Civic Report No. 39 November 2003
Why is Manhattan So Expensive?
Endnotes
 The price data plotted in Figure 1 are for the metropolitan area as a whole. See the notes to the figure for the details.
 See Saiz (2003a,b) for estimates of the impact of immigration on rents.
 See Glaeser and Gyourko (2003) for evidence consistent with housing supply being very elastic in growing cities with new construction. Large population increases are not associated with large house price increases in most of these areas.
 It is noteworthy that these prices are not even adjusted for the age of the building. A rough accounting for the depreciation on older structures suggests that ‘as if new’ values would be from 2540 percent higher. In terms of the condominium sample, mean and median values are in the $600$625 per square foot range.
 Our research is part of a distinguished literature on the impacts of regulation on property markets. The interested reader should see Fischel (1985) for an extensive economic analysis and review of the likely motivations for and effects of zoning. Hamilton (1978) was among the first to view zoning as a tool of public sector monopoly power, and he also provided early empirical evidence supporting the contention that such restrictions could raise house prices throughout a given metropolitan area. Much of the other empirical work in this area focuses on the interjurisdictional effects of local land use controls on values. Katz and Rosen (1987) estimated that house prices were from 2040 percent higher in San Francisco Bay Area communities that had enacted growth moratoria or imposed growth management control plans. Speyrer (1989) and Pollakowski and Wachter (1990) followed with confirmation of significant capitalization of land use controls into property prices in other market areas. In addition, the impacts of Portland’s unique (for the US) growth boundary have been widely studied and discussed. See the recent article by Downs (2002) and the cites therein for a review of this market and an analysis of recent conditions. Finally, there have been a variety of studies of specific types of regulation, often with the goal of identifying influences on developers or a specific type of development. Fu and Somerville (2001) and Thornes (2000) are two recent examples.
 Of course, as Glaeser and Gyourko (2002) emphasize, there is no reason why housing prices cannot fall below construction costs in declining areas. In those places, there will be no new building in the market.
 Recall from above that the use of this term is for ease of exposition only. Government intervention that creates real limits to construction activity can take many forms. See Salama, Schill, and Stark (1999) for a recent study of the myriad zoning rules and other restrictions applicable in New York City.
 Condominiums and cooperatives in buildings with multiple units are excluded from this particular analysis.
 Two publications are particularly relevant for greater detail on the underlying data: Residential Cost Data, 19th annual edition, (2000) and Square Foot Costs, 21st annual edition (2000), both published by the R.S. Means Company.
 This assumption of modest, but not really low, quality is also reflected in our assumptions that the costs are for a onestory house with an unfinished basement and the average costs associated with four possible types of siding and building frame. In addition, we develop cost estimates for small (<1,500ft2), medium (1,5501,850ft2), and large (>1,850ft2) homes in terms of living area.
 Two adjustments are made to the AHS data before comparing house prices to construction costs. These are to account for the depreciation that occurs on older homes and to account for the fact that research shows owners tend to overestimate the value of their homes. To account for the latter factor, we follow Goodman and Ittner (1992) and presume that owners typically overvalue their homes by 6 percent. Empirically, the more important adjustment takes into account the fact that the vast majority of homes are not new and have experienced real depreciation. Depreciation factors are estimated using the AHS as follows. First, house value per square foot (scaled down by the Goodman & Ittner correction) in the relevant year is regressed on a series of age controls and metropolitan area dummies. The age data are in interval form so that we can tell if a house is from 05 years old, from 610 years old, from 1125 years old, from 2536 years old, and more than 45 years old. As expected, the coefficients on the age controls are each negative and represent the extent to which houses of different ages have depreciated in value on a per square foot basis. We then adjust the reported values to account for the estimated depreciation so as to compare the value of a unit as if it were new with its replacement cost. See our 2002 working paper for more detail on this procedure.
 Four different hedonic specifications were estimated. Generally, the specifications are of the following form: House Price = p*Land Area + z*Other Controls, where the model is estimated separately for each metropolitan area. The other controls include the number of bedrooms, the number of bathrooms, the number of other rooms, an indicator variable that takes on a value of one if the home has a fireplace, an indicator variable that takes on a value of one if the home has a garage, an indicator variable that takes on a value of one if the home is in the central city of the metropolitan area, an indicator variable that takes on a value of one if the home has a basement, an indicator variable that takes on a value of one if the home has central airconditioning, and the age of the home. Two of the specifications use the logged value of house price, while two are purely linear in nature. In addition, two of the models (one logged, the other not) use the data on interior square footage to capture the size of the home, while the other two use the detail on the number of bedrooms, bathrooms, and other rooms. In general, the results were quite consistent, although there is some variability in estimated land prices across the different hedonic models. The land prices in Table 1 are the mean of the values from the two specifications yielding the second and third highest prices (i.e., we rank the specifications from lowest to highest implicit land prices estimated and report the average of the prices from the second and third specifications). The key conclusions regarding the zoning tax are not sensitive to this choice. All the underlying results are available upon request.
