Do growth, investment, and trade encourage water use or water conservation

Lett Spat Resour Sci (2008) 1: 127–146DOI 10.1007/s12076-008-0013-5

O R I G I NA L PA P E R

Do growth, investment, and trade encourage water useor water conservation?

John P. Hoehn · Kwami Adanu

Received: 6 March 2008 / Accepted: 12 September 2008 / Published online: 2 October 2008© Springer-Verlag 2008

Abstract A two-sector, country-wide model of production, consumption, and regu-lation is developed to examine the linkages between water use and economic activity.The model indicates that water use is influenced by economic scale, composition, na-tional income, openness to trade, and climatic factors. The relationships described bythe model are tested using UNESCO data on water withdrawals and water consump-tion. Results indicate that increases in economic scale lead to increases in water usewhile trade intensity, capital investment, and national income reduce water use. Nu-merical simulations using the estimated results show that equal percentage increasesin scale, personal income, and capital investment lead to increases in water use. Suchincreases are readily offset by increased trade intensity. The simulations show thatimprovements in trade intensity have the potential to switch the sign on water usetrends from increases of 6 to 8 percent to reductions of 4 to 13 percent over thecourse of a decade.

Keywords Economic globalization · Resource use · Water use · Waterconservation · Trade

JEL Classification F18 · Q25 · Q28

1 Economic globalization and resource use: do growth, investment, and tradeencourage water use or water conservation?

Studies of water use and water consumption show that economic activity is a primarydeterminant of global water use (Vorosmarty et al. 2000). Understanding the exactrelationships between economic activity and water use, however, has proven difficult.

J.P. Hoehn (�) · K. AdanuDepartment of Agricultural, Food, and Resource Economics, Michigan State University, EastLansing, MI 48864, USAe-mail: [email protected]

mailto:[email protected]

128 J.P. Hoehn, K. Adanu

Past trends in some regions of the world imply that water use is positively correlatedwith economic scale. In Asia, water use shows a long-term positive trend with morerapid growth in use during economic expansions (Shiklomanov and Rodda 2003).In contrast, water use in both Europe and North America peaked in the early 1980sdespite continuing economic expansion (IHP 2005). These contrasting results suggestthat the relationship between water use and the economy may involve other variablesin addition to gross domestic product. Moreover, forecasts of future water use basedon such linear trends are poor predictors of future water use and usually overstateactual water use by substantial margins (Gleick 2000).

The structure and composition of different economies appear to have a strong in-fluence on water use. Agriculture accounts for two-thirds of global water use (Shik-lomanov and Rodda 2003) and regional water use is greater where economic share ofagriculture is larger. In Asia, agriculture accounts for 80 percent of water use and to-tal water use is growing. In Europe and North American, the commercial, industrial,and residential sectors are the primary water users and water use is stable or declin-ing (Shiklomanov and Rodda 2003). While these data link water use with economiccomposition, other economic factors, such as economic scale, openness to trade andpublic policy, may also be at work.

The analysis herein builds on recent research to model the relationship betweenwater use and economic activity. The model indicates that water use is influenced byeconomic scale, economic composition, per capita national income, trade intensity,and climatic factors. The derived relationships are examined empirically using UN-ESCO data on water withdrawals and water consumption. Results indicate that tradeintensity, capital investment, and increases in national income are water conservingwhile increases in economic scale increase water use.

2 Literature review

Grossman and Krueger (GK) (1991) ignited recent interest in the relationship be-tween natural resource use and country-level economic activity. GK examined therelationship between pollution levels and national income and formulated the hypoth-esis of an environmental Kuznets curve (EKC). Under the EKC hypothesis, pollutionlevels follow an inverted-U relationship with income; pollution levels rise with na-tional income at lower levels of per capita income, but peak and then decline withincreases in income at higher levels of income. While the EKC turned out to lackgenerality (Dasgupta et al. 2002; Harbaugh et al. 2002), the GK analysis led to sub-stantial theoretical and empirical work on the economic factors that influence the useof natural resources.

GK identified three economic factors—scale, composition, and production tech-nique—as the primary economic factors affecting natural resource use. Scale capturesthe idea that as an economy expands, resource use increases with additional economicgrowth. Composition refers to the relative size of different economic sectors within aneconomy. Composition affects resource use since resource intensity varies across sec-tors. Production technique refers to production processes used by firms. Productiontechnique matters since technologies differ in their input and resource requirements.

GK argued that all three effects—scale, composition, and technique—were func-tions of national income. According to their analysis, national income was a direct

Do growth, investment, and trade encourage water use 129

measure of the overall scale of an economy. Composition was also viewed as relatedto national income, since lower income economies are likely to produce differentgoods than high income economies. Production techniques were viewed as respond-ing to factor prices and environmental regulation. GK hypothesized that as incomerose, citizens would seek to consume more of both market and environmental goodsand put pressure on regulators to require less polluting production techniques. Themixing of these three effects of income led to the EKC hypothesis: the scale effectdominates at lower levels of income while at higher incomes the income-techniqueeffect dominates the scale effect, yielding the inverted-U relationship between na-tional income and pollution.

Initial empirical work by GK supported EKC (Dasgupta et al. 2002). Other re-searchers replicated the EKC with deforestation (Ehrhardt-Martinez et al. 2002;Koop and Lise 1999), air pollutants (Cole and Rayner 2000), water pollution (Gross-man and Krueger 1995; Lim 1997), and water consumption (Cole 2004).

Two types of problems emerged. First, the estimated inverted-U relationship be-tween resource damage and income was empirically fragile. Removing questionableobservations and including more recent data made the inverted-U unstable, and evendisappear (Harbaugh et al. 2002). Second, the emphasis on income obscured the sep-arate effects of scale, composition, and technique. There was a need to find specificvariables for measuring the scale, composition, and income-technique effects as wellas for incorporating other economic factors such as trade liberalization (Dasguptaet al. 2002).

Barbier (2004) made progress with the income and regulation issues using a singlesector growth model. Barbier modeled water as an input into goods production, withthe input price of water controlled by a regulatory authority. The model leads to aninverted-U relationship between water use and national income similar to the EKC.Barbier’s empirical results confirm the negative effect of income-induced regulatorydecisions on aggregate water use.

