Download as doc, pdf, or txt
Download as doc, pdf, or txt
You are on page 1of 56

Living Costs in the world’s largest cities

Employee-compensation data, collected yearly by Mercer HRC, are used to analyze

living costs in the world’s largest cities. Nominal cost-of-living scores (those released

by Mercer), which are sensitive to currency levels, and real cost-of-living scores,

which are generated for this paper, are compared during the period 2000-2009.

While some fluctuations do occur, the annual city rankings exhibit considerable

stability on both indices over the 10-year period. Using pooled data, linear

regression models indicate that living costs in large cities reflect compensating

differentials for amenities. Specifically, large-city living costs are shown to be

determined by population size, world city-ness, local climate, national economic

development, and distance to world markets.

Keywords: Cost of living, largest cities, amenities, currency levels, Mercer

Prepared for Special Session at the North


American Meetings of the Regional Science
Association International, November 2009
Introduction

Each year a number of cost-of-living surveys are undertaken across the world’s

largest cities. While these surveys do generate a lot of international media attention, it

seems that the factors determining the differences in those living costs have not been

systematically analyzed. In fact, the topic of living costs is hardly mentioned in the well-

known urban collections assembled by Bourne and Simmons (1978), Dear and Scott

(1981), Knox and Taylor (1995), and Scott (2001), or in the volumes on world cities

prepared by Hall (1984) and Sassen (1991, 2000). Of course, these authors were largely

driven by other interests or agendas, where considerably more thought was devoted to

urban production than to urban consumption. In recent times, though, more attention has

been given to consumption issues, due in no small part to the research of people like

Glaeser (1998), Costa and Kahn (2000), and Glaeser, Kolko, and Saiz (2001). This

general shift has in turn renewed interest in the valuation of natural and human-created

amenities in urban environments (Brueckner, Thisse, and Zenou 1999).

Using employee-compensation data collected by Mercer Human Resource

Consulting (2009a), this paper analyzes household living costs across the world’s largest

and most important cities during the time period 2000-2009. In March of each year

Mercer undertakes a detailed price survey in 144 world-class cities, and a panel data set

consisting of surveys for 139 cities is available for all ten years.1 In each survey a basket

of goods and services, including housing, is identified that reflects the typical household

budget of a corporate employee. Place-specific price data are collected for each of these

items, and then prices are calculated for wider categories such as health or transportation.

1
An overall cost-of-living score is next generated for each place, where New York City is

assigned the benchmark value of 100 in each year.2 Mercer has offices in some 40

countries, so these data are widely trusted by private and public agencies that compensate

workers for cost-of-living differences existing both within and across nations. Being

valuable, these data (especially in detailed form) are expensive to obtain. While some of

the living-cost data examined in this paper were initially acquired directly from Mercer,

all of the overall city scores examined in this paper are now available online.3

Cost-of-living studies, much like quality-of-life surveys, are sometimes viewed

with suspicion because they presume there is a representative household, when it is

widely known that households are diverse in both composition and behavior. Some

studies are questionable because there is no guarantee that either the variety or the quality

of goods and services is the same everywhere (Hoch, 1972). Along similar lines,

adjustments often are not made for differences in government policy existing across the

surveyed places (Smith, 1994). Cost-of-living studies involving Third World nations pose

special concerns because so many transactions fail to take place in the formal economy;

in fact, there is much evidence that the size of the informal economy in part reflects

taxation policy (Ray, 1998). Finally, these problems tend to be compounded in

longitudinal studies because both the attributes and behaviors of economic agents change

over time.

In light of these issues, the online Mercer scores appear to be the best freely- and

widely-available indicators of relative living costs in the world’s largest metropoli.4

However, it is useful to be reminded that the expatriate employee targeted by Mercer

represents a very special type of household—one that is educated, white collar, mobile,

2
and relatively affluent. Although there are some indications, at least in the U.S., that the

rich and the poor are affected very much alike by price changes to their market baskets, it

would be hazardous indeed to extrapolate the findings of this paper to those other

socioeconomic groups residing in the world’s largest cities (Garner et al., 1996).

. The main purposes of the paper are fourfold. First, the nominal Mercer HRC

scores are transformed into real scores by taking into account the annual movements of

the appropriate national currencies. Second, both these nominal and real living-cost

scores are examined briefly for longitudinal stability and convergence. Then, third, the

pooled observations are used to estimate nominal living costs in the world’s largest cities.

As might be expected, the overall cost of living is shown to be determined by certain

city-specific attributes, such as population size, world city-ness, and location, and by

other nation-wide attributes, including level of development. Here it is shown that cities

with low (high) levels of natural and human amenities have relatively high (low) living

costs, a result that is known as compensating differentials. Finally, the various cities

surveyed by Mercer are grouped into different types based on the varying roles of these

attributes in determining the annual cost of living.

Background

Ever since Fuchs (1967) and Hoch (1972), U.S. analysts have recognized that

both the earnings and the living costs of urban households climb with increasing city

population size.5 Glaeser (1998), for one, estimated that a 1% change in population size

elicits a 0.10% increase in wages and a 0.16% increase in overall living costs for the

residents of America’s 50 largest cities. He provides other evidence that housing costs

respond faster to population size than do other urban living costs. Although the higher

3
wages of large U.S. cities are more than offset by higher living costs, households are not

leaving the largest cities in droves and seeking employment elsewhere. This suggests that

large cities must enjoy significant non-pecuniary advantages over small cities and rural

areas. Some of the linkages persisting between earnings, living costs, and city population

size are now briefly summarized.

Land Markets

Following Alonso’s (1964) pioneering work, Mills (1972) developed one of the

first comprehensive urban models, one that made numerical predictions within a closed

neoclassical context. Here he demonstrated that any increase in household wages leads

directly to an increase in the demand for housing, which in turn creates larger cities

having lower densities. Land rents and commuting costs are increased at all residential

locations, thus confirming that earnings and living costs are directly linked in the

capitalist city. However, Oates (1969) and Cadwallader (1993) provided alternative

evidence that urban property values are sensitive to a wide variety of factors in the local

economy. More recently, DiPasquale and Wheaton (1996) suggested that property values

are driven by city population size, recent city growth, and local construction costs. In any

event, observers now recognize that the relationship between urban earnings and overall

living costs is very complex, and that the strength of the relationship varies from city to

city depending upon local economic and demographic conditions, the behavior of local

government, and the incidence of human and natural amenities.

4
Externalities

5
Henderson (1974, 1988) modified Alonso by shifting interest to the production

externalities (information spillovers, lower transport costs) appearing in the city center

and the consumption externalities (social preferences, neighborhood quality) appearing in

the urban periphery. While the economies of scale enjoyed in production would be

industry-specific, the diseconomies arising with city size would be largely independent of

the city’s economic base. Consequently, cities tend to specialize in that narrow range of

export industries providing the strongest local external economies. Thus a national spatial

equilibrium evolves where cities of different population sizes exhibit different functional

specializations. Larger cities might specialize in business and finance activities while

smaller cities might specialize in manufacturing. However, the neoclassical logic dictates

that individuals living in these large places must on average enjoy the same net (real)

benefits as those living in small places, else they would relocate. Subsequent debate has

focused on specifying the precise nature and benefits of these spatial externalities. Jacobs

(1968), for one, argued that urbanization economies (firms in all industries) appeared to

be more important than localization economies (firms in similar industries) for long-term

employment and population growth (Glaeser et al., 1992). In any case, agglomeration in

space allows the marginal productivity of urban labor to rise, which in turn allows urban

households to earn higher nominal wages than their rural counterparts. But these urban

households then face higher living costs because land values are bid up everywhere and

daily commuting becomes more onerous. Moreover, continuing dissatisfaction with

uneven consumption externalities, reflected in poor schools and high crime rates, leads to

the relocation of households, either within the cities themselves or between cities. So

while nominal living costs certainly rise as individual cities grow in population size, the

6
real living costs experienced across the entire system of cities eventually converge on

some sort of spatial equilibrium.

Hedonic Models

Rosen (1974) is widely cited for advancing the hedonic approach in urban

analysis where implicit prices for location-specific amenities like climate and public

goods, and disamenities like crime and pollution, are separately estimated. Roback (1982,

1988) then extended this idea to include the combined effects of interacting land and

labor markets on the pricing decisions of urban households. The hedonic approach

ensures that the marginal prices of both natural and human-made amenities are related to

annual earnings, but where econometric allowances are made for geographically variable

housing costs. Households having different attributes are shown to pay an extra amount

for each of these different location-specific amenities, where prices are then combined in

order to impute quality-of-life differences across all cities. Those cities with the highest

quality of life are simply those places where the aggregate imputed prices are deemed to

be the greatest. Perhaps the best-known hedonic study is that of Blomquist et al. (1988),

who analyzed the 1980 variation of amenities both within and between major U.S.

metropolitan areas.6 Gyourko and Tracy (1991) extended the approach to include local

fiscal conditions, arguing that the differential quality of local government services was

usually neglected in such studies. Both papers concluded that quality of life was often

highest in those small- and medium-sized places found across the nation’s lower tier of

Sunbelt states. The hedonic literature is important to note because urban living costs are

seen to be associated not only with household earnings but with a host of other natural

and human-created factors; for a summary of this literature see Mulligan et al. (2004).

