IMPORTED MACHINERY FOR EXPORT COMPETITIVENESS. Ashoka Mody * Kamil Yilmaz *

Transcription

1 IMPORTED MACHINERY FOR EXPORT COMPETITIVENESS Ashoka Mody * Kail Yilaz * The World Bank Koç University Washington, D.C. Istanbul, Turkey January 1998 Revised: March 2001 Abstract We analyze the relationship between investent in achinery and export copetitiveness, allowing for iperfect substitution between doestically produced and iported achinery. A translog export price function is estiated for developed, export-oriented developing, and iport-substituting developing countries eploying cointegration techniques for panel data. Between 1967 and 1990, and especially fro about the id-1970s, iported achinery helped lower export prices for export-oriented developing countries. For ost of this period, iported achinery was not a substitute for doestic achinery, although the distinction between the two blurred over tie. Iport-substituting developing countries were unable to harness iported achinery and since the id-1970s, even doestic achinery to reduce costs. JEL classification codes: F12, D24. Keywords: Equipent Investent, Export-oriented growth, Iport-substitution, Cost Reduction. * Extensive coents fro two referees and the editor are gratefully acknowledged. The views expressed in the paper are those of the authors and should not be attributed to the World Bank and its affiliates.

2 1 1. Introduction This paper builds on two recent lines of research: investent in equipent as a source of econoic growth and iported goods as conduits for the international diffusion of technology. We cobine these two thees to assess the effectiveness of iported achinery in increasing export copetitiveness and hence in stiulating growth. 1 Underlining the iportance of achinery in the developent process, De Long and Suers (1991, 1992a, 1992b, 1993) find strong epirical support for a causal relationship between equipent investent and growth in a cross-section of developing and developed countries. In particular, they find that a one-percentage point increase in the share of equipent investent in gross doestic product (GDP) raises the GDP growth rate by 0.34 percentage points. 2 They infer that the doestic R&D and learning activities associated with the production and installation of equipent create generalized benefits for that econoy. The De Long/Suers results are also consistent with the possibility that the foreign knowledge ebodied in iported equipent ay be of significant value to the econoy buying the equipent. Studying the spillovers of knowledge across national boundaries, Coe and Helpan (1995) and Engelbrecht (1997) find that international spillovers are ediated through iported goods: the greater the iports, the higher is benefit of the stock of foreign knowledge. Engelbrecht notes that these papers do not distinguish between different types of iports; such a distinction is likely to be iportant since consuer goods, interediate inputs, and equipent are likely to convey spillovers to differing degrees. Coe, Helpan, and Hoffaister (1997) extend these earlier studies and find specifically that iported capital goods are critical conduits of international knowledge. 1 Several studies suggest that ore extensive trade is associated with higher productivity growth (e.g., Pack and Page 1994 and Srinivasan 1995 and 1999). However, Rodriguez and Rodrik (1999) and Rodrik (1999) are sceptical of such results where they are based on cross-sectional growth regressions. Srinivasan and Bhagwati (1999) express general concern about cross-sectional growth analyses and conclude that nuanced and in-depth studies of individual countries, over a period of tie, provide the clearest evidence in favor of the beneficial effects of greater trade-orientation. 2 In the traditional neoclassical odel, an increase in the investent rate raises output but has no long-run effect on growth rates. The endogenous growth literature identifies conditions under which increased investent has external effects and hence raises growth rates. De Long and Suers go further and find evidence that the external effects are strongest when the investent is in achinery rather than in buildings and structures.

3 2 In this paper, we exaine epirically the differential efficiency of doestically produced and iported achinery. Differences in efficiency arise on account of the access ade possible to the international pool of knowledge and the incentives to effectively use that knowledge (see Table 1). With inforation freely accessible to all, there would be no difference in the efficiency of doestic and iported achines. However, in practice, inforation is not freely available. Even in the absence of foral intellectual property protection, doestic producers can be at a handicap relative to international producers because knowledge is tacit (see, for exaple, Mody 1989 for a review of the econoic and anageent literature on tacitness). International producers can ore readily tap into a uch broader pool of knowledge. As a consequence, doestic and iported achinery trigger different fors of learning in the doestic econoy. In soe instances, innovative new achinery is produced within a developing country. More often, however, the doestic production of achines is associated with adaptive R&D, i.e., developent activity for tailoring foreign achinery. In contrast, iported achinery is bundled with knowledge in various fors: blueprints, installation support, quality control software, and services of trained engineers and supervisors. Such knowledge absorption is less glaorous than the developent, or even the adaptation, of achines. However, because it fors a ore coprehensive package, it can lead potentially to greater short-run efficiency and stronger absorptive capacity in the long-run. Iported achinery will also be ore efficient because it is typically of newer vintage than doestically produced achinery. 3 We use the country s trade regie to proxy the incentives to deploy knowledge. Not all econoies (and firs within the) are able to take advantage of the bundled software and training and of the greater efficiency built into the new vintages of iported achinery. To stay copetitive, firs in countries with high export orientation are likely to have strong incentives to exploit the knowledge flows associated with iported achinery. Iport substitution was based on the preise that, with teporary protection, doestic producers would also have the incentive to tap into and internalize internationally available knowledge. The extent to which this process 3 We are aware that other echaniss for knowledge transission can be iportant, including through foreign buyers of exported goods who provide technical and arketing support in the context of long-ter relationships (Westphal, Rhee, and Purcell 1981, Egan and Mody 1992, Mody and Yilaz 1997).

