During the 1990s, the number of regional trade agreements (RTAs) in effect in the world more than doubled. The number of RTAs in effect by August 1998 was 220, compared with 95 in 1990. [2] This has occurred despite the rather pessimistic assessment of the performance of regional trade arrangements in the some parts of the developing world during the previous two decades (see de Melo and Panagariya, eds., 1993 ; or Bhagwati and Panagariya, 1996). Although assessments of the comparative static effects of regional trade arrangements may sometimes be dominated by trade diversion, proponents may argue that dynamic effects will result in gains. But if the comparative static effect is negative, there will be less income for both current consumption as well as investment, and the dynamic impact could be worse. A second question we examine in this paper is the contention that dynamic effects necessarily lead to a superior welfare result for preferential trade arrangements.

We examine these questions with data and a model of Chile. Chile undertook a major reform of its trade policy between 1975 and 1979 during which it converted its non-tariff barriers to tariff barriers. There was a temporary setback in its movement toward liberal trade in the early 1980s, due to a fixed and overvalued exchange rate regime at that time (see Shatz and Tarr, 2001). Since 1984, the hallmark of Chile’s trade policy has been its uniform tariff, which has been progressively reduced over time. [3] In 1998, the Chilean Parliament (supported by testimony from Chilean business interests) voted to reduce its then eleven percent uniform tariff by one percent a year until the tariff reaches a uniform six percent in 2003. In the 1990s, however, Chile adopted a strategy in which it appears willing to enter into a free trade agreement with virually all of its trading partners. In this paper we examine the impact of the regionalism strategy by examining Chile’s free trade agreement with MERCOSUR and its potential free trade agreement with NAFTA. We ask how do various modeling assumptions impact on the quantitative assessment of the results.

The starting point of this paper is our earlier comparative static analysis of Chile’s trade policy options (Harrison, Rutherford and Tarr, 2002). In that paper we employed a comparative static multi-region computable general equilibrium (CGE) model with perfect competition and constant returns to scale. Our first new construction in this paper is a 24 sector small open economy (SOE) comparative static, computable general equilibrium (CGE) model of Chile which replicates the results of our 24 sector multi-region trade model. While it is customary to employ an open economy model to study trade issues for a small economy, in our earlier work we adopted a multi-sectoral framework in order to assure that we could account for any changes in terms-of-trade induced by Chilean accession to MERCOSUR or NAFTA. In the present analysis, we begin by implementing a small open economy model which virtually replicates the results of a multi-region trade model. This key to this reconciliation is the inclusion of the terms-of-trade effects (derived from improved access in the regional arrangements) in the export demand functions for the small open economy. This reconciliation is the first result of our paper. To our knowledge this is the first time a small open economy CGE model has been constructed that replicates the results of a multi-region CGE model including the terms-of-trade effects.

After having validated the small open economy model, we construct a fully intertemporal model with which we evaluate Chile’s free trade arrangement with MERCOSUR and its possible free trade arrangement with NAFTA. We examine the models in our preferred elasticity case that we call a central elasticity case as well as and a low elasticity scenario. Our preferred elasticities are high by the standards of many CGE models, but we believe they are justified by the work of Reidel (1988) and Athukorala and Reidel (1994). [4] We compare results across the multi-region trade, small open economy static model and small open economy dynamic model.

Our principal result is that the dynamic SOE model does not produce welfare estimates significantly different from the comparative static SOE model. Simply allowing dynamic capital accumulation in the style of Ramsey does not result in very different welfare estimates than a comparative static model. On the other hand, we have shown (see Rutherford and Tarr, 2002), that an endogenous growth model can result in significantly larger estimated gains than a comparative static model. But this paper shows that not any dynamics will do for large welfare gains.