 Our 2002 paper performed a similar analysis using the 1999 national file of the American Housing Survey. The results are similar quantitatively, with the coastal metropolitan areas having much higher marginal prices of land. However, the estimates here tend to be smaller and much more precisely estimated. Given the far fewer number of observations at the metropolitan area level in the national file, we are much more confident of the reliability of the results reported in Table 1.
 There are 43,560 square feet in an acre of land, so $0.13*10,890=$1,416 and $0.15*10,890=$1,634.
 We assigned a value of zero for the zoning tax if physical construction costs exceeded the average house price in the metropolitan area. In those cases, the computed zoning tax would be negative. We would not expect to see new construction in areas with production costs above market prices. In all areas but Philadelphia and Pittsburgh, a positive zoning tax results if we assume construction costs associated with the lowest quality units (i.e., economy homes) tracked in the Means Company data. Effective house quality may, in fact, be lower than we presume in these areas, but we think it preferable to employ assumptions that lead our zoning tax estimates to be conservative rather than aggressive.
 In general, we took every possible precaution to guard against our zoning tax estimates being biased upward. For example, we restrict the underlying samples from the AHS to homes with no more than two acres of land (i.e., with less than 87,120 square feet of lot). Including observations with larger amounts of land reduces estimated land prices below those reported in Table 1, suggesting that there is some downward bias of the hedonic estimates associated with large lot residences being developed in parts of the metropolitan area with cheap land. In addition, as discussed above, the hedonic estimate of land value will overestimate the free market value in markets with binding zoning constraints.
 Thus, the fee differentials are not due entirely to different maintenance levels or requirements.
 None of the prices reported in Tables 24 reflect values ‘as if new’. Ideally, these prices should be adjusted (upward) to account for the depreciation that occurs on older units before any comparison with replacement costs. Unfortunately, the condo sales data from the First American Real Estate Corporation do not include any information on the age of the buildings in which the units are situated. The NYCHVS does report age in discrete form, including the following five categories:
(a) built in the 1990s; (b) built in the 1980s; (c) built between 19601979; (d) built between 19301959; and (e) built prior to 1930. To try to estimate the impact of depreciation on older structures, we regressed reported condo value in the NYCHVS on a set of age dummies to determine the average impact of age on value. Adjusting the reported prices to reflect the regression estimates of depreciation yielded a mean of $625/ft2 and a median of $601ft2 for the 156 Manhattan condominiums in the NYCHVS, which is 2025 percent higher than the reported values. However, the R2 from the regression was only 0.04, not all the coefficients were statistically significant, and it was not the case that average values were monotonically lower the older the structure. Similar analyses with the cooperative unit and rental unit observations for Manhattan from the NYCHVS yielded slightly higher upward adjustments (in percentage terms), as the apartment sample in particular tends to be a bit older. However, these regression results also were somewhat imprecise. Given these uncertainties, we decided not to work with adjusted (i.e., higher) prices. While ‘as if new’ prices must be somewhat higher than the values we use, this decision is a conservative one in the sense that it guards against any upward bias in our estimates of the zoning tax.
 We also used the NYCHVS to examine the rental market and comment briefly on those results below. Because we only observe monthly rents on those units, their asset values must be imputed. We did so utilizing the socalled 'cap rate' methodology employed in the real estate industry. Excluding the small fraction of fully rent controlled apartments for which rents in no way reflect market forces, we transformed data on monthly rents for rent stabilized and truly free market units into asset values from the landlord's perspective. Specifically, the value of the rental unit was calculated as the annual rental payment adjusted for operating costs and divided by the capitalization (or cap) rate. [This is the inverse of the pricetoeamings ratio on a stock.] We assumed that operating costs were 30 percent of gross operating income, a figure that is standard among equity real estate investment trusts which own and operate apartments throughout the country. Based on some published data on apartment builders and owners in New York City and our own admittedly informal survey, we chose a cap rate of 7 pecent to convert the observed net rental flow into an asset price.