4 3 actually occurs is an epirical question. Incentives in ore protected iport-substituting econoies are likely to be weaker on account of the relatively sall size of doestic arkets and the less deanding doestic users of that achinery (see Srinivasan and Bhagwati 1999 for a review of how incentives are blunted in an iport-substituting regie). Thus, while alternative explanations are possible, the results of this study are consistent with the proposition that iportsubstituting regies create weaker incentives to invest in technological iproveents that can help increase their greater presence as exporters in international arkets In Figure 1, we plot the change in the volue of exports against the change in the capital stock in the previous year. In export-oriented countries, an increase in exports is strongly associated with an increased stock of iported achinery. A positive relationship also exists between increases in exports and the doestic achinery stock. A siilar set of relationships is found for developed countries. In contrast, for iport-substituting developing countries, iported equipent and export growth are negatively related. Our epirical analysis focuses on the price of exports rather than on export volue. This forulation specifies a link fro iported achinery to reduced costs and prices, which, in turn, lead to greater exports. The relationship between achinery investent and export copetitiveness is analyzed using a odel of iperfect copetition in international arkets, and allows for iperfect substitutability between doestically produced and iported achinery. An export price function is derived based on the deand for exports and the costs of producing the exported goods. A short-run cost function with variable labor and aterials costs and fixed stocks of iported and doestically produced achinery is used. Higher levels of capital stock are expected to lower short-run production costs and hence export prices. Higher productivity of iported achinery would be reflected in greater cost reduction than can be achieved through the use of doestic achinery. The export price equation is estiated for developed countries (DCs) and for lessdeveloped countries (LDCs), and within the latter for export-oriented and iport-substituting econoies. For each country group, the fixed-effects procedure on panel data is used. But first, the t-bar test recently developed by I, Pesaran and Shin (1996) is used to identify, in a panel data context, the presence of stochastic trends in export price and explanatory variables. We find that the variables do have stochastic trends (unit roots) but are also cointegrated, i.e., the regression with the variables easured in their levels results in a stationary error ter. Thus, the export price

5 4 function is estiated in first differences with the stationary error-correction ter fro the levels equation. The estiates obtained describe the relationship in the variations of the variables around their long-run trends. The long-run (or cointegrated) relationship is of less interest since variables such as price and capital stock are all trended upwards. In contrast, the results in first differences reflect the forces that operate when the variables depart fro their long-run trends. The results show that the relationship between iported achinery and equipent and export prices has evolved over tie. We, therefore, estiate the export price function for several windows. For the period , an increased stock of iported achinery is found to lower export prices for developed countries (elasticity: 0.218) and for export-oriented developing countries (elasticity: 0.475) but not for iport-substituting developing countries. Where iported achinery had a cost reducing ipact, did it ainly substitute for doestically produced achinery? For developed countries, substitutability between doestic and iported achinery cannot be ruled out. For export-oriented developing countries, doestic achinery was not a substitute for iported achinery fro the id-1960s to the id-1970s; however, the degree of substitutability increased thereafter. The plan of the paper is as follows. The underlying odel and the epirical specification of the price function are presented in the next section. In Section 3 the data and the econoetric ipleentation are discussed. Results are presented in Section 4. Section 5 concludes and outlines possibilities for future research. Data sources, the definition of variables, descriptive statistics, and the unit root test results are presented in the Appendix. 2. A Model of Iperfect Copetition in Export Markets In setting up the odel, we are guided by the following intuition. The significance attached to achinery investent by soe countries, especially the Newly Industrializing Countries of East Asia (NICs), was not accidental but was dictated by the adoption of an export-oriented strategy and the resulting discipline of international copetition. To aintain arket presence, exporters had to continually reduce production costs and/or enter into the production of higher quality products. Both strategies required substantial investent in new vintages of achinery and equipent. Initially, doestic achinery had lower productivity and, consequently, the scope for substitution of doestic for iported achinery was sall. Over tie the ore advanced countries

6 5 developed the technological capability to produce achinery that could copete with iports fro developed countries. This intuition can be tested by estiating a cost function that includes achinery stock as an explanatory variable. However, data on production costs are difficult to obtain. For this reason, we estiate an export price function, which is based on both deand and cost function paraeters. In a odel of iperfect copetition, anufacturers choose their export price, given deand and cost conditions. While deand depends on copetitors prices and on incoes in target arkets, production costs depend on input prices, output levels, and other variables that shift the cost function, such as the stocks of iported and doestic achinery. We assue that production for doestic and export arkets are two independent decisions and focus on exports. 4 Firs produce export goods through an hoothetic production function, using two variable inputs, labor and aterials (including raw aterials, fuel and electricity), and a quasi-fixed input, the capital stock. Firs are assued to be price takers in input arkets. Consequently, the short-run cost function is separable in variable input prices, on the one hand, and the quasi-fixed input and output, on the other. We assue that each fir exports a differentiated product and chooses its export price to axiize its profits at a point in tie, given the deand curve for its product and the cost of production. 5 When the second-order condition for profit axiization is satisfied, it is possible to solve for the profit-axiizing price, by inverting the first-order condition. The profit-axiizing price is a function of all variables that enter the cost function (wage rate, w, price of aterials, p, and the capital stock, K) plus variables that shift the deand function (naely the copetitors 4 For a siilar assuption and epirical ipleentation, see Feenstra (1989). If arginal costs of production for the doestic arket and export arkets are not flat, influences in one arket will influence the other. Essentially, an oitted variable bias would arise, where the oitted variables refer to deand influences in the doestic econoy. If doestic deand were to shift exogenously, then the arginal costs of production would change, leading to a change in prices charged in both the doestic and international arkets. We believe that these exogenous shifts would be reflected in the doestic input prices (of wages and aterials). However, it is possible that a bias still reains. The direction of this bias is not clear, though. If increased doestic activity leads to ore investent but also higher arginal costs, a larger stock of private capital would be associated with higher export prices the opposite of the relationship that we are hypothesizing.

