Employment Responses of Skilled and Unskilled Workers at Mexican Maquiladoras: The Effects of External Factors

World Development Vol. 37, No. 7, pp. 1285–1296, 2009 Ó 2008 Elsevier Ltd. All rights reserved 0305-750X/$ - see front matter www.elsevier.com/locate/worlddev

doi:10.1016/j.worlddev.2008.10.008

Employment Responses of Skilled and Unskilled Workers at Mexican Maquiladoras: The Effects of External Factors ANDRE´ VARELLA MOLLICK * University of Texas – Pan American, Edinburg, TX, USA Summary. — This paper compares a standard labor demand model to another augmented by the real exchange rate (RER) when studying the growth of dynamic Mexican maquiladoras from 1990 to 2006. The basic question is whether the real value of the peso changes the responses of employment to wages and to capital. In the augmented models, skilled labor employment becomes very sensitive to its own wages. Also, the user cost of capital has a larger negative impact on skilled employment, and a weaker currency has a negative impact on unskilled employment only. Overall, skilled labor benefits relatively more under globalization. Ó 2008 Elsevier Ltd. All rights reserved. Key words — employment, labor demand, maquiladoras, Mexico, real exchange rate

cal specialization and intra-firm trade among the NAFTA partners as surveyed by Kose, Meredith, and Towe (2005). 2 This article differs from previous studies in three major ways. First, a special role is assigned in this paper to the real exchange rate (RER). This variable captures all information in goods and financial markets, especially after the peso crash of December 1994. In fact, the evolution of the real exchange rate is very telling, as the evidence in Fajnzylber and Maloney (2005) has suggested for Mexico. Fullerton and Sprinkle (2005) further suggest that error–correction models capture trade flows between the United States and Mexico well, responding in heterogeneous ways to changes in income, prices, and the exchange rate. Additionally, RER influences the Latin American unemployment rate with a pooling coefficient of 0.57, as evidenced by Frenkel and Ros (2006), implying that an appreciation of the RER is associated with an increase in the unemployment rate two years later. For Mexico only, the estimated coefficient was 0.86, which suggests that a stronger peso leads to a higher unemployment rate (or lower employment). Theoretical models of the impact of RER on wages include Faria and Carneiro (2003). This set of results linking Mexican labor and financial markets warrants further investigation with respect to the links between employment and the real exchange rate. There are no studies examining these links for Mexican maquiladoras, despite the presumption that the impact of real exchange rate movements should be more significant for relatively more open sectors. 3 While there is no consensus on the perfect measure of trade, Rodrı´guez and Rodrik (2000) have argued that indicators of ‘‘openness” used by researchers are poor measures of trade barriers or are highly correlated with other sources of bad economic performance. Due to the nature of the maquiladoras, tariffs are not the correct price measure of trade (or financial) liberalization; we follow representative works on the Mexican economy that employ RER as the competitiveness indicator, such as Kamin and Rogers (2000).

1. INTRODUCTION In a global world, foreign capital moves quickly to profit opportunities. Empirical evidence in Ramı´rez (2006) shows that foreign capital is an important determinant of Mexican wages in the long run through its effects on labor productivity. At the same time, the Mexican peso has become increasingly more responsive to market forces after removal of the peg to the US dollar. This paper examines these two issues within the context of hiring by Mexican maquiladoras, which import inputs (mostly from the United States), process them, and send the product back to the country of origin. 1 Embedded in this research strategy is the idea that international trade can increase the own-price elasticity of demand for labor. Slaughter (2001) examines US manufacturing from 1961 to 1991 and concludes that demand for US production labor became more elastic overall, a fact not observed for non-production labor. Rodrik (1997) provides a model in which the elasticity of demand for domestic labor increases (in absolute value) with the international mobility of physical capital. Inspired by the Le Chatelier-Samuelson principle, the demand for any factor (e.g., labor) becomes more elastic when other factors (e.g., capital) can respond to changes in the economic environment with greater ease. The decision to examine this question within the Mexican maquiladoras can be justified by the sector’s growing importance in the national economy. Figure 1 provides an idea of the increasing relevance of the maquiladora sector to the Mexican economy, as captured by the relative share of maquiladora exports to total Mexican exports (EXPORTSHARE) and of imports (IMPORTSHARE) for the period 1980–2006. The period depicted in Figure 1 is one decade longer than that covered by the empirical work, but it does provide a longer-term view of the increasing importance of maquiladora activity in total Mexican trade. Figure 1 documents the jump from about 15% in the early 1980s to close to 50% of total exports in the early 2000s; maquiladora imports display similar behavior. In order to answer how the demand for labor within the maquiladora sector has changed with an increasingly more open Mexican economy, in this paper we estimate several forms of a standard microeconomic labor demand model (with no trade-based considerations) for the maquiladora sector over the period 1990–2006. The period conveys changing trade under globalization, with Mexico’s exports shifting toward manufactured goods and a substantial increase in verti-

* The author wishes to acknowledge comments from Joa˜o Faria, Miguel Leo´n-Ledesma, and Jose´ Paga´n on previous versions of this article. Two anonymous referees of this journal provided useful insights on the contents and exposition of a previous version. The usual disclaimer applies. Final revision accepted: October 8, 2008. 1285

1286

WORLD DEVELOPMENT

plied in the empirical analysis. Section 4 contains the main findings of this article and Section 5 summarizes these results.

