You’ve probably heard the words “Case-Shiller” mentioned hundreds of times by now … it’s not surprising. The Case-Shiller Index (CSI) is now touted as one of the most respected and reliable means of tracking the health of the housing market. Headlines are littered with references to this magical index, so we thought we would take a peek at how relevant the CSI is as it relates to our Home Sweet Home: New York City.
The purpose of the index is to solve for the basic dilemma faced in tracking home values over time: the particular mix of properties sold at any point in time can create the illusion of a change in property values. In other words, if larger, more expensive homes sell in one month, values appear to increase, even if the price of every home in the market has actually remained flat. The same dynamic applies in the opposite direction. Welcome to the fun, fun world of reading housing data.
So let’s look a bit at the methodology employed by the index. First, it uses the repeat sales method of tracking changing home values. In order to capture the true appreciated value of a property, it only tracks data on those units that have sold at least twice. The index looks to sales pairs to determine the price change for same property. Further, it tries to discount for changing sizes or quality by down-weighting price change observations that are dramatically out of whack with surrounding area comps. Lastly, and importantly, its focus is single-family homes. This means that the index (reporting on the Metropolitan Statistical Area of NYC, or MSA) does NOT include co-ops and condos.
What are the pros and cons of the CSI as it relates to NYC? Let’s start with the cons, first (numbers crunched from StreetEasy, Miller Samuel and Property Shark data).
Cons
- Does not include co-ops: Among purchase properties, co-ops represent approximately 65% of Manhattan inventory, yet represented 45% of sales that took place in 2009.
- Does not include condos: Among purchase properties, condos represent approximately 30% of Manhattan inventory, yet represented over 50% of sales in 2009. [Note: Shiller has created a condo-specific index recently, though these numbers are not incorporated in the CSI, itself.]
- One large chunk of data omitted: When we combine co-ops and condos, as a percentage of 2009 activity, they represent 99% of sales in Manhattan, 46% in Brooklyn, 42% in the Bronx and 40% in Queens. That’s quite a significant amount of data that’s in no way included in the CSI.
- Does not include new developments: New development sales (all of which are condos or condops) made up fewer than 20% of overall sales in Manhattan. Therefore, new construction in established markets or emerging market activity is not captured.
- Does not include multi-family dwellings: While less of a factor for Manhattan, multi-family dwellings are certainly material in number throughout the outer boroughs.
- Skewed towards volume: The theory goes that smaller homes tend to have more turns per period. Therefore, since larger homes turn less often, data will be skewed in either direction by when these larger homes do move.
- Excessively large footprint: The counties included in the NYC metro area are many. Should conclusions about Manhattan trends be made off a single-family home sold in Passaic, NJ?
- Connecticut: Fairfield and New Haven
- New Jersey: Bergen Essex, Hudson, Hunterdon, Ocean, Passaic, Morris, Monmouth, Middlesex, Somerset, Sussex, Union, and Warren
- New York: Bronx, Dutchess, Kings, Nassau, New York, Orange, Putnam, Queens, Richmond, Rockland, Suffolk, and Westchester
- Pennsylvania: Pike
Pros
- Same to same: It actually tracks changes in value of the same home over time. Average and median prices usually published on a quarterly basis are difficult to interpret based on the size and quality of what sold in that particular time period.
- Not including new construction in the mix is a good thing: Overall, new development units tend to be larger and more expensive, thereby skewing trends upwards the greater percentage of sales they make up. (The same goes for remodeling and house upgrade trends, nationwide.) Analyzing average sales prices alone may therefore provide a false market high because it does not fully reflect the cost of improvement. Furthermore, the fact that contracts were signed on average 12-18 months prior to closing makes it an excessively lagging market indicator.
- It bypasses the size issue altogether: Other methodologies of tracking housing values divide house prices by the number of square feet to create even comparison of homes by size. Yet, in NYC, as we’ve recently read, square feet are far from static and can vary quite significantly.
Takeaways
- A Manhattan market reflection it is not: The mere fact that 99% of the sales in Manhattan are excluded, along with about half of the sales in the outer boroughs, makes the CSI tangential, at best, in analyzing NYC trends. What is interesting to ponder upon, however, is the relationship between the single-family home sales and condo/co-op activity. To the extent that anomalous times require a more macro, national perspective, for as local as real estate is, the CSI serves as a valuable yet limited starting point for overlaying national trends on top of local NYC dynamics.
- Best for anticipating longer-term swings: While there is arguably little predictive value of this or any index for near-term conclusions, the index becomes much more useful for analyzing long-term trends. This is particularly the case during times when we are far away from historical norms and equilibrium levels, as we have been.
- Far from perfect yet still the most relied upon: Making sense of housing statistics is no easy feat. As such, every methodology will have its pitfalls, as does the CSI. That said, as imperfect as it is, the repeat sales method continues to be regarded as the gold standard. The key is to understand its strengths and limitations as you read the conclusions drawn from headline numbers.
- Laudable intentions: We must not forget the intentions behind the tracking and analysis of such data: adding transparency into the system and empowering all end-users to make better decisions, whether it’s buyers, sellers, agents or even the government as it looks to create policy. The great news is that the trend is heading towards more and better data, and we must recognize and encourage that … it’s just not perfect quite yet.