We have estimated that Chile’s agreement with MERCOSUR will result in losses for Chile under our central elasticity assumptions. [5] We estimate, however, that in one of our models the estimated losses from the dynamic SOE model are larger in absolute value than in the comparative static SOE model. Our second result then is that it is possible for a dynamic model to produce welfare estimates for a preferential trade area that are welfare inferior than those from a comparative static model. Thus, regional trade agreements that are immizerizing when evaluated in a comparative static framework do not necessarily have improved evaluations in a dynamic framework and these evaluations can be worsened.

In addition, we examine the impact of comparative steady-state models. In an attempt to obtain estimates of the long run benefits of trade liberalization without actually constructing a fully dynamic model, in recent years a number of authors have produced results using steady-state models (this includes Harrison, Rutherford and Tarr (1996, 1997a) ; Francois, McDonald and Nordström (1996) ; and Baldwin et al. (forthcoming)). Whereas static models hold the capital stock fixed and allow the rental rate on capital to adjust, the steady-state model allows the capital stock to adjust with a fixed ratio of the rental rate on capital to the cost of producing the capital good. Since the comparative steady-state model does not account for the foregone consumption of achieving a possibly larger capital stock, we have emphasized that (when the capital stock increases) estimates from the comparative steady-state model are upper bound estimates of results from a dynamic model. We show that in our basic steady-state model, the estimated gains are about four to five times larger than the gains from the comparative static model.

We have also developed estimates from a model we refer to as the calibrated steady-state model. This model corrects for inconsistencies in an unadjusted comparative steady-state model to assure that the model meets dynamic steady-state equilibrium conditions. These adjustments result in a lower initial capital stock and a smaller increase in the capital stock after the shock (when the capital stock increases) and lower estimated gains. The estimated gains are then closer to true gains which come from the dynamic model. These results suggest that if modelers wish to take the shortcut of using a comparative steady-state model to estimate the gains from trade liberalization, in order to get more accurate estimates, it is necessary to do further calibration to assure that the dynamic steady-state conditions are satisfied.

In the next section we describe the model in some detail. Detailed results and interpretations are in the following section.

where the domestic-import composite is defined :

in which is the elasticity of substitution between domestic and imported goods.

This elasticity is greater than unity, so domestic and imported varieties remain net substitutes in demand.

s.t.

in which government income includes value-added taxes, import tariff revenue (accounted on a bilateral basis) and the sum of all other direct and indirect taxes (TX). The tax rate on factor inputs in sector i equals in which τ is a tax adjustment multiplier which increases from unity to replace declining tariff revenues and hold government expenditure constant.

In this equation parameters and are the base year (benchmark) levels of output to the domestic and export markets, and is the benchmark value share of domestic sales in total output for sector i.

Production of this composite is associated with a nested production function which has a Leontief aggregation of intermediate inputs and value added. In turn, value added is a constant elasticity composite of capital, labor, land and (for extractive industries) sector-specific resources. Intermediate demand for good j is a CES Armington aggregation of domestic and imported varieties. Hence, we have :

and

s.t.

g_i(D_i,E_i) = Y_i

Domestic and imported intermediate demands solve :

s.t.

and primary factor inputs solve :

min

s.t.

f_i(K_i, N_i, L_i) = Y_i

Regional shares of total imports are determined by cost-minimization, given a constant elasticity Armington aggregation of the form :

and the cost-minimizing composition of commodity i imports is determined by :

s.t.

in which is the base year value share of transport costs on exports of good i to region r, is the ad-valorem tariff on commodity i exports to region r, is the export tax, and the reference prices are defined such that in the benchmark equilibrium :

The market clearance condition for domestic output is : [7]

and market clearance for composite imports is :

There are factor markets for labor, capital, and land :

Finally, there is a trade balance condition which relates the value of imports to the value of exports less any exogenous capital in ows  :