7 6 average price, p c, and the world incoe, Y). Export price ay also be a function of the exchange rate (e), as discussed below. p = p( p, Y, w, p, e, K) (1) c The elasticity of export price with respect to variable input prices, prices of copeting products, and the capital stock depends on the paraeters cost and deand functions. When the second-order condition for profit axiization is satisfied, a positive elasticity of arginal cost with respect to input prices is sufficient to generate a positive elasticity of the export price with respect to input prices. In other words, the exporter will increase its price following an increase in input prices. Again, with the second-order condition satisfied, decreasing arginal costs with increasing achinery stock is both a necessary and a sufficient condition for the price function to be a decreasing function of achinery stock. Consequently, if the estiated price elasticity with respect to achinery stock is negative, then it follows that technology ebodied in new achinery has a cost reducing effect. For the purpose of epirical estiation and following Mann (1986 and 1989), we siplify the deand function by substituting world price, p w, for copetitors price, p c, and world incoe, Y. The world price variable reflects the influence of pricing decisions of all copetitors and of changes in the world incoe. Thus, using the reduced for price equation, we analyze the elasticity of export price with respect to world price (which cobines the influences of copetitors prices and world incoe ), two input prices, and the two kinds of achinery stock. 6 In the epirical analysis, we assue that the export price decision is best suarized by the translog price function: 5 Since the analysis is restricted to the cost reducing effect of the technology ebodied in existing achinery, the odel is a static one and does not incorporate investent deand for doestic and iported achinery. Analytically, it is not difficult to incorporate the deand for achinery through a dynaic odel. However, due to lack of data on cost of production and the rental price of capital stock, it is not possible to estiate factor deand functions of the long-run odel. 6 Local currency wages and price of aterial inputs were obtained by dividing the corresponding variables denoinated in US dollars by the annual average exchange rate.

8 7 log p = λ + β log p + β log w + β log p + β loge + β log K + β log K Pw w w P e K 2 Pw, Pw w Pw, w w Pw, P w Pw, e w ψ (log p ) + ψ log p log w + ψ log p log p + ψ log p log e d 2 + ψ log p log K + ψ log p log K ψ (log w) + ψ log w log p Pw, K w 1 Pw, Kd w 1 ww w, P d + ψ log wlog e + ψ log w log K + ψ log w log K ψ (log p ) we w, K 1 w, Kd 1 P, P d 1 Kd 1 d + ψ log p log e + ψ log p log K + ψ log p log K ψ (log e) P, e P, K 1 P, Kd 1 d 2 e, K 1 e, Kd 1 K, K 1 + ψ log e log K + ψ log e log K ψ ( log K ) d d 2 K, Kd 1 1 Kd, Kd 1 + ψ log K log K ψ ( log K ) ee 2 2 (2) where K d and K are the doestically produced and iported achinery stock. Note also that a variable with subscript -1 is lagged one period. We analyze the cost reducing effects of the iported and doestically produced equipent separately and test whether they are substitutes or not. Finally, previous studies analyzing export price behavior under iperfect copetition have noted that exchange rates often exercise an independent influence on the price of traded goods. In other words, even if all variables on both sides of the equation are easured in the sae currency, exchange rate oveents see to have a significant ipact on the price of exports (see Feenstra 1989, Ohno 1989, and Mann 1986). By representing the input prices in local currency ters and including the exchange rate as a separate variable, we allow for the possibility that exchange rate changes are not perfectly passed through to export prices. The ain results of this paper reain unchanged if instead we easure the input prices in dollars and drop the exchange rate variable. 3. Epirical Specification and the Data The export price equation is estiated for a cross-section of 14 developed countries (DCs) and 25 less-developed countries (LDCs). The definitions of variables and data sources are presented in Table A.1 in the appendix. Since the odel is derived based on profit-axiizing assuptions for an individual fir, it would be best to use fir- or industry-level data to estiate the price function in equation 2. However, it is not possible to take that route because of data constraints. While data on export prices, input prices and investent can be found for soe anufacturing sub-sectors in soe countries, it is not possible to obtain data on doestic and iported coponents of investent undertaken by each industry. We are forced, therefore, to aggregate all anufactured exports fro a country. Aggregation, however, presents its own probles. The higher the level of aggregation