7
Productivity and Earnings

Today’s urban wage premium—which apparently remains greater than 30% in the

U.S.—exceeds other well-known wage gaps due to race or union membership (Glaeser,

1998). Standard thinking once suggested that city wages were necessarily higher because

there is a greater demand for most types of labor in urban markets. But then Rauch

(1993) borrowed the hedonic approach to provide a more convincing perspective. He

argued that cities with higher human capital, based on resident workers having more

education or better skills, should always enjoy higher wages—an expectation that he was

able to confirm. Subsequent debate has focused on clarifying the actual mechanism for

this human-capital effect, and remains basically split between those supporting wage

levels (through a variety of static externalities) and those supporting wage growth

(through temporal learning and efficient job-matching). Glaeser and Maré (2001) have

recently suggested that urban wage premiums are in fact some combination of both

effects. Young workers who come to large cities do not immediately receive the urban

wage premium, but they eventually enjoy higher wages than their non-metropolitan

counterparts because they are better able to upgrade their education or improve their skill

base. On the other hand, those people who leave large cities give up only a portion of the

wage premium they have built up over time. In short, there is increasing evidence that

average urban earnings are both tied to household demographics, especially age, and to

the differential abilities of cities to provide successful learning environments.

Shortcomings

The literature just addressed is largely restricted to the behavior of the average or

representative urban household. However, another more radical literature has given

8
attention to the unequal distribution of costs and benefits that arise in growing capitalist

cities (Cox, 1973, Harvey, 1973; Smith, 1977). The alternative view held here is that

households use their scarce resources to locate among a mosaic of overlapping and ever-

shifting externality fields. Nearness to both private (shopping centers) and public

(schools, hospitals) facilities is valued because this lowers the aggregate costs for urban

users, and thereby enhances their net benefits. As households continue to bid for this

multi-good accessibility, where goods have different externality gradients, urban land

costs typically rise at certain focal points. Over time a complex surface of net benefits

evolves across the ever-expanding city. Very large cities appear especially susceptible to

these inequalities because of their special needs for the services of casual or poorly-

trained labor (Fainstein, 2001). In fact, many factors—operating at the individual,

institutional, and wider cultural levels—work against the free spatial adjustment of the

global city’s land and labor markets. The stylized result is that the very large cities of

today are perhaps about the best size for the wealthy but they are certainly too large for

the poor.

Summary

Wages and salaries are higher in cities than in rural areas simply because urban

industries are more productive at the margin than their rural counterparts (Lloyd and

Dicken, 1977). Likewise, the industries of large cities are usually more productive than

those of small cities, so wages and salaries tend to climb regularly with city size. By their

very nature cities bring industries (suppliers) and workers (consumers) closely together in

space, thereby maximizing the density of transactions in both factor and final markets.

Moreover, cities not only offer a wide variety of employment opportunities on the

9
production side but they provide a wide array of both private and public goods on the

consumption side (Dicken and Lloyd, 1981). Large cities simply offer more of these

opportunities and goods than do small cities. So while it is widely agreed that human

well-being improves on average with increasing city population size, unfortunately the

resource gap between the best-off and the worst-off in capitalist societies also widens as

their cities become more populous (Smith, 1977).

In any case, the earnings and living costs of urban households are widely viewed

as being endogenous, because both phenomena grow together (at least in nominal terms)

as cities increase in population size. But the precise effect of increasing population size

on earnings and living costs varies from one place to the next in accordance with the

demography and lifestyles of urban residents, the varying infrastructure and technical

aspects of each city’s economic base, the differential quality of city governance, and the

list of location advantages (or disadvantages) enjoyed by each place (Haynes, 2006). In

fact, the current thinking is that no optimal city size actually exists—instead there is only

an efficient city size, one that depends in part upon what the city produces and where it is

functionally positioned in the wider urban system (Capello, 2004).

The Mercer Data

Aggregate Results

The upcoming discussion is based on two different cost-of-living data sets. The

first of these, referred to as the nominal data NOCOL, is comprised of the raw data

collected and distributed by Mercer each year. The second data set, referred to as the real

data RECOL, was created by making annual currency exchange-rate adjustments to those

nominal data. This adjustment alternatively could have been made with purchasing power

10
parity (PPP) figures but, for reasons that are soon disclosed, a currency-based adjustment

was chosen.7

*** TABLE 1 ***

Several methods exist for transforming the raw data but the preferred method is

one that is applied to all of the city observations as a pooled group and not separately to

individual cities over the study period. For this reason, ordinary least-squares (OLS)

regression was chosen where the nominal data NOCOL were estimated using three

different types of variables. The first of these is a city-specific variable that captures the

average value, AVCOL, of the nominal scores over time. The second variable, STCUR, is

a standardized currency variable that is calculated by forming the ratio between each

nation’s year-specific exchange rate with the U.S. dollar and that same nation’s average

exchange rate over the study period. This variable exceeds unity (outside of the U.S.)

when a national currency is weak and it falls short of unity when that national currency is

strong. Finally, a series of time dummies (using 2000 as the excluded year) is used, which

improves the annual estimates of NOCOL but does not disturb the annual rank-ordering

of cities due to the other two variables, AVCOL and STCUR. Then, as a concluding step,

the pooled regression estimates are transformed into the real scores RECOL by ensuring

that New York City’s index equals 100 each year.

The comparative results are shown in Table 1, where Model I is based on

purchasing power parity rates and Model II is based on currency exchange rates. The

variable STPPP is the counterpart to STCUR and can be interpreted in the same way. The

regression estimates are based on only seven years, 2000 through 2006, as PPP data were

not yet available for 2007. Of the two, quite clearly the currency-based model is superior:

11
note it has a smaller standardized regression error, 5.30 versus 6.92, and a larger adjusted

R-squared, 0.899 versus 0.829. Model 3 then extends the superior currency-based

approach to all eight years of the study period, where the estimates of this model are the

basis for generating the real cost-of-living scores RECOL.

*** TABLE 2 ***

Table 2 indicates that the year-specific mean of the nominal Mercer scores varies

considerably, ranging from a low of 69.40 in 2003 to a high of 81.87 in the most recent

year, 2007. In fact, the higher means during recent years reflects the steady weakening of

the U.S. dollar against many foreign currencies, a movement which has inflated living

costs (in $U.S.) for many white-collar employees living in those cities located outside the

U.S. Not surprisingly, the variation in the mean values for the currency-adjusted scores is

much lower, ranging only from a low of 71.59 in 2002 to a high of 77.42 in 2005.

There is also evidence that the within-year variation of the nominal scores has not

remained constant over time. The range (maximum minus minimum) is highest (132.80)

in 2000 and lowest (80.40) in 2006, where the annual figures suggest that variation in the

raw Mercer scores has diminished since 2001. However, when the real scores are

examined the longitudinal pattern is very different. Here the range increases from 83.44

in 2000 to 102.70 in 2004, but then declines to 95.16 in 2007. These findings are

confirmed by looking at either the standard deviation or, preferably, the coefficient of

variation. The latter statistic, which is often used as a preliminary descriptive indicator of

convergence, indicates that while the mean-adjusted variation in the nominal scores

certainly diminished over time, the mean-adjusted variation in the real scores remained

roughly constant instead.

12
Placed together, these two sets of figures provide only weak evidence that the

overall cost of living was becoming increasingly similar across the world’s largest cities

during the first few years of the 21st century. However, it is entirely possible that these

cities were alternatively sorting themselves into a small number of separate cost-of-living

clubs, and the entire issue of convergence calls for a much more thorough analysis (Barro

and Salai-i-Martin, 1992; Tondl, 2001). So those myriad forces commonly associated

with recent globalization—dramatic improvements in international communications,

enhanced levels of factor mobility, and greater volumes of international trade and

investment—appear not to have widened pre-existing inequalities in the compensation

levels received by urban-based corporate workers around the world (Economist, 2006;

Frieden, 2006).

*** TABLE 3 ***

The 140 cities surveyed by Mercer were then rank-ordered by living costs for

each year of the study period, first for the raw version and then for the currency-adjusted

version of the data. Spearman correlation coefficients were next computed across all pairs

of years, where the two sets of estimates are shown in Table 3. As was expected, the

inter-year rank correlations generally declined as the time lag was increased, indicating

that the difference in living costs between any pair of cities tended to widen with

intervening years. But some anomalies do occur, at least for the nominal scores.

Moreover, as was also anticipated, those lag-based declines were somewhat greater for

the nominal scores NOCOL than for the real scores RECOL. These results constitute

further evidence that the within-year variation of the nominal scores is greater than that of

the real scores. In addition, the Spearman correlation coefficient was calculated between

13
NOCOL and RECOL during each year of the study period. These coefficients ranged

between a low of 0.913 in 2007 and a high of 0.985 in 2003, providing evidence that the

association between the raw scores and the real scores weakened along with the U.S.

dollar during the later years of the study period.

Disaggregate Results

Table 4 shows the 15 cities with the highest cost-of-living scores during each year

of the time period 2000-2007. These high-end rankings are given for both the unadjusted

scores and the currency-adjusted scores. The main finding is that both annual lists tend to

be dominated by pretty much the same very large cities, especially Tokyo, Moscow,

Hong Kong, London, Seoul, and Osaka. Furthermore, among the big-four world cities,

Paris is conspicuously absent from several of those yearly lists.