Openness to trade is a final economic factor potentially important to country-levelresource use (Copeland and Taylor 2004; Dasgupta et al. 2002). Openness to interna-tional trade allows production to shift to areas with lower production costs. Suchshifts potentially lead to efficiency gains and net water conservation, with origincountries experiencing a reduction in water use and receiving countries an increase inwater use. Sector-level analysis appears to confirm the net water conserving effectsof trade. De Fraiture et al. (2004) finds that trade liberalization reduces global wateruse in cereal production by more than 6 percent and net water use in irrigation by11 percent. Hoekstra and Chapagain (2008) detail how trade in virtual water shiftsuse from water scarce counties to those with greater water availability.

Copeland and Taylor (2003) and Antweiler, Copeland and Taylor [ACT] (2001)develop a model of resource use that identifies specific and separate economic vari-ables with the scale, composition, and income-technique effects. These analyses be-gin with a two sector, country-level model of production and trade where environmen-tal impacts are controlled by a regulatory authority. The country engages in interna-tional trade consistent with its comparative advantage and the trade frictions such astransportation costs and trade policies. The regulatory authority controls pollution inview of the tradeoff between goods output and environmental impacts.


ACT evaluates the empirical consequences of the model data for sulfur dioxide(SO2) pollution. The analysis confirms the effects of scale, composition, technique,and openness on pollution. SO2 pollution increases with economic scale and capi-tal intensity. Through income effects on the demand for environmental quality andregulation, increases in national income reduce SO2 pollution. Trade intensity alsoreduces SO2 pollution. Cole and Elliott (2003) obtain similar results, though the ef-fects of trade vary across different pollutants. Cole and Elliott find that trade intensityreduces SO2 and water pollution, increases carbon emissions, and has no significanteffect on oxides of nitrogen.

3 Theoretical framework

The theoretical model developed below adapts ACT two-sector, open economy modelto country-level water use. In the model, water use provides services as an input intogoods production and as a non-market environmental amenity. Such non-market ser-vices may include habitat, in situ services for plants, animals, and humans, transporta-tion, and aesthetic features. Water use in production results in reduced non-marketservices. As in the ACT model, the market versus non-market tradeoffs are controlledby an regulatory authority. The model’s central feature is that it identifies, as hypothe-ses, the relationships between water use and key macroeconomic economic variablessuch as scale, capital investment, income, and trade. The econometric analysis eval-uates these hypotheses and compares the estimated relationships with those of anenvironmental Kuznets curve.

3.1 Goods production and water

The analysis begins with an economy composed of a water using sector and a sectorthat, for simplicity, uses no water. The water using sector produces a good x and thenon-water using sector produces the good y. Production technologies are constantreturns to scale. The domestic price of the good produced by the water using sectoris proportional to the world price,

p = βpw (1)

where β represents the effect of trade frictions and pw is world price. A countryexports x if β is less than one and imports x if β is greater than one. Since onlyrelative prices matter, the price of y is normalized to 1.

Water use in the water using sector is denoted by z and is proportional to the outputof x,

z = e(θ,ω)x (2)

where e(θ,ω) is the intensity of water use per unit of output, θ is the amount of waterconservation done by the water using sector, and ω represents climatic factors thatinfluence water conservation. Water conservation reduces water intensity per unit ofoutput so ∂e/∂θ < 0. Water conservation is achieved by allocating units of output x towater conservation. The amount of output used in conservation is xa and the amount


of water conservation effort per unit of output is θ = xa/x. Some water conservationis worthwhile even when water is abundant, so ∂e(0,ω)/∂θ = −∞ and some wateris always needed, so e(1,ω) > 0.

A governmental regulatory authority supplies water infrastructure and charges afixed fee or tariff, τ , for each unit of water used in the production of output. Thenet producer price per unit of x is pN = p(1 − θ) − τe(θ,ω); the per unit sale ofoutput that is not used for water conservation, p(1 − θ), minus the cost of water,τe(θ,ω). Firms choose the level of water conservation in order to maximize per unitprofits. The first order condition for maximizing net producer price, pN with respectto conservation, θ , leads to

p = −τ∂e(θ,ω)/∂θ (3)

so that, using equation (1), water conservation is a function of the cost of water rela-tive to the world price of x,

e = e(τ/p,ω). (4)

Combining (2) and (4), overall demand for water use is the product of water conser-vation per unit of product and total production of the water using sector,

z = e(τ/p,ω)x = e(τ/βpw,ω)x. (5)

Water demand is a function of the water tariff, the domestic price of x, climatic fac-tors, and the total production of x. Alternatively, demand may be written as a functionof world price times the trade friction parameter β in place of the domestic price.

Firms use capital, K , and labor, L, purchased at prices r and w, respectively,to produce either x or y using the appropriate constant returns to scale technology.Equilibrium on the production side of the economy satisfies (2), (3), and the zeroprofit conditions,

pN = cX(w, r), 1 = cY (w, r),

(6)K = cX

r x + cYr y, L = cX

wx + cYwy.

The constraints on national income are a country’s capital investment, available labor,and amount of water used in production. National income from the private sector is afunction

I = I (pN,K,L). (7)

The first derivatives of I with respect to pN defines the level of output supplied perworker IpN = x (Copeland and Taylor 2003).

3.2 Water pricing and regulation

The level of water use and its environmental impact are controlled by a regulatoryauthority. The regulatory agency determines water use policy by making an optimiz-ing tradeoff between the benefits and costs of water use. Benefits arise from wateruse as an input to production. These production benefit raise national income. Water


use incurs two types of costs. The monetary costs of supplying water (cf. Barbier2004) and the non-market costs of environmental impacts (Vorosmarty and Sahagian2000). From an analytical perspective, water use policy is characterized by the reg-ulatory agency’s choice of a water tariff.1 The selected tariff is fixed to water usersand thereby determines the level of water use. By using the relationships derived inthe previous section, it is shown that water use and the water tariff are functions oftrade friction, world price, national income, and climatic factors.

