Environmental and Resource Economics 22: 133–156, 2002. © 2002 Kluwer Academic Publishers. Printed in the Netherlands.
133
Assessing the Employment Impacts of Environmental and Natural Resource Policy* PETER BERCK1 and SANDRA HOFFMANN2
1 Dept. of Agricultural and Resource Economics and Giannini Foundation, University of California, Berkeley, USA; 2 Resources for the Future, Washington, DC, USA
Abstract. This paper provides an introductory guide for environmental and resource economists to methods of assessing the impact of environmental and natural resource policy on employment. It examines five basic approaches to evaluating the effect of a policy action on employment: 1) supply and demand analysis of the affected sector; 2) partial equilibrium analysis of multiple markets; 3) fixed-price, general equilibrium simulations (input-output (I-O) and social accounting matrix (SAM) multiplier models); 4) non-linear, general equilibrium simulations (Computable General Equilibrium (CGE) models); and 5) econometric estimation of the adjustment process, particularly time series analysis. The basic modeling structure and data requirements for each of these approaches are described. Simple examples of their application to evaluation of environmental and natural resource policy are developed and the relative merits and applicability of each are discussed. Key words: computable general equilibrium, employment impact, input-output, natural resource policy, social accounting matrix
A ban on thin plastic grocery bags was due to go into effect in June 2000 as part of South Africa’s solid waste management efforts. The ban was delayed pending an assessment of employment impacts due to fears of possible job losses (Africa News 2001). Across Britain, newspapers have run stories since 1999 warning of job losses that might be expected from Britain’s Climate Change Levy (The Journal 1999; Sunday Telegraph 2000; Sunday Times 2001). Since 1944, the U.S. Forest Service’s forest management mandate has included maintaining job stability in timber regions (Sustained Yield Forest Management Act of 1944). Many other national and provincial governments have similar policies (see e.g., British Columbia 2002; Finland 1999; Tanzania 1998). Following adoption of the Clean Air Act Amendments of 1990, communities in the Appalachian coal region of the U.S. feared that shifts from high to low sulfur coal would wreak economic havoc in their mining economy (New York Times, February 15, 1996, A1). The impact of natural resource and environmental policy on employment seems an unending public concern. As environmental and natural resource economists, we are frequently called on to assess the impact of environmental and natural resource policy on employment or to interpret others’ assessments of such policies. The purpose of this paper is to provide an introductory guide for environmental and resource economists new to
134
PETER BERCK AND SANDRA HOFFMANN
this area of analysis. It is not intended to be a comprehensive review of the existing literature, which is extensive. It is intended to provide the basic intuition behind the models and empirical practices most widely used in these assessments and to provide examples of how they are used in evaluating environmental and natural resource policy. Overview There are five basic approaches to evaluating the effect of a policy action on employment: 1. supply and demand analysis of the affected sector; 2. partial equilibrium analysis of multiple markets; 3. fixed-price, general equilibrium simulations (input-output (I-O) and social accounting matrix (SAM) multiplier models); 4. non-linear, general equilibrium simulations (Computable General Equilibrium (CGE) models); and 5. econometric estimation of the adjustment process, particularly time series analysis. In theory, the choice among these alternatives should be dictated by the type and scale of the action being evaluated. As a practical matter, the cost of the evaluation, the time available to complete the analysis, and the convenience of using the tools at hand often dictate the method used. Environmental and natural resource policies can affect employment through two major paths of action: actions that change the availability of a factor and actions that change the cost of production. Environmental restraints on logging and development are examples of restrictions in the availability of different types of land, a factor of production. Requirements for catalytic converters on automobiles or scrubbers on coal fired power plants are examples of increasing the cost of production. In the extreme, either of these actions can change the industrial structure of a region. The type of action needs to be considered in choosing the method used for its evaluation. The scope of proposed actions ranges from very small – reformulating spray paint – to very large – reducing emissions from every source in a whole country. For a very small action, it makes no sense to assess changes in any but the most immediate markets. For a very large action, such as national carbon taxes to address global warming, economy-wide assessment including assessment of price adjustments is imperative. Finally, the cost of analysis and the need for timely analysis are factors in the choice of assessment methodology. Building large customized models to evaluate a small regulatory action usually is not financially feasible. Conversely, the temptation to use the tool at hand, be it a linear multiplier or a CGE model, for all questions that come to the analyst usually is great. In making employment assessments, time is often of the essence. This weighs on the side of using tools
ASSESSING THE EMPLOYMENT IMPACTS
135
at hand. Yet these methods will not always provide appropriate analysis or may provide results that must be interpreted with care and presented with appropriate caveats. Geographic scope of the analysis can also affect results. The larger the region, the larger the potential economy-wide effect simply because there is wider scope for impact. On the other hand, it also may be more difficult to detect econometrically a policy’s impact in a larger geographic area. The modeling approaches discussed in this paper have been used for local analysis, such as that for U.S. counties and metropolitan areas, as well as for multi-national analysis. These five approaches differ in the way they capture adjustment and economywide interactions. By definition, single and multi-market analysis do not access economy-wide impacts and will underestimate effects if used in situations where sectors other than those modeled are affected. Depending on model specification and the nature of the data set, these models estimate either short-run or long-run demand. I-O, SAM and CGE models represent a continuum of closely related models. I-O and SAM models provide an upper-bound on employment impacts because their Leontief production functions do not allow for adjustment through factor substitution. For the same reason, they can be thought of as simulating very short-run adjustment. CGE models allow for factor substitution in response to changes in relative price. At an extreme, a perfectly neoclassical CGE model will have no aggregate change in employment, and therefore represents a lower bound on possible aggregate employment effects (Robinson and Holst 1987). Even with no aggregate employment change, CGE models simulate labor movement between industries and therefore impacts on labor income. More commonly, CGE models include migration or labor force participation equations that allow aggregate employment to change in response to changes in wages. Since CGE models allow capital to be substituted for other inputs, they represent long-run equilibria. Dynamic CGE models that explicitly model changes in population and capital stock simulate very long-run equilibria. General equilibrium models simulate the adjustment process; time series analysis estimates the effect of the actual, typically long-run, adjustment experience. These models may examine aggregate adjustment or focus on one or a few sectors. All of the models discussed in this paper focus on the effects of policy on the number of jobs or workers’ earnings, rather than the unemployment rate. This may seem odd given macro-economic’s focus on the employment rate. At least in the United States, this focus on jobs rather than employment rates is due, in part, to poor subnational estimates of the size of the active labor force and labor force participation. It also is due to state and local policy makers’ interests. These policy makers have more explicit influence on job creation than on the employment rate, which depends on regional migration. Policy makers are as interested in the types of jobs created or lost as in aggregate employment. The simulation models discussed below focus on industrial-level changes in employment, as well as changes in aggregate employment.