*** TABLE 4 ***

14
More interesting, though, are the cost-of-living changes ∆NOCOL and ∆RECOL

that took place between 2000 and 2007. As for the nominal scores, the western European

capitals tended to climb the most, while many of the northern Asian cities tended to fall

the most. In the first group, places like Copenhagen (#1, 34.40), Madrid (#2, 28.10),

Rome (#3, 27.20), and Amsterdam (#5, 26.40) had nominal scores rise by more than 25

points, while in the second group places like Tokyo (#1, -42.80), Beijing (#2, -42.40),

Shanghai (#3, -35.90), and Osaka (#4, -35.20) had scores fall by more than 35 points!

However, the composition of these two groups is very different when the regression-

based real scores are analyzed. Now the largest increases were incurred in Central and

Eastern European cities like Bratislava (#1, 23.98), Prague (#2, 20.44), Budapest (#3,

17.03), and Warsaw (#4, 16.40), where real scores climbed by at least 15 points. As for

declines in real living costs, the greatest changes occurred in South America, the

Caribbean, and parts of Africa where places like Buenos Aires (#1, -34.83), Montevideo

(#4, -22.33), San Jose (#5, -20.71), and Kingston (#6, -16.09) experienced very

substantial drops. One obvious observation that follows from these trends is that broad

regionalization appears to characterize the short-run cost-of-living changes experienced

by global cities, where the most extreme shifts would seem to vary, geographically

speaking, depending upon whether nominal or real data are addressed.8

Cost-of-Living Estimates

The annual Mercer cost-of-living figures were next estimated in an explanatory

framework. In fact, a whole family of linear models was developed where step-by-step

improvements were made in explanatory power, based on the introduction of entirely

new constellations of variables. Moreover, this incremental approach sheds light on the

15
robustness of certain key variables, such as population size. In all cases the raw data

NOCOL were estimated using ordinary least-squares (OLS) regression procedures.

Descriptive statistics for all variables are available in the Appendix (see Table A1).

First, though, the issue of addressing annual fluctuations in the nominal data has

to be briefly revisited. Linear models, each with six time dummies, were designed for

both currency adjustments and purchasing power parity adjustments as before. But then a

new explanatory variable—city population size, POPUL—was introduced and NOCOL

was estimated in both models using only the 980 pooled observations for the time period

2000-2006 (recall the PPP figures were not yet available for 2007).

Once again the currency-based model proved to be much superior to the PPP-

based model: it exhibited both a higher adjusted R-squared (0.218 versus 0.136) and a

lower standardized regression error (14.79 versus 15.54). Then the model was re-

estimated with the outliers removed but the goodness-of-fit results changed very little. In

fact, this performance gap was maintained for all the more complicated explanatory

models that are outlined in the upcoming discussion.

City-Specific Attributes

Table 5 and Table 6 show the regression results from a family of models where a

pooled total of 1119 observations was used in each run. The first of these shares the same

properties as the currency-adjusted model just estimated, except that now the study

period covers the full eight years of the Mercer data. The two key explanatory variables

are the standard currency ratio, STCUR, and population size, POPUL. City population

size was certainly expected to have a positive impact on living costs. The annual

population levels for the 140 cities during 2000-2007 were simply interpolated from the

16
2000, 2005, and 2010 urbanization figures given by the United Nations (2004). POPUL,

measured in millions of persons, had a mean value 4.31 over the full study period.

Although city size appears to have a non-linear relationship with urban cost of living, its

effect is easier to interpret in a model restricted to linear variables. This very simple

model indicates that (at the mean) an increase of 1 million residents raises the raw Mercer

score by 0.92, so the difference between a city of 2 million persons and 7 million persons

would be approximately 4.6 nominal points due to population size alone.

*** TABLE 5 ***

The next variable considered is the degree of world city-ness, as proposed by the

Loughborough group (Beaverstock et al., 1999; 2000).9 This index captures the

importance of global cities in terms of corporate service provision, focusing on

accountancy, advertising, banking, and law. Although in some ways a fairly crude

instrument, it is now a very widely cited measure of private sector dominance. The index

creators argue that it not only addresses global city corporate attributes, but it also

captures those corporate linkages existing throughout global city system (Hall, 2001).

The variable, WCITY (mean = 3.05), which ranges in this study between a low of 0

(indicating an absence of world-city corporate attributes) and a high of 12 (representing

the 4 dominant Alpha world cities of London, New York, Paris, and Tokyo), was

expected to have a positive effect on living costs as high-order commercial relationships

worked through both the land and labor markets of world cities. While this variable is

significantly correlated with population size (Spearman coefficient = 0.465), not all large

cities are globally important and some small cities do have an important international

profile. The results from Model 2 indicate that world city-ness does have this expected

17
effect on living costs where, based on the magnitudes of the two t-scores, the strength of

its relationship with living costs is much greater than that of population size. It is worth

noting that the Model 2 estimate on POPUL is reduced by two-thirds to 0.302, indicating

that the estimate for population size in Model 1 is not robust at all.

Coastal location and capital city status were next introduced as dummies into

Model 3. Previous results from U.S. hedonic studies indicate that cities enjoying coastal

locations typically have higher living costs, either because households bid up residential

property values for coastal amenities or because businesses bid up industrial property

values for port locations. Similarly, capital cities typically exhibit superior human

amenities in the form of national cultural centers and have a much wider array of private

and public opportunities (especially in education, medicine, and entertainment), all of

which tend to bid up urban living costs. The estimates for Model 3 endorse the first

proposition but provide only weak evidence for the second proposition.

Local Climate

There is ample evidence from various hedonic studies that climate variation

should be accounted for when estimating cost-of-living differences among world cities.

In this study, data were collected for seven variables: mean summer temperature SUMAV,

mean winter temperature WINAV, days over 90 degrees F OVR90, days under 32 degrees

F UND32, extreme temperature ratio EXTRE (proportion of days over 90 or below 32), a

summer temperature regression residual RESID, and annual rainfall YRAIN. The first two

variables address seasonal mildness, the next three address extreme temperature

conditions, and the fourth also addresses extreme conditions by controlling summer heat

for winter conditions.10 Data were also collected for elevation ELEVA, which can affect

18
climate in different ways depending upon city location. Unfortunately, reliable data for

humidity and wind speed were not available for all places at all times.

These climate variables were then factor analyzed in order to create orthogonal

components. A Varimax rotation procedure was used and the scores for five independent

factors, explaining 98.8% of the variance, were retained. Table A2 in the Appendix shows

the most significant loadings for each of the five factors, which assist in arriving at the

following interpretations: CLMF1: winter temperature (low scores: Montreal, Moscow;

high scores: Accra, Kingston); CLMF2: summer temperature (low scores: Glasgow,

Quito; high scores: Kuwait, Tehran); CLMF3: extreme summer conditions (low scores:

Beirut, Brisbane; high scores: Bangkok, Madras); CLIMF4: annual rainfall (low scores:

Cairo, Manama; high scores: Hong Kong, Panama); and CLIMF5: elevation (low scores:

Seattle, Vancouver; high scores: Denver, Mexico City).

Model 4 demonstrates the improvement that is made to the earlier city-attribute

models when climate variation is included in the regression estimation. Note first that the

adjusted R-squared improves by some 36%, rising from 0.354 to 0.482, and the standard

error of the estimate improves by some 10%, declining from 13.36 to 11.96. Four of the

five factors are highly significant (0.001 level) and the other, Factor 3, just fails to be

significant at the 0.10 level. Taking into account both the signs of the factor scores and

those of the regression coefficients there is strong evidence that NOCOL is bid up higher

in those cities with (1) very cold winters, (2) very hot summers, (4) high levels of rainfall,

and (5) lower elevations (even when controlling for coastal locations). It seems that

Factor 3 simply recalibrates the temperature-driven effects of Factors 1 and 2, in order to

19
avoid double-counting in those places that are blessed with mild temperatures both in the

winter and in the summer.

National Development

National development was next addressed by collecting data on an array of

economic, demographic, and geographic variables: land area LAREA, population size

NAPOP, population density POPDN, birth rate BRATE, death rate DRATE, population

growth rate POPGR, gross domestic product NAGDP, GDP per capita GDPPC, GDP

density GDPDN, GDP growth rate GDPGR, human development HUDEV, GNP divided

by GDP NPODP (a measure of net foreign investment), trade openness OPENS,

importance of government in the economy GOVER, and importance of informal

transactions INFOR. Nearly all of the data pertain to the year 2003. These data were then

factor analyzed, again allowing a reduced set of variables to be introduced into the

regression estimation of the raw Mercer scores. As before, a Varimax rotation was used

to extract 7 orthogonal factors from the original 15 variables. The more important

loadings, which assisted in making the various factor interpretations, are shown in Table

A3.