9 8 the ore difficult it becoes to obtain price indices that reflect fir-level pricing decisions. An aggregate price easure incorporates changes in the coposition of the coodity basket, as well as the arket price of each coodity in the basket. Aggregation can be justified by assuing either a representative fir (as in Feenstra 1989 and Ohno 1989) or a translog aggregate production technology for anufacturing exporters (Pindyck and Roteberg 1983). Note that our focus is on cost reduction. To the extent that changes in the coposition of exports fro one year to the next are iportant, the cost reduction ipact will be blurred. Indeed, if products are oving up the quality ladder, we would expect to find no cost reduction effect. Hence a finding of cost reduction despite that possibility provides soewhat greater confidence in our results. Since our LDC saple includes countries with substantially different developent strategies, we divide LDCs into two groups, export-oriented and iport-substituting, based on the classification used by the World Bank (World Bank 1986; see also Balasubraanya et al ) Changes over tie in trade policy orientation can be expected to introduce probles in ters of classification. However, as can be seen in Table A.2, between the and no ajor shift occurred in the outward-orientation of the countries included in our saple. Before estiating the export price function we test for non-stationarity of the variables using the t-bar statistic proposed by I, Pesaran and Shin (1996) for heterogeneous panels. This is a well-known crucial first step in tie series odels. When a tie series equation contains a non-stationary variable, then the results based on this estiation will be spurious. I, Pesaran, and Shin (1996) have recently extended the stationarity tests to cross section, tie series odel. The test procedure is siple. It is an extension of the widely used Augented Dickey-Fuller (ADF) test to panel data fraework and allows for heterogeneity across groups included in the panel. First, the average ADF unit root test statistics for the panel is obtained as the ean of individual ADF unit root statistics. Next, the expected value and the standard error of the average ADF test statistic under the null hypothesis of a unit root are obtained through Monte Carlo siulation. The t-bar statistic is calculated as the average ADF inus its expected value divided by its standard error. I, Pesaran, and Shin (1996) show that under the null hypothesis of a unit root, t-bar statistic has a standard noral distribution for sufficiently large nuber of countries, N, and nuber of tie periods, T, while N/T goes to zero. Using Monte Carlo ethod they show that t- bar test has ore power than ADF tests applied to each individual in the panel separately.

10 9 The results of the I-Pesaran-Shin test are presented in Table A.3 in the Appendix. We cannot reject the null hypothesis of a unit root for all variables of the price function for all country groups. Consequently, estiating the export price function in levels (equation 2) would generate spurious results. Next, we test for unit roots in the first-differenced variables and reject nonstationarity. This allows us to estiate the equation in first differences. However, one further step is needed before estiating the price equation. If the equation specified in levels is cointegrated, i.e., if the residuals fro the levels equation are stationary, then the lagged residual fro the levels equation needs to be incorporated as an additional variable in the first-differenced equation to allow for adjustent following departures (errors) fro the equilibriu levels. That specification is referred to as the error correction odel. Residuals fro the price equation estiations in levels turn out to be stationary for all country groups (Table A.3). Therefore, we conclude that the price equation variables are cointegrated and the correct specification for the export price function is the error correction odel. 4. Epirical Results Given the results of cointegration tests we estiate the following translog price function where all variables are in first-differences: 7 log p = λ+ β log p + β logw+ β log p + β loge + β log K + β log K Pw w w P e K ψ ( log p ) + ψ log p logw+ ψ log p log p + ψ log p loge Pw, Pw w Pw, w w Pw, P w Pw, e w d 2 + ψ log p log K + ψ log p log K ψ ( log w) + ψ logw log p Pw, K w 1 Pw, Kd w 1 ww w, P + ψ logw loge + ψ logw log K we w, K d 1 w, Kd 1 P, P d P, e P, K 1 P, Kd 1 d 2 e, K 1 e, Kd 1 K, K 1 d d 2 K, Kd 1 1 Kd, Kd 1 1 d 1 Kd 1 + ψ log w log K ψ ( log p ) + ψ log p loge + ψ log p log K + ψ log p log K ψ ( log e) + ψ loge log K + ψ loge log K ψ ( log K ) + ψ log K log K ψ ( log K ) + δε ee 2 2 (3) where ε 1 is the lagged residual fro the cointegration relation. The first-differenced export price equation is estiated using the fixed-effects procedure, which aounts to assuing that countries 7 We do not estiate the first difference of the translog function in equation 2. Instead we define the translog price function in first-differenced log variables, so that we can derive elasticity estiates.