136
PETER BERCK AND SANDRA HOFFMANN
Single Market Analysis The analysis most familiar to economists is the analysis of supply and demand in the single primary affected market. This type of analysis is appropriate when the policy is unlikely to affect other markets and the policy can be stated as a change in marginal cost or a shift in demand. To make such an analysis, one needs an estimate of the supply of, and demand for, the product in the affected region. But often the needed estimates or even the data to do the estimates do not exist. For example, the California Air Resources Board recently needed to know the employment affects of a requirement to reformulate spray paint in California. Data on prices and quantities of spray paint sold in California was unavailable. When estimation is impractical, one can produce a “back of the envelope” calculation by taking the demand elasticity from aggregate studies of consumer demand and supply elasticity from aggregate industrial studies. (One could simply assume constant returns to scale, so that the entire price increase was passed to consumers and the quantity cutback was maximal). Letting p be price, q be quantity, and c be cost, s be supply, d be demand, a simple comparative statics exercise gives the change in output quantity as: εq = −
εs εd c , (εd − εs )p
where εi is an elasticity. The result gives the percent change in output given a one percent change in cost. For example, with a demand elasticity of –0.7, elasticity of supply of 1, a 10% change in costs, and again assuming constant returns to scale, there would be a quantity change of 4%. On the heroic assumption that labor demand is proportional to output, one would then calculate the number of jobs likely to be directly lost by the regulation. This is an extremely inexpensive type of analysis and, if the number of jobs is de minimus, it is probably all that is politically required. Of course, it is also wrong. Compliance with regulations may create additional jobs that are not accounted for. For example, if one knew the make up of the increased cost of producing the reformulated spray paint, one might find that additional jobs were needed to produce the new paint. In our experience, this second step is all that is needed to complete a reasonable estimate of employment impacts induced by very small-scale regulations. Multi-Market Analysis When government policy has a significant enough impact to affect prices in primary markets, there also may be affects in secondary markets for complements and substitutes in consumption or production. In our spray paint example, if the reformulated product has a significantly higher price than the original one, furniture and appliance manufacturers may find a substitute for spray paint, perhaps brushed
ASSESSING THE EMPLOYMENT IMPACTS
137
on finishes. With a small scale action, consideration of closely related markets for substitutes and complements in consumption or production probably is more than sufficient for a good answer – there is little likelihood of changing the wage rate or any other macro parameter by a measurable amount.1 But where the scope of the impact is large enough to affect a large number of industries in the economy, an economy-wide model may be needed. Expanding the scope of analysis to more markets requires either more data to estimate more supply and demand curves or assumptions about functional form, such as linearity.
Economy-wide, Linear Simulation Models By far the most widely used methods of assessing employment impacts are linearsimulations, either in the form of input-output (I-O) analysis or, in recent years, social accounting matrix (SAM) multiplier analysis.2 This class of models has a long history of use in evaluating both natural resource and environmental policy (Leontief 1970; Isard et al. 1972; Pan and Kraines 2001). These are fixed-price, fixed-coefficient, demand-driven, economy-wide simulation models. They provide projections for industry, government, and households that can be highly disaggregated. As a result, this class of model is a useful analytical tool for tracing the sectoral impact of changes in policy, such as a restriction on land use to protect endangered species, or a tax on fossil fuels, on an entire region’s economy. These models can be used alone or linked to models describing the physical and biological environment (Xie 2000). Behind most conventional I-O or SAM multiplier simulations is the conceptual framework of a demand-driven, economic-base model of a region’s economy (Krikelas 1992). In an economic-base model, economic activity is driven primarily by export sales, X, and investment, I, which are taken as exogenous. Domestic production, E, is allocated to production for domestic consumption, export and investment: E = C + X + I. Consumption of domestically produced goods is assumed to be a linear function of regional income, Y, i.e., C = cY. In equilibrium, regional income equals regional expenditures: Y = E. A more complete model would include savings and purchases of imports. In these models, the financial infusions into the region’s economy from exogenously determined export sales are viewed as forming the “economic base” of the region’s economy. Export demand generates increased labor demand, and therefore increased earned income by workers in the region. To the extent that this income is spent on domestically produced goods, rather than saved or spent on imports, it generates more demand for regional production. This increased product demand further increases regional labor demand and earned income. This cycle of increased product demand, factor demand, factor income, and consumption “multiplies” initial infusions of export spending or investment. Solving this small model yields Y = (1/(1 – c)) (X + I). That is, regional income, Y, is a multiple of exogenous flows into the economy, (X + I). The economic-base multiplier, dY/dX = 1/(1 – c), is the
138
PETER BERCK AND SANDRA HOFFMANN
factor by which regional income will change in response to a change in exogenous demand. The obvious attraction of this model is that once X has been identified, one can use regression (or even ratios) to find the multiplier.3 Given the multiplier, any policy that can be thought of as changing exports then can be evaluated with a single multiplication. The attractiveness of this approach, if only one has a good definition of X, cannot be overstated. Usually, environmental and resource economists will be interested in the multiplier on a single industry or small cluster of industries, not a region’s entire economic base. Multiplier analysis that is free of the need to partition activities into basic and non-basic is done with a social accounting matrix (SAM) or I-O model. SAMs extend earlier I-O tables, which focused on interactions among production sectors, to include all flows that would be included in a circular flow diagram. Typically SAMs also have more detailed disaggregation of households, by income or demographic characteristic, and institutions, such as levels of government, than do I-O tables. The SAM is a tabular record of all transactions that occur in an economy at a particular time. These transactions occur among the agents in the economy: producers, owners of primary factors of production, households, government, and rest of the world. Table I presents a conceptual SAM. In the SAM, producers are represented as industrial sectors (aggregates of similar industries), which include manufacturing, service and natural resource industries. Households usually are disaggregated by income or other demographic characteristics, and government is disaggregated by level of government. It is the inclusion of households, government and other institutions that primarily distinguishes SAMs from the older input-output models.4 Primary factors of production include labor and capital and also may include land or even multiple natural resource factor accounts, such as stumpage, agricultural land, and water. The columns of the SAM represent payments by these actors and the rows represent their receipts. The first row of Table I shows industry receipts for intermediate inputs from other industries and the rest of the world, for capital equipment from the investment account, and for final products from households, government, and the rest of the world. The first column of Table I shows industry payments to other industries and the rest of the world for intermediate inputs, to primary factors of production for their services, and to government as taxes. Since the SAM is based on current accounts, capital equipment is sold to an “investment account” which has transfers and savings as its income sources. Industries purchase the current use of capital equipment and plant from the investment account. Industries buy labor, land and capital from primary factor accounts, which in turn pay households and other owners of primary factors of production. Total payments equal total receipts, so the row totals and column totals for each agent are equal. For developed countries, SAMs usually are adapted from prepared I-O or SAM tables constructed by public or commercial modelers such as IMPLAN or Regional
Table I. A conceptual social accounting matrix
ASSESSING THE EMPLOYMENT IMPACTS
139
140
PETER BERCK AND SANDRA HOFFMANN
Input-Output Modeling System (RIMS II) in the U.S. or GRIMP in Australia (Minnesota IMPLAN 2002; U.S. Dept. of Commerce 2002; West 1988). These prepared tables should be tailored to specific policy analysis needs and in some cases should be augmented by secondary public data on non-interindustry transactions. Data for the underlying input-output tables is drawn from detailed national surveys of manufactures, which allows for quite disaggregated industrial sectoralization based on standardized industrial codes. In the U.S., national I-O tables can be disaggregated to 528 industrial sectors (Minnesota IMPLAN 2002). Data on non-interindustry transactions is generally taken from national income accounts or their regional counterparts. Regional SAMs, such as state, county and metropolitan area SAMS, are constructed from national input-output (I-O) tables and regional data. In the U.S., state and county SAMs are available from IMPLAN and RIMS II. The analyst should be aware that, at least in the U.S., these regional I-O tables conventionally reflect average national technology for each industry. In most cases, this will be a reasonable approximation for regional industries. At times, however, industry level technology and costs may differ significantly across regions. For example, coefficients on agriculture that reflect a national average that includes both rain fed and irrigated agriculture may be inappropriate for analysis of a dam project in an arid region of the country. If this appears to be critical to the analysis, the analyst will have to use regional data to adjust the prepared SAM. As with all empirical work, the analyst must be aware of how a secondary data set has been gathered or constructed and must be prepared to make appropriate adjustments. Table II is a SAM for Del Norte and Humboldt counties in California. These counties are heavily timbered in redwood and have experienced rapid declines in harvest, due partially to environmental regulation and partially to timber depletion. The SAM was aggregated down to 9 sectors for explanatory purposes. A working model may have up to 500 or more sectors. We used the IMPLAN system to produce this SAM. The SAM in Table II includes four industrial sectors (accounts 1–4 in the table). We have aggregated the retail and service sectors to form a single local trade sector. In input-output and SAM models, the retail sector provides retailing services to other industries so that households and government purchase final goods from an industry, rather than from the retail sector. For example, the forest, forestry and mills industry in this SAM sells doors to households. These doors are “bundles” of the intermediate inputs, retail and professional services, primary factors and institutional inputs required to produce and market the good. Sector 5 is combined local, state and federal government. Sectors 6 and 7 are factors of production, labor and capital, although sector 7 also includes the earnings of proprietors and enterprises that are not necessarily attributable to capital earnings and investment accounts. In a more disaggregated SAM, this would not be the case. Sector 8 is households. Sector 9 is trade, or the rest of world.