The tariff, τ , is selected by the regulatory authority to optimize the tradeoffsbetween the benefits and costs of water use. The net benefit objective functionfor water use is a function of national income including water sales revenues, thecosts of water supply, and the environmental costs of water use, b(I, τ,ω, z) =λ[I + τz − c(z,ω)] − z.2 Net benefits are composed of an index function, λ[·], thatis a function of national income net of the costs of water supply and an index, −z,representing the environmental impact of water use. The index function λ[·] maps themoney values of national income plus water sales, I + τz, and water supply costs,c(z,ω), into a metric that is comparable to the index of environmental impact. Thus,the index function λ[·] puts money values on comparable terms with environmen-tal impact so that changes in net national income may be compared with and tradedoff with changes in environmental impact. For simplicity, the index of environmen-tal impact is set equal to the negative of water use, −z, so that λ[·] is allowed torepresent the nonlinearity between money income and environmental impact. The in-dex function, λ[·], is increasing, λ′ > 0, to reflect a positive valuation of income andstrictly concave, λ′′ < 0, to indicate that the marginal benefits of money income arediminishing relative to the environment.

The regulatory authority’s first order condition for optimizing the net benefit func-tion for water use with respect to regulated water tariff, τ ∗, is

(IpN

dpN

dτ+ z + τ

dz

dτ− c′ dz

dτ

)λ′ − dz

dτ= 0. (8)

From (7), IpN = x. From the definition of pN , pN = p(1 − θ) − τe(θ), dpN

dτ= −e.

Bringing together these last two equalities, IpNdpN

dτ= −ex, and ex = z by (5). Using

the last relationship, the first two terms within the parentheses in (8) cancel out. Withthe latter cancellation, divide (8) by dz

dτand rearrange the result to obtain

τ = ∂c

∂z+ 1/λ′. (9)

1Though the analysis of the regulatory agency and water policy is developed in terms of selecting a waterprice, it is analytically equivalent to selecting a water quantity. Selecting a water quantity is the primalproblem and selecting a water price is the dual problem. The dual solution is a water price that results inthe same level of water use that would be derived from the primal problem. Once either the dual or primalproblem is solved analytically, actual water policy may be implemented as a regulated quantity of wateruse or a regulated fixed price. The regulatory problem is specified here as the dual problem since the dualapproach is more tractable with the ACT Heckscher-Ohlin framework.2Similar to Barbier (2004), the cost of supplying water reduces national income, I (pN ,K,L), by anamount c(z,ω) that increases with water supplies and varies with climatic factors, ω.


The right hand side of (9) describes the water supply tariff as the sum of marginalwater supply cost, ∂c

∂z, and the marginal money value of environmental impact of

water use, 1/λ′. Marginal supply cost is increasing in the amount of water suppliedand also varies with climatic factors. The money value of the environmental impact,1/λ′, increases in the national income since the denominator, λ′, falls as nationalincome rises due to the concavity of λ[·]. The latter implies that the money valueof the environmental impact, 1/λ′, is lower at lower levels of income and higher athigher levels of income, ceteris parabus. Thus, one may expect that lower incomecountries set a lower water tariff due to a lower relative valuation of environmentalcosts while higher income countries set a higher water tariff due to a higher relativevaluation of environmental costs. Other things equal, the latter observation suggeststhat higher income levels lead to a higher water tariff and a lower level of water use.

To make explicit the economic factors that determine it, the water tariff is writtenas a function of the predetermined variables that influence marginal water supplycost, ∂c

∂z= ∂c(z,ω)

∂z, and the marginal benefits of national income, λ′ = λ′(τ,β,pw, I ).

The water tariff as a function of predetermined variables is

τ = τ(β,pw, I,ω). (10)

3.3 The economic determinants of water use

The foregoing results are used to identify the predetermined variables that determinewater use. Equation (5) describes the demand for water as a function of output andwater per unit of output. Equation (10) describes the supply tariff of water as a func-tion of predetermined variables. These two equations are combined to derive the de-terminants of water use.

Substituting the water tariff function into the water conservation function resultsin e = e(β,pw, I,ω). Substituting the latter result into (5) leads to the equation forthe equilibrium quantity of water use,

z = e(β,pw, I,ω)x. (11)

Equation (11) is a reduced form equation since it combines the parameters of thewater demand equation (5) and the regulatory authority’s water price (supply) equa-tion (10).

Equation (11) may also be used to write country-level water use as a function of theoverall output scale of the economy, S, and the share, ϕ, of the output attributable towater using commodity x (Antweiler et al. 2001). By the Heckscher-Ohlin theorem,the relative size of the water using sector and the water using sector’s share of totaloutput, ϕ, is a function of a country’s capital per worker, k = K/L. Thus, we canwrite the water using sector’s share of output as a function of capital per worker,ϕ = ϕ(k) where ϕ is the water using sector’s share of output. The output of the waterusing sector, x, can therefore be written as the scale of the overall economy timesthe share of output produced by the water using sector, x = Sϕ(k). Using the latterresult, an alternative version of the reduced form (11) is

z = Sϕe = Sϕ(k)e(β, I,ω) = z(S, k,β, I,ω) (12)


where the world price, pw , is left implicit since it is constant across countries.3 Thefirst line of (12) confirms the GK hypothesis for water use; water use is influencedby economic scale, composition, and production technique. In this case, productiontechnique is the technology of water conservation. The second and third lines definea reduced form equation that describes a set of empirically measurable determinantsof water use.

Equation (12) suggests hypotheses about how water use may be influenced by itseconomic determinants. Water use is proportional to scale so water use is expectedto rise and fall with economic scale. Water use is expected to be smaller in a countrywith a smaller economy and larger in a larger economy.

An increase in capital per worker and a shift toward manufacturing is likely to re-duce water use. Agriculture is a major water using sector and is less capital intensivethan economic sectors such as manufacturing. Also, capital investments in agricul-ture, such as drip irrigation and other water saving investments, are likely to reducewater use.

A reduction in trade frictions, β , could result in either an increase or reductionis water use depending on a country’s comparative advantage in agriculture or innon-water and less water intensive sectors. However, on average across a numberof countries, a reduction in trade barriers is likely to shift production to areas withthe respective comparative advantage, including a comparative advantage in watercosts and pricing. The latter result is suggested by the recent results where freer tradein cereal production reduced global water use by 6 percent. We expect this resultto hold generally within economies so that water use is likely to decline and waterconservation is likely to increase with trade intensity.