11 10 do differ in ters of the trend coefficient, which could be interpreted as disebodied technical change. 8 For all country groups in our analysis, the translog paraeters are estiated using the data for period. The paraeter estiates are presented in Table 2. The specification test for functional for indicates that the translog function provides a better approxiation of the export price decision than the Cobb-Douglas function. However, the paraeters of the translog function cannot be interpreted directly. Instead, one needs to derive the elasticity estiates of the export price function with respect to input prices, exchange rate, and the iported and doestic achinery using the underlying paraeters of the translog function. These elasticities are obtained by taking the partial derivative of the first-differenced log of export prices with respect to the firstdifferenced log of each variable in the price function in equation 3. Using the ean of each variable over tie and across countries (indicated by a bar over the variable), one can obtain the elasticity estiates: E = β + ψ log p + ψ logw+ ψ log p + ψ loge + ψ logk + ψ logk d Pw Pw Pw, Pw w Pw, w Pw, Pw, e Pw, K 1 Pw, Kd 1 d w w Pw, w w ww w, P we w, K 1 w, Kd 1 E = β + ψ log p + ψ logw+ ψ log p + ψ loge+ ψ logk + ψ logk E = β + ψ logp + ψ logw+ ψ log p + ψ loge + ψ logk + ψ logk d P PwP, w wp, PP, Pe, P, K 1 P, Kd 1 E = β + ψ p + ψ w+ ψ p + ψ e+ ψ K + ψ K d e e Pw, e log w we log P, e log ee log e, K log 1 e, Kd log 1 (4) d K = βk + ψpw, K log w + ψwk, log + ψp, K log + ψe, K log + ψk, K log 1 + ψk, Kd log 1 E p w p e K K Kd = βkd + ψpw, Kd log w + ψw, Kd log + ψp, Kd log + ψe, Kd log + ψk, Kd log 1 + ψkd, Kd E p w p e K logk d 1 The standard error of each elasticity is estiated using the δ-ethod (for a ore detailed treatent see Rao 1973, pp ). Excluding the intercept and the error correction ter, the translog function in equation 3 has 27 paraeters to be estiated. Consequently, one can write the elasticities in equation 4 in atrix notation: E = Z Y, where Y is the 27x1 vector of translog function paraeters and Z is a 6x27 atrix of zeros, ones and the eans of the first-differenced 8 Alternatively, one could assue that individual effects occur on a rando basis rather than being fixed. This iplies that, instead of the constant ter, the individual effect is part of the rando disturbance.

12 11 log variables, as given in equation 4. Using this atrix notation we obtain the variance-covariance atrix of the elasticity atrix E, S E = Z Sy Z, where S y is the variance-covariance atrix of the paraeter estiates for equation 3, excluding the intercept and error correction ter. The estiation results are presented in two parts. First, the full saple period, is discussed; this is the period for which we have coplete data for the variables of interest. 9 The paraeter estiates of the translog price equation are presented in Table 2. The adjusted-r 2, su of squared residuals, Durbin-Watson statistic, as well as Wald tests for hypotheses of interest are also reported. The elasticity estiates derived fro the translog price function are reported in Table 3. Next, in order to undertake a ore detailed analysis of the data, we repeat the estiations of the translog function for sub-saple windows, where each tie we drop one observation fro the beginning of the saple. We present the sub-saple elasticity estiates for different country groups in Tables 4 through Full Saple Period: For all country groups, the error correction ter is negative and significantly different fro zero at the one-percent significance level (Table 2). As iplied by the cointegration test results, there is a long-run equilibriu relationship between the export price and its deterinants. However, in the short-run, deviations do occur fro the long-run relationship. The estiates show that if in the previous year the price was set above the level consistent with the long-run equilibriu, the error will be taken into account and a downward adjustent in price level will be ade in the current year. A coefficient of for LDCs indicates that if an LDC exporter were to set its price at 10 percent above the equilibriu price in the previous year, it will keep this year s change in price at 3.52 percent lower than that iplied by the export price equation. World price elasticity is high when own-price elasticity of deand is high and/or there are significant diseconoies of scale in production. Indeed, world price elasticity approaches one as own-price elasticity approaches infinity, i.e., the deand curve for the country s products is infinitely elastic. As expected, the world price elasticity estiate is the lowest for developed countries (0.34), which face the least elastic deand curve and where diseconoies of scale are However, this is not justified in our case because we did not saple countries on a rando basis. 9 The binding data constraint is iposed by the use of achinery investent data fro the Penn World Tables, which ends in 1990.

13 12 likely to be the weakest. The test result supports the hypothesis that world price elasticity is significantly different fro one for developed countries. World price elasticity for LDCs, on the other hand, is 0.86, which is quite close to one. Taken separately, world price elasticity is 0.86 for export-oriented LDCs and significantly different fro one, whereas it is 0.97 for iportedsubstituting LDCs and not statistically different fro one. Lower wage and aterial price elasticities for LDCs are consistent with their price-taking role in the world arket. A price-taking fir cannot increase its prices to fully reflect increases in unit costs. In contrast, for a fir with arket power, which can influence the export price of its products, the wage and aterial price elasticity would be significantly different fro zero. Wage elasticity is the highest, 0.243, for developed countries. Wage elasticity for LDCs is 0.01 and statistically not different fro zero. The result for LDCs is driven ainly by iport-substituting LDCs. While their wage elasticity is 0.05 and not significantly different fro zero, the wage elasticity for the export-prooting countries is and statistically significant. The aterials price elasticity is significantly different fro zero all groups except for the iport-substituting countries. It is the highest for the developed country group (0.15), and approxiately 0.04 for LDC groups. What is the evidence for the cost-reducing role of achinery and equipent stock? Elasticity estiates for the entire period show that iported achinery has a cost reducing ipact for export-oriented LDCs, while doestic achinery is associated with a cost-reducing ipact for iport substituting LDCs. In both cases, however, the elasticities are significant only at the 10 percent significance levels. For developed countries, the elasticity estiates with respect to doestic and iported achinery are statistically insignificant. The statistical insignificance of these estiates can be due to a ulticollinearity proble. In particular, if doestic and iported achinery perfectly substitute for each other, their cost reducing ipact ay be blurred by considering the as separate variables. The nonnested Davidson & McKinnon (1981) J-test is used to test whether iported and doestically produced achinery are perfect or iperfect substitutes in ters of their cost reduction effect (see also Greene 1997 for a description of the test). If they are iperfect substitutes, then we need to consider their cost reduction effects separately and the price equation with both iported and doestic achinery as separate right-hand-side variables, as in equation (2) above, is appropriate. However, if they are perfect substitutes, then we need to include their