Export and Import Maquiladora Share

.55 .50 .45

2. THE THEORETICAL FRAMEWORK

.40

As in Milner and Wright (1998), we assume the maquiladora firms (indexed by i) in Mexico share a Cobb–Douglas production function as follows:

.35 .30

Y i ¼ Ac K ai Lbi ;

.25

where Y represents real output, K is the stock of capital, L is the number of employees, and A is a productivity factor, whose parameter (c) allows for changes in the efficiency of the production process. The a and b parameters are factor share coefficients of the inputs. The profit-maximizing firm (the i subscript is omitted henceforth) will employ K and L at levels such that:

.20 EXPORTSHARE

IMPORTSHARE

.15 90

92

94

96

98

00

02

04

Figure 1. Maquiladora shares of Mexican exports (EXPORTSHARE) and of Mexican imports (IMPORTSHARE).

Second, the model estimated herein is a time series version of the panel data methods in Milner and Wright (1998), Greenaway, Hine, and Wright (1999), Mollick (2003), and Fu and Balasubramanyam (2005), with an important modification. Our cointegration-based framework not only handles the non-stationarity features of the data, but does not assume that variations in the user cost of capital can be captured by time dummies under perfect capital markets. Therefore, we allow for US real interest rates to affect the activity of the maquiladoras. As Neumeyer and Perri (2005, p. 345) put it, ‘‘the need for working capital to finance the wage bill makes the demand for labor sensitive to the interest rate.” Our modification is particularly important for Mexico since Ramı´rez (2006) reports that private capital and foreign capital have positive effects on the long-run rate of labor productivity growth. Third, we allow for cross-wage effects on labor demand. While Robertson (2007) finds that Mexican and US production workers are complements, no study has looked at the effects of cross-wages within the maquiladora industry. Given the absence of (public) data on perks and incentives, we conjecture that the decision to hire unskilled workers may also depend on the increasingly higher costs of more qualified workers. Skilled wages could thus serve as good proxies for other types of payments. The management literature in Miller, Hom, and Gomez-Mejia (2001) documents that some types of non-wage payments (profit sharing and saving plans) lower turnover at maquiladoras as high-tech firms use advanced manufacturing technologies more intensively. Focusing on the impressive growth of Mexican maquiladoras from 1990 to 2006, this paper examines whether the competitive peso changes the response of employment to wages and to the price of capital in what we call the augmented model. Several new findings are reported. First, the augmented model adds significant information to the baseline model. Second, the augmented specifications lead to higher price-elasticities of skilled employment to wages. Third, fluctuations in the user cost of capital yield robust complementarities between capital and skilled labor. Fourth, employment responds differently to cross-wage movements: unskilled employment falls with increases in skilled labor wages, but the reverse is not true. Fifth, real exchange rate depreciations contribute to a drop in unskilled employment. Overall, these findings are consistent with the maquiladora labor adjustment being felt more severely by the less skilled. This paper is organized as follows: Section 2 presents the theoretical framework and Section 3 introduces the data to be ap-

ð1Þ

MRP L ¼ w ¼ @Y =@L ¼ Ac K a bLb1 ; c

MRP K ¼ r ¼ @Y =@K ¼ A aK

a1

ð2aÞ

b

ð2bÞ

bL :

b

c

a

From (2b), we know that L = Kr/(A aK ), which substituted into (2a) yields: K ¼ ðwaLÞ=ðbrÞ ¼ ðw=rÞðaLÞ=b:

ð3Þ

Finally, substituting in (1) yields the equation for the i-th firm’s output: Y i ¼ Ac ½ðaLi =bÞðw=rÞa Lbi :

ð4Þ

Applying logarithms to both sides of (4) leads to the following equations: log Y i ¼ c log A þ a log a þ a log Li  a log b þ a logðw=rÞ þ b log Li ; log Y i ¼ ðc log A þ a log a  a log bÞ þ ða þ bÞ log Li þ a logðw=rÞ;

ð5Þ ð6Þ

Dividing through by (a + b) and rearranging yields: log Li ¼ ½1=ða þ bÞðc log A þ a log a  a log bÞ  ½a=ða þ bÞ logðw=rÞ þ ½1=ða þ bÞ log Y i

ð7aÞ

or log Li ¼ b0 þ b1 logðw=rÞ þ b2 log Y i ;

ð7bÞ

where b0 = [1/a + b] (c log A + a log a  a log b; b1 = [a/ a + b]; and b2 = [1/a + b]. Both w and r measure cost factors and Y, if measured by value added in maquiladoras (VA), comprises further items such as raw materials, profits and general expenses. We will employ below W as real wages paid in the maquiladoras and R (the user cost of capital) as the relevant cost of capital for maquiladoras operating in a global environment under global production sharing. To avoid endogeneity problems, we consider US industrial production (YUS) instead of value added in the maquiladoras. This ensures that our measure of demand is completely exogenous to the industry. 4 If the technical efficiency of the production process increases over time and the competitiveness index measured by the real exchange rate (RER) systematically affects overall production efficiency, the parameter A takes the form: At ¼ ed0T RERtd1 ;