Income enters equation (12) through its effect on water pricing. An increase in in-come tends to raise the regulatory price of water. A higher price reduces water use byincreasing the level of water conservation. As the country-level value of water rises,one would expect a greater investment in water conserving institutions and technol-ogy. Accordingly, increases in income are expected to reduce water use, holding othervariables unchanged.

Equation (12) suggests no definite hypotheses for climatic factors. Climatic fac-tors enter as determinants of water use through both water price and the technologyof water conservation, with each having countervailing effects on water use. For in-stance, an increase in temperature may increase the costs of providing water due towater scarcity and greater evaporation, leading to a higher implicit regulatory priceof water. At the same time, increased temperature may increase the value of marginalproduct of water in agriculture and lead to greater water use and reduced conser-vation. The cost effect of temperature tends to reduce quantity of water demanded

3It is interesting to consider the specification of the water use (12) when both sectors use water. In thelater case, there are two water demand equations similar to (5), zi = ei (τ/βpw,ω)xi where i = (1,2).Following the analysis that led to (12), aggregate water use, Z, when each sector uses water is the sum ofwater use in each sector,

Z = Sϕ(k)ei (β, I,ω) + S[1 − ϕ(k)]ei (β, I,ω) = Z(S, k,β, I,ω).

Hence, aggregate water use when both sectors use water is a function of the same variables as water use inthe simpler model where only one sector uses water.


while the productivity effect tends to increase the quantity of water demanded. Thenet effect may only be measured empirically.

Notably, the theoretical concepts summarized in (12) place no restrictions on theoverall functional form connecting water use with the predetermined factors. Equa-tion (12) may have an inverted-U relationship with income, but there is nothing inthe theory requiring equation (12) to be an inverted-U in income. The only hypothe-sis about income is that it reduces water use due to its effect on regulation and waterprice policy. Unlike the GK approach, scale and composition effects enter the analysisthrough their own, distinct variables, S, and k, so there is no implicit netting out ofthe scale and income-technique effects. Scale, composition, income-technique, andopenness to trade are each represented separately and each is tested separately for itseffect on water use.

4 Data

The empirical measures of water use are water withdrawals and water consumption(Shiklomanov and Penkova 2003). Water withdrawals measure the amount of wa-ter diverted from water resources for residential, municipal, industrial, and agricul-tural uses. Most water withdrawn is returned to the terrestrial hydrological systemin some form. Water consumption refers to water that is removed from a country’savailable water resources and not returned. Water is consumed through evaporation,due to leakages from distribution systems, and when it becomes embodied in a man-ufactured or agricultural product (Shiklomanov and Penkova 2003). Given their dis-tinct definitions, withdrawals and consumption may have different linkages to eco-nomic activity. Hence, both withdrawals and consumption are used as dependent vari-ables.

The primary sources for water use data are the AQUASTAT database main-tained by the Food and Agriculture Organization (2005), the World Resource In-stitute summary database EarthTrends (2005), and the International HydrologicalProgramme (IHP) database maintained by UNESCO (2005). AQUASTAT focusesprimarily on agricultural water use and has significant missing data problems formost countries. EarthTrends lists withdrawal and infrastructure data from multiplesources, including AQUASTAT, but data are missing for many countries and timeperiods.

The IHP database contains consistently measured data on water resources, with-drawals, and consumption. The data set focuses on regional summaries, but data areavailable for a large number of selected countries. The data are compiled for a consis-tent set of years, covering country-level annual estimates for 1900, 1940, 1950, 1960,1970, 1980, 1990, and 1995.

The IHP data were used to formulate the dependent variables, per capita waterwithdrawals and water consumption. To match the economic and demographic dataavailable, use of the IHP data was limited to 1970, 1980, and 1990. Per capita with-drawals and consumption were calculated from the IHP data by dividing total waterwithdrawals and consumption by country-level population. Population data were ob-tained from the Penn World tables 5.6 (Heston et al. 2002).


Country-level economic data were taken from the Penn World Table (PWT), ver-sion 5.6 (Heston et al. 2002). The PWT data were used to formulate each of the eco-nomic variables identified by theoretical analysis; scale, capital-labor ratio, income,and trade. Following ACT, gross domestic product (GDP) was used as the measureof economic scale and gross national income (GNI) was used to measure nationalincome. GDP and GNI are often thought of as similar, but GDP measures the totalproduct produced within a country’s borders while GNI measures the income re-ceived by a country’s residents wherever it may be produced. The capital-labor ratiowas measured by capital stock per worker. The scale and income variables were bothdivided by population to convert them to scale and income per capita. Scale, income,and the capital-labor ratio were adjusted to US dollars at the first quarter, 2005 pricelevel.

The trade literature uses a variety of measures to approximate trade frictions andopenness to trade (Cipollina and Salvatici 2008; Harrison 1996). One measure istrade intensity, computed as the sum of exports and imports divided by gross do-mestic product. ACT show that trade intensity is directly correlated with the the-oretical measure of trade friction, β , in the ACT Heckscher-Ohlin. Accordingly,trade intensity derived from the PWT data is used as the measure of openness totrade.

Five climatic variables were included in the empirical analysis. Mean annual pre-cipitation and temperature for 1970, 1980, and 1990 were derived from data compiledby the Global Historical Climatology Network (Vose et al. 1992). The latter datawere available as monthly precipitation and temperature values across the reportingstations in a given country. Country level annual means were computed by averagingthe monthly data for stations within a given country and year. Three climatic dummyvariables were also included to denote specific climatic features contained in standardclimatic descriptions (CIA 2005). A monsoon dummy denoted countries subject tothe Asian monsoon. An arid dummy variable denoted countries with arid or very dryconditions. A cold dummy variable denoted countries with very cold conditions, suchas artic or subartic conditions.

The full dataset included the 2 dependent variables and the 9 independent variablesfor 32 countries for 1970, 1980, and 1990 and one additional country for 1980 and1990. The result was 98 observations for both water withdrawals and water consump-tion. The dataset included three African countries, seven Asian countries, six Euro-pean countries, three North American countries, twelve South and Central Americancountries, and two Oceania countries.