*** TABLE 6 ***

Model 5 then includes these nation-specific development variables along with the

city-specific variables discussed above (Table 6). This marginally improves both the

standardized regression error, as it falls to 11.77, and the adjusted R-squared, as it rises to

0.499. The regression estimates indicate that five of the seven factors appear to play a

significant role in influencing urban cost of living: (factor 1) overall human development,

which differentiates nations by the quality of their human capital; (2) geographic size,

20
where large nations typically enjoy greater resources and more economic efficiencies

than small nations; (3) overall density, which determines the levels of both land rents and

housing costs; (4) age structure of the population, where youthful nations usually have

greater shortages of private and public infrastructure; and (6) recent economic growth,

again implying a variety of shortages in those nations that are growing rapidly. Based on

the signs of the factor scores and those of the related regression coefficients, each of the

five significant cost-of-living effects conforms to prior expectations. Greater levels of

compensation are apparently required for urban corporate employees living in countries

that have (1) a high level of development (affecting workers living in places like Brussels

and Sydney), (2) a very small economy (Bandar, Douala), (3) a very dense economy

(Hong Kong, Taipei), (4) a relatively youthful population (Dubai, Singapore), or (6) a

high rate of economic growth (Hanoi, Kuwait City).

Location

It has long been recognized that city location can be viewed in both absolute and

relative terms. Absolute location is captured by using global coordinates and relative

location is captured by considering distance to major world markets. In the former case,

latitude and longitude have been used in empirical studies not only to capture local or

regional processes but also to address place-specific unobserved effects (Mulligan, 2006).

The impact of coordinates, though, must be interpreted with some caution because their

inclusion typically diminishes the significance of otherwise key variables in the

regression estimation. As for relative location, the approach follows Gallup, Sachs, and

Mellinger (1999), who measured nearness to world markets by taking aggregate great-

circle distances to three of the four alpha cities—London, New York, and Tokyo. Now,

21
however, shortest distance to one of four global cities—the three cited above but also Los

Angeles—is used in the regression estimation. This lends more of a balance to the

relative importance of the North Atlantic and North Pacific basins in shaping global

living costs. The prior expectation is that any increase in distance (measured in thousands

of kilometers) to the nearest world market will induce lower economic rents, thus serving

to reduce white-collar compensation.

Model 5 only introduces relative location while Model 6 includes absolute

location as well. Table 6 indicates that the goodness-of-fit statistics are once again

improved by adopting location variables in the estimation. Model 6’s standard estimation

error is 11.23 and its adjusted R-squared is 0.544, while Model 7’s standard estimation

error is 10.80 and its adjusted R-squared is 0.578. The Model 7 figures represent

improvements of more than 8% and 15%, respectively, on the counterpart statistics of

Model 5. The standardized regression error is approximately 14.3% of the mean of the

dependent variable, NOCOL, making the estimation an entirely satisfactory one

according to the widely applied 15% rule-of-thumb figure (Pindyck and Rubinfeld,

1991). The results of both models indicate that minimum distance to world markets does

indeed have a strong negative impact on the living costs of large cities. As for absolute

location, latitude has a weak positive effect and longitude a strong positive effect on

urban living costs.

Model 8 is the partially-specified version of Model 7. This model is more

parsimonious because it removes eight variables that failed to be statistically significant

(at the 0.05 level) in the earlier model. Evidently, two time dummies in the middle of the

study period, the status of being a national capital, one climate factor, and three

22
development factors do not play a highly significant role in affecting living costs in

global centers. Moreover, this model exhibits remarkably little collinearity because so

many of the variables are designed to be orthogonal. In fact, only distance to world

markets poses a possible problem but its variance inflation factor is 2.13—well below the

value of 5 used by many analysts to indicate the presence of a collinearity problem

(Griffith and Amrhein, 1997).

The estimates of the partially-specified model are worthy of some discussion.

First, the estimates for POPUL and WCITY are 0.343 and 1.484, respectively, where the

t-score on world city-ness is nearly three times that on population. Coastal location

remains an important determinant of urban living costs but being a capital is not

important once city location is accounted for. Each of the climate factors retains the same

sign in Models 5-7 but each of the effects is generally diminished compared to Model 4.

Nevertheless, only rainfall turns out to be sufficiently unimportant to be dropped from

Model 8. However, in Models 5-7 there is some instability in the signs of the national

development factors and the significance levels of these variables are clearly affected by

the introduction of global coordinates. In fact, only three of the seven development

factors are sufficiently important to be retained in Model 8. Outside of the currency-

exchange ratio, the variables WCITY and LONGI have the highest t-scores in the

partially-specified model. In fact, the mix of variables retained in Model 8 suggests that

the living costs of the Mercer cities depend at least as much on city location as they do on

the more recognizable city-specific attributes like population size and commercial

importance.

Other Comments

23
Unfortunately, the various estimates for Model 8 in Table 6 do not convey much

useful information about effects because the variables are measured in different units. To

get a more accurate picture of the relative importance of each regression variable, the

various Beta coefficients must be scrutinized. Those variables with the largest Beta

estimates are the year-specific standard currency ratio STCUR (-0.303), world city-ness

WCITY (0.297), distance DIST4 (-0.254), and longitude LONGI (0.251). The second of

these estimates indicates that a 1 standard deviation increase in world corporate

importance increases living costs by 0.30 standard deviations (once location is accounted

for). The other Beta coefficients worth noting are associated with latitude (0.202), city

population (0.102), and the two climate factors, CLMF2 (0.124) and CLMF3 (-0.092).

So, contrary to Model 1, once other key variables are included a 1-standard-deviation

increase in city population of 4.9 millions, which is considerable, induces only a modest

increase of 1 point in the raw Mercer score. The (absolute) values of the Beta coefficients

for the remaining climate and development factors all range between 0.50 and 0.90,

indicating that these factors are not so important in determining living costs as are the

currency exchange rate, city location, and world city-ness.

An alternative log-linear specification was also estimated, which directly

generates estimates of elasticities for each of the variables. However, in all cases the

goodness-of-fit statistics of the simpler, linear model proved to be marginally superior to

those of the log-linear version; for example, the adjusted R-squared of 0.578 in Model 7

declined to 0.548 in its log-linear specification. Nevertheless there are some properties of

the alternative model that are worthy of mention. To begin with, coastal location is

dropped and one other national development factor, DEVF4, is added at the 0.05

24
significance level. All three of the location variables become more important with the

logarithmic transformation and now, except for the currency ratio, LATIT has the highest

elasticity estimate (0.478) of all. Moreover, the relative importance of world city-ness

and population size shift somewhat, where the elasticity for WCITY (0.013%) is now less

than twice that of POPUL (0.008%). In any event, at the international level the effect of

population size on living costs evidently is much lower than the national figure of 0.16%

estimated for the U.S. (Glaeser, 1998). Generally, though, the alternative specification

endorses the very same variables that were selected for the partially-specified linear

model.

Classification

The estimates for Model 8 in Table 6 were next used to generate a cost-of-living

typology for the Mercer cities where the various city groupings reflect in-group

similarities and between-group differences in the underlying composition of urban living

costs. This typology was generated by first recognizing the five major living-cost

categories of Model 8: (category i), comprised of three city-specific attributes; (ii) four

climate factors; (iii) three national economic development factors; (iv) three location

variables; and (v) year-specific effects, based on the currency exchange ratios and five

time dummies. The next step was to multiply the observation-specific values by their

corresponding estimates in Model 8 and then sum the correctly signed figures over all

variables in that living-cost category. So, to take an example, Tokyo’s category-wide

score for its city-specific specific attributes was 0.343*35.68 + 1.484*12 + 1.987*1 =

32.01. This figure, when added to the regression constant and the other four category-

25
wide scores, leads to the following estimate for Tokyo’s overall living costs in 2007:

108.11 + 32.01 + 3.65 – 1.51 + 12.35 – 35.30 = 119.31.

*** TABLE 7 ***

Once all of these category-wide scores were computed for each of the 1119

observations, the raw scores were all transformed into z-scores. So Tokyo’s raw score of

32.01 became the very high z-score of 4.11, and that city’s other category-specific z-

scores were, in order: 1.21, -0.76, 1.84, and -0.63. Next, Ward’s clustering algorithm was

then applied across all of the standardized observations. This algorithm is noted for

generating typologies whose groups have approximately the same numbers of members.

After some experimentation a final classification of the Mercer cities was produced.

Given that two of the cost-of-living categories are not time invariant, there was a certain

amount of shifting evident in the overall scores over the study period. As a result, cities

could migrate across the various hierarchical groupings, which meant that providing the

results for just one year was perhaps the best way to summarize matters. So the upcoming

discussion is specific to the 2007 cost-of-living classification. The group averages for

each of the five category means (measured in z-scores) are shown in Table 7.

The year-specific classification shown in Table 8 has seven groups and the

general properties of each group are indicated by the mean scores across that group’s five

cost-of-living categories. There were n = 18 cities, comprising nearly 13% of the Mercer

total, allocated to Group 3. This group had a mean population size of 4.02 million and a

mean Mercer score of 77.31, both of which were slightly below the overall means of 4.53

and 81.87, respectively. Cities in this group, which included many in Australasia, had

their living costs bid down mainly due to extremely peripheral locations (z-score= -1.665)

26
relative to world markets. In comparison, the 16 cities of Group 2 had very high living

costs (mean = 97.04)—all of these cities are large in population size (averaging just under

10 million people) and they exhibit world dominance in corporate matters. In part Group

1 cities (many found in the Middle East) had their living costs bid up because of bad

climates, but the declining U.S. dollar led most of the American cities to be allocated to

this group as well. Group 7 cities had their living costs bid up due to prevailing national

economic conditions, most notably recent economic growth. During 2007, currency ratios

played a very different role in two of the city groups. Those cities in Group 4, which

included many West European capitals, had their living costs rise because their national

currencies continued to strengthen against the U.S. dollar; however, in the cities of Group

6 living costs fell because those national currencies significantly weakened against the

benchmark currency. Finally, the cities of Group 5 had low living costs largely because

of low levels of national development.