ð8Þ

EMPLOYMENT RESPONSES OF SKILLED AND UNSKILLED WORKERS AT MEXICAN MAQUILADORAS: THE EFFECTS OF EXTERNAL FACTORS

A

1.000,000

800.000

600.000

400.000 L

log Lt ¼ /0 þ b0 trend þ b1 log W t þ b2 log Rt þ b3 log YUS t

LU

200.000

ð9Þ

where /0 ¼ ½1=a þ bða log a  a log bÞ; b0 ¼ d0 ½c=a þ b; b1 ¼ b2  ½a=a þ b; b3 ¼ ½1=a þ b; and b4 ¼ d1 ½c=a þ b. The baseline specification (7b) has no trade- or capital-based variables, and Eqn. (9) contains business cycles and RER factors. Based on the standard theory of labor demand and related literature for (9), we expect b1 < 0, b2 < 0 (if capital and labor inputs are complements in production), b3 > 0, and the sign on b4 is open and will be discussed in detail below. Eqns. (7b) and (9) will be estimated by cointegration methods below. We also estimate labor demand equations separately and allow for cross-wage effects changing the impact of skilled labor on unskilled labor demand and vice versa. This allows for substitution of labor types. We conjecture that the decision to hire unskilled workers may also depend on the increasingly higher costs of more qualified workers. The management literature in Miller et al. (2001), for example, documents that some types of non-wage payments (profit sharing and saving plans) lower turnover at maquiladoras. See Sargent and Matthews (2004) for reasons other than technology for maquiladora firms to keep operating in mature, higher cost zones. Partial elasticities are defined in the literature of demand for labor, such as Draper and Manders (1997). For example, for a vector of inputs xi, gij = o log xi/o log pj represents the ordinary (compensated) elasticity of factor demand i with respect to the price of factor j and git = o log xi/o log (trend) represents the relative change in factor demand i due to technical progress. We say two factors are complements if their cross-price elasticity is negative and two factors are substitutes if their cross-price elasticity is positive.

3. THE DATA All labor data from the maquiladoras are from Mexico’s INEGI in the ‘‘Banco de Informacio´n Econo´mica” (BIE) section, under the subsection on IME (industria maquiladora de exportacio´n) (http://www.inegi.gob.mx). As measures of an industry’s employment (L), non-production workers (Ls) and production workers (Lu) are considered, as can be seen in Figures 2A and B. Figure 2A depicts the growing trend of employment figures at Mexican maquiladoras, with a visible comovement between total employment and unskilled labor employment. Figure 2B displays skilled labor employment over time. For the real monthly wage variable in pesos in Figure 3, the deflator is Mexico’s INPC, with third and fourth weeks of June 2002 = 100. The notation is as follows: W cap-

90

B

92

94

96

98

00

02

04

110.000 100.000 90.000

Number of Workers

þ b4 log RERt þ et

1.400,000

1.200,000

Number of Workers

where d0, d1 > 0, and T is the time trend. The real exchange rate operates as an important shift factor of the demand for labor. See Fullerton and Schauer (2001), Mollick (2003), and Robertson (2003). 5 Other theoretical frameworks include the recursive model by Faria and Carneiro (2003) showing that changes in aggregate output and in the real exchange rate precede changes in real wages. Frenkel and Ros (2006) discuss three conceptually different mechanisms as well. Trade-based modifications have played important roles in recent studies. See Greenaway et al. (1999) for treatments with import and export penetration ratios in the United Kingdom and Fu and Balasubramanyam (2005) for export penetration and FDI inflows into China. Substituting (8) into (7b), separating wage from user cost factors, and introducing the white-noise error term (et), the augmented equation becomes:

1287

80.000 70.000 60.000 50.000 40.000 LS 30.000 1990

1992

1994

1996

1998

2000

2002

2004

Figure 2. (A and B) Employment in Mexican maquiladoras: total labor (L); skilled labor (Ls); and unskilled labor (Lu).

tures total wages; Ws refers to wages for white-collar workers: ‘‘sueldos” for ‘‘empleados”; and Wu refers to wages for bluecollar workers: ‘‘salarios” for ‘‘obreros.” Figure 3 also shows that real wages (especially those of unskilled workers) fall with the peso crisis of 1995 and that real wages have risen since then. The output measure of maquiladora production is VA (real value added of exports, charged for maquiladora services), deflated by Mexico’s equivalent of the consumer price index (CPI), the I´ndice Nacional de Precios al Consumidor (INPC, December of 2003 = 100). Its source is Mexico’s INEGI (http://www.inegi.gob.mx). In order to mitigate endogeneity problems, we employ US industrial production (YUS) as the demand variable in the employment equation. US output at monthly frequency is proxied by the Industrial Production Index (code INDPRO) from the Board of Governors of the US Federal Reserve (http://www.frbstlouis.com/), not seasonally adjusted, 2002 = 100. The empirical section below includes 11 seasonal dummies to deal with seasonal adjustment. Any measure of the cost of capital that excludes the inflation rate is misleading. There is also a component due to the corporate tax rate as discussed in Gale and Orszag (2005), but this is difficult to incorporate and is typically overlooked in practice. For these reasons, we calculate the real user cost of capital (R) by the difference between the nominal interest rate and the rate of inflation, adjusted for the depreciation rate (of 6% in Figure 4). After experimentation, we employ the real user cost

1288

WORLD DEVELOPMENT 120 YUS

7.200 6.800 6.400 6.000 5.600 5.200

W

4.800 1990

1992

1994

1996

1998

2000

2002

2004

U.S. Index of Industrial Production

Real Monthly Wages in Pesos

7.600

90

80

70

60

16.000

1990

1992

1994

1996

1998

2000

2002

2004

15.000 14.000

11

13.000 12.000 11.000 10.000

WS

9.000 1990

1992

1994

1996

1998

2000

2002

2004

3.400

Real Monthly Wages in Pesos

100

17.000

3.200 3.000 2.800

U.S. Real User Cost of Capital (in % p.a.)