Figures are millions of 1998 $US.
Other manufacturing Agriculture and mining Forest, forestry, and mills Local trade and services Government Employee compensation Capital, enterprise income, etc. Households Trade Totals
Sector 1 2 3 4 5 6 7 8 9
Sector no. 15.99 6.47 2.88 96.28 12.41 72.68 53.93 0.22 213.34 474.20
1 1.69 7.46 0.07 20.00 6.50 61.10 119.03 0.00 39.59 255.44
2 2.34 1.81 171.15 137.13 20.12 173.77 118.05 0.00 229.85 854.22
3 29.44 5.93 6.38 732.70 268.95 934.08 844.74 1.11 508.09 3331.42
4 6.19 1.24 0.23 229.82 2157.18 497.61 650.72 765.37 104.45 4412.81
5 0.00 0.00 0.00 0.00 209.53 0.00 0.72 1462.88 66.10 1739.23
6 1.59 0.01 0.05 346.46 1118.54 0.00 580.11 1215.27 –10.21 3251.82
7
Table II. Social account matrix for Del Norte and Humboldt counties, California
50.12 5.99 1.51 1594.73 590.83 0.00 214.50 89.45 991.85 3538.98
8
366.84 226.46 672.00 174.33 28.88 0.00 669.94 4.58 0.00 2143.03
9
474.20 255.37 854.27 3331.45 4412.94 1739.24 3251.74 3538.88 2143.06 20001.15
Totals
ASSESSING THE EMPLOYMENT IMPACTS
141
142
PETER BERCK AND SANDRA HOFFMANN
The third column and row of Table II corresponds to the very broadly defined forestry sector. All figures are quoted in 1998 $US. At the bottom of the column is the total of all purchases by the forestry sector ($854 million (m)). The major purchases are from itself ($171 m) (for instance, the logging industry selling to a mill), local services ($137 m), employee compensation ($174 m), capital ($118 m), and trade ($229 m). Almost all the sales, found by reading across the third row, are either to itself or to trade, which is to say out of the two-county area. Row and column 8 of Table II represent household transactions. Reading across row 8, households receive payments mainly from government (transfers $765 m), employee compensation (earned income $1463 m), and capital (dividends and capital gains $1215 m). Households receive few transfers from actors outside the region ($89 m). Reading down column 8, households spend mainly on local trade and services ($595 m), taxes ($591 m), and imported goods ($992 m). To derive multipliers from a SAM, the SAM is converted into a linear model of the economy’s transactions. For simplicity, let a J + 1 × J + 1 SAM matrix, S, be partitioned into a J × J matrix of endogenous transaction accounts, T = [tij ], bordered to the right by a column vector of exogenous demand, x = [x1 , . . . xi , . . . xJ , 0] = [s1,J+1 , . . . si,J+1 . . . sJ,J+1 , sJ,J ] and at the bottom by a row vector of exogenous accounts, l = [l1 , . . . lj , . . . lJ , 0] = [sJ+1,1 , . . . sJ+1,j . . . sJ+1,J , sJ,J ]. As a practical matter, the SAM is simply rearranged so that the exogenous trade column and row are at the far right and bottom of the SAM. Let y be the vector of row totals of S representing total receipts by each actor in the SAM. Total receipts for any one of the J accounts in the region’s economy (yi ) is made up of payments from other actors or institutions in the economy (ti j¯ ), and payments from exogenous demand (xi ): yi =
tij + xi .
(1)
j =1,J
In order to convert the SAM into a linear model, the endogenous part of the SAM, T, is transformed into a matrix of fixed “requirements coefficients”, A, with elements aij = tij /yj . Substituting for t, yi =
aij yi + xi ;
j =1,J−1
or in matrix notation, y = Ay + x. Inverting, y = (I − A)−1 x;
(2)
ASSESSING THE EMPLOYMENT IMPACTS
143
that is, regional, sectoral and household income (output) are a multiple of exogenous demand. The matrix (I – A)−1 is generally referred to as the multiplier matrix. In our example, the counties have experienced a decrease in timber harvest. Conventionally, in economic analysis of U.S. forest policy, this decrease in supply due to environmental regulation and timber depletion is interpreted as a corresponding decrease in demand for timber exports from the region (Burton and Berck 1996; Alward and Palmer 1983). This assumes that the region is a small supplier of timber that cannot affect timber prices and that the entire reduction was sold to export. The question of how environmental regulation and depletion of timber supply affects employment is addressed by looking at the affect of the corresponding change in timber export demand, x, on the equilibrium represented by the SAM. In this case x would be a vector with only one non-zero element, the change in the level of forest product exports. The change in economic activity (or output), y, is then computed as y = (I – A)−1 x. Finally, one would convert the change in output to a change in jobs by dividing the appropriate elements of y by the average number of dollars of sales per job, w. One could then add together the elements of y that correspond to industrial outputs to get a total output multiplier (or, if one had converted to jobs, a total job multiplier.) In this simple illustrative case, closure of the model (deciding which accounts are endogenous and which are exogenous) is virtually ignored. In practice, decisions about model closure critically affect the results obtained from SAM or I-O multiplier models. As a broad generalization, SAM models treat more accounts as endogenous than do I-O models. One commonly used SAM closure treats industry, households, and local and state or provincial government as endogenous and national government and export demand as exogenous (Schaffer 1999). Since SAM models treat more accounts as endogenous, there is more money circulating in a SAM model of a region’s economy than in an I-O model. As a result, SAM multipliers tend to be slightly larger than I-O multipliers. Clearly, the type of closure used should reflect the regional scale of the modeling exercise. Closure for a national SAM model should be different from that developed for a metropolitan region. Table III provides a numerical example of the decrease in exports of the forestry sector using the SAM in Table II. First, find the matrix of multipliers (I – A)−1 , which are shown in Table III. Column 3 of the multiplier table corresponds to the forestry sector. Based on this analysis, a dollar decrease in forestry exports has no appreciable effect on other manufacturing or on agriculture, and reduces forestry output by $1.25 and local services by $1.30. This produces an output multiplier of 2.6 or $2.6 lost for each dollar of decrease. In this example, government is endogenous and decreases by $1.58, employee and capital compensation decrease by 80 cents and $1.01 respectively and household income decreases by $1.36. To find the employment effects, one needs the number of employees per dollar of sales. The numbers of jobs per million dollars output are: Other, 5.7; Ag, 18; Forestry,
144
PETER BERCK AND SANDRA HOFFMANN
Table III. SAM multipliers for Del Norte and Humboldt counties, California, 1998 Sector Other manufacturing Agriculture and mining Forest, forestry, and mills Local trade and services Government Employee compensation Capital, enterprise income, etc. Households
Sector no.