*** TABLE 8 ***

Conclusions

This paper has analyzed overall living costs in 140 of the world’s largest cities

during the years 2001-2007. Using data generated each year by Mercer HRC, both

nominal and real (currency-adjusted) levels of living costs were examined. Mixed

evidence existed for global convergence in urban living costs during this relatively short

time period. In any event, the annual fluctuations in living costs were higher for the

nominal scores than for their real, currency-adjusted, counterparts. Urban cost of living

was shown to be a function of various constellations of variables, including several city-

specific properties, attributes of local climate, features of national economic

27
development, location, and year-specific currency-exchange levels relative to the U.S.

dollar.

Significant improvements might be made to the paper’s estimation of living costs

by incorporating other factors into the regression models. Positioning in various global

networks, including airlines and telecommunications, would be especially interesting to

include as explanatory variables (Hall, 2001; Derudder and Witlox, 2005). It might also

be informative to incorporate other Mercer data regarding urban quality of life, and

ascertain just how cost of living and quality of life are interrelated over time and through

space (Mercer, 2007a). Perhaps the bi-directional adjustment model, now very popular in

regional science, could be adopted to undertake this task (Mulligan et al., 2004).

Furthermore, the model developed in this paper could be re-estimated using the recently

generated MasterCard index of world-city commercial importance. While use of these

newer data would restrict the analysis to 50 cities, or 400 pooled observations, the

alternative index would permit attention being given to the various dimensions of

commercial performance, including knowledge creation and ease of doing business.

These new perspectives might also assist in a more rigorous analysis of global

convergence in living costs. Enhanced globalization should equalize rewards and costs

across nations but exacerbate disparities within nations. One approach would be to

examine the pair-wise qualitative shifts in living costs that took place during the study

period (Webber et al., 2005). A second approach would be to construct a formal -

convergence regression model, thereby addressing the effects of different city-specific

and countrywide variables on international convergence in global-city living costs (Barro

and Salai-i-Martin, 1992; Tondl, 2001). It is quite possible that a small number of living-

28
cost clubs are now emerging, groups that might even be similar in composition to those

identified in the city typology outlined at the very end of this paper.

29
References
Alonso, W (1964) Location and Land Use. Harvard University Press, Cambridge MA.

Barro, R and Sala-i-Martin (1992) Convergence. Journal of Political Economy 100, 223-
251.

Beaverstock, J V, Smith, R G and Taylor P J (1999) A roster of world cities. Cities 16,
445-458.

Beaverstock, J V, Smith, RG and Taylor, PJ (2000) World city network: a new

metageography? Annals of the Association of American Geographers 90, 123-134.

Blomquist, G C, Berger, M C and Hoehn, J P (1988) New estimates of quality of life in

urban areas. American Economic Review 78, 89-107.

Cadwallader, M (1993) Inter-metropolitan housing value differentials: the United States,

1960-1980. Geoforum 24, 307-313.

Capello, R (2004) Beyond optimal city size: theory and evidence reconsidered. In Urban

Dynamics and Growth, (eds.) R Capello and P Nijkamp, pp.57-85. Elsevier, Amsterdam.

Cheshire, P C and Magrini, S (2006) Population growth in European cities: weather

matters—but only nationally. Regional Studies 40, 23-37.

Cox, K (1973) Conflict, Power and Politics in the City: A Geographic View. McGraw-

Hill, New York.

Derudder, B and Witlox, F (2005) An appraisal of the use of airline data in assessing the

world city network: a research note on data. Urban Studies 43, 2371-2388.

Dicken, P and Lloyd, P E (1981) Modern Western Society. Harper and Row, New York.

DiPasquale, D and Wheaton, W C (1996) Urban Economics and Real Estate Markets.

Prentice-Hall, Upper Saddle River, NJ.

30
The Economist (2006) The New Titans: A Survey of the World Economy. September 16

issue, 30 pp.

The Economist (2007) Economics Focus: Misleading Misalignments. June 23 issue, p.

86.

Fainstein, S S (2001) Inequality in global city-regions. In Global City-Regions: Trends,

Theory, Policy (ed.) A J Scott, pp. 285-298. Oxford University Press, Oxford.

Frieden, J A (2006) Global Capitalism, W. W. Norton, New York.

Fuchs, V R (1967) Differentials in Hourly Earnings by Region and City Size. Columbia

University Press, New York.

Gallup, J L, Sachs, J D and Mellinger, A D (1999) Geography and economic

development. International Regional Science Review 22, 179-232.

Garner T I, Johnson D S, Kokoski M F (1996) An experimental consumer price index for

the poor. Monthly Labor Review 119, 32-42.

Glaeser, E L (1998) Are cities dying? Journal of Economic Perspectives 12, 139-160.

Glaeser, E L and Maré, D C (2001) Cities and skills. Journal of Labor Economics 19,

316-342.

Glaeser, E J, Kallal H, Scheinkman J and Shleifer A (1992) Growth in cities. Journal of

Political Economy 100, 1126-1152.

Graves, P E (1980) Migration and climate. Journal of Regional Science 20, 227-237.

Graves, P E (1983) Migration with a composite amenity: the role of rents. Journal of

Regional Science 23, 541-546.

Griffith, D A and Amrhein, C G (1997) Multivariate Statistical Analysis for

Geographers. Prentice-Hall, Upper Saddle River, NJ.

31
Gyourko, J and Tracy, J (1991) The structure of local public finance and the quality of

life. Journal of Political Economy 99, 774-806.

Hall, P (2001) Global city-regions in the twenty-first century. In Global City-Regions:

Trends, Theory, Policy (ed.) A J Scott, pp. 59-77. Oxford University Press, Oxford.

Harvey, D (1973) Social Justice and the City. Edward Arnold, London.

Haynes, K E (2006) Infrastructure: The Glue of Megacities. Megacities Foundation,

Amsterdam.

Henderson, J V (1974) The sizes and types of cities. American Economic Review 64, 640-

656.

Henderson, J V (1988) Urban Development: Theory, Fact, and Illusion. Oxford

University Press: Oxford.

Heston, A, Summers R, and Aten, B (2006) Penn World Table (Version 6.2). Center for

International Comparisons. University of Pennsylvania, Philadelphia.

Hoch, I (1972) Income and city size. Urban Studies 9, 299-328.

Hogan, T D and Rex T R (1984) Intercity differences in cost of living. Growth and

Change 15, 16-23.

International Monetary Fund (2007) World Economic Outlook Database. September

2006 edition. See www.imf.org.

Jacobs, J (1968) The Economy of Cities. Vintage Books, New York.

Knox, P L and Taylor, P J (eds.) (1985) World Cities in a World-System. Cambridge

University Press, Cambridge.

Kurre, J A (2003) Is the cost of living less in rural areas? International Regional Science

Review 26, 86-116.

32
Lloyd, P E and Dicken, P (1977) Location in Space, 2nd ed. Harper and Row, New York.

MasterCard Worldwide (2007) Worldwide Centers of Commerce Index. See

www.mastercard.com/us/company/en/wcoc/pdf/index_2007_us.pdf

McGranahan, D A (1999) Amenities drive rural population change. Agricultural

Economic Report #781. U. S. Department of Agriculture, Washington.

Mercer Human Resource Consulting (2007a). See www.mercerHR.com.

Mercer Human Resource Consulting (2007b). Cost of Living Press Release. June 18.

London.

Mills, E S (1972) Studies in the Structure of the Urban Economy. Resources for the

Future, Washington.

Molin, A and Mulligan, G F (2004) Geographic patterns in U.S. urban inflation: 1990-

2000. Yearbook of the Association of Pacific Coast Geographers 66, 1-19.

Mulligan, G F (2006) Logistic population growth in the world’s largest cities.

Geographical Analysis 38, 344-370.

Mulligan, G, Carruthers, J and Cahill M (2004) Urban quality of life and public policy: a

survey. In Urban Dynamics and Growth, (eds.) R Capello and P Nijkamp, pp.729-802.

Elsevier, Amsterdam.

Oates, W (1969) The effects of property taxes and local public spending on property

values: an empirical study of tax capitalization and the Tiebout hypothesis. Journal of

Political Economy 77, 957-971.

Oanda (2007) FXHistory: Historical Currency Exchange Rates. See www.oanda.com.

Pindyck, R S and Rubinfeld D L (1991). Econometric Models and Economic Forecasts.

McGraw-Hill, New York.

33
Rauch, J E (1993) Productivity gains from geographic concentration of human capital:

evidence from the cities. Journal of Urban Economics 34, 380-400.

Ray, D (1998) Development Economics. Princeton University Press, Princeton, NJ.

Roback, J (1982) Wages, rents, and the quality of life. Journal of Political Economy 90,

1257-1278.

Roback, J (1988) Wages, rents, and amenities: differences among workers and regions.

Economic Inquiry 26, 23-41.

Rosen, S (1974) Hedonic prices and implicit markets: product differentiation in pure

competition. Journal of Political Economy 82, 34-55.