Empirically, the distribution of imputed values of rental apartments is somewhat cheaper than the values we found for owned units, but that should not be surprising. It often is the case that rental buildings are not as high quality as condominium structures, and standard agency issues may lead to more rapidly declining quality in rental structures. Nonetheless, the landlord's value of free market apartments typically is well above $300 per square foot, and 75 percent of free market rental units are estimated to be worth at least $202 per square foot. While there is not a thick upper tail of extremely high value rental apartments, the typical unit still is quite valuable. Even though we have tried to be conservative when making these estimates, the very need to impute market prices for rental units introduces measurement error that inevitably renders the results less reliable for this sector of the housing market. Hence, our focus on owneroccupied units.
 It is not clear from these data whether costs are higher or lower for taller buildings. The Means Company notes that larger buildings tend to have lower costs per square foot, primarily due to economies of scale and the decreasing contribution of the exterior walls. However, it is possible that buildings much taller than 24 stories could have higher costs since they may require stronger structural reinforcements and higher quality materials.
 One can obtain a slightly higher marginal cost by making different assumptions regarding the underlying cost function. For example, if we were to begin by assuming the average costs reported by Means pertain to the 2nd, 5th, and 15th floors (i.e., to the midpoints of the three height categories), and then presumed that the average cost function is quadratic and goes through those midpoints (i.e., $142,2; $144,5; $160,15), the total and marginal cost functions can be derived. In this particular example, marginal costs rise from around $140/ft2 on the lower floors to about $180/ft2 for floors 1015. To be conservative in our zoning tax calculations, we will use a value of $200/ft2 value for construction costs in the analysis below.
 As a point of reference, physical construction costs above $150/ft2 are at least as high as those pertaining to the highest quality singlefamily residences. In its single family construction cost files, the Means Company estimates that a large 2,200 square foot, custom quality home with an unfinished basement would have cost about $132/ft2 to build in the New York metropolitan area. The analogous figure for a luxury quality home, which is the highest quality category in the Means data, is $159/ft2. [No land costs are included in these figures.] Hence, the per square foot physical construction costs associated with the average Manhattan apartment are at least as high as those for the best quality single family structures anywhere in the metropolitan area.
 There are too few observations on units in newly constructed buildings to create statistically meaningful samples.
 As another source, we also turned to data from the New York City permit office. This office reports an estimate of the total value of structures put in place by builders along with the total number of units. Dividing the value per unit by the average square footage per unit from our condominium data yielded an average cost of $89/ft2. While this price is not totally implausible, especially if many of the units were somehow controlled or stabilized, it seems quite low. One explanation for this low value is that builders must pay taxes based on the value of construction they report. Therefore, they have an incentive to underestimate true construction costs. In any event, we take the permit data as further confirmation of our view that $200 per square foot is a reasonably generous estimate of construction costs for luxury apartment buildings in Manhattan.
 The picture is quite different for condominiums in boroughs outside of Manhattan. Only 20 percent of the condo units in the outer boroughs have values above construction costs. Clearly, the zoning tax does not have a large impact in these areas. Those results are available upon request.
 The analogous plot for rental units finds 75 percent of the apartments in the free market sector are valued above construction costs. Despite the fact that rent stabilized apartments are much cheaper, it still is the case that 42 percent of them have implied asset values in excess of construction costs. This result is particularly striking given that our estimate of construction costs is almost surely an overestimate for older, lower quality buildings. Details regarding our findings for the rental sector are available upon request.
 From the top row of Table 2, we know that the median condominium price is $455/ft2. Assuming $200/ft2 in construction costs, the implied zoning tax is $255/ft2. The ratio of 255/455 equals 56 percent.
 The results for the rental sector depend upon whether the unit is stabilized or not. In the free market sector, it is clear that the zoning tax is economically meaningful for the typical unit. For the median rental unit in that sector, which we estimate to be worth $325/ft2 to the building owner, the zoning tax of $125/ft2 amounts to nearly 40 percent of total value. There is no meaningful zoning tax for the typical rentstabilized unit, as the $168/ft2 price is less than our presumed construction costs. That said, 42 percent of these units have pricetocost ratios above one. Therefore, the magnitude of the zoning tax still could be meaningful for a significant fraction of units in this sector. In addition, it seems likely that construction costs for stabilized units generally will be below the $200 level we assume in our calculations.
 A more sophisticated analysis might compute the user cost of owning (e.g., see Poterba (1984) for example). This term includes factors such as local property taxes and maintenance expenses (along with a liquidity premium possibly) that also are thought to influence the cost of owning. For simplicity alone, we assume these costs are orthogonal to the zoning tax and exclude them from our analysis. To the extent they are positively correlated, our annual zoning cost estimates are lower bounds.