1
2
3
4
5
6
7
8
1 2 3 4 5 6 7 8
1.06 0.02 0.01 1.09 1.26 0.62 0.80 1.07
0.05 1.04 0.01 1.51 2.16 0.93 1.58 1.79
0.04 0.01 1.25 1.30 1.58 0.80 1.01 1.36
0.05 0.01 0.01 2.63 2.12 0.99 1.33 1.74
0.05 0.01 0.01 1.69 4.09 0.95 1.43 2.09
0.05 0.01 0.01 1.58 2.08 1.69 1.04 2.22
0.05 0.01 0.01 1.79 2.84 0.84 2.46 2.17
0.05 0.01 0.01 1.64 1.88 0.68 1.03 2.34
5.8; and Local, 16. So, for a $1 million decrease in forestry output, the model projects that 28 jobs will be lost. The example we have presented is analysis of an exogenous change in demand. It is more difficult to impose a change in supply. Natural resource projects or policies, such as opening a mine or restricting harvesting in a forest, do not really follow the logic of these demand-driven multiplier models. In these cases, it is not a change in the demand for a sector’s exports that is in question; it is a change in the actual level of activity in the sector. A change in the activity level in an industry translates in these linear, fixed-coefficient models into a proportional change in all inputs used in that activity. For example, if one scaled up all the inputs of the forestry column by a given percentage, as an experiment in increasing forestry output by that percentage, it is apparent from equation (2) that one also would scale up all the outputs of the economy by that same percentage. This is an odd result and shows some of the limitations of the multiplier methods. One can achieve much the same effect in a roundabout fashion. If, in the above example, one wanted to reduce forestry output by $1 m, one could start by noting that a $1 m reduction in export demand for forestry products results in a $1.25 m reduction in regional forestry output. So in order to simulate a $1 m decrease in forestry output, one could reduce export demand for forestry products by $0.8 m. Creation of a major new facility, such as constructing a dam or opening a mine, may require including a new account in the model. Since this account has not existed in the region in the past, it will have to be constructed based on similar industries in other regions and on local interviews. Construction of a new account requires significant experience and judgment. Similarly, analysis of an environmental regulation in the SAM framework requires knowledge of how the regulation changes the I-O matrix or limits output. A regulation, such as the California requirement to use cleaner, reformulated gasoline, requires new technology. This will likely change the input mix required by the petroleum sector. Modeling introduction of a new technology requires an understanding of how intermediate
ASSESSING THE EMPLOYMENT IMPACTS
145
input demand will change. This can require extensive field investigation. Even with such investigation, changes in this and other sectors due to behavioral adjustment are usually difficult to capture. The effect of the change can be simulated by looking at output in the SAM with the old technology and the new technology. While this is a conceptually simple exercise, it is unsatisfying. The underlying assumption is that the new, and possibly more expensive, product sells just as well as the older, possibly less expensive, product. (Of course, the more common mistake in analysis is to assume that the new product sells less well, yet has the same factor demands.) Introduction of a new facility points to some of the limitations of linear simulation models, like the SAM multiplier model. The linearity of these models requires constant prices, fixed-proportion production and linear demand. Because there is no substitution in consumption or input use, and no price adjustment in the model, linear SAM or I-O multiplier models are known to overstate regional multipliers. In a small economy with policies that do not affect relative prices, linear models are more likely to provide good approximations of actual changes than in situations where the policy impact is large enough to affect relative prices. In a study for the California Energy Commission, Hoffmann et al. (1996) empirically explore the extent of this bias. Their 24-sector, economy-wide model uses closures that allow the model to function under one set of closures as a closedeconomy CGE model that adjusts through changes in relative prices, and, under another set of closures, as a SAM multiplier model that adjusts through changes in quantity. The model has three classes of labor: industrial, service, and professional. Under the closed-economy CGE variant, a 40% cut in federal defense spending in California resulted in almost no change in gross state product (GSP), a 1 to 2 percent decrease in industrial and professional wages, and a 1.3 percent increase in service labor wages. At the other extreme, the open-economy SAM variant, where both labor and capital were assumed to be free to move in and out of use, a 40% decrease in defense expenditures resulted in a GSP multiplier of 4.7 and a 9 to 12 percent decrease in all classes of employment. An intermediate model variant, more typical of regional CGE models, allowed labor migration but was closed to movement of real capital plant. This model variant resulted in a small GSP multiplier of 1 and wage/rent decreases of 1–2 percent for professional employment, a 6 percent decrease in industrial employment, and a 2.4 percent increase in services employment. Job losses for this intermediate case fell in the neighborhood of those projected by the California Commission on State Finance using econometric forecasting models. In the case of policies that have large, widespread impacts, like carbon taxes to address global warming, the exogeneity of prices implicit in the linear model can lead to significant inaccuracy in policy analysis. To achieve a greater degree of reality in these cases, one must accept a more complicated level of modeling, such as a CGE with endogenously set relative prices.