Sassen, S (2000) Cities in a World Economy, 2nd ed. Pine Forge Press, Thousand Oaks,

CA.

Savageau, D (2000) Places Rated Almanac: Millennium Edition. Hungry Minds, New

York.

Scott, A J (ed.) (2001) Global City-Regions: Trends, Theory, Policy. Oxford University

Press, Oxford.

Showers, V (1989) World Facts and Figures, 3rd ed. John Wiley, New York

Smith, D M (1977) Human Geography: A Welfare Approach. Edward Arnold, London.

Smith, D M (1994) Geography and Social Justice. Blackwell, Oxford.

Tondl, G (2001) Convergence After Divergence? Regional Growth in Europe. Springer-

Verlag, New York.

United Nations (2004) World Urbanization Prospects: The 2003 Revision. United

Nations, New York.

United Nations (2005) Human Development Report 2005. Oxford, New York.

34
Webber, D J, White, P and Allen, D O (2005) Income convergence across U.S. states: an

analysis using measures of concordance and discordance. Journal of Regional Science

45, 565-589.

35
Table 1 Estimates for Real Cost of Living
________________________________________________________________________

Model

Variable I II III
________________________________________________________________________

Constant 1.229 38.686 37.814


(0.954) (22.28) (22.31)
AVMER 0.996 0.990 0.988
(66.14) (85.94) (87.71)
STPPP -0.369
(-1.72)
STCUR -37.140 -36.662
(-26.28) (-27.15)
DUM01 -5.401 -4.211 -4.212
(-6.52) (-6.63) (-6.46)
DUM02 -5.543 -2.513 -2.165
(-6.70) (-3.33) (-3.27)
DUM03 -1.756 -0.589 -0.670
(-2.12) (-0.93) (-1.03)
DUM04 2.239 1.084 1.021
(2.70) (1.71) (1.57)
DUM05 5.488 3.097 2.984
(6.62) (4.84) (4.55)
DUM06 4.236 2.497 2.359
(5.10) (3.92) (3.61)
DUM07 4.726
(7.19)

SEE 6.925 5.301 5.443


ADJ RSQ 0.829 0.899 0.893
________________________________________________________________________

Note. N=980 for Models I (PPP adjustment) and II (currency adjustment); N=1119 for
Model 3 (currency adjustment). All t-scores are shown in parentheses. The year 2000 is
the excluded year for the time dummies

36
Table 2 The Mercer Data: Overall COL Scores
________________________________________________________________________

Year 2000 2001 2002 2003 2004 2005 2006 2007


________________________________________________________________________

Nominal Data: NOCOL

Minimum 32.10 29.30 34.40 36.50 42.70 40.30 43.50 50.00


Maximum 164.90 134.00 124.20 126.10 130.70 134.70 123.90 134.40
Mean 74.96 69.53 69.40 73.17 77.15 80.38 79.06 81.87
SD 19.85 17.97 15.47 15.14 15.31 15.58 13.41 14.39
CV 0.264 0.258 0.223 0.207 0.198 0.194 0.169 0.176

Real Data: RECOL

Minimum 49.38 49.38 33.49 35.06 30.21 35.06 35.06 33.74


Maximum 132.82 129.62 124.09 128.41 132.91 133.65 129.36 128.90
Mean 74.98 73.77 71.59 73.87 76.15 77.42 76.72 77.38
SD 14.17 14.35 14.83 15.33 15.78 15.42 15.29 15.64
CV 0.188 0.195 0.207 0.208 0.207 0.199 0.199 0.202
________________________________________________________________________

Note. Statistics are based only on those 140 cities having data collected each year (one
observation, Harare, is omitted in 2007).The nominal data are made public each year and
the real data are based on currency adjustments.

37
Table 3 Inter-Year Correlations of Cost-of-Living Scores
________________________________________________________________________

Year 2000 2001 2002 2003 2004 2005 2006 2007


________________________________________________________________________

2000 1.000 0.957 0.847 0.815 0.674 0.582 0.650 0.545

2001 0.977 1.000 0.918 0.805 0.599 0.502 0.580 0.453

2002 0.938 0.970 1.000 0.785 0.541 0.445 0.543 0.402

2003 0.846 0.869 0.896 1.000 0.911 0.850 0.798 0.767

2004 0.768 0.774 0.783 0.962 1.000 0.965 0.855 0.888

2005 0.740 0.746 0.747 0.939 0.986 1.000 0.858 0.912

2006 0.769 0.779 0.787 0.950 0.984 0.991 1.000 0.931

2007 0.712 0.721 0.729 0.930 0.977 0.989 0.985 1.000


________________________________________________________________________

Note. Coefficients measure Spearman’s rank-order correlation. Figures above the main
diagonal represent the raw scores NOCOL and figures below the main diagonal represent
the currency-adjusted scores RECOL.

38
Table 4 The 15 most expensive Mercer cities: 2000-2007
________________________________________________________________________

Rank 2000 2001 2002 2003

________________________________________________________________________

1 Tokyo Tokyo Hong Kong Tokyo


Tokyo Tokyo Tokyo Tokyo
2 Osaka Moscow Moscow Moscow
Moscow Moscow Moscow
Moscow
3 Hong Kong Hong Kong Tokyo Osaka
Hong Kong Hong Kong Hong Kong Hong Kong
4 Beijing Beijing Beijing Hong Kong
Osaka Osaka Osaka Osaka
5 Moscow Osaka Shanghai Beijing
Seoul Beijing Beijing Beijing
6 Shanghai Shanghai Osaka Geneva
Beijing Seoul Seoul London
7 Seoul Guangzhou New York London
London St Petersburg Shanghai Seoul
8 St Petersburg New York St Petersburg Seoul
St Petersburg London London Shanghai
9 Guangzhou Guangzhou Seoul Zurich
Shanghai Shanghai New York New York
10 London Seoul London New York
New York New York St Petersburg St Petersburg
11 Taipei Hanoi Guangzhou Shanghai
Geneva Istanbul Guangzhou
Geneva
12 Shenzhen Taipei Hanoi St Petersburg
Taipei Geneva Geneva Zurich
13 New York London Shenzhen Oslo
Istanbul Guangzhou Hanoi Oslo
14 Hanoi Shenzhen Ho Chi Minh Hanoi
Zurich Hanoi Shenzhen Copenhagen

15 Singapore Ho Chi Minh Taipei Copenhagen


Hanoi Tel Aviv Taipei Guangzhou

39
________________________________________________________________________

Rank 2004 2005 2006 2007


________________________________________________________________________

1 Tokyo Tokyo Moscow Moscow


Tokyo Tokyo Tokyo Tokyo
2 London Osaka Seoul London
Moscow Moscow Moscow Moscow
3 Moscow London Tokyo Seoul
Hong Kong Hong Kong Hong Kong Hong Kong
4 Osaka Moscow Hong Kong Tokyo
Osaka Osaka Osaka Osaka
5 Hong Kong Seoul London Hong Kong
London London Seoul Seoul
6 Geneva Geneva Osaka Copenhagen
Beijing Seoul London London
7 Seoul Zurich Geneva Geneva
Seoul Beijing Beijing Beijing
8 Copenhagen Copenhagen Copenhagen Osaka
St Petersburg St Petersburg St Petersburg St Petersburg
9 Zurich Hong Kong Zurich Zurich
Shanghai Geneva Shanghai Shanghai
10 St. Petersburg Oslo Oslo Oslo
Geneva Shanghai New York Geneva
11 Beijing Milan New York Milan
New York Zurich Geneva New York
12 New York Dublin St Petersburg St Petersburg
Zurich New York Zurich Zurich
13 Milan Paris Milan Paris
Copenhagen Oslo Oslo Oslo
14 Dublin New York Beijing Singapore
Oslo Copenhagen Copenhagen Milan
15 Oslo St Petersburg Istanbul New York
Milan Milan Milan Copenhagen
________________________________________________________________________

Note. Rankings based on the raw scores NOCOL are shown on the top lines; rankings
based on the currency-adjusted scores RECOL are shown on the bottom lines.

40
Table 5 Cost of Living Estimates: I
________________________________________________________________________

Model

Variable 1 2 3 4
________________________________________________________________________

Constant 111.949 108.237 106.624 106.949


(28.70) (30.24) (29.26) (32.61)
POPUL 0.922 0.302 0.272 0.380
(10.41) (3.30) (2.96) (4.23)
WCITY 1.995 2.030 1.681
(14.67) (14.87) (12.57)
COAST 2.126 1.020
(2.61) (1.16)
CAPIT 0.885 2.364
(1.05) (3.06)
STCUR -40.187 -40.016 -40.026 -39.973
(-11.07) (-12.04) (-12.07) (-13.46)
DUM01 -4.150 -4.119 -4.117 -4.125
(-2.37) (-2.57) (-2.57) (-2.88)
DUM02 -1.948 -1.891 -1.886 -1.904
(-1.09) (-1.16) (-1.16) (-1.31)
DUM03 -0.726 -0.622 -0.616 -0.637
(-0.41) (-0.39) (-0.39) (-0.44)
DUM04 0.691 0.843 0.849 0.826
(0.39) (0.53) (0.53) (0.58)
DUM05 2.478 2.672 2.680 2.652
(1.41) (1.65) (1.48) (1.84)
DUM06 1.865 2.093 2.419 2.913
(1.06) (1.30) (1.48) (1.99)
DUM07 4.276 4.508 4.756 5.032
(2.42) (2.78) (2.92) (3.44)
CLMF1 -4.178
(-10.81)
CLMF2 2.590
(6.79)
CLMF3 -0.572
(-1.57)
CLMF4 1.247

41
(3.45)
CLMF5 -3.446
(-8.14)

SEE 14.63 13.39 13.36 11.96


ADJ RSQ 0.225 0.351 0.354 0.482
________________________________________________________________________

Note. N=1119. All t-scores are shown in parentheses. The year 2000 is the excluded year
for the time dummies. The 5 variables denoted by CLMF are city-specific climate factors;
see the text for clarification.