 As we write, the current interest rate on 30year, fixed rate mortgages is approximately 5.6 percent (depending upon factors such as points paid and the like) while inflation over calendar year 2002 was only 1.4 percent according to the annual average personal consumption expenditures chainprice index published by the Bureau of the Economic Analysis. The simple difference between these two figures is 4.2 percent.
 These results are sensitive to the values of the real cost of capital and true construction costs used in the calculation. However, absent much higher construction costs (e.g., $100 per square foot higher) or much lower real interest rates (e.g., Japanese levels of near zero), the annual cost to condominium owners is substantial. In addition, housing prices in part reflect expectations about future prices. If future prices are expected to rise dramatically, then this might explain some of the high prices in New York and would suggest that our simple calculation of multiplying the gap between housing prices and construction costs with the cost of capital overstates the true zoning tax. Of course, if by chance, New York City apartment prices were expected to fall (a not totally implausible possibility) then our calculation understates the true zoning tax.
 A similar calculation can be made for renters in the free market sector. Using the imputed asset values of rented units, we follow the same procedure outlined above to find the annual cost of the zoning tax. We then divide this figure by twelve so that it will be comparable to a monthly rental payment. The average burden on renters in the free market is estimated to be $156 per month, with a wide interquartile range running from $6 per month to $306 per month. Since many of the rent stabilized apartments had imputed asset values below our $200/ft2 cost estimate, we compute a negative zoning tax for a significant portion of this sample. Nonetheless, a full quarter of renters in subsidized units face a zoning tax burden of more than $84 per month. The figures for the rent stabilized sector are available upon request.
 We use lagged prices on the righthand side on the assumption that it takes some time for developers to be able to respond to price changes, even in an unregulated market. The house price series is for the New York metropolitan area. See the notes to Figure 4 for the details. There is not an analogous price series for Manhattan proper (i.e., for New York County).
 Both of the regressionline slopes are significantly different from zero at the 95 percent confidence level. However, the estimation error is large enough that we cannot confidently conclude the slope is greater for the 195569 period. All regression results are available upon request.
 While both slopes are slightly negative, we cannot reject the null that each is zero.
 The assumption of a fixed price for each square foot of space is purely for analytical convenience. All our key conclusions hold if (say) price is an increasing function of building height (which one might reasonably expect), as long as price is the same function of height for all buildings in the same neighborhood.
 This will not be exactly true if people have a taste for views, in which case, developers will build up beyond the degree that would be predicted by this condition.
 We focus on recently built structures because it is primarily since the 1980s that we believe limits on supply to have had a significant effect. The dataset includes too few structures that were built in the 1990s, so we cannot reliably use that time period.
 In principle, this will be true with heterogeneous consumers as well as long as there are enough apartments of differing types. If a new apartment blocks the view of someone who values his view more than the market does, this person can just move apartments and acquire a new (identical) apartment with a view paying the market cost of a view. Of course, we recognize that urban housing markets are rarely that fluid.
 Unit quality is not fully captured by controlling for unit size (and we do not have other variables in our data set). If interior quality is superior on higher floors, as we suspect is the case, our estimate of view is biased upward. We are largely unconcerned by this because our goal is not to provide a precise measure, but to determine whether views and other externalities could reasonably justify the gap between values and costs that we currently see in Manhattan. An overly generous estimate of the value of a view serves that purpose well.
 Because the indicator variable is not continuous, the derivative relating price to floor is not well defined. Using the adjustment suggested by Halvorsen and Palmquist (1980) for such cases yields the 25 percent figure.
 The upper bound on the loss occurs when a new apartment blocks one view completely. In that case, the value of lost views accounts for 25 cents per dollar of new construction.
 This sort of specification will generate the elasticity that we are interested in if, for example, U=W/P + A across cities, where U is reservation utility, W is wages, P is the price level (which will be proxied for with rents) and A equals the amenity level.
 The regression also included a control for the percent of residents over the age of 25 with less than 9 years of schooling.
 That said, other specifications indicated a smaller negative or even positive effect of population. For example, controlling for land area turns the population coefficient positive, although not statistically significantly so. [The coefficient on land area is significantly negative, as expected.] We also tried controlling for density directly and found denser areas had higher prices. At face value, this suggests that crowding is good. However, this probably reflects omitted variables having to do with supply constraints or local amenities. In addition, attempts to instrument for population and density in the crowding regressions yielded slightly positive coefficients on population. In any event, all the analyses we have done indicates that –5 percent is an upper bound on this particular externality.
 In a comprehensive survey of the literature, Hamermesh (1993) concludes that the elasticity of employment with respect to wages lies in the range of 0.15 to 0.75. These values imply an elastic response of wages with respect to employment.