Real Monthly Wages in Pesos

18.000

110

R 10 9 8 7 6 5 4 3

2.600

1990

2.400 WU 2.200 1990

1992

1994

1996

1998

2000

2002

1992

1994

1996

1998

2000

2002

2004

Figure 4. US industrial production (YUS) and US real user cost of capital (R).

2004

Figure 3. Real monthly wages in Pesos in Mexican maquiladoras: total (W); skilled labor (Ws); and unskilled labor (Wu). Index of Mexican Real Exchange Rate

of capital in the US as the appropriate cost of capital for maquiladoras operating in a global environment under production sharing. Villarreal (2005) sustains that the majority of maquiladora plants have US parent companies, which clearly facilitates access to US financial funding. 6 The 1-month US certificate of deposit (CD) rate at monthly frequency (CD1M) comes from the Board of Governors of the Federal Reserve of the United States (available at: http:// www.frbstlouis.com/); the monthly annualized inflation rate in the US measured by the CPI-all items (1982–184 = 100) was taken from the BLS (available at: http://www.bls.gov/). Figure 4 shows the decreases in R right after the recessions of the early 1990s and 2001–02, which are also captured by movements in YUS. Yet, note the imperfect correlation between YUS and R in Figure 4. This is in agreement with real interest rates being acyclical and lagging the business cycle in developed economies as documented by Neumeyer and Perri (2005) for several countries. Maquiladora exports, non-maquiladora exports, and total exports presented in Figure 1 are obtained from Banxico (http://www.banxico.org.mx). Figure 5 has the real exchange rate (RER) as the multilateral exchange rate, also from Banxico. In Figure 5, one sees the appreciation (decrease in the index) until the peso crash of December 1994 and sustained

140 130

RER

120 110 100 90 80 70 60 50 90

92

94

96

98

00

02

04

Figure 5. The real exchange rate (RER) in Mexico.

appreciation since then. In 2000–01, the peso gained against other currencies after a surge in capital inflows and economic stability; the peso peaked at 55.69 in March 2002. Table 1 combines the descriptive statistics of all series. Total maquiladora employment is highly correlated with US industrial production (sample correlation coefficient of 0.977), with maquiladora output (0.958), as well as with maquiladora exports (0.936). There is also a mild positive correlation be-

EMPLOYMENT RESPONSES OF SKILLED AND UNSKILLED WORKERS AT MEXICAN MAQUILADORAS: THE EFFECTS OF EXTERNAL FACTORS

1289

Table 1. Descriptive statistics

L Ls Lu W Ws Wu R RER YUS

Mean

Median

Max

Min

Std. Dev.

Skewness

Kurtosis

874,298 65,766 702,970 6,013 12,845 2,786 7.500 81.843 89.305

973,074 68,102 792,301 5,831 12,483 2,763 7.700 77.943 93.635

1,347,803 100,888 1,093,278 7,535 17,291 3,332 10.180 138.316 109.767

424,652 31,017 341,879 4,937 9,497 2,239 3.910 55.686 67.117

291,717 23,197 230,912 642 1,685 255 1.636 15.163 13.811

0.180 0.087 0.189 0.427 0.475 0.123 0.249 0.885 0.255

1.481 1.391 1.541 2.028 2.171 2.211 1.917 3.648 1.495

Notes: Employees (L, Ls, and Lu) are given in number of employees; monthly real wages (W, Ws, and Wu) are in pesos; US real user cost of capital (R) is in%; and Mexican peso real exchange rate (RER) and US industrial production (YUS) are indexes.

tween employment and total wages (0.512), a low correlation between employment and the user cost of capital (0.119), and a negative correlation between employment and the real exchange rate (0.587). The user cost of capital has a mild correlation with total wages (0.640) and a low correlation with US output (0.205). The real exchange rate has mild correlation coefficients with all variables (ranging from 0.455 to 0.631). Therefore, serious correlation problems among the explanatory variables of (9) are unlikely. One criticism in Campbell and Perron (1991) is that the relatively short time span could induce power concerns in unit root testing. Table 2 reports three unit root tests for the series in levels and in first differences. The ADF (k) test by Ng and Perron (1995), the KPSS tests by Kwiatkowski, Phillips, Schmidt, and Shin (1992), and the M-tests by Ng and Perron (2001) reported in the table all suggest that the series are nonstationary in levels but are stationary after first-differencing. 7 The DF-GLS result was very similar to the ADF and is omitted. The ADF test does not reject the unit root in levels but does reject at standard confidence levels in first-differences; similar results were found by the more powerful Ng and Perron (2001) tests. The KPSS test confirms the I (1) decision, rejecting the stationary in levels assumption and not rejecting in first differences. Based on Table 2, we proceed under the premise that all series can be fairly represented as I (1) stochastic processes. 4. RESULTS In order to check empirically the appropriateness of the model above, we first apply block exogeneity Wald tests for Granger Causality in the first-differenced VARs. The use of impulse dummies for the RER during the Mexican financial crisis was in general not statistically significant and did not change the results. We thus proceeded with the VAR for the basic series discussed in Section 3, and the VARs are estimated with three lags determined by a combination of lag-exclusion tests of significance and lag-selection criteria. The upper part of Table 3 contains the VAR with skilled labor (Ls), and the bottom part displays the VAR with unskilled labor (Lu). For each equation in the VAR, the v2-statistics are for the joint significance of lagged endogenous series in that equation with p-values in brackets. For skilled employment in column (1), for example, one can strongly reject the hypothesis of zero coefficients for skilled wages, unskilled wages, and US industrial production. Also reported are the statistics for the joint significance of all other lagged endogenous variables in the equation, in which there is a strong rejection of zero coefficients for the first column for skilled labor in the upper