Figure 1 describes gross domestic product and water consumption per capita in1990 for the 33 countries included in the water data set. Countries are arrayed fromlowest to highest gross domestic product per capita. As a point of comparison, thecountries included in the water data set were compared to a ranking of 172 coun-tries for which 1990 gross domestic product data were available in the World Bank’sWorld Development Indicators (TWB 2007). The 172 countries were sorted into threecategories of 57 low, 57 medium, and 58 high income countries. Based on the lattercategories, the water data set contains 4 low income countries (Madagascar to Indiain Fig. 1), 16 medium income countries (Honduras to Chile in Fig. 1), and 13 highincome countries (Argentina to USA in Fig. 1).


Fig. 1 Gross domestic product and water consumption

5 Econometric model

The objective of the econometric analysis is to estimate an empirical form of (12) andtest the hypotheses regarding the role of scale, capital per worker, income, and tradeopenness as measured by trade intensity. The empirical form of (12) is specified as

wit = xitγ + ci + uit (13)

where wit represents either water withdrawals or consumption per capita, xit is thevector of explanatory variables, γ is a vector of coefficients to be estimated, ci rep-resents unobserved country effects, uit denotes idiosyncratic disturbances that varyacross time and country, and the subscripts i = {1, . . . ,33} represent individual coun-tries and the subscripts t = {1, . . . , Ti} represent years.

A troubling problem in country-level analysis is accounting for all the relevantvariables that affect the dependent variables while preserving the degrees of freedomrequired for estimation and hypothesis testing. It is always possible that potentiallyimportant variables may be omitted from the analysis. There are two econometricprocedures that deal explicitly with the potential impact of omitted variables on es-timation and testing with panel data. These are the fixed effects (FE) and randomeffects (RE) estimators.

A FE estimator treats the unobserved country effects, ci as constant over time, butvarying across countries. A FE estimator deals with one of the problems potentiallyintroduced by omitted variables, correlation between the unobserved effect and theexplanatory variables. The FE estimator is asymptotically consistent in the presenceof such correlation.

The RE estimator treats the country-level effect as a stochastic variable. The sto-chastic specification allows for omitted variables that may be constant over time but


differ between countries, and those that may be fixed between countries but vary overtime. In contrast to the FE estimator, the RE model is consistent when the unobservedeffect is uncorrelated with the observable explanatory variable. The stochastic unob-served effect and the idiosyncratic disturbance term may then be added to form acomposite error term, vit = ci + uit . Because ci appears in the composite error forevery time period t , the composite error term is serially correlated. The RE modelis estimated using generalized least squares to address serial correlation (Wooldridge2002).

Equation (13) was estimated for both water withdrawals and water consumptionin three stages. At the first stage, equations for water withdrawals and consumptionwere estimated using a RE estimator and tests were conducted for heteroskedasticityin the data using a likelihood ratio (LR) test. Results rejected homoskedastic errorsfor the RE estimates at the 99 percent level of significance. At the second stage, waterwithdrawal and consumption equations were estimated using a standard FE estimatorwith robust standard errors and an RE estimator allowing for heteroskedastic errors.To allow estimation of the FE, the FE and RE estimates were obtained without thedummy variables for Asian monsoon, arid conditions, and cold climate. The lattervariables could not be included in the FE equations since each dummy is constantacross time and therefore drops out of the FE within estimator.

The RE estimates were evaluated for consistency relative to the FE estimates us-ing a Hausman (1978) test. The Hausman test evaluated the null hypothesis that theestimated coefficients of the random effects model were the same as the fixed effectsmodel estimates. The test resulted in a Chi-squared statistic of 13.7 with 7 degreesof freedom. The test statistic indicates no significant difference between the RE andFE estimates at the 95 percent level of significance, meaning that correlation betweenthe country level effect and the independent variables is not statistically significant.Hence, attention is focused on the RE estimator and estimates.

The final stage estimated two forms of water use equations. The first form followedthe conventional Kuznets curve approach and used a single function quadratic inincome to capture the summary effect of the economy on water use,

wit = γ0 + γI Incomeit + γI 2 Income2it +

5∑g=1

γωgωgit + vit (14)

where the Incomeit is the ith country’s gross national income at time t , and ωgit are the

climatic variables discussed in the last section. In contrast to the Kuznets relationship,(12) identified four specific and measurable factors that influence water use: scale,composition, income-technique, and trade intensity. Equation (12) therefore leads toan alternative empirical relationship,

wit = γ0 + γSScaleit + γkCapitalit + λOOpennessit + γI Incomeit + γI 2 Income2it

+5∑

g=1

γωgωgit + vit (15)

where Scaleit was specified as gross domestic product, Capitalit represented thecomposition effect with capital per worker, and Tradeit was trade intensity as dis-


cussed in the last section. Equation (15) retained the quadratic in income portion inorder to provide a nested comparison with the Kuznets curve equation (14). The finalequations were estimated using a generalized least squares random effects estimatorthat allowed heteroskedasticity with the unbalanced panel data.

6 Results

Table 1 describes the variables used in the empirical analysis of water use. The de-pendent variables were water withdrawals, PCww, and water consumption, PCwc.Table 1 lists two columns of means. The first column of means lists means computedby giving the data for each country an equal weight. The second column of meanslists means computed by weighting the data for each country by the country’s popu-lation. The population weighted means give more weight to country level values forIndia than a country with a smaller population, such as Spain and Japan.

Table 2 lists the estimated coefficients for the water withdrawal and water con-sumption equations. The table reports coefficients for both the Kuznets Curve (KC)specified in (14) and full model derived from the analysis of the two sector trademodel and specified in (15). In the discussion below, the full model specified in (15)is referred to as the Two-Sector Trade Model (TSTM). In Table 2, there are two sets ofestimated coefficients using the TSTM, one for water withdrawals and one for waterconsumption. Table 2 shows that almost all of the estimated coefficients are statisti-cally different from zero at the 99 percent level of significance. The coefficients thatare not statistically different from zero are the Income squared coefficients in the KCequations and the TSTM withdrawal equation, the coefficients for Cold in the KCequations, the Temperature and Temperature squared coefficients in the consumptionequation, and the Constant coefficient in the TSTM equation for water consumption.