146
PETER BERCK AND SANDRA HOFFMANN
Economy-wide, Non-Linear Simulation (CGE) Models Like the SAM multiplier model, a Computable General Equilibrium (CGE) model is a general equilibrium model of an economy.5 It uses the SAM as its database, and represents the same transactions as the SAM model. But unlike the SAM, CGE models permit non-linear relationships between actors in the economy and adjust through changes in relative prices rather than quantity (Rose 1995). As a result, the CGE allows for substitution among inputs in production and goods in consumption. This permits a more realistic representation of the adjustment process and results in less extreme assessments of employment impact. However, this potentially more realistic representation of relationships in the economy comes at the cost of significantly greater data and modeling requirements. Each relationship represented in the economy must be modeled. In the case of our California CGE, the Dynamic Revenue Analysis Model (DRAM), there are more than 1100 equations (Berck et al. 1999). This is computationally expensive, and only recently has the availability of computer capacity allowed for the widespread use of CGE models.6 Use of nonlinear relationships also means more decisions must be made about functional form and choice of parameter values. In practice, while these choices are sometimes based on estimated relationships, they more often draw on the modeler’s judgment and a stylized understanding of the economy being examined.7 This has been a major criticism of CGE models (Abler et al. 1999). Yet the parallel criticism of linear SAM models is that these same relationships are modeled under the arbitrary assumption that they are linear. CGE modeling can advance in accuracy over time through more extensive estimation of the parameter values needed to describe the economy. A correlate of the fact that CGE models generally have a neoclassical closure, and that markets clear through changes in relative prices, is that there is no involuntary unemployment. In contrast, fixed-price, linear SAM models adjust through changes in quantity, including changes in total employment. This can be thought of either as labor migration or, on the downside, unemployment. Since all relationships are specified in the CGE model, intermediate cases with stickiness in labor market adjustment, and therefore unemployment, can be created. Before deciding to embark on CGE analysis of some policy, one needs to examine the scale of the economy. DRAM models the $800 billion California economy, an economy a little bigger than that of France. Air quality regulations that change the cost of spray paint by 10% will make no noticeable difference in the results from such a model. Only policies that are of the scale of $100 million, in our experience, have a chance of having effects larger than rounding errors. The investment, migration and trade equations that specify the closure of the economy to the rest of the world receive particular attention in CGE models. In the case of DRAM, none of the equations needed to connect California to the rest of the world were ideal. Within the U.S., there is very limited data on the flow of goods between states. As a result, trade flows must be inferred from other data. California sectoral output was derived from the I-O portion of a U.S. SAM by
ASSESSING THE EMPLOYMENT IMPACTS
147
using California’s share of national sectoral employment to determine California’s share of national sectoral output. U.S. Consumer Expenditure Survey data was used to estimate consumer demand functions for goods and services. The difference between output and demand was taken as trade. In addition, imported and domestically produced goods are not perfect substitutes in DRAM. The model uses an Armington specification that requires an elasticity of substitution between foreign and domestic goods used in the production of domestic consumption goods. This elasticity cannot be estimated from U.S. data and is instead taken from international trade studies. Similarly, migration between U.S. states or within a state is not directly measured. Capital plant and equipment investment is not tracked by the state in which it is installed. In the U.S., significant efforts are being made to estimate interstate migration from tax records. Capital investment is not traceable by state from primary sources. Of investment, migration, and trade, it turns out that DRAM is most sensitive to the assumptions one makes about imports, particularly the substitutability of foreign and domestic goods. To find the change in jobs implicit in a policy, one could run the CGE model twice, once with and once without the policy in place. The model would track the labor market: how many are working, how many are not working (the wage rate changes labor force participation), and how many migrated. Take, for example, a restriction in timber harvest. It is common in forest policy analysis to assume that forestry sector output is a linear function of stumpage. It appears straightforward to model a restriction on timber harvest by adding a primary factor, stumpage, to the model (enlarge the I-O table by one column and row) and running the model twice with two different levels of stumpage. But this would be a fair amount of work and the model could not be guaranteed to solve. The more likely way to model such a restriction would be to take advantage of the duality of prices and quantities. Since it is assumed that output is produced in fixed proportion to stumpage input, one can model a decrease in stumpage as an increase in stumpage price through a tax on timber output. In actuality, most of the money paid for stumpage goes to out-of-state capital owners. Modeling the stumpage price increase as a federal excise tax has the same effect: the net price of the output goes down and the money raised leaves the state with no effect on any other variable. To find the right level of the tax, one runs the model with different size taxes and finds the resulting size of the timber sector. One picks the tax that corresponds to the timber sector size one wants to achieve. The resultant tax take divided by the stumpage input use is the increase in the price of stumpage that would occur if the quantity were restricted. Table IV shows the effects of a 10, 20, and 30 percent tax on the “Other Primary” industrial sector, the sector that includes non-energy, minerals, fishing, and forestry. From the table, one can see that to achieve a 50% reduction in sectoral output, one would require a tax of a little over 30%. The industry would have roughly 3400 fewer workers, while the state as a whole would have only 1800 fewer people in the workforce. This is the opposite of the logic in the multiplier
148
PETER BERCK AND SANDRA HOFFMANN
Table IV. Impacts of an ad valorem tax on the ‘energy, minerals, fishing and forestry’ sector California environmental computable general equilibrium model Base Case 10%
Tax 20%
30%
Effects on the whole state economy Personal income (billion $US) percent change
892.96
892.69 –0.00030
892.47 –0.00055
892.31 –0.00073
Pop (million families) percent change
23.15
23.15 –0.00005
23.14 –0.00008
23.14 –0.00011
Labor demand (millions) percent change
14.05
14.05 –0.00007
14.05 –0.00011
14.05 –0.00013
2961.09
2956.04 –0.0017
2951.42 –0.0033
2947.25 –0.0047
Reactive organic chemicals (tons/day) percent change
Effects on the ‘other primary’ industrial sector Output (billion $US) percent change
1.48
1.24 –0.16
1.01 –0.32
0.78 –0.47
Labor demand (millions) percent change
0.007
0.006 –0.16
0.005 –0.32
0.004 –0.48
Imports (billion $US) percent change
2.03
2.20 0.08
2.38 0.17
2.55 0.26
models. Here, many of the people who stop working in the other primary sector start working in other sectors. The employment multiplier is less than 1. Again looking at the table, there is a considerable increase in imports as a result of the tax. If the elasticity of substitution between foreign and domestic produced goods were higher, this effect would be more pronounced, as would the job losses in the primary sector. An analysis of this sort takes a good half hour, except that the existing DRAM model will rarely have the sectoralization demanded for the particular analysis desired. Sectoralization should be tailored to reflect greater detail in the sectors most likely to be affected or of concern for other policy reasons. In either the SAM or CGE models, one could spend several days to break an existing sector into two – for instance, in DRAM separating forestry from non-energy minerals and fishing. Another possibility is to find the percent of the sector’s output that is forestry and multiply the cutback in forestry by that percent to find the cutback in the sector. The latter option is obviously very expeditious if data is available.
ASSESSING THE EMPLOYMENT IMPACTS
149
As another example, consider California implementation of the federal Clean Air Act. Compliance with the California State Implementation Plan under the Clean Air Act has required that consumer chemicals and engines be changed to produce less pollution. This reflects a change in production technology. This change can be represented in a CGE model by increasing intermediate input demand of the affected industry (scaling up that industry’s column of the matrix of input-output coefficients). To evaluate the impacts of California’s implementation of the federal Clean Air Act we use EDRAM, a version of the DRAM customized for environmental analysis. EDRAM links DRAM to a matrix of air pollution coefficients for each industrial activity and appropriately focuses the model’s sectoralization on industries affected by the State Implementation Plan (SIP) requirements. The EDRAM model run simulates the SIP by increasing annual intermediate requirements of the consumer chemicals and engine sectors by $184 million/year (14.9%) and $882 million/year (13.38%) respectively. Statewide output drops by roughly $461 million (0.035%). Statewide income drops by roughly $511 million (0.057%). Income drops by more than output due to lower returns to factors. Not only is overall state output falling, but in-state labor and capital are receiving smaller shares of it. As a result, some families, jobs, and investment move out of state: 2,822 (0.012% of) families leave California, 3,832 (0.027% of) jobs are lost, and the state’s capital stock falls by $269.5 million (0.019%). The higher prices of cleaner consumer chemicals (+5.3%) and engines (+3.0%), along with an outward shift in demand for (and thus a rise in the price of) the intermediate inputs required to make them, nudges the consumer price index up 0.03%. The building and upkeep of DRAM, which is based upon both Bureau of Economic Analysis numbers and California state data, is not a trivial undertaking. The model was developed under a legislative requirement to evaluate the economic impact of all money bills over $10,000,000 that were introduced into the legislature. Producing and maintaining a model of this complexity makes some sense if there is a steady stream of work, such as that generated by this legislative mandate. A realistic alternative to building one’s own model for impact analysis is to rent or buy a model from a commercial provider such as Region Economic Models Inc.