panel. The evidence of the VAR with unskilled labor in the bottom panel is not so strong. However, in both sets of VARs, there is not much evidence that supports reverse causality either: there is never much evidence that wages, user cost, US output, or real exchange rates are strongly affected by other series. We thus conclude that the assumption of employment being endogenous is not refuted, but there seems to be preliminary evidence that the relationship appears stronger for skilled labor than for unskilled labor. Given the unit roots found for all series in levels, we proceed in two steps. In the first stage, the Johansen cointegration method is used for estimation of the long-run vector with eleven seasonal dummies. In the second stage, lagged one period residuals from the first stage are used in differenced form in an error–correction model. Table 4 contains estimates of the baseline Eqn. (7b). Diagnostic tests on the VECM methodology do not indicate problems, as can be verified by the LM-tests of serial correlation on the residuals for six lags. The results were not sensitive for different lag orders, and the chosen VAR always had very good specification properties as can be seen by the LM statistic not rejecting the null hypothesis of no serial correlation. Also reported at the bottom of Table 4 are the two tests for existence of cointegration: the trace and maximum eigenvalue tests. Both reject the of no cointegration only for the skilled labor specification, in which there exists only one vector, suggesting a stronger relationship for skilled employment as already suggested by the Granger causality tests. Equations are reported with and without the time trend in each column. Focusing on the (7b) specification without the time trend first, the real wage coefficient suggests a stronger effect on the unskilled labor estimation (1.479) compared to skilled labor (1.282). This implies that increases in real wages of unskilled workers affect unskilled employment relatively more than in the skilled employment case. Fluctuations in the user cost of capital have negative effects on skilled labor (0.221), but are not statistically significant in total labor type (0.093) and in unskilled labor (0.032). According to the strong definition above, capital and (skilled) labor are found to be complements. The effect of US industrial production on employment is positive as expected, varying from 2.212 for unskilled labor to 3.062 for skilled labor. For total labor, the b3-coefficient is estimated at 2.574. The more than proportional effects of US industrial production on maquiladora employment confirms previous works on the maquiladora industry, such as Gruben (2001), Fullerton and Schauer (2001), and Mollick (2003), in which increases in US output have a very strong effect on maquiladora employment growth. This finding is not confined to the maquiladoras since Haouas, Yagoubi, and Heshmati (2003) find that Tunisian

1290

WORLD DEVELOPMENT Table 2. Unit root tests on monthly data Trend

ADF (k)

KPSS (4)

MZa (k)

MZt (k)

L

Series

Yes

1.564 (14)

0.609***

7.678 (4)

1.910 (4)

D(L)

No

2.962 (13)**

0.315

10.619 (5)**

2.296 (5)**

Ls

Yes

2.045 (5)

0.448***

10.050 (5)

2.240 (5)

D(Ls)

No

Lu

Yes

***

3.626 (4)

1.496 (14) **

2.988 (13)

0.178 0.636***

**

9.180 (6)

6.876 (4) **

11.041 (5)

2.131 (6)** 1.791 (4) 2.349 (5)**

D(Lu)

No

W

Yes

2.150 (14)

D(W)

No

*

2.812 (14)

0.044

Ws

Yes

2.345 (14)

0.590***

2.374 (6)

1.028 (6)

0.028

66.766 (0)***

5.746 (0)***

0.691***

1.543 (5)

0.710 (5)

D(Ws)

No

Wu

Yes

3.270 (13)

***

2.406 (14) **

D(Wu)

No

3.464 (14)

R D(R) YUS D(YUS) RER D(RER)

Yes No Yes No Yes No

1.548 (14) 2.884 (14)** 2.018 (4) 3.359 (14)** 2.446 (8) 4.551 (7)***

0.329 0.719**

0.061 0.635*** 0.112 0.599*** 0.087 0.218*** 0.048

0.847 (6) 65.296 (0)

***

0.513 (6) 5.678 (0)***

***

4.980 (0)***

4.324 (2) 39.107 (5)*** 3.228 (2) 63.572 (0)*** 10.823 (4) 81.115 (1)***

1.441 (2) 4.382 (5)*** 1.262 (2) 5.636 (0)*** 2.322 (4) 6.362 (1)***

50.510 (0)

Notes: Data are of monthly frequency for Mexican maquiladoras from 1990:01 to 2006:3. L refers to total employment (s and u capture skilled and unskilled workers, respectively), W refers to total monthly real wage paid, YUS refers to US industrial production index, R refers to the real user cost of capital in the US based on 1-month CD rate minus the rate of inflation and allowing for a 6% depreciation rate; and RER stands for the real exchange rate. The symbol D refers to the first-difference of the original series. We include the deterministic trend only when testing in levels as suggested from graph inspection. ADF (k) refers to the Augmented Dickey-Fuller t-tests for unit roots, in which the is that the series contains a unit root. The lag length (k) for ADF tests is chosen by the Campbell and Perron (1991) data dependent procedure, whose method is usually superior to k chosen by the information criterion, according to Ng and Perron (1995). The method starts with an upper bound, kmax = 14, on k. If the last included lag is significant, choose k = kmax. If not, reduce k by one until the last lag becomes significant at the 5% level. If no lags are significant, then set k = 0. Next to the reported t-value, the selected lag length appears in parentheses. The KPSS test follows Kwiatkowski et al. (1992), in which the is that the series is stationary and k = 4 is the used lag truncation parameter. The MZa (k) and MZt (k) are two of the M-tests developed by Ng and Perron (2001) with good size and power. In both cases, the autoregressive OLS-detrended spectral estimation method is used, together with modified akaike information criterion (MAIC) for selection of lag-length (k) with maximum number of lags set at 6. * Rejection of the at 10% levels, respectively. ** Rejection of the at 5% levels, respectively. *** Rejection of the at 1% levels, respectively.