Collinearity between these pairs of variables makes the coefficient-by-coefficienttests less prone to reject the null hypothesis compared to tests with low levels ofcollinearity. Accordingly, joint tests of significance were conducted for Income andIncome squared coefficients as well as between consumption Temperature and Tem-perature squared coefficients. Chi-squared tests for the KC and withdrawal coeffi-cients showed that the Income and Income squared coefficients were jointly differentfrom zero at the 99 percent level with 2 degrees of freedom. The Chi-squared val-ues for these tests were 167 with the KC withdrawal coefficients, 8.4 with the TSTMwithdrawal coefficients, and 42.4 with the KC water consumption coefficients. A sim-ilar Chi-squared test for water consumption temperature coefficients resulted in aChi-squared value of 10.9 with two degrees of freedom, indicating that the Temper-ature and Temperature squared coefficients were jointly different from zero at the 99percent.

The KC is nested within the TSTM insofar as it may be derived by setting theScale, Capital, and Trade coefficients equal to zero in the TSTM. These restrictionsare tested with a likelihood ratio test for the pair of equations for water withdrawalsand the pair of equations for water consumption. The test statistic for restricting theScale, Capital, and Trade coefficients in the water withdrawals TSTM to zero is 26.The test statistic for the restricted water consumption TSTM is 20. Both of the latter


Table 1 Variable names, means, and descriptions

Variable name Variable meansa Description

Equally Population

weighted weighted

country data country data

PCww 710(489)

857(593)

Per capita water withdrawals, cubic meters per year

PCwc 349(232)

422 Per capita water consumption, cubic meters per year

Scale 8.52(7.20)

8.50(9.04)

Per capita gross domestic product in $US 1,000,2005 price level

Capital 20.0(17.4)

17.5(19.5)

Capital per worker in $US 1,000, 2005 price level

Income 8.33(7.16)

8.46(9.09)

Per capita gross national income measure in $US1,000, 2005 price level

Incomesquared

120(181)

153(245)

Per capital income squared

Trade .461(.209)

.267(.165)

The sum of exports and imports divided by grossdomestic product

Temperature 292 291(6.49)

Mean annual temperature measured in degreesKelvin

Temperaturesquared

85405(406)

85069(381)

Temperature squared

Precipitation 929(476)

1033(372)

Mean annual precipitation measured in millimeters

Monsoon .122(329)

.527(.502)

A value of 1 if a country is subject to an Asian mon-soon climate; 0 otherwise

Arid .306(.463)

.275(.449)

A 1 if a country’s standard climate description indi-cates an arid or very dry climate; 0 otherwise

Cold .153(.362)

.109(.313)

A 1 if a country’s standard climate description indi-cates a very cold climate (e.g., artic or subartic); 0otherwise

aStandard deviations are given in parentheses

test statistics exceed the 99 percent level Chi-squared critical value of 11.3 with 3degrees of freedom. Hence, the Kuznets Curve specifications for water withdrawaland water consumption are rejected in favor of the TSTM specifications. For waterwithdrawal and consumption, the detailed TSTM specifications that assign specificvariables to the scale, composition, income-technique, and trade intensity factors ex-plain the data better than the KC specification.

The TSTM coefficients also are consistent with the variable specific hypotheses.The positive coefficients on Scale in the water withdrawal and consumption equations


Table 2 Water use equation estimatesa

Variable Water withdrawalsb Water consumptionc

Kuznets Two-sector Kuznets Two-sector

curve trade curve trade

model model

Scale – 349∗∗ – 206∗∗(88.9) (63.8)

Capital – –32.7∗∗ – −10.4∗∗(3.92) (2.59)

Trade – −706∗∗ – −290

(91.7) (43.5)

Income 55.5∗∗ −212∗∗ 21.4∗∗ −151∗∗(11.9) (85.4) (5.99) (61.1)

Income squared −.266 −.521 −.354 −.642∗∗(.511) (.412) (.224) (.223)

Precipitation −232∗∗ −.150∗∗ −.155∗∗ −.118∗∗(.063) (.044) (.033) (.021)

Arid 414∗∗ 169∗∗ 137∗∗ 61.1∗∗(50.2) (49.0) (31.0) (32.0)

Monsoon 448∗∗ 215∗∗ 260∗∗ 207∗∗(51.1) (41.1) (28.5) (29.3)

Cold −249 −402∗∗ −54.9 −130∗∗(129) (90.1) (62.0) (50.7)

Temperature −997∗∗ −1097∗∗ 309∗∗ 170

(233) (229) (129) (121)

Temperature squared 1.70∗∗ 1.87∗∗ −.531∗∗ −.287

(.398) (.391) (.220) (.207)

Constant 146562∗∗ 160816∗∗ −44515∗∗ −24807

(33974) (33526) (18813) (17767)

Chi-squared 545 1346 210 218

Degrees of freedom 8 11 8 11bv

Number of observations 98 98 98 98

aThe standard errors are given in parentheses. A “∗” indicates that a coefficient is statistically differentfrom zero at the 95% level. A “∗∗” indicates that a coefficient is statistically different from zero at the 99%levelbThe dependent variable is PCwwcThe dependent variable is PCwc


indicate that both forms of water use increase with economic scale. The Scale coef-ficient for withdrawals is about 1.7 times the size of the coefficient for consumption,so scale has larger impact on withdrawals than on consumption.

The estimated coefficients on Capital are negative and statistically different fromzero at the 99 percent level. The estimated Capital coefficient in the water withdrawalTSTM is about 3 times more negative that the water consumption TSTM. The neg-ative signs are consistent with the hypothesis that capital investment reduces wateruse by either substituting capital investment for water in the agricultural sector orindicating the comparative advantage of the manufacturing and commercial sectorsrelative to agriculture.

The estimated Trade coefficients are also statistically different from zero at the99 percent level in both the water withdrawal and consumption TSTM equations.The negative signs on the Trade coefficients are consistent with the hypotheses thatopenness allows water using activities to move from countries with high water coststo countries where water costs are low.

The estimated Income coefficients are perhaps the most striking difference be-tween the KC and TSTM estimates. The positive Income coefficients and negativeIncome squared coefficients in both KC equations confirm the inverted-U hypothesiswhen all economic effects—scale, composition, technique, and trade intensity—aresummarized through the single variable income. In contrast to the KC equations,the TSTM equations partition the four effects to separate variables—Scale, Capital,Trade, and Income and Income squared.