Time Series Methods If it is true, as economic base theory suggests, that there is a long run relationship between employment in basic industries and employment in other industries, then time series methods can be used to uncover that relationship. The simplest such relationship is the base multiplier equation: T – mB = ε, where T is total employment, B is basic employment, m is the base-employment multiplier, and ε is a stationary error term. When both B and T are nonstationary time series (have trends), but a linear combination of them, namely T – mB is stationary (shows no trend),8 the equation T – mB is called a cointegrating relation. In a
150
PETER BERCK AND SANDRA HOFFMANN
cointegrated time series model, the variables will tend towards satisfying the cointegrating relationships in the long run. Thus the cointegrating relationships are the stochastic equivalent of the long run equilibria of the system. Brown et al. (1992), whose work is described below, exploit this fact to find the best definition for basic industries. The generalization to more than one cointegrating equation, which naturally occurs if one has more than two variables of interest, can be made using Johansen’s maximum likelihood techniques (Johanssen 1995). In addition to identifying long run equilibrium conditions, which are the cointegrating equations, time series methods also permit the estimation of the adjustment process that takes an economy from its current state to the long run equilibrium conditions. We describe both the multivariate methods and the estimation of adjustment below.9 Time series methods and SAM (or CGE) multipliers provide different information and have different data requirements. SAM-based multipliers provide a great deal of detail about long run relationships such as the likely distribution of impacts of a policy among industries, households, and other actors in the economy. These economy-wide models provide a better understanding of the structure of economic interactions in the economy, but simulate, rather than estimate, multipliers. Time series methods are practically limited to far fewer sectors, but allow direct estimation of long run relationships, of adjustment and of the standard error of the estimates. SAM multipliers can be created from a single year’s worth of data, while time series estimates require long time series. Ideally, then, time series analysis and structural simulation or econometric modeling should be seen as complements, not substitutes. Compared to much of the data used by environmental and resource economists, the availability of time series data on employment is remarkably good. Employment data is often collected as a necessary part of administering government programs. In the U.S., firm level employment data is collected monthly at the metropolitan and county level in order to administer the unemployment insurance program and then is aggregated up to state and national levels (U.S. Dept. of Labor 2002). Several decades-long time series on employment are available at the state and national levels at fine levels of sectoralization. But data retention varies considerably at the sub-state level, and sectoral disaggregation in areas smaller than the state is often limited to prevent disclosure of information on individual firms. Most commonly, employment time series data has been used to estimate aggregate base multipliers. The most straightforward way to find this relationship is to identify the region’s basic industries and to regress total employment on aggregate base employment. The regression coefficient is an estimate of the base multiplier. In highly specialized resource dependent economies, where the region’s economic base is made up of a single industrial sector that is large relative to other sectors in the areas economy, such as mining or forestry, this relatively simple approach may be helpful to the resource economist. Bivariate analysis also can be used to examine the impact of changes in a single sector on other employment in
ASSESSING THE EMPLOYMENT IMPACTS
151
an economy, though the change in that sector must be large enough relative to other sectors in the economy or the impact will be lost in the measurement noise. Brown et al. (1992) use bivariate cointegration analysis as a way of identifying the Philadelphia metropolitan statistical area’s economic base, and in the process estimate the base multiplier for the region. Brown, Coulson and Engle compare five different methods of identifying basic industries. The aggregate basic employment data series generated by each of these methods is non-stationary as is total employment. Stationarity is tested using an augmented Dickey-Fuller test. Brown, Coulson and Engle find a base employment multiplier of 2.43, where basic employment was defined as total employment in manufacturing sectors with location quotients greater than one, plus excess regional employment in service sectors with location quotients greater than one. This multiplier is consistent with long run base employment multipliers obtained using other methods. This is an elegant and simple method and therefore has much to recommend it if policy analysis demands an analysis of the base multiplier or a simple estimate of a long run multiplier for a single large sector. More recent developments in multi-variate vector autoregressive (VAR) models allow economists to address a richer array of questions about long run economic adjustment to a policy than can be addressed by simple bivariate analysis.10 To see the basic structure of testing for cointegration in a VAR model, consider a hypothetical VAR of a very simple economy with only three sectors, timber (T), other manufacturing industries (M), and non-basic service sectors (S). A VAR is a set of simultaneous equations in which each variable in the system is regressed on its own lagged values and the lagged values of all other variables in the system. The estimated model might take the form: St −1 a11 a12 ε1t St Mt = a21 a22 1 −1 −2 Mt −1 + ε2t 0 1 −1 Tt 0 0 Tt −1 ε3t This is obviously a simplified system, missing possible lagged differences and constant terms. The notation is the first difference of the variable, the α’s are coefficients and the ε’s are normal error terms. The matrix with (1 –1 –2) as its first row is the matrix of the cointegrating relationships. In the long run, the system of equations describing the variables, S, T, M, will tend towards both of these relationships. The matrix of α’s is the speed of adjustment to the long run relationships. For instance, let u1t = St −1 – Mt −1 – 2Tt −1 be the discrepancy between the state of the system at t − 1 and the long run relationship, S – M – 2T. Similarly, u2t = Tt −1 – Mt −1 . Doing the matrix multiplication, St = a11 u1t + a12 u2t or the change in S is a function of the discrepancies in both cointegrating equations, the adjustment rate being the estimated α’s. This hypothetical model illustrates properties that reflect testable hypotheses. First, S is missing (excluded) from the second cointegrating equation. Exclusion restrictions can be used to test the hypothesis that S is not part of a long-run
152
PETER BERCK AND SANDRA HOFFMANN