{ 12 comments… read them below or add one }
One can do basic hedonic regression analysis to account for all of the cons that you’ve noted to adjust the CSI–in particular, the condo/coop issue. You’ll find, as I (and I suspect they) have, that the index is highly predictive, which is the standard that ought to be applied. Even better, in places like Manhattan, is to control for rentals, in particular, rentals of the same apartments over time. I got a lease at the crest of the boom (circa mid 2007) and renegotiated at the crest of the bust (circa mid 2009) and wrestled the monthly rate down by 15%, which included a fresh painting of the entire apartment. In other words, this anecdote tracks nicely with the CSI.
(Psuedo) John Hicks, PhD
Thank you for your comment, John. Perhaps the correlation may be there … though that doesn’t equate into causation. As for highly predictive piece, I would have to disagree. This, like any sales-related index, is both backward looking and lagging. In it of itself, I belive its predictive powers are limited, particularly over the short term. Over the long term it could carry more weight, since then we’d be looking at more macro dynamics at play.
I can’t say I understand your point with respect to rentals, as the index tracks matching pairs of single-family homes. Certainly you can say the index shows economic distress, therefore the landlord should take that into account; but you could point to many other indicators in your negotiations to make the economic point.
The correlation/causation issue is a bit of a dodge with regard to something like the CSI. Neither Case nor Shiller would claim that their index is estimated using a structural econometric model of housing that separately estimates demand and supply equations and then determines the endogenous price equilibrium. I think you make a very valid point about whether the index they have developed, which is based on observed market outcomes rather than structural market estimation, is relevant and valid for Manhattan. I was just making the simple point that the index can be adjusted using factors that are specific to Manhattan, at which point it, of course, ceases to be the CSI. As with most indices of this type, however, they are more (not less) predictive in the short run.
As an economist, I can tell you that the real estate bubble was the most obvious empirical manifestation I’ve observed in my professional life, especially in areas that are nothing but sprawl like Las Vegas. Such places essentially have an infinity supply of land, so values should almost perfectly track construction costs.
My little anecdote was supposed to a toss-off, so I shouldn’t even have made it.
Anyway, great blog. I will definitely add it my blog roll. (And sorry for any typos. My keyboard is acting up today.)
(Pseudo) John Hicks, PhD
Thank you for the kind words, John. Much appreciated! ( …and no worries, we’re never spelling sticklers with others – just ourselves)
Your point regarding the infinity supply of land is one that Shiller, himself, made: http://theapplepeeled.com/buyers/up-close-with-robert-shiller-part-2/ … we noted it in one of our past posts — indeed, values do track construction costs, even in more “space constrained” areas of the country.
As for your comment regarding the greater predictability of such indices in the short term, could you elaborate? I’m intrigued. (clearly I need to revisit my stat analysis class notes
)
John, great comments. Have to run and will get back and throw some thoughts out later, but in meantime, what is your blog? Great discussion Ana!
A very nice quote from Shiller. I wholeheartedly agree with his point about not using real estate to get filthy rich.
On the predictability point, I guess I should be more precise in my language. The point I wanted to make was the ability of the CSI (or frankly any index or statistical estimate) to forecast, in a time sense, housing prices typically diminishes the longer the time horizon over which it is used to forecast. This is because the forecast error tends to grow with time. Consider a naive forecast: tomorrow’s price is today’s price plus a small positive trend as well as a little bit of noise. (Technically, this is called a unit-root process with a trend or an integrated process of order one.) Even if the noise is, on average, zero over time (a customary assumption), the further out I go, the larger is my forecast error because, put simply, the noise comes to dominate. The general principle holds even when we do something more sophisticated, like developing some type of model that gets away from the naive approach that tomorrow’s price is today’s price.
Anyway, the sun is shining! And I hope the links work.
(Pseudo) John Hicks, PhD
John, great comments. Have to run and will get back and throw some thoughts out later, but in meantime, what is your blog? Great discussion Ana!
Sadly, no blog. Because of my job,which isn’t in the area of real estate, I have to stay anonymous to the outside world. Of course, like many people in NYC, I avidly follow real estate issues. I can be reached at mrkeynesandtheclassics@gmail.com.
(Pseudo) John Hicks, PhD
Ah – yes … completely understood now. Of course, the nearer we are to the point of prediction (and the data that supports it), the more accurate the prediction. I personally still (and respectfully) disagree on the predictability capability of the index … looking forward to UrbanDigs’ comments on it
As a newby, I’m curious. You may have posted on this before, so forgive my ignorance. Suppose we take an arbitrary set of housing units in New York City with a variety of sizes, locations, amenities, and legal structures. Suppose this hypothetical set sells in 2000, sells again in 2006, and sells again today. The median sales price of this hypothetical set is X% higher in 2006 than in 2000, but Y% lower today than in 2006. I’m curious what you think are reasonable values for X and Y and how they compare to the CSI values for NY.
Oh, and I should note that I have nothing to do with Case, Shiller, their index, or whoever owns it at the moment. I’ve met Robert professionally once or twice.
It’s a great question, John. I’ve actually relied on others’ data for precisely this sense. It appears that, for the most part, we are back to 2006 levels right now. In fact, as per Miller Samuel via The Real Deal Data Book, the Manhattan median and average sales prices in 2000 were $399k and $710.8k, respectively, compared to 2006′s $799k and $1.22M, compared to 2009′s $810k and $1.23M. (I will make this note due to your background: 2000 data is year end, while 2006 and 2009 numbers are based on 4th Quarter numbers — perhaps a mild skew from 2000, but apples to apples for 2006 and 2009)
Great article and fascinating discussion. I believe, as John and Honeycrisp indicated, that there is a strong correlation between Manhattan RE and the CSI. Since we are dusting off stat books, I am curious about the betas between the CSI, Manhattan, and the boroughs. Do you have a sense for what they might be?