The signs of the TSTM Income and Income squared coefficients have negativesigns. The income coefficients are also jointly statistically different from zero as in-dicated above. Together, the negative TSTM Income and Income squared coefficientsmean that increases in income reduce water use through the technique effect andthat the size of the reduction increases with income. The coefficients suggest thatincreases in income lead to more strict water regulation, leading to more water con-serving production techniques.

The weather and climate variables have impacts on water withdrawals and con-sumption that are consistent with intuition. The negative signs and statistically signif-icant coefficients for Precipitation indicate that precipitation is a substitute for humanwater withdrawals and consumption—the more rainfall a country enjoys the less wa-ter it needs to withdraw and distribute through human water infrastructure. The signsand significance of Arid are also consistent with the latter interpretation. Arid coun-tries lack the natural distribution of water through rainfall and therefore withdraw,distribute, and consume more water than non-arid countries. Nevertheless, the Asianmonsoon indicates that water withdrawals and consumption is greater in countrieswith a seasonal abundance of water from monsoon rains. The Cold dummy coef-ficients indicate that water withdrawals and consumption are less in countries withshort growing seasons and less evaporation.

According to the above joint tests of significance for the temperature coefficients,temperature has different effects on water withdrawals and water consumption. TheTSTM equation for water withdrawals is convex in temperature, so water withdrawalsacross countries decrease at a decreasing rate as mean temperature rises. The TSTMwater consumption equation is concave and increasing at a decreasing rate with re-


Table 3 Water use impacts of a 10% increase in the 1990 economic variables

Independent variable %Change in water %Change in water

changing by 10% withdrawala consumptiona

Mean Median Mean Median

Scale 56.9 51.8 92.2 50.4

(5.77) (24.4)

Capital −14.4 −11.7 −14.0 −7.07

(1.90) (4.46)

Income −35.0 −33.1 −71.5 −36.3

(3.52) (17.9)

Trade −11.4 −6.33 −19.3 −5.02

(1.97) (11.1)

Scale, Capital, and 7.56 6.61 6.67 3.99

Income (1.38) (2.92)

Scale, Capital, Income, −3.89 .012 −12.7 −1.33

and Trade (1.76) (8.34)

aThe standard errors for the estimated elasticities are given in parentheses below the coefficient estimates

spect to temperature. Thus, temperature reduces overall water withdrawals, but in-creases water consumption.

The estimated TSTM equations were use to simulate water use changes aseconomies grow. The TSTM equations were used to simulate the impacts on waterwithdrawals and water consumption of a 10 percent change in the economic variablesfrom their 1990 levels. A 10 percent change in Scale and Income is equivalent to adecade of growth for an economy growing at 1 percent per year and 3 years of growthfor an economy growing at just over 3 percent. In terms of Trade, a 10 percent changefrom the Population Weighted mean Trade (Table 1) is less than a 3% increase in thevalue of exports and imports relative to overall gross domestic product.

Table 3 lists the water use impacts of a 10 percent change in the economic factorsof Scale, Capital, Income, and Trade. A 10 percent increase in Scale alone leadsto a simulated increase in water withdrawals and consumption of almost 60 per-cent. The economic pressures exerted by Scale are offset by the countervailing pres-sures of Capital, Income, and Trade. Investment resulting in 10 percent more capitalper worker reduces water withdrawals and water consumption by about 14 percent.A 10 percent increase in Income reduces water withdrawals by 35 percent and waterconsumption by almost 72 percent.

A real economy is unlikely to experience a 10 percent increase in Scale or Capital,without an increase in Income as well. Table 3 shows that a 10 percent change inScale, Capital and Income increase mean water withdrawals by 7.56 percent andmean water consumption by 6.67 percent. The latter estimates are roughly similar to


those of the IHP which forecast a change of 4.9 percent for water withdrawals and achange of 5.2 percent in water consumption between 1990 and 2000 (IHP 2005).

The simulation results in Table 3 show that a 10 percent change in Trade re-duces water withdrawals by 11.4 percent and water consumption by 19.3 percentfrom their 1990 levels. When combined with a 10 percent increase in Scale, Cap-ital and Income, a 10 percent change in Trade results in a net 3.89 percent re-duction in water withdrawals and a net 12.7 percent reduction in water consump-tion. This latter result is qualitatively consistent with resource accounting studiesof trade in virtual water. Trade allows countries with economically scarce water re-sources to import virtual water through the importation of water intensive products,thereby reducing the demands on domestic water resources (De Fraiture et al. 2004;Hoekstra and Chapagain 2008).

7 Conclusions

The analysis demonstrates that national economic structure has profound conse-quences for global water use. It is not enough to know whether a country is highor low income, is cold or hot, or has more or less precipitation. Economic organi-zation and incentives have a significant impact on water use. Water withdrawals andwater consumption vary with economic scale, composition, national income, and thedegree of openness to trade in the world economy. Increases in economic scale un-equivocally increase water use, but scale is offset by changes in the other economicfactors. Increases in capital investment, improvements in national income, and na-tional policies that encourage openness to trade have important effects on reducingwater use and improving water conservation.

Results indicate that openness to trade as measured by trade intensity is a keypolicy variable in reducing water use and improving water conservation. Hold-ing trade intensity constant, a decade of economic growth of 1 percent per yearin scale, capital, and income increases water withdrawals by 7.6 percent and wa-ter consumption by 6.7 percent. However, a 10 percent increase in trade inten-sity, maintained over the course of a decade, converts the 7.6 percent increaseinto a 3.9 percent reduction in water withdrawals. For water consumption, thesame increase in trade intensity converts the 6.7 percent increase in water con-sumption into 12.7 percent reduction. The simulation results extend and reinforcethe resource accounting results for trade in virtual water (De Fraiture et al. 2004;Hoekstra and Chapagain 2008). Water conservation could be improved by policiesthat encourage capital investment in water saving technologies.