manufacturing employment responds most to output in the long run, followed by changes in capital stock and least by wages. In addition to the long-run estimations, the bottom of the table contains the coefficient associated with the dynamic ECM specification. Note first that the ECM term is negative and statistically significant in all cases. For the skilled labor case, in particular, when current employment is higher than the estimated employment figures, the response is negative and strongly so at 0.119. The latter implies that around 12% of the deviations are corrected in the following month, a fairly high rate of adjustment. The fit of these estimations, however, as measured by the adjusted R2, is relatively weaker for the skilled labor case, at around 24%, than for unskilled labor, at over 50%. Table 4 also contains the modifications of the baseline model for the introduction of a time trend that presumably captures technology. The coefficient on the time trend is found to be negative in the unskilled labor case, which would suggest less usage of unskilled workers over time. In other words, labor saving practices appear to occur in the unskilled labor case, which also implies labor saving in total employment. For wage elasticity, the estimated coefficients are negative as before, but do suggest a more than proportional effect to wage changes, particularly in skilled labor.

It is useful to check whether these results are maintained when we use RER as the shift factor of labor demand in Table 5. The trace and maximum eigenvalue tests reject the null of cointegration for all specifications. For skilled labor, the evidence could be consistent with two cointegration vectors, although the other possible vector does not differ much from the one reported in the table. 8 The coefficient of the time trend now becomes statistically significant only in the total labor specification. The most important result is perhaps that real wages affect employment negatively with a stronger impact for skilled labor: wage elasticities of 4.930 or 4.816 versus 1.949 or 2.153 for unskilled labor. Introducing real exchange rates as a shift factor in the production function, employment seems to fall more with wage increases in the skilled labor case than in the unskilled labor case. Put differently, skilled labor becomes much more sensitive to changes in its own price under the RER-modified version. These findings can be reconciled with recent studies in the literature started by Rodrik (1997). Examining UK labor demand from 1982 to 1996, Hijzen, Go¨rg, and Hine (2005) show that international outsourcing has had a strong negative impact on the demand for skilled labor. Fajnzylber and Maloney (2005) find that labor demand elasticities do change greatly in magnitude over time and with levels of openness to interna-

EMPLOYMENT RESPONSES OF SKILLED AND UNSKILLED WORKERS AT MEXICAN MAQUILADORAS: THE EFFECTS OF EXTERNAL FACTORS

1291

Table 3. VAR granger causality tests Excluded Regressors ;

D(Ls)

VAR with skilled labor (Ls) D(Ls) **

D(Ws)

D(Wu)

D(R)

D(YUS)

D(RER)

0.187 [0.980]

4.916 [0.178]

0.960 [0.811]

4.519 [0.211]

3.881 [0.275]

**

D(Ws)

10.856

D(Wu)

15.679*** [0.001]

6.158 [0.104]

2.247 [0.523]

3.389 [0.335]

2.571 [0.463]

***

2.199 [0.532]

1.065 [0.786]

D(R) D(YUS) D(RER) All

11.878

[0.013]

[0.008]

*

[0.016]

1.162 [0.762]

1.286 [0.732]

1.152 [0.764]

2.325 [0.508]

2.379 [0.498]

0.071 [0.995]

3.885 [0.274]

1.002 [0.801]

4.407 [0.221]

4.743 [0.192]

5.822 [0.121]

1.038 [0.792]

6.414 [0.093]

2.082 [0.556]

1.425 [0.700]

36.435*** [0.002]

20.109 [0.168]

19.783 [0.180]

16.061 [0.378]

16.473 [0.351]

10.404 [0.794]

1.985 [0.576]

3.250 [0.355]

1.513 [0.679]

11.344*** [0.010]

4.189 [0.242]

2.544 [0.467]

1.104 [0.776]

0.393 [0.942]

2.614 [0.456]

1.268 [0.737]

0.227 [0.973]

4.212 [0.239]

1.229 [0.746]

3.440 [0.329]

4.123 [0.249]

23.791* [0.069]

10.724 [0.772]

VAR with unskilled labor (Lu) D(Lu) D(Ws)

10.334

**

2.979 [0.395]

8.041 **

[0.045]

D(Wu)

1.754 [0.625]

D(R)

0.665 [0.881]

3.291 [0.349]

2.691 [0.442]

***

2.726 [0.436]

1.615 [0.656]

4.702 [0.195]

D(YUS)

11.525

[0.009]

8.743

[0.033]

*

D(RER)

0.104 [0.991]

6.525 [0.089]

2.519 [0.472]

1.236 [0.744]

All

21.218 [0.130]

22.128 [0.105]

17.968 [0.264]

16.665 [0.339]