14 13 su, the total achinery stock, as a right-hand-side variable. The usual nested test does not apply in this situation because an alternative to the null hypotheses cannot be constructed based on restricting the paraeters iplied by the null. Because of this property of the odel, iperfect and perfect substitution are nonnested hypotheses. The J-test is used in such situations but because it is a two-way test its use ay lead to inconclusive results. In the first stage (which we call hypothesis test H2) iperfect substitution is the null hypothesis and perfect substitution is the alternative hypothesis. 10 The procedure works as follows. First the predicted export price is obtained under the assuption of perfect substitution (doestic and iported achinery are added to for one capital stock variable). Then this predicted export price is included as an additional variable in the export price estiation under the iperfect substitution assuption. If the coefficient on the predicted export price variable is significantly different fro zero, then iperfect substitution is rejected and perfect substitution accepted. The J-test aounts to testing whether the estiate of the dependent variable obtained under the alternative perfect substitution specification has any explanatory power in the null iperfect substitution specification of the export price function. If it does, then we reject the iperfect substitution hypothesis. The p-values reported in Tables 3 through 6 refer to the significance of the coefficient on the predicted price estiated fro the alternative hypothesis. Next, we take perfect substitution as the null hypothesis and iperfect substitution as the alternative and again conduct the J-test (H3). If the test fails to reject the null perfect substitution hypothesis, then we conclude that the two types of achinery are perfect substitutes. If, instead, the test rejects the null perfect substitution hypothesis, then the iplication is that the two types of achinery are iperfect substitutes. As noted, if the J test rejects the null hypothesis in both directions, then we cannot deterine whether the iported and doestic equipent are perfect or iperfect substitutes. This happens to be the case for developed countries for the full period. Both H2 and H3 are rejected by the J-test. For both export-oriented and iport-substituting LDCs, the J-test rejects the null hypothesis of the perfect substitution at the 5 percent significance level, but not the iperfect substitution hypothesis. 10 We thank an anonyous referee for suggesting the use of nonnested hypothesis tests.

15 14 Finally, as noted, all the variables in our odel have long-ter trends and, as such, estiating the in levels could lead to a spurious regression. However, it turns out that these (stochastic) trending variables are cointegrated i.e., the residual fro the estiated relationship is stationary (non-trending). This iplies that the long-run trends of the different variables and their transforations in the translog function are tied together by a linear relationship. Specifically, the results show that for developed and export-oriented countries, as the stock of iported capital goods grows over tie (responding to the growing needs of the econoy), the export price also trends upwards (reflecting, perhaps, quality iproveents and changes in the coposition of the export basket). We have not presented these level regressions, in part, because though the estiates are consistent, their t-statistics are not reliable because the coefficient estiates follow non-standard distributions. But ore iportantly, this is not the technological or behavioral relationship we are seeking to identify. The trends in levels are being influenced by several forces that are not of iediate interest to this paper. Though the cointegration is soeties referred to as an equilibriu relationship, this equilibriu can reflect the prediction of an econoic odel or, as in the present case, a reduced-for statistical relationship. However, the results in first differences (with the error-correction ter included) are those that our odel seeks to identify. These results iply (especially for the export-oriented econoies, as discussed below in ore detail) that when iported capital goods grow at a pace ore rapid than their long-run trend, then export prices grow at a rate slower than their long-run trend. This is followed by a period of slower growth in iported capital goods, leading to a reversion to the longrun trend, during which tie export prices catch up. Thus, iported capital goods offer a costreduction benefit in periods of accelerated growth. Over the longer-haul those benefits are assiilated and the econoy oves to a new level in quality and/or a different coposition of goods, also supported by the growing stock of capital goods. 4.2 Sub-saple Windows Considerable changes occurred in the extent of arket power and the technology absorption capacity of different countries over the period Sub-saple windows regressions help us study the evolution of elasticity estiates. We start with the full saple. Then, we drop the observation for 1967 and estiate the odel for the sub-saple. Next, we drop the observation for 1968 and estiate the odel for and so on up to the subsaple window. In this fashion, we obtain 13 different estiates of elasticity. In the reainder of