Based on the original work of Hansen and Koopmans (1972) and Dantzig and Manne (1974), we assume that given the rate of return on capital and the cost of producing a unit of the investment good in the initial equilibrium, the capital stock in each country is optimal. That is, increases in the rate of return on capital relative to the cost of a unit of the capital good would induce an increase in investment until the marginal productivity of capital is driven down to the initial ratio of the rate of return on capital to the cost of producing the capital good. A change in trade policy will produce a new equilibrium, where for many of the changes we consider, the rate of return on capital increases (relative to the cost of investment) due to a more efficient allocation of resources. This implies that in a dynamic sense a fixed capital stock can no longer be optimal in the new equilibrium of the comparative static model-investment would be forthcoming until the marginal productivity of capital is reduced to the long run equilibrium where the ratio of rate of return on capital to the cost of the capital good is restored to its initial value. In the comparative static model we allow the price of capital to vary, while holding constant the aggregate stock of capital. The steady-state calculation essentially reverses this : we allow the capital stock (and investment demand) to be endogenously determined while holding constant the price of capital. [8]

Since our calculation ignores the forgone consumption necessary to obtain the larger capital stock, we believe that this calculation measures an upper bound on potential welfare gains in a long run classical Solow type growth model. [9] Of course, it could be an underestimate of the long run gains since it fails to capture endogenous growth effects. This approach to steady-state evaluation in a multi-regional trade model was implemented in Harrison, Rutherford and Tarr (1996 ; 1997a), and has also been used by Francois, McDonald and Nordström (1994 ; 1996).

Since we observe an increase in the rental rate of capital relative to the cost of the investment good in virtually all of the scenarios we examine, the capital stock must grow in the steady-state version to keep the ratio at its benchmark value. This expansion of the capital stock then works through our model like an “endowment effect”, generating larger welfare gains since there are more resources to be employed.

and investment demand is scaled proportionally :

Both of these effects enter into the representative consumer’s budget constraint :

max U(c^D,c^M)

s.t.

Let qrepresent the marginal cost of a unit of new vintage capital. In our model this cost function has the form :

In the steady-state model, the capital stock adjusts so that the ratio of the rental rate on capital to the cost of producing a unit of the capital good is constant :

implying that in the long-run equilibrium, the return to capital is equal to the sum of the discount rate on future consumption plus depreciation.

In this equation parameter controls the intertemporal elasticity of substitution [10] and ∆ is the single period discount factor. As in the static and steady-state models, aggregate consumption in a given period C_t is a Cobb-Douglas aggregate of consumption of domestic and imported final goods :

where

The intertemporal and within period consumption decisions are weakly separable. Thus, the typical static first order condition applies on consumption decisions within a time period, given a decision on how much to spend on consumption in any period. In the standard manner, the intertemporal decision is based on the maximization of utility subject to the constraint that the present value of income less expenditures is zero :

max U(C^D,c^M)

s.t.

In this expression, all prices are defined in present value terms, discounted to period 0 (1992). The present value of income includes the value of the entering (period 0) capital stock, the present value of land rents, the value of sector-specific resource rents, the present value of wage income and all other taxes and transfers, respectively. (The final term includes the net value of transportation sales and purchases which are fixed at benchmark level).

Table 1
Sectors in the model

Table 2
Sector-specific elasticities

Table 3
Benchmark tax rates

Table 4
Other parameters

Table 5
Welfare effects for Chile in NAFTA : alternative models

Table 6
Welfare effects for Chile in MERCOSUR : alternative models

In the case of the MERCOSUR agreement, with low elasticities the results of the static and dynamic models are the same. With high elasticities, however, the losses are slightly larger with the dynamic model. The reason for larger losses in the dynamic model is that with high elasticities, MERCOSUR induces a loss to Chile in the static model (due to trade diversion costs and terms-of-trade losses as partner countries increase their prices of exports) and the return to capital declines in the static model ; but the static model holds the capital stock fixed. In the dynamic model, the adverse impact on the return to capital in Chile from MERCOSUR induces a decline in investment and the capital stock declines relative to a steady-state capital stock prior to MERCOSUR. This shows that a dynamic model may produce larger losses when there are estimated losses from a static model.