42
Table 6 Cost of Living Estimates: II
________________________________________________________________________

Model

Variable 5 6 7 8
________________________________________________________________________

Constant 107.411 112.313 107.124 108.113


(33.15) (35.91) (28.04) (30.10)
POPUL 0.539 0.454 0.311 0.343
(5.17) (4.56) (3.19) (4.14)
WCITY 1.400 1.478 1.503 1.484
(8.70) (9.62) (10.10) (11.55)
COAST 1.211 2.190 2.267 1.967
(1.38) (2.59) (2.79) (2.51)
CAPIT 1.678 1.041 1.276
(1.93) (1.25) (1.59)
STCUR -39.946 -38.955 -37.925 -38.423
(-13.66) (-13.96) (-14.11) (-14.43)

DUM01 -4.135 -4.163 -4.189 -4.243


(-2.93) (-3.09) (-3.24) (-4.00)
DUM02 -1.925 -2.007 -2.085 -2.112
(-1.34) (-1.47) (-1.59) (-1.95)
DUM03 -0.666 -0.681 -0.687
(-0.47) (-0.51) (-0.53)
DUM04 0.789 0.841 0.907
(0.56) (0.54) (0.70)
DUM05 2.607 2.697 2.808 2.697
(1.83) (1.99) (2.16) (2.52)
DUM06 2.612 2.461 2.645 2.085
(1.80) (1.78) (1.99) (1.96)
DUM07 4.727 4.610 4.830 4.372
(3.28) (3.35) (3.65) (4.09)
CLMF1 -3.874 -1.364 -1.073
(-7.03) (-2.36) (-1.87)
CLMF2 2.130 2.435 1.757 2.068
(4.86) (5.81) (4.29) (5.92)

43
CLMF3 -0.946 -1.081 -1.592 -1.528
(-2.51) (-3.00) (-4.31) (-4.21)
CLMF4 0.726 0.879 1.073 1.199
(1.91) (2.42) (2.83) (3.33)
CLMF5 -2.897 -2.025 -1.100 -0.933
(-6.55) (-4.71) (-2.59) (-2.27)
DEVF1 -1.356 -0.161 0.658
(-2.18) (-0.27) (1.10)
DEVF2 -0.859 -1.252 0.317
(-1.99) (-3.02) (0.73)
DEVF3 1.495 1.023 0.305
(3.89) (2.76) (0.84)
DEVF4 0.747 0.066 0.446
(1.73) (0.16) (1.09)
DEVF5 -0.009 0.425 -1.051 -0.997
(-0.02) (1.19) (-2.74) (-2.67)
DEVF6 1.382 2.419 1.458 1.492
(3.58) (6.35) (3.73) (3.93)
DEVF7 0.265 0.055 -0.864 -0.931
(0.72) (0.16) (-2.47) (-2.70)
DMIN4 -0.002 -0.002 -0.002
(-10.41 (-3.96) (-4.26)
LATIT 0.115 0.129
(2.85) (3.46)
LONGI 0.056 0.055
(8.98) (9.86)

SEE 11.77 11.23 10.80 10.81


ADJ RSQ 0.499 0.544 0.578 0.578
_______________________________________________________________________

Note. N=1119. All t-scores are shown in parentheses under the estimates. The year 2000
is the excluded year for the time dummies. The 7 variables denoted by DEVF are nation-
specific development factors; see the text for clarification. Model 8 is the partially-
specified version of Model 7 at the 0.05 significance level.

44
Table 7 Cost-of-living groups in 2007: means for z-scores
________________________________________________________________________

COL Category

Gr No Col Pop 1 2 3 4 5
________________________________________________________________________

1 35 80.1 4.4 -0.235 0.877 0.269 0.334 -0.456


2 16 97.0 10.0 1.992 -0.063 -0.139 0.388 0.159
3 18 75.7 4.0 -0.056 0.447 0.229 -1.665 0.157
4 33 86.0 2.0 -0.218 -0.429 -0.399 0.774 0.860
5 10 77.6 6.5 -0.331 0.091 -1.928 -0.237 -0.196
6 17 68.4 4.2 -0.646 -1.019 0.091 -0.914 -1.289
7 10 91.4 4.2 -0.116 -0.725 1.960 0.443 0.607

Total 139 82.1 4.5 0.000 0.000 0.000 0.000 0.000


________________________________________________________________________
Note. There are five cost-of-living categories: 1, city attributes; 2, climate; 3, national
development; 4, city location; and 5, currency ratio. Gr designates each group and No
indicates the number of cities assigned to each group. Col indicates the average living
cost (using NOCOL) and Pop indicates the average population of those assigned cities.

45
Table 8 Cost-of-living groups in 2007: the 50 largest Mercer cities
________________________________________________________________________

Group 1 Group 2 Group 3 Group 4

Shanghai Tokyo São Paulo Manila


Osaka Mexico City Rio de Janeiro Montréal
Beijing New York City Lima
Dhaka Los Angeles Santiago
Seoul Jakarta Sydney
Tianjin Paris
Istanbul Chicago
Tehran London
Miami Hong Kong
Ho Chi Minh City Toronto
Riyadh Milan
Washington Singapore
Guangzhou Madrid
Detroit Frankfurt
Hanoi
Houston

Group 5 Group 6 Group 7

Mumbai Buenos Aires Moscow


Delhi Cairo Lagos
Chennai Karachi Bangkok
Bangalore Bogotá St Petersburg
Lisbon

________________________________________________________________________

Note. Cities are arranged in descending order by their 2007 population. Places among the
20 most expensive Mercer cities (139 in all) are shown in boldface and places among the
20 least expensive Mercer cities are shown in italics. The groups are based on applying
Ward’s algorithm to the z-scores for five cost-of-living categories.

46
Appendix Tables

Table A1 Descriptive Statistics for Variables


________________________________________________________________________

Variable Mean St Deviation Minimum Maximum


________________________________________________________________________

NOCOL 75.71 16.62 29.30 164.90


POPUL 4.31 4.94 0.06 35.68
WCITY 3.05 3.32 0 12
COAST 0.54 0.50 0 1
CAPIT 0.51 0.50 0 1
STCUR 1.01 0.13 0.46 1.91
CLMF1 0.00 1.00 -2.16 1.79
CLMF2 0.00 1.00 -2.26 2.80
CLMF3 0.00 1.00 -1.65 3.02
CLMF4 0.00 1.00 -2.13 4.64
CLMF5 0.00 1.00 -0.94 5.09
DEVF1 0.00 1.00 -1.50 2.84
DEVF2 0.00 1.00 -2.76 2.03
DEVF3 0.00 1.00 -0.54 11.09
DEVF4 0.00 1.00 -3.97 2.24
DEVF5 0.00 1.00 -2.41 3.65
DEVF6 0.00 1.00 -2.90 4.84
DEVF7 0.00 1.00 -4.86 8.92
DMIN4 3.08 2.56 0.00 9.27
LATIT 26.77 26.04 -41.25 60.15
LONGI 9.94 75.45 -157.85 174.75
________________________________________________________________________
Note. Based on N=1119 pooled observations.

47
Table A2 Climate Factor Loadings
________________________________________________________________________

Variable CLMF1 CLMF2 CLMF3 CLMF4 CLMF5


________________________________________________________________________

SUMAV 0.502 0.806 0.253


WINAV 0.932 0.269
OVR90 0.619 0.338 0.690
UND32 -0.945 0.238
EXTRE 0.242 0.954
RESID 0.962 0.217
YRAIN 0.222 0.970
ELEVA 0.982
________________________________________________________________________
Note. Loadings are based on the Varimax transformation, where only those scores with
absolute values exceeding 0.20 are shown. Labels and percentages of the variance
explained for the five factors are: CLMF1: winter temperature (30.9%); CLMF2: summer
temperature (22.4%); CLMF3: extreme temperature (19.7%); CLMF4: annual rainfall
(13.1%); and CLMF5: elevation (12.7%). Data sources: www.worldclimate.com;
Showers (1989).

