danhammer]

@WheelerDR has generated a probability distribution for SLR by 2100, based on climate projections. It is a linear transformation to get the probability distribution of dates by which a given property goes under water.

The question, now, is how to incorporate the full distribution of uncertainty into the estimates. For now, we can use the maximum likelihood rate, but it seems more sensible to account for the range of SLR values (now pegged to a probability).

[danhammer]

I have integrated over the probability distribution that @WheelerDR specified. The code is embedded in this script.

[glenearthgenome]

I've been pondering this and wanted to offer some thoughts, especially as it relates to discount rates we should use in our analysis.

Let me start with an analogy: cars and how that asset (like a house) is valued over time.

A car always has a PV (present value) and a FV (future value). When buying new, that PV is your purchase price of that car before you drive it off the lot. The FV of that car at any time is the retained value of the car based on car quality and condition (important to note that cars are almost always a depreciating asset...the car value decreases over time). In addition to the PV and FV of the car, there are also operating costs (car insurance, gasoline, maintenance like tune-ups and oil changes, etc). Think of these costs as cash outflows in your cashflow diagram (e.g. monetary losses over time).

In this car analogy: -Discount rate. How a FV is related to PV is not a macro-economic discount rate. It is the car specific depreciation and condition (e.g. total miles on the odometer, age, whether there is rust on the car, etc). As you know, some cars better retain their values than others. That updated value is a market based transaction, as tracked by companies like Kelly Blue Book who collect data on what cars are being sold at what price given what condition. So applying a macroeconomic discount rate is incorrect for cars (just like it is for homes) -Total cost of ownership. A financially wise car owner would track the PV and FV of a car, but they would also sum up the total operating costs of the car...what some people call Total Cost of Ownership (which often includes purchase price). For these operating costs (insurance, gasoline, maintenance) the appropriate discount rate is a cost of capital rate, such as a macro-economic discount rate. Although the amounts will vary based on type of car (e.g. some cars require more gasoline given they are less fuel efficient or require more expensive premium gasoline), the money spent on those operating costs is cash and should be discounted similarly

Now, here is what I think that means for us: -For home PV and FV. We use Zillow estimate (zestimate) and then gross up or down based on Case Shiller index prices for homes by major metropolitan area. So in our situation, the Case Shiller is our equivalent for Kelly Blue Book. https://en.wikipedia.org/wiki/Case%E2%80%93Shiller_index. If data were more available by county or zip code that would be even better because it would be more specific, but I'm not sure that is available. I don't think it exists for that far out (maybe beyond a year or so forecasting) but we may have to just use historical to create a future looking rate. We do this by taking the historical % yearly change (home price increases or decreases) that has been seen by any metropolitan area over a certain period of time and make that our discount rate for homes in that area. It is important to note that for homes they can both appreciate or depreciate (unlike cars that just depreciate). So in some case, PV > FV or sometimes PV < FV -For operating costs. Homes have operating costs too (e.g. property taxes, insurance, etc). But notably for sea-level rise you also have the real threat of recurring damage to your home given flooding during extreme weather events. I personally believe these kinds of costs should be calculated use a macro-economic discount rate, like car operating costs should be done. I think the key here is for us to use a risk based assessment of damage akin to what actuaries in insurance companies use. They try to estimate the probability (%) and impact ($) and frequency (t) of exact events occurring to any asset. Same is true for liability insurance, earthquake insurance, flooding insurance, etc (also same is true for medical insurance for people). I think for this we need to research how insurance companies (or maybe FEMA) do these estimates

Hope the above is clear. Happy to discuss live.

[danhammer]

This is perfect, @glenearthgenome. There are some actionable insights in your last post. I think we're ready to put this into equations, so that we can start assembling the relevant variables from available data (see this post on #7 for some available data sets).

In order to operationalize your accounting insights from above, it seems like we need a web service to grab information from:

1. The Case Schiller Index (possibly regional, if it exists)
2. The aggregate zillow estimates of neighborhood value over time, like this.
3. The specific factors that are used to calculate disaster insurance, and especially coastal flood insurance.
4. Information on local property taxes.

Do we need Item 1 if we have Item 2? Seems like we should try to remain internally consistent with home values, so if Zillow publishes their estimates, then we should choose that over Case Schiller. Right?

[glenearthgenome]

Yes, agreed. I only suggested Case Shiller because it is the undisputed "gold standard" for home price changes in the U.S. If we are doing this for Zillow, we should use Zillow numbers. If we are pitching to someone else beyond Zillow, we may want to use Case Shiller

[danhammer]

Got it. Ok. I will pull both. Still, I am unsure how to assemble this information into a meaningful value. I will work with you, @glenearthgenome, to figure out that equation.

[danhammer]

Ok, @glenearthgenome, I've identified the web service for the Zillow Home Value Index 1-Yr change for a neighborhood. Interestingly, it is much more difficult to incorporate the national Case-Shiller index into our application than the local, Zillow estimate. I will build this later.

Do you need local mortgage rates? Or any information about comparable home sales in order to calculate the PV/FV?

Any guidance on what insurance information you need for operating costs would be helpful. I can easily get information on property taxes, but the real value added, it seems, would come from identifying potential changes in flood insurance rates -- as specified by FEMA. Available data that I've found is here, but I am sure there is more data that is not listed.

[glenearthgenome]

Great comments.

I just read the link on ZHVI. They are exactly right...sales mix is key on "comparable" home sales. I think we should initially use the ZHVI instead of Case-Shiller.

Local mortgage rates? NO, we do not need them. Mortgage rates are independent of home value and are simply a reflection of risk of lending money to any home owner (e.g. based on that person's FICO scores). More importantly, mortgage rates are tied to the Fed setting interest rates (and value of mortgage backed securities) and that is not tied to the value of any particular home.

Comparable home sales? Are directly incorporated into either ZHVI and Case-Shiller. You don't need them separately.

[danhammer]

I think that this issue thread has run its near-term course, so I am closing it. This is not to suggest that the financial modeling has been fully specified, but rather that development has incorporated or grown out of the original premise. I will open a new issue now about FEMA Flood Insurance mapping, which will incorporate financials more directly.