We have also developed estimates from a model we refer to as the calibrated steady-state model. The first step in building the dynamic model is to construct a steady-state equilibrium growth path. This requires reconciliation of data on the value of capital earnings and the value of investment with depreciation, growth and interest rates. As has been emphasized by Mercenier and Michel (1995), data on the value of capital’s earnings are typically too large for consistency, in part because the value of proprietor’s income is attributed entirely to capital. Thus, we reduce capital’s earnings and the implied capital stock in the initial data which allows us to calibrate a consistent reference steady-state growth path for the dynamic model. In particular, when the model is calibrated to a 2 growth rate, 7% capital depreciation rate and 5% interest rate, the capital value share is reduced by 33%. We impose the same adjustment in the benchmark equilibrium for the calibrated steady-state model. With a lower initial capital stock, the induced increase in the capital stock from NAFTA is also smaller, so that the gains from the calibrated steady-state model are less than in the unadjusted steady-state model. We conclude that it is important when using a steady-state model to calibrate the data set for dynamic consistency. In particular, there is a relationship between the interest rate, the depreciation rate and the growth rate that is consistent with the conditions necessary for steady-state equilibrium. In addition to theoretical consistency, doing so will typically produce estimates closer to the true estimates of the fully dynamic model.

Our primary conclusion in this subsection is methodological. In practical general equilibrium work it makes sense to initially analyze policy equations using a static general equilibrium model. These results can then be compared to the corresponding steady-state model, recognizing the importance of constructing a consistent steady-state growth path with plausible underlying parameter values for the interest rate, growth rate and depreciation rate. A dynamically consistent steady-state model is likely to produce estimates closer to the true dynamic model.

If the static and steady-state equilibrium results are close, then it may be ill-advised to spend time developing a fully dynamic model of the transition path. If they are quite different, then the transition effects may play a crucial role in evaluating the efficiency effects of a policy change.

Let q represent the marginal cost of a unit of new vintage capital. In our model this cost function has the form :

In the steady-state model, the capital stock adjusts so that the ratio of the rental rate on capital to the cost of producing a unit of the capital good is constant :

implying that in the long-run equilibrium, the return to capital is equal to the sum of the discount rate on future consumption plus depreciation.

When this ratio rises as a result of a shock, there will have to be an increase in the capital stock to reduce the marginal productivity of capital to restore equilibrium. A fall in this key ratio, however, will induce a fall in the long run equilibrium capital stock.

It is known result in the Hecksher-Ohlin literature that the rental rate on capital relative to the wage rate can go up or down depending primarily on whether the shock induces an increase or decrease in the relative price of the capital intensive good. In general, trade liberalization could favor labor intensive industries as easily as capital intensive ones, so there should be no presumption that the rental rate of capital will increase relative to the wage rate. In some developing countries, trade liberalization would be expected to favor agriculture, which is likely to be labor intensive. In these countries we would expect to see a fall in the rental rate of capital relative to the wage rate.

But it is not the rental rate on capital relative to the wage rate that is important for the long run capital stock. The relevant denominator in the ratio which determines the long run equilibrium level of the capital stock is the price of a unit of capital, q. In a model of homogeneous goods, the Stolper-Samuelson theorem implies that if the rental rate falls relative to the wage rate, the rental rate must also fall relative to the price of the goods in the model. That is, if the rental rate falls relative to the wage rate, then the ratio r_K/q would also fall in a Hecksher-Ohlin model of trade.

But we employ the Armington assumption, in which imported and domestic goods are imperfect substitutes. Due to the Armington assumption, it does not follow that the rental rate must fall relative to the price of all goods, if the rental rate falls relative to wages. As indicated in the cost function above, the capital good is produced by both domestic and imported inputs as well as labor and capital, and a trade liberalization will reduce the price of imported inputs. Thus, there is a general presumption that q will decline. Even when trade liberalization shifts resources to labor intensive industries and induces a fall in the rental rate of capital relative to the wage rate, the price of a unit of capital, q, could fall more inducing a rise in the capital stock. Although the ratio r_K/q could fall, we would normally expect it to rise. This explanation justifies the presumption that a steady-state model will produce greater estimated gains than a static model, when the static model produces gains.