relationship with other variables in the system. Second, the a’s for T are both zero, so T does none of the adjusting towards equilibrium; it is weakly exogenous. In an economic base model, base industries should be weakly exogenous. Multipliers also can be calculated from estimated VARs. In this example, a long run multiplier of 5 for timber can be calculated directly from the estimated model. One more timber job results in one more manufacturing job from the second cointegrating equation; from the first equation, there are two more service jobs from the additional timber job and one more service job from the manufacturing job. The total is one more timber job, one more manufacturing job, and three more service jobs for a total of five jobs. The hypothetical model just presented had three variables and two cointegrating equations. Logically, there can be zero, one, two or three long run relationships among these variables. There are long-run employment multipliers only in the case where there are n variables and n − 1 cointegrating relationships. With no relationships, there are no multipliers. With one cointegrating relationship, there could be one multiplier equation (e.g., M = 1T) or simply a relationship among the three variables (e.g., S = 2T + M). In the latter case, one could not tell whether increased M led to decreased T or increased S, so it is not a multiplier relationship. In the case of two linearly independent relationships (e.g., M = 1T and S = 2T + M) one could solve the system for S as a function of T and M. Two linearly independent relationships (more generally n − 1 linearly independent relationships among n variables) is the case that leads to a full set of multipliers. Finally, three relationships result in a stable system – increases in any variable are only temporary. Thus one can test for the existence of multipliers by testing that the rank of the cointegrating space is n − 1. Berck et al. (2000) study the long run impacts of timber harvest restrictions on poverty and employment in northern California using the maximum likelihood methods of Johansen to estimate the error correction form of a VAR model. The VAR structure allowed testing of the hypothesis that state, rather than local, economic conditions drove poverty in this timber region during the “spotted owl crisis.” During this period, local perception was that protection of the spotted owl through federal restrictions on timber harvest threatened local economic stability. The estimated VAR in this study can be viewed as a solution to a two-sector, structural model. The structural model includes five equations representing: 1) equilibrium in local timber labor markets (T); 2) equilibrium in local non-timber labor markets (N); 3) local poverty as a function of local labor demand, population, migration, and state poverty (POV); 4) local population as a function of prior population and migration; and 5) local migration as a function of state employment (SE), local employment, local poverty and state poverty (SPOV), where abbreviations represent the systems endogenous variables. Written in error correction form, the VAR model regresses first differences of all the endogenous variables in
ASSESSING THE EMPLOYMENT IMPACTS
153
the model on constants φ, seasonal dummies (D), the lagged variables, and lags of the first difference of these same variables: yt = φ + 1 yt −1 + . . . + k−1 yt −k+1 + yt −1 + Dt + εt , where yt is a 5 × 1 column vector of the model’s five endogenous variables (N, T, SE, SPOV, and POV) at time t. The VAR model was used to test for the existence of long-run relationships between the variables. For example, it was used to test the existence of a long run relationship between local timber employment and local poverty, which many area officials feared. The model also was used to estimate long-run multipliers of a change in local timber employment on local non-timber employment. The VAR was estimated first by identifying the number of cointegrating relationships among the five endogenous variables using trace and eigenvalue tests. Given the number of cointegrating relationships, the VAR model was then estimated. This estimated model was then used to calculate multipliers. The estimation used CATS in the time series software RATS (Estima 2002). The estimated VAR also was used to test for the existence of long-run relationships. Exclusion tests were used to see if some series could be excluded from any long run relationships with other variables. Weak exogeneity tests were used to test for variables that do not adjust towards equilibrium and therefore are unaffected by changes in other variables. In our county models, poverty was frequently weakly exogenous and hence unaffected by changes in local employment. State poverty and employment variables had a strong, significant long-run relationship with county poverty variables in most counties, suggesting that poverty in these rural counties was more related to broader macroeconomic trends and poverty policy than to local economic conditions. The local timber employment did have a multiplier effect on other local employment, but it is smaller than the average SAM multiplier for the study counties, as should be expected since the model captures adjustment. In no case did an additional timber job induce an additional job in the non-timber sector, a frequent assumption in the spotted owl impact studies (see Sample and LeMaster 1992). Conclusion The evaluation of potential changes in employment incident upon regulatory action or the proposal of projects is often required as part of the evaluation process. Small projects should be evaluated using methods that concentrate on the markets most directly affected. Supply and demand in just a few markets should be studied. Larger scale actions require methods that can account for economy wide effects. Multiplier models of the base or SAM types are extremely easy to use, but impose linearity on all parts of the economy. They do not permit great flexibility in the manner of the evaluation and necessarily assume that the proposed action changes the size of a variable (usually external demand) once and for all. The
154
PETER BERCK AND SANDRA HOFFMANN
CGE class of models requires considerably more effort to build or customize, but permits an almost unlimited amount of flexibility in structuring the analysis. These models work particularly well for examining changes in technology (such as requiring scrubbers on power plants) or taxes and can be made to work for resource constraints (such as limiting forest harvesting.) Time series models lack the detail of SAM’s or CGE’s, but permit direct estimation of the adjustment to changes in the target sector employment. Each of these methods has its place in assessing the employment effects of environmental and natural resource policy. Notes * Senior authorship is not assigned. The authors thank Steve Vogel of the U.S.D.A. Economic Research Service, Sherman Robinson of the International Food Policy Research Institute, and Jeff LeFrance and Jeff Perloff of the University of California, Berkeley for very helpful discussions in developing this paper. 1. Analysis can become more complicated when markets are distorted through market power, regulation, or sticky adjustment. Boardman et al. (1996) provide a practical overview of analogous problems with cost-benefit analysis in primary and secondary markets. Sadoulet and de Janvry (1995) provide a good presentation of partial equilibrium analysis of policies with more extensive multi-market effects. 2. Basic references for input-output and SAM models include Bulmer-Thomas (1982), Miller and Blair (1985), Isard et al. (1998). 3. There is a large literature on identification of a region’s economic base (see e.g., Brown et al. 1992). Schaffer (1999) provides a good historical review. 4. One will also encounter extended input-output models in the literature. These are extensions of input-output models moving in the direction of a full SAM and were developed prior to SAMs (personal communication with Adam Rose, Pennsylvania State University, Dec. 2001). 5. Dervis et al. (1982), Sadoulet and de Janvry (1995), and Löfgren et al. (2001) provide good introductions to CGE modeling. 6. CGE models have a long history of use in international trade and national policy analysis. In the U.S., they have recently begun to be used for state and multi-state analysis. Lack of availability of data, uncertainty about parameter estimates, and greater appropriateness of linear functional form has limited their use at the sub-state level (Partridge and Rickman 1998). 7. CGE models are usually created by adapting existing models to current needs. This reduces the modeling burden, but can add opaqueness and makes it easy to adopt prior modeling choices uncritically. 8. To be stationary a series must have mean, variance, and auto covariances that do not change over time. 9. Kennedy (1998) provides a very readable, intuitive explanation of recent developments in the time series econometrics used in employment impact modeling. Hamilton (1994) and Hendry and Richard (1997) are definitive reference books. 10. A very different approach is to use state space analysis to estimate long run relationships between time series. State space analysis deals directly with the matrix of auto-correlations with varying lag structures. Kraybill and Dorfman (1992) use state space analysis to model the Georgia economy in value added terms. The article provides an excellent explanation of the use of state space methods in regional multiplier analysis.