Notes: Reported are block exogeneity Wald tests for Granger Causality in the first-differenced VAR. Statistical significance at the 5% level or lower is highlighted in bold. For each equation in the VAR, the v2-statistics are for the joint significance of lagged endogenous series in that equation; p-values are in brackets. Also reported are the statistics for the joint significance of all other lagged endogenous variables in the equation. The VARs are estimated with three lags determined by a combination of lag-exclusion tests of significance and lag-selection criteria.

tional trade. They observe for Mexico statistically significant increases in blue-collar elasticities across the liberalization period, amplifying the findings by Hanson and Harrison (1999) and others. Fluctuations in the user cost of capital continue to have negative effects on skilled labor only. The estimated values show very strong effects for skilled labor (between 0.859 and 0.805). Capital and skilled labor are found to be complements as in the baseline model. The estimated responses of employment to US industrial production are again strong and very robust, with higher effects for skilled labor. The response of employment to real exchange rates, however, contrasts markedly and deserves further elaboration. Increases in the real exchange rate (a weaker Mexican peso) affect skilled labor employment positively (between 0.886 and 0.829) but the b4-coefficient is estimated to be negative for unskilled labor at 0.727 or 0.822. The total employment effect of a weaker peso across the industry is negative and statistically significant at 0.586 or 0.742, suggesting that overall employment loses with real peso depreciations. This overall finding is consistent with Kamin and Rogers (2000), who documented aggregate Mexican output benefits with a strong RER over 1980–96. Panels of Mexican states reported by Mollick (2003) for maquiladoras account for a negative relationship (weaker peso leads to less employment) over the years 1990–2001 as well. The rationale in these papers is through increasing uncertainty and the perverse output effects of sudden currency depreciations. The above explanation would thus fit into the macroeconomic channel discussed in Frenkel and Ros (2006). Another interpretation is to recall the labor intensity channel discussed in Frenkel and Ros (2006). 9 The RER is an important factor of the labor/capital goods relative price since capital goods have a significant portion of imported parts. The

weakening of the real value of the peso makes the amount of imports decrease, moving down the K/L ratio. The lower level of capital intensity leads to lower employment in the sector. On the other hand, a stronger peso leads to more capital imports, pushing up the K/L ratio and leading to higher employment. The estimates suggest that a movement towards more skilled workers could have happened at Mexican maquiladoras for given movements in RER. Our findings are consistent with a weakening of the peso bringing down K/L overall with some adjustment in the skilled/unskilled workers share. How credible are these estimates with labor responding only to its own price? Table 6 explores the possibility of cross-wage effects on the demand for labor. We now have each labor type as dependent variable and two coefficients for wages: b1 and b2. We also explored a subset of the estimations in Table 6 with cross-wages and without the RER, which yielded very similar results and are omitted for space constraints. For each equation, b1 captures the impact of Ws on each type of labor demand and b2 captures the impact of Wu. Earlier findings are confirmed, especially with respect to the strong complementarity of capital and skilled labor when the real exchange rate is used: b3 now suggests more than proportional responses for skilled labor when the RER is included. On the cross-wage elasticities, changes in unskilled labor wage have no impact on the hiring decisions of skilled workers: the b2-coefficients in the skilled labor columns are insignificant. However, the same does not hold for the decisions to hire unskilled workers. Responding to increases in the price of skilled labor, there is evidence suggesting that maquiladora firms cut back hiring of unskilled workers by more than proportional amounts: 1.156 and 1.207, both statistically significant at 1% or 5% confidence levels. From the right columns in Table 6, real exchange rate depreciations contribute to a fall in unskilled labor employment

1292

Table 4. Vector ECM of maquiladora employment in the baseline model logðLÞt ¼ b1 logðW Þt þ b2 logðRÞt þ b3 logðYUSÞt þ eit

ð7aÞ

logðLÞt ¼ b0 trend þ b1 logðW Þt þ b2 logðRÞt þ b3 logðYUSÞt þ eit

ð7bÞ

Regressors

Dependent variable skilled labor log Lst

Total labor log Lt 0.001 (0.001)

b0

Unskilled labor log Lut 0.004*** (0.002)

0.001 (0.001)

b1

1.248*** (0.283)

1.000*** (0.262)

1.282*** (0.198)

1.408*** (0.216)

1.479*** (0.328)

0.990*** (0.241)

b2

0.093 (0.085)

0.122 (0.101)

0.221*** (0.056)

0.158* (0.084)

0.032 (0.104)

0.186 (0.121)

Lag-Length & Wald test on lag excl. (k = 3)

2.574

(0.105)

***

2.943

(0.438)

***

3.062

(0.117)

***

2.667

(0.370)

***

2.212

(0.108)

3.479*** (0.556)

2 9.817 [0.876]

2 9.827 [0.875]

2 15.847 [0.464]

2 15.475 [0.490]

2 21.103 [0.175]

2 21.164 [0.172]

LM-test on k = 6 [p-value]

9.373 [0.897]

8.763 [0.923]

22.437 [0.130]

20.997 [0.179]

19.796 [0.230]

16.483 [0.420]