16 15 this section, we analyze and discuss these results in detail. As we ove fro the first ( ) to the last ( ) window, we obtain a better fit for the regressions (the adjusted-r 2 increases fro 0.87 to 0.93 for DCs, fro 0.63 to 0.65 for export-oriented LDCs, and fro 0.45 to 0.56 for iport-substituting LDCs). All elasticity estiates, including the ones for achinery and equipent, evolve in discernible and interesting directions. The downward trend in the world price elasticity estiate is apparent in developed countries. That elasticity declines fro a high of 0.34 in the full saple window to 0.20 in the window (Table 4). The null hypothesis that the world price elasticity is equal to one is rejected for all sub-saple windows considered. This result suggests that anufacturing goods exporters fro these countries have, over tie, enjoyed increasing arket power in international arkets. For LDCs in the full saple period, we do not reject the hypothesis of unit world price elasticity. As we reduce the saple period fro to , the p-values for the Wald test (H1) decrease, and we start to reject the unit world price elasticity even at the one percent level. When we analyze the world price elasticity for the exportoriented and iport-substituting LDCs, we observe a siilar pattern (Table 5). In export-oriented LDCs, wage elasticity estiates are always significantly different fro zero. Furtherore, the wage elasticity displays an upward trend fro to 0.191, as we drop annual observations. Even though the wage elasticity estiate drops slightly as we ove fro to window, it still stays around 0.14 (Table 5). Wage elasticity estiates for iport-substituting LDCs are not significantly different fro zero. We interpret this result to be a consequence of the low wage rates in these LDCs. Hence, sall wage rate increases do not appear to affect the unit cost of production in a significant anner. Materials price elasticities reain in a relatively narrow range for developed countries ( ) and at a lower level for the two LDC groups. Finally, we analyze the elasticity estiates for iported and doestically produced achinery. For the windows, , , and , the elasticity estiates for iported achinery is negative and significantly different fro zero; though the value of the coefficient reains around 0.20, the statistical significance falls off thereafter. Thus, the data suggests that the iported achinery has had a cost reducing effect in developed countries, but priarily in the 1970s. Doestic achinery has been less iportant in this respect. However, for developed countries, the Davidson and McKinnon J-test continues to give inconclusive results. It rejects both

17 16 the null of perfect substitution (between iported and doestic achinery) and also the iperfect substitution null hypotheses for the entire period (see Table 4). If we assue that the two types of capital goods are substitutes and, hence, replace the individual capital goods ters by the su of the two, the coposite capital good is seen to have a consistently negative and significant sign, iplying a cost reduction effect coing fro the total rather than fro either doestic or foreign capital goods. The ore interesting results pertain to trends in the iported and doestic achinery elasticity estiates for the two groups of developing countries. Consider first the export-oriented LDCs. When we include the data for 1967 through 1974 in our estiation, the Davidson and MacKinnon J-test shows doestic and iported achinery to be iperfect substitutes. The p- values for iperfect-substitution null hypothesis (H2) exceed 5 percent for the sub-saple windows fro to In contrast, the p-values for the perfect substitution null hypothesis (H3) are lower than one tenth of one percent. Once we drop the period fro our saple, in the last seven sub-saple windows ( to ) the p-values for the iperfect substitution null hypothesis test fluctuate between 1 and 11 percent. However, since the p-values for the perfect substitution null hypothesis (H3) are always lower than the p-values for the iperfect substitution null hypothesis (H2), we can conclude that iported and doestic achinery are iperfect substitutes for the export-oriented developing countries. Also, for export-oriented developing countries, a substantial difference exists between the iported and doestic achinery elasticities. Iported achinery has significant cost reduction effects in all windows at the 5 percent or higher significance level. The cost reduction ipact of the doestic achinery is not significantly different fro zero. For the ost part, the elasticity estiates for the doestic achinery are positive. These results show that for export-oriented LDCs, investent in iported achinery has helped keep their average production costs under control. For iport substituting LDCs, the windows estiates of elasticities are no different fro the ones obtained for the whole saple. Iported achinery has either no significant effect on export price, or, when it has, it has an adverse cost increasing effect. The elasticity of the doestic achinery is in general negative in iport-substituting LDCs and significantly different fro zero in the sub-saple windows that included data for the period.

18 17 Thus, in both LDC groups we obtain support for iperfect substitution between doestic and iported achinery. Yet, the cost reduction effects of the two types of achinery are copletely different in the two groups. Iported achinery has a significant cost reduction ipact in export-oriented countries but not in iport-substituting LDCs. We interpret this result as follows. The cost reduction effect of a specific type of achinery is not only deterined by the technology ebodied in that achinery. If that were the case, then we should have obtained significant cost reduction of iported achinery in both countries. The incentive regie and hence the trade policy orientation of the country is also crucial. The onset of a sustained exportprootion policy occurred soetie in the id-1970s; our results indicate that the use of iported technology ebodied in equipent becae an increasingly significant source of export copetitiveness fro about the sae point. Though, over tie, exporters were able to obtain doestic substitutes, iported achinery continued to have a significant effect. In contrast, in iport-substituting econoies the trade-orientation liited investent in iported achinery and encouraged the investent in doestic achinery. Iport-substituting countries tended to have a ore significant cost reduction effect through investent in doestic achinery in the earlier years, though that effect faded by the late-1960s. 5. Conclusions The standard odel of pricing decision allowed us to specify an export price function incorporating different degrees of international arket power and dependent on cost function variables including wages and aterials prices, and shift variables such as stock of achinery and equipent. We estiated a translog anufactured goods export price equation using pooled data for developed and less-developed countries. The results show that, while developed countries enjoy substantial arket power, over tie certain developing countries have also coe to acquire a liited ability to influence their export prices. More iportantly, in this paper we provide epirical evidence on the relationship between export copetitiveness and the stock of achinery, allowing for the possibility of iperfect substitution between doestically produced and iported achinery. In the wake of increasing labor costs, countries adopting an export-oriented strategy, especially East Asian NICs, relied heavily on achinery iports to acquire odern technology. Governents and private businesses supported the absorption and adaptation of iported technology through local R&D and