 One would have to make an extreme assumption about the share of tradeable goods across areas to get no local output effects.
 Empirical work in this area is fraught with various potential specification biases, so accurately pinning down an impact is very difficult. For example, regression estimates of the populationwages relationship using crosssectional variation may be upwardly biased because more productive places (with higher incomes) end up attracting more people and, therefore, have larger populations. Regressions using changes in population and income may also be biased if increases in urban population are driven by increases to the productivity of the metropolitan area.
 This conclusion does suggest that there are benefits to the New York City budget from allowing new entrants into Manhattan. Such a detailed calculation is beyond the scope of this paper, but it is worth doing given the current fiscal situation in the city.
References
Benson, Earl, Julia Hansen, Arthur Schwartz, Jr., and Greg Smersh. “Pricing Residential Amenities: The Value of a View”, Journal of Real Estate Finance and Economics, Vol. 16, no. 1 (1998): 5573.
Brueckner, Jan. “The Structure of Urban Equilibria: A Unified Treatment of the MillsMuth Model” in Handbook of Regional Economics (Peter Nijkamp, Ed.). NorthHolland Press. Vol. 2, 1986: pp. 821854.
————————. “Growth Controls and Land Values in an Open City”, Land Economics, Vol. 66 (August 1990): 237248.
Ciccone, Antonio and Robert Hall. “Productivity and the Density of Economic Activity”, American Economic Review, Vol. 6, no. 1 (1996): 5470.
Fischel, William. The Economics of Zoning Laws. Baltimore: Johns Hopkins University Press, 1985.
Fu, Yuming and Tsur Somerville. “Site Density Restrictions: Measurements and Empirical Analysis”, Journal of Urban Economics, Vol. 49, no. 2 (2001): 404423.
Glaeser, Edward and Joseph Gyourko. “The Impact of Zoning on Housing Affordability”, NBER Working Paper No. 8835, March 2002.
——————————————————————. “Urban Decline and Durable Housing”, NBER Working Paper No. 8598 (revised version, 2003).
Glaeser, Edward and David Mare. “Cities and Skills”, Journal of Labor Economics, Vol. 19, no. 2 (2001): 316342.
Goodman, John C. and John B. Ittner. “The Accuracy of Home Owners’ Estimates of House Value”, Journal of Housing Economics, Vol. 2, no. 4 (1992): 339357.
Hamilton, Bruce. “Zoning and the Exercise of Monopoly Power”, Journal of Urban Economics, Vol. 5 (1978): 116130.
Halvorsen, Robert and Raymond Palmquist. “The Interpretation of Dummy Variables in Semilogarithmic Equations”, American Economic Review, Vol. 70 (June 1980): 47475.
Hamermesh, Daniel S. Labor Demand. Princeton: Princeton University Press, 1993.
Katz, Lawrence and Kenneth Rosen. “The Interjurisdictional Effects of Growth Controls on Housing Prices”, Journal of Law & Economics, Vol. 30, no. 1 (1987): 149160.
Marshall & Swift Valuation Service. Commercial Cost Explorer CD, January 2003.
Pollakowski, Henry and Susan Wachter. “The Effects of LandUse Constraints on Housing Prices”, Land Economics, Vol 66, no. 3 (1990): 315324.
Poterba, James. “Tax Subsidies to OwnerOccupied Housing: An Asset Value Approach”, Quarterly Journal of Urban Economics, Vol. 99, no. 4 (1984): 729752.
R. S. Means. Residential Cost Data, 19th Annual Edition, R.S. Means Company, 2000.
———————. Square Foot Costs, 21st Annual Edition, R.S. Means Company, 2000.
Rubinfeld, Daniel. “Suburban Employment and Zoning: A General Equilibrium Analysis”, Journal of Regional Science, Vol. 18 (1978): 3344.
Saiz, Albert. “Room in the Kitchen for the Melting Pot: Immigration and Rental Prices”, Review of Economics and Statistics, forthcoming, 2003.
_________. “Immigration and Housing Rents in American Cities”, Federal Reserve Bank of Philadelphia working paper, 2003.
Salama, Jerry, Michael Schill and Martha Stark. “Reducing the Cost of New Housing Construction in New York City”. Working paper, The New York University School of Law, Center for Real Estate and Urban Policy, 1999.
Thornes, Paul. “Internalizing Neighborhood Externalities: The Effect of Subdivision Size and Zoning on Residential Lot Prices”, Journal of Urban Economics, Vol. 48, no. 3 (2000): 397418.
 U.S. Bureau of the Census. American Housing Survey, various National and Metropolitan Area Files, 19981999.