ASSESSING THE EMPLOYMENT IMPACTS
155
References Abler, D. G., A. Rodriguez and J. S. Shortle (1999), ‘Parameter Uncertainty in CGE Modeling of the Environmental Impacts of Economics Policies’, Environmental and Resource Economics 14, 75–94. Africa News (2001), ‘South Africa Ban on Plastic Bags Delayed’, June 22, Business Day, News, Documents and Commentary. Alward, G.S. and C.J. Palmer (1983), ‘IMPLAN: An Input-output Analysis System for Forest Service Planning’, Proceedings of the First North American Conference of Forest Sector Modeling. Oxford: AB Academic Publishers. Berck, P., C. Costello, L. Fortmann and S. Hoffmann (2000), ‘Poverty and Employment in TimberDependent Counties’. Resources for the Future Discussion Paper 00-52. Washington, DC. Berck, P., E. Golan, B. Smith, J. Barnhart and A. Dabalen (1999), ‘Dynamic Revenue Analysis for California’, for the California Department of Finance. Available at http://www. dof.ca.gov/html/fs_data/dyna-rev/dynrev.htm (accessed 31 January 2002). Boardman, A., D. Greenberg, A. Vining and D. Weimer (1996), Cost-Benefit Analysis: Concepts and Practices. Upper Saddle River, New Jersey: Prentice Hall. Bulmer-Thomas, V. (1982), Input-output Analysis in Developing Countries: Sources, Methods, and Applications. New York: Wiley. British Columbia, Ministry of Forests (2002), Forest Renewal Plan. www.gov.bc.ca/for (accessed 3 February 2002). Brown, S. J., N. E. Coulson and R. Engle (1992), ‘On the Determination of Regional Base and Regional Base Multipliers’, Regional Science and Urban Economics 22, 619–635. Burton, D. and P. Berck (1996), ‘Statistical Causation and National Forest Policy in Oregon’, Forest Science 42, 86–92. Dervis, K., J. de Melo and S. Robinson (1982), General Equilibrium Models for Development Policy. Cambridge, MA: Cambridge University Press. Estima (2002), CATS in RATS (Cointegration Analysis Software). Evanston, Illinois. www.estima.com/catsinfo.htm (accessed 31 January 2002). Finland, Ministry of Agriculture and Forestry (1999), Finland’s National Forest Programme 2010. Publications 2/1999. www.mmm.fi/KMO (accessed 10 December 2001). Hamilton, J.D. (1994), Time Series Analysis. Princeton, New Jersey: Princeton University Press. Hendry, D. F. and J. F. Richard (1997), The Econometric Analysis of Economic Time Series. Cheltenham, U.K.: Elgar. Hoffmann, S., S. Robinson and S. Subramanian (1996), ‘The Role of Defense Cuts in the California Recession: Computable General Equilibrium Models and Interstate Factor Mobility’, Journal of Regional Science 36, 571–595. Isard, W., C. L. Choquill, J. Kissin, R. H. Seyfarth and R. Tatlock (1972), Ecologic-Economic Analysis for Regional Development. New York: The Free Press. Isard, W., I. J. Azis, M. P. Drennan, R. E. Miller, S. Salzman and E. Thorbecke (1998), Methods of Interregional and Regional Analysis. Brookfield, Vermont: Ashgate Publishing Company. Johanssen, S. (1995), ‘Likelihood-based Inference’, in Cointegrated Vector Auto-Regressive Models. Oxford, England: Oxford University Press. Jorgenson, D. and P. J. Wilcoxen (1990), ‘Intertemporal General Equilibrium Modeling of U.S. Environmental Regulation’, Journal of Policy Modeling 12, 1–30. Kennedy, P. (1998), A Guide to Econometrics. Cambridge, MA: MIT Press. Kraybill, D. S. and J. H. Dorfman (1992), ‘A Dynamic Intersectoral Model of Regional Economic Growth’, Journal of Regional Science 32, 1–17. Krikelas, A. C. (1992), ‘Why Regions Grow: A Review of Research on the Economic Base Model’, Federal Reserve Bank of Atlanta Economic Review 77, 16–29.
156
PETER BERCK AND SANDRA HOFFMANN
Leontief, W. (1970), ‘Environmental Repercussions and Economic Structure: An Input-Output Approach’, Review of Economics and Statistics 52, 262–271. Löfgren, H., R. L. Harris and S. Robinson (2001), ‘A Standard Computable General Equilibrium (CGE) Model in GAMS’. Trade and Macroeconomics Division Discussion Paper 75, International Food Policy Research Institute. Washington, D.C. www.ifpri.cgiar.org (accessed 15 February 2002). McKitrick, R. R. (1998), ‘The Econometric Critique of Computable General Equilibrium Modeling: The Role of Functional Forms’, Economic Modeling 15, 543–573. Miller, R. E. and P. C. Blair (1985), Input-0utput Analysis: Foundations and Extensions. Englewood Cliffs, New Jersey: Prentice-Hall, Inc. Minnesota IMPLAN Group Inc. Stillwater, Minn. www.implan.com (accessed 29 January 2002). Pan, X. and S. Kraines (2001), ‘Environmental Input-Output Models for Life-Cycle Analysis’, Environmental and Resource Economics 20, 61–72. Partridge, M. D. and D. Rickman (1998), ‘Regional Computable General Equilibrium Modeling: A Survey and Critical Appraisal’, International Regional Science Review 21, 205–248. Pyatt, G. and J. I. Round (eds.) (1995), Social Accounting Matrices, A Basis for Planning. Washington, DC: The World Bank. Regional Economic Models, Inc., http://www.policyinsight.com (accessed 2 Feb. 2002). Rose, A. (1995), ‘Input-Output Economics and Computable General Equilibrium Models’, Structural Change and Economic Dynamics 6, 295–304. Robinson, S. and D. W. Holst (1987), ‘Macroeconomic Structure and Computable General Equilbirium Models’. University of California at Berkeley Department of Agricultural and Resource Economics Working Paper 450. Sadoulet, E. and A. de Janvry (1995), Quantitative Development Policy Analysis. Baltimore, Maryland: Johns Hopkins University Press. Sample, A. V. and D. C. Le Master (1992), ‘Economic Effects of Northern Spotted Owl Protection’, Journal of Forestry 90, 31–35. Schaffer, W. A. (1999), ‘Regional Impact Models’, in Regional Research Institute, West Virginia University, Web Book of Regional Science. www.rri.wvu.edu/regscbooks.htm (accessed 30 January 2002). Sunday Telegraph (2000), ‘Tax on Pollution Is Threat to Seed Firms’, by Jo Knowsley. London: February 17, p. 9. Sunday Times (2001), ‘Bosses Fume as They Count Cost of Energy Tax’, by Terry Murden. London: April 1, Business. Tanzania Ministry of Natural Resources and Tourism (1998), National Forestry Policy. www.leat.or.tz/publications (accessed 10 December 2002). The Journal (1999), ‘1,500 Jobs Threatened by Energy Tax Burden’, by Gordon Morris. Newcastle, UK: September 17, Business News p. 45. U.S. Congress (1944), ‘Sustained Yield Forest Management Act,’ Ch. 146, 58 Stat. 132. U.S. Dept. of Commerce, Bureau of Economic Analysis (2002), ‘Regional Input-Output Modeling System (RIMSII)’. www.bea.doc.gov/bea/regional/rims/brfdesc.htm (accessed 29 January 2002). U.S. Dept. of Labor, Bureau of Labor Statistics (2002), ‘Covered Employment and Wages (ES-202) Series’. www.bls.gov/cew/cewnote.htm (accessed 30 January 2002). West, G. R. (1988), ‘GRIMP, Input-Output Analysis Computer Program. Versions 6.0. User’s Reference Manual’. St. Lucia, Australia: University of Queensland, Department of Economics. Xie, J. (2000), ‘An Environmentally Extended Social Accounting Matrix: Conceptual Framework and Application to Environmental Policy Analysis in China’, Environmental and Resource Economics 16, 391–406.