48
Table A3 National Development Factor Loadings
________________________________________________________________________

Variable NECD1 NECD2 NECD3 NECD4 NECD5 NECD6 NECD7


________________________________________________________________________

LAREA -0.219 0.786 0.228


NAPOP 0.671 0.379 0.442
POPDN 0.984
BRATE 0.963
DRATE -0.947
POPGR 0.909 0.387
NAGDP -0.220 0.769 0.469
GDPPC -0.668 0.266 -0.430 -0.263
GDPDN 0.985
GDPGR 0.949
HDEVI -0.868 0.305 -0.227
NPODP 0.989
OPENS -0.737 0.272
GOVER 0.928
INFOR 0.627 -0.391 -0.381

________________________________________________________________________
Note. Loadings are based on the Varimax transformation, where only those scores with
absolute values exceeding 0.20 are shown. Labels and percentages of the variance
explained for the seven factors are: NECD1: level of development (23.3%); NECD2:
overall economic size (16.3%); NECD3: density (16.1%); NECD4: age structure of the
population (9.2%); NECD5: level of government (9.0%); NECD6: economic growth
(7.8%); and NECD7: level of net foreign investment (6.8%). Data sources: Heston et al.
(2006), see pwt.econ.upenn.edu; United Nations (2004, 2005).

49
Endnotes

1. Mercer collects their yearly cost-of-living data each March and uses the exchange rates
on the first day of that month to convert local living costs to U.S. dollars. From time to
time Mercer changes its list of cities but normally a total of 144 places are surveyed each
year. A consistent panel set comprised of the same 140 cities was surveyed every year
during 2000-2007, except in 2007 when Harare was excluded.

2. Mercer collects information over thirteen major categories in computing overall living
costs: housing, education, business travel, food at home, alcohol and tobacco, domestic
supplies, personal care, clothing, home services, utilities, food away from home, local
transportation, and sports and leisure. Cost-of-living studies generally apply fractional
weights to these major categories when calculating an overall score. Mercer evidently
generates an updated system of international weights that is applied to all of the cities that
are surveyed each year. Unfortunately, the weights used by Mercer are not disclosed.
These weights often vary across different cost-of-living studies, making it difficult to
compare the overall scores compiled by different agencies. In any event, the implicit
assumption by Mercer is that expatriates adjust their spending habits to local conditions
(i.e., prices and goods availability) instead of maintaining their prior spending behavior.

3. Mercer (2007b) kindly provided data for the last two years of the study period, 2006
and 2007. The various websites (with appropriate survey years) used to acquire the
overall cost-of-living data for the other years were:

www.legal-recruitment.net/salary/living.asp (2000, 2001)


www.rebuz.com/research/0701research/worldwide-cost-of_living-survey-7-24-
01.htm (2000, 2001)
noctalis.com/nocturne/reality/col_globe.shtml (2001, 2002)
www.pensija. com/MHRC%20inf%20pranesimai/2004-06- 14%20Mercers
%20news.DOC (2003, 2004)
experts.about.com/q/France-140/Paris-expensive-London.htm (2003, 2004)
www.factbook.net/Cost-of_living.htm (2004)
www.why-thailand.com/cost_of_living.html (2003, 2004, 2005, 2006)
www.finfacts.com/costofliving.htm (2005, 2006)

Other useful websites exist showing only some of the cost-of-living data; in such cases
the scores might be confined to cities in a specific region or they might only list the
world’s 50 most expensive cities. Cross-checking was accomplished by comparing the
data from at least two websites in each survey year.

50
4. Mercer undertakes field surveys several times a year and also provides a direct city-to-
city living cost comparison that is frequently updated. Many observers comment on the
fact that Mercer administers very detailed questionnaires in the field where a range of
low- and high-price businesses are included in estimating item-specific prices. One other
agency, the Economist Intelligence Unit, provides an alternative international living-cost
index using many of the same cities as Mercer. However, the EIU data are not so freely
available on worldwide websites. Moreover, the EIU method does not include either
health expenses or housing costs in the overall index, although their prices are noted.
Both of these cost categories always comprise a large portion of urban living costs and
evidence exists that housing costs increase faster than overall living costs with increasing
city size. One measure of consistency that was undertaken involved estimating the rank
correlation for 17 of the U.S. cities. The 2000 Mercer scores were compared to the 1999
scores (adjusted for taxes) shown in the Places Rated Almanac (Savageau, 2000). The
Spearman coefficient was 0.535, indicating a highly significant association existed
between the two cost-of-living estimates.

5. In the U.S., cost-of-living data are regularly provided by the Bureau of Labor
Statistics, but these data only pertain to a small number of large cities. Moreover, most of
these cities are found throughout the nation’s traditional industrial heartland. In order to
address living costs over more places and a wider area, some social scientists have made
use of data regularly provided by the American Chamber of Commerce Researchers
Association (ACCRA). Hogan and Rex (1984), in examining some 230 cities, found that
1980 living costs were positively influenced by both population size and levels of
disposable personal income, although this relationship did not hold across all of the major
cost categories. Kurre (2003) later demonstrated that overall living costs in 1997 were
positively affected by population density, recent population growth, the cost of
government services, and the cost of utilities. Alternative specifications of his model
were given, some using population size as a predictor, but these had lower explanatory
powers.

6. Urban hedonic models were pioneered in the U.S. and nearly all applications have
been made there. However, it makes little sense to estimate these models unless there is
relatively free movement of labor and capital across space, so that compensating
differentials can be estimated in both land and labor markets. A closely-related literature,
pioneered by Graves (1980, 1983), argues that migration is an income-compensating
process that drives households to seek out areas with superior natural or human
amenities. This literature is again dominated by U.S. contributions, although Cheshire
and Magrini (2006) have recently applied the approach to European cities. Their findings
were that climate factors significantly affected city growth, but only within nations.

7. Purchasing power parity computes the number of units of a currency that are required
to purchase the same representative basket of goods that a U.S. dollar would buy in the
United States. It is based on the notion that, over the long run, exchange rates should
equalize prices across nations. More complex PPP models adjust for differences in
productivity or income, because prices are generally recognized to be lower in low-
income countries. However, this approach ignores such factors as current-account

51
properties and capital flows, so a number of alternative ways have been devised for
determining the “correct” value of a currency. One school of thought has a more
behavioral approach where past exchange rates are modeled for historic levels of
productivity, terms of trade, net foreign assets, etc., and then current values of these
variables are used to estimate a currency’s correct value. Due to this uncertainty, The
Economist (2007) claims that the brokerage firm Morgan Stanley now uses 13 different
models to value currencies. The data sources used were: purchasing power parity,
International Monetary Fund (2007); currency exchange rates, Oanda (2007).

8. In light of the findings of Molin and Mulligan (2004), this regionalization in living-
cost shifts is not surprising at all. In examining 170 U.S. cities during the 1990s, they
found that regionalization was especially apparent in the shifts of the various living-cost
categories like housing, health care, and transportation.

9. Other measures of city importance exist. One such index, addressing the so-called
Worldwide Cities of Commerce, has been recently developed by MasterCard Worldwide
(2007). Here an overall index is generated for 50 large cities, all of which are surveyed by
Mercer. This index ranges from 77.79 (London) to 39.50 (Moscow), and correlates very
highly (Spearman coefficient = 0.581) with the Loughborough team’s earlier index of
world city-ness. However, there are some advantages in using the older scores. First, the
overlap with the cities surveyed by Mercer is much higher—of the 122 cities targeted by
the Loughborough team, 91 (a full 75%) are found in the Mercer group. Second, given
present purposes the size distribution of the world city-ness scores is preferable—the
regression estimates of real living costs are not biased when zero values are assigned to
those nearly 50 Mercer cities that were not indexed by the Loughborough team.

10. A useful background to this approach is provided by McGranahan (1999).

52
Table 1 Annual Fluctuations in Nominal Costs: PPP versus Currency
________________________________________________________________________

Model

Variable I II III
________________________________________________________________________

Constant 75.678 76.959 115.191


(79.68) (40.09) (31.30)
STPPP -1.959
(-1.33)
STCUR -38.899
(-11.62)
DUM01 -6.013 -5.339 -4.106
(-3.66) (-2.78) (-2.25)
DUM02 -6.365 -5.770 -2.106
(-3.87) (-3.01) (-1.15)
DUM03 -2.261 -1.617 -0.472
(-1.38) (-0.84) (-0.26)
DUM04 1.680 2.384 0.967

53
(1.02) (1.24) (0.53)
DUM05 4.855 5.602 2.770
(2.95) (2.91) (1.51)
DUM06 3.583 4.371 2.233
(2.18) (2.27) (1.22)
DUM07 6.409 7.245 4.500
(3.90) (3.75) (2.45)
DUM08 9.241 10.137 4.800
(5.62) (5.23) (2.56)

SEE 15.824 15.986 15.193


ADJ RSQ 0.081 0.088 0.176
________________________________________________________________________

Note. N=1251 (9 years times 139 cities). All t-scores are shown in parentheses. The year
2000 is the excluded year for the time dummies

Table 2 Calculating Real Living Costs for Tokyo


________________________________________________________________________

Year Ex Rate St Ex Rate Mercer Costs Real Costs


________________________________________________________________________

2000 106.1 0.95 164.9 132.4

2001 116.1 1.04 134.0 125.1

2002 133.4 1.20 117.5 121.5

2003 119.9 1.08 126.1 127.5

2004 105.8 0.95 130.7 133.5

2005 103.5 0.93 134.7 136.2

2006 116.9 1.05 119.1 131.6

2007 118.4 1.06 122.1 133.5

54
2008 100.0 0.90 127.0 139.6

2009 97.9 0.88 143.7 135.6

GMean 111.3 131.5 131.5

________________________________________________________________________

55

You might also like