In Harrison, Rutherford and Tarr (1997a, table 5) we estimated the gains from the Uruguay Round on 24 regions in both comparative static and steady-state models, under four different policy trade liberalization policy shocks, i.e., 96 numerical simulations were presented. In the vast majority of the cases we found that the steady-state gains were larger than the comparative static gains. Other authors, such as Francois, McDonald and Nordstom (1996) found similar results. There were a few cases, however, where we found that the gains in the comparative steady-state model were smaller than the gains from the comparative static model. [11] These exceptional cases illustrate that the rental rate on capital can decline by even more than the cost of capital such that the capital stock and the welfare estimates decline in the steady-state. But the fact that the large majority of cases result in larger gains in the steady-state reflects the decline in the cost of the capital good due to the trade liberalization.

We developed a fully dynamic small open economy model of Chile with constant returns to scale in all sectors The principal difference between the static and dynamic models is that in the dynamic model the capital stock is optimized along the steady state growth path based on the consumption-investment tradeoff. Our key result is that this kind of dynamic model need not produce estimated welfare gains significantly different from a static model. The reason is that since the capital stock is already optimized on the steady state path based on the consumption-investment tradeoff, trade liberalization that induces a larger capital stock does not necessarily improve welfare. So not any kind of dynamics is sufficient to produce larger gains from trade liberalization and gains from a regional arrangement that is welfare reducing in a static model. Moreover, in the case of the Chile-MERCOSUR agreement, we estimate slightly larger losses in the fully dynamic model that we estimate in the static model. Thus, when dynamic impacts are taken into account, there may be larger losses than in a static model.

We estimated the impact of Chile-NAFTA and Chile-MERCOSUR in steady state models. We found little difference in the static and steady state results for Chile-MERCOSUR. For Chile-NAFTA, the gains are substantially larger in a comparative steady state model. It is known that comparative steady state models systematically overestimate the gains compared to a fully dynamic model with the same features, since the foregone consumption costs of attaining a higher capital stock are ignored. But we show that an additional bias is that the calibration may be inconsistent with a steady state growth path. When we reduce the calibrated capital stock to be consistent with a steady state growth path, the estimated gains in a properly calibrated comparative steady-state growth model may be significantly less than in a comparative steady-state growth model that is not consistently calibrated.

Going beyond the models of this paper, proponents of regional trade agreements might argue that learning and technology transfer effects from RTAs would result in much larger gains, and would produce gains even if the static effects are dominated by trade diversion. That is, an endogenous growth model would produce estimated gains for Chile forming a free trade arrangement with MERCOSUR even though a comparative static model does not. Rutherford and Tarr (2002) have estimated that when these learning and technology effects are taken into account endogenously in a fully dynamic model, the estimated gains from trade liberalization will be many multiples of the estimated gains from a static model. Moreover, foreign direct investment could increase as a result of a RTA, and this could have additional benefits.

But there is a dynamic form of trade diversion in models that allow for productivity impacts and technology transfer from imports as well. That is, while regional preferences will encourage additional varieties and technology imports from regional partners, it will discourage imports and additional varieties from the rest of the world. It is likely to be very important in this context to question how technologically advanced and how big is the prospective partner. For large developed regions like the European Union or NAFTA, the additional technology imports are likely to be sufficiently large to offset the losses from the rest of the world. Then the dynamic model will produce gains from regional arrangements with technologically advanced partners that are several multiples of the estimated gains from static models. On the other hand, if a RTA is made with a technologically less advanced region, the diversion of new technologies or varieties from the rest of the world could hinder productivity advances in the home country. On balance growth and welfare may be reduced and it may result in losses several multiples of the estimated static losses [12].

T.F. R. & D.G. T.

Article received on March 6, 2002

Accepted on May 12, 2003