19 18 engineering efforts. Our results show that over tie, and especially since the id-1970s, iported achinery has had an iportant cost reducing effect. The results also show that doestic achinery was not a substitute for iported achinery, though the distinction has begun to blur in recent years. In contrast, in countries where iport-substitution had been the doinant trade strategy, exporters were not able to use iported achinery to iprove their copetitiveness. Neither was doestically produced achinery a sustained aid to international copetitiveness in such countries. An interpretation of the De Long and Suers papers is that since additions to the stock of achinery spur growth, governent policies should support the rapid increase of equipent stock. However, the authors theselves were cautious in this regard and were ore inclined to favor a liberal iport regie, which while rewarding entrepreneurial behavior, would facilitate the inflow of iported equipent and hence foster growth. This paper certainly supports that view. Further work along these lines would benefit fro disaggregated tie series data on anufacturing subsectors. Data on sectoral achinery investent as well as achinery prices would enable endogenizing the use of achinery. Further, our results point to the iportance of trade as a vehicle for transfer of knowledge, identifying capital goods as the conduit. However, recently, Keller (2000) and Branstetter (2001) have argued that while international knowledge transfer through trade ay occur, knowledge spillovers within a country are quantitatively ore iportant. Our results suggest that the relative iportance of internal and external knowledge spillovers ay change over tie as the international environent changes and also as the doestic incentives and absorptive capacity evolve. A further exploration of the deterinants of internal and external knowledge spillovers is also likely to be a fruitful avenue of research.

20 19 REFERENCES Balasubraanya, V. N., M. Salisu, and D. Sapsford, 1996, Foreign Direct Investent in Export-Prooting and Iport-Substituting Countries, Econoic Journal 106: Branstetter, Lee, 2001, Are Knowledge Spillovers International or Intranational in Scope? Microeconoic Evidence fro the US and Japan, Journal of International Econoics 49: Coe, David T. and Elhanan Helpan, 1995, International R&D Spillovers, European Econoic Review 39: Coe, David T., Elhanan Helpan, and Alexander Hoffaister, 1997, North-South R&D Spillovers, Econoic Journal 107: Davidson, R., and J. MacKinnon, 1981, Several Tests for Model Specification in the Presence of Alternative Hypotheses, Econoetrica 49: De Long, J. Bradford and Lawrence H. Suers, 1991, Equipent Investent and Econoic Growth, Quarterly Journal of Econoics 106(2): De Long, J. Bradford and Lawrence H. Suers, 1992a, Equipent Spending and Econoic Growth: How Strong is the Nexus? Brookings Papers on Econoic Activity De Long, J. Bradford and Lawrence H. Suers, 1992b, How Robust Is the Growth- Machinery Nexus? Rivisita Di Politica Econoica 92(11): De Long, J. Bradford and Lawrence H. Suers, 1993, How Strongly Do Developing Econoies Benefit fro Equipent Investent? Journal of Monetary Econoics 32(3): Egan, M.L., and A. Mody, 1992, Buyer-Supplier Links in Export Developent, World Developent 20(3): Inc. Enders, Walter, 1995, Applied Econoetric Tie Series. New York: John Wiley & Sons,

21 20 Engelbrecht, H. J., 1997, International R&D Spillovers, Huan Capital and Productivity in OECD Econoies: An Epirical Investigation, European Econoic Review 41: Feenstra, Robert C., 1989, Syetric Pass-Through of Tariffs and Exchange Rates Under Iperfect Copetition: An Epirical Test, Journal of International Econoics 27: Feenstra, Robert C., Joseph E. Gagnon, and Michael M. Knetter, 1996, Market Share and Exchange Rate Pass-Through in World Autoobile Trade, Journal of International Econoics 40: Giovannini, Alberto, 1988, Exchange Rates and Traded Goods Prices, Journal of International Econoics: 24: Greene, Willia, 1997, Econoetic Analysis. Third Edition. Upper Saddle River, New Jersey: Prentice Hall. Hsiao, C., 1986, Analysis of Panel Data. Cabridge: Cabridge University Press. I, Kyung S., H. Hashe Pesaran, and Yongcheol Shin, 1996, Testing for Unit Roots in Heterogenous Panels, ieo, University of Cabridge. Keller, Wolfgang, 2000, Geographic Localization of International Technology Diffusion, National Bureau of Econoic Research Working Paper Lall, Sanjaya, 1992, Technological Capabilities and Industrialization, World Developent, 20: Mann, Catherine L., 1989, Effects of Exchange Rate Trends and Volatility on Export Prices: Industry Exaples fro Japan, Gerany, and the United States, Weltwirtschaftliches Archiv 129: Mann, Catherine L., 1986, Prices, Profit Margins, and Exchange Rates, Federal Reserve Bulletin 72: Mody, Ashoka, 1989, Fir Strategies for Costly Learning, Manageent Science 35(4):