The results show that an understanding of economic structure and organizationis essential to improving water use forecasts and policy. The poor performance offorecasts noted by Gleick (2000) is not surprising given the substantial impacts onwater use by scale, capital investment, income, and openness to trade. Changes ineconomic factors, such as capital investment, and changes in economic policy, suchas trade liberalization, work to reduce water use and invalidate even the best forecastswhen such forecasts rely only on past trends in water use. Past water use depends onpast economic conditions, conditions that change as economies develop and the worldeconomy becomes more integrated.


The same economic and policy variables that make water use trend analysis errorprone also hold out promise for reducing human water withdrawals and water con-sumption. Capital investment can substitute improved processes and technology forincreased quantities of water. Continued reductions in trade barriers encourage wa-ter conservation by allowing the movement of water intensive production activitiesto countries with lower water costs and environmental conditions that facilitate wa-ter conservation. Both capital investment and openness to trade have the potential toshift the global water balance from continuing increases in water use to continuingreductions and improvements in water conservation.

References

Antweiler, W., Copeland, B.R., Taylor, M.S.: Is free trade good for the environment? Am. Econ. Rev.91(4), 877–908 (2001)

Barbier, E.B.: Water and economic growth. Econ. Rec. 80(248), 1–16 (2004)Central Intelligence Agency (CIA): World Factbook. National Technical Information Service, Springfield

(2005)Cipollina, M., Salvatici, L.: Measuring protection: mission impossible? J. Econ. Surv. 22(3), 577–616

(2008)Cole, M.A.: Economic growth and water use. Appl. Econ. Lett. 11, 1–4 (2004)Cole, M.A., Elliott, R.J.R.: Determining the trade-environment composition effect: the role of capital, labor

and environmental regulation. J. Environ. Econ. Manag. 46(3), 363–383 (2003)Cole, M.A., Rayner, A.J.: The Uruguay round and air pollution: estimating the composition, scale and

technique effects of trade liberalization. J. Int. Trade Econ. Dev. 9(3), 339–354 (2000)Copeland, B.R., Taylor, M.S.: Trade and the Environment: Theory and Evidence. Princeton University

Press, Princeton (2003)Copeland, B.R., Taylor, M.S.: Trade, growth, and the environment. J. Econ. Lit. 42, 7–71 (2004)Dasgupta, S., Laplante, B., Wang, H., Wheeler, D.: Confronting the environmental Kuznets curve. J. Econ.

Perspect. 16(1), 147–168 (2002)De Fraiture, C., Cai, X., Amarasinghe, U., Rosegrant, M., Molden, D.: Does International Cereal Trade

Save Water? The Impact of Virtual Water Trade on Global Water Use. Comprehensive AssessmentSecretariat, International Water Management Institute, Consultative Group on International Agricul-tural Research, Colombo, Sri Lanka (2004)

Ehrhardt-Martinez, K., Crenshaw, E.M., Jenkins, J.C.: Deforestation and the environmental Kuznets curve:a cross-national investigation of intervening mechanisms. Soc. Sci. Q. 83(1), 226–243 (2002)

Food and Agriculture Organization (FAO): AQUASTAT Water Resources, Development and ManagementService, January 20, 2005

Gleick, P.H.: The World’s Water: 2000–2001. Island Press, Washington (2000)Grossman, G.M., Krueger, A.B.: Environmental Impacts of a North American Free Trade Agreement.

NBER Working Paper No. 3914, National Bureau of Economic Research, Cambridge, MA (1991)Grossman, G.M., Krueger, A.B.: Economic-growth and the environment. Q. J. Econ. 110(2), 353–377

(1995)Harbaugh, W.T., Levinson, A., Wilson, D.M.: Reexamining the empirical evidence for an environmental

Kuznets curve. Rev. Econ. Stud. 84(3), 541–551 (2002)Harrison, A.: Openness and growth: a time-series, cross country analysis for developing countries. J. Dev.

Econ. 48, 419–447 (1996)Hausman, J.A.: Specification tests in economics. Econometrica 46(6), 1251–1271 (1978)Heston, A., Summers, R., Aten, B.: Penn World Table Version 5.6. Center for International Comparisons

at the University of Pennsylvania (CICUP), Philadelphia (2002)Hoekstra, A.Y., Chapagain, A.K.: Globalization of Water: Sharing the Planet’s Water Resources. Black-

well, Hobokin (2008)International Hydrological Programme (IHP): Information about World Water Use and Availability. United

Nations Educational, Scientific, and Cultural Organization (2005)Koop, G., Lise, T.: Is there an environmental Kuznets curve for deforestation? J. Dev. Econ. 58(1), 231–244

(1999)


Lim, J.: The effects of economic growth on environmental quality: some empirical investigations for thecase of South Korea. Seoul J. Econ. 10(3), 273–292 (1997)

Shiklomanov, I.A., Penkova, N.V.: Methods for assessing and forecasting global water use and water avail-ability. In: Shiklomanov, I.A., Rodda, J.C. (eds.) World Water Resources at the Beginning of the 21stCentury. Cambridge University Press, Cambridge (2003)

Shiklomanov, I.A., Rodda, J.C. (eds.): World Water Resources at the Beginning of the Twenty-First Cen-tury. Cambridge University Press, Cambridge (2003)

The World Bank (TWB): World Development Indicators. The World Bank, Washington (2005)United Nations Educational Scientific and Cultural Organization (UNESCO): World Water Resources and

Their Use. International Hydrological Programme, 1999 (2005)Vorosmarty, C.J., Green, P., Salisbury, J., Lammers, R.B.: Global water resources: Vulnerability from cli-

mate change acid population growth. Science 289(5477), 284–288 (2000)Vorosmarty, C.J., Sahagian, D.: Anthropogenic disturbance of the terrestrial water cycle. BioScience 50(9),

753–765 (2000)Vose, R.S., Schmoyer, R.L., Steurer, P.M., Peterson, T.C., Heim, R., Karl, T.R., Eischeid, J.: The Global

Historical Climatology Network: long-term monthly temperature, precipitation, sea level pressure,and station pressure data (1992)

Wooldridge, J.M.: Econometric Analysis of Cross Section and Panel Data. Cambridge University Press,Cambridge (2002)

World Resource Institute (WRI): EarthTrends: The Environmental Information Portal. World ResourcesInstitute, Washington (2005)

Do growth, investment, and trade encourage water use or water conservation

Documents

Transcript of Do growth, investment, and trade encourage water use or water conservation