Coint. tests versus critical values

Trace 44.9 < 47.9 23.6 < 29.8 Max. Eig. 21.4 < 27.6 15.3 < 21.1

Trace 56.6 < 63.9 34.9 < 42.9 Max. Eig. 21.7 < 32.1 16.7 < 25.8

Trace 62.1 > 47.9** 19.0 < 29.8 Max. Eig. 43.1>27.6** 10.7 < 21.1

Trace 70.1 > 63.9 25.7 < 42.9 Max. Eig. 44.4>32.1 12.3 < 25.8

Trace 45.1 < 47.9 24.0 < 29.8 Max. Eig. 21.1 < 27.6 17.9 < 21.1

Trace 60.2 < 63.9 37.3 < 42.9 Max. Eig. 22.9 < 32.1 18.1 < 25.8

0.052*** (0.013)

0.056*** (0.014)

0.119*** (0.027)

0.113*** (0.025)

0.042*** (0.014)

0.056*** (0.010)

0.553

0.549

0.237

0.236

0.570

0.549

ECM in VECM (std. error) Adj. R2 in VECM

Notes: Data are of monthly frequency from 1990:1 to 2006:3. The total number of observations is 192. Standard errors are reported next to the coefficients. The method of estimation is the vector error– correction model with eleven seasonal dummies. In the first stage, the Johansen cointegration method is used for estimation of the long-run vector. In the second stage, residuals from the first stage are used in differenced form. For lag-length selection, information criteria were checked together with lag-exclusion Wald-tests of one more lag than the lags actually chosen. The LM t-stat. is a Lagrange Multiplier test on the residuals of the regression, calculated under the hypothesis of no serial correlation on up to 6 lags. * Rejection of the hypothesis of zero coefficients at 10% level, respectively. ** Rejection of the hypothesis of zero coefficients at 5% level, respectively. *** Rejection of the hypothesis of zero coefficients at 1% level, respectively.

WORLD DEVELOPMENT

b3

***

logðLÞt ¼ b1 logðW Þt þ b2 logðRÞt þ b3 logðYUSÞt þ b4 logðRERÞt þ eit

ð9aÞ

logðLÞt ¼ b0 trend þ b1 logðW Þt þ b2 logðRÞt þ b3 logðYUSÞt þ b4 logðRERÞt þ eit

ð9bÞ

Regressors

Dependent variable

Total labor log Lt Skilled labor

b0 b1 b2 b3 b4 Lag-Length &Wald test on lag excl. (k = 3)

LM-test on k = 6 [p-value] Coint. tests versus critical values

ECM in VECM (std. error) Adj. R2 in VECM

**

1.935*** (0.252) 0.178*** (0.071) 2.636*** (0.091) 0.586*** (0.077) 2 15.252 [0.935] 17.077 [0.879] Trace 99.0 > 69.8** 50.0 > 47.9** 23.6 < 29.8 Max. Eig. 49.0 > 33.9** 26.4 < 27.6 0.006 (0.010) 0.503

0.0034 (0.0017) 2.412*** (0.323) 0.052 (0.117) 1.616*** (0.537) 0.742*** (0.100) 2 14.057 [0.961] 17.467 [0.864] Trace 113.4 > 88.8** 61.0 < 63.9 Max. Eig. 52.4 > 38.3** 26.4 < 32.1

4.930*** (0.787) 0.859*** (0.227) 5.548*** (0.493) 0.886*** (0.248) 2 31.845 [0.163] 27.622 [0.326] Trace 113.3 > 69.2** 59.2 > 47.9** 32.0 > 29.8** Max. Eig. 54.1 > 33.9** 27.2 < 27.6

0.005 (0.009) 0.503

0.027*** (0.006) 0.222

log Lst 0.001 (0.004) 4.816*** (0.790) 0.805*** (0.328) 5.279*** (1.427) 0.829*** (0.254) 2 31.873 [0.162] 27.558 [0.329] Trace 125.1 > 88.8** 71.1 > 63.9** 38.1 < 42.9 Max. Eig. 54.1 > 38.3** 33.0 > 32.1** 23.7 < 25.8 0.028*** (0.007) 0.221

Unskilled labor log Lut 1.949*** (0.250) 0.105 (0.073) 2.188*** (0.084) 0.727*** (0.094) 2 23.852 [0.528] 22.453 [0.610] Trace 92.2 > 69.8** 44.4 < 47.9** 19.6 < 29.8 Max. Eig. 47.87 > 33.9** 24.8 < 27.6

0.002 (0.002) 2.153*** (0.293) 0.034 (0.131) 1.658*** (0.622) 0.822*** (0.114) 2 24.511 [0.490] 22.367 [0.615] Trace 104.6 > 88.8** 56.2 < 63.9 8.3 < 15.5 Max. Eig. 48.4 > 38.3** 25.3 < 32.1

0.005 (0.010) 0.502

0.005 (0.009) 0.502

Notes: Data are of monthly frequency from 1990:1 to 2006:3. The total number of observations is 192. Standard errors are reported next to the coefficients. The method of estimation is the vector error– correction model with 11 seasonal dummies. See notes to Table 4 for details. * Rejection of the hypothesis of zero coefficients at 10% level, respectively. ** Rejection of the hypothesis of zero coefficients at 5% level, respectively. *** Rejection of the hypothesis of zero coefficients at the 1% level, respectively.

EMPLOYMENT RESPONSES OF SKILLED AND UNSKILLED WORKERS AT MEXICAN MAQUILADORAS: THE EFFECTS OF EXTERNAL FACTORS

Table 5. Vector ECM of Maquiladora Employment Modified by RER

1293

1294

WORLD DEVELOPMENT Table 6. Vector ECM of maquiladora employment with two labor types i

logðL Þt ¼ b1 logðW i Þt þ b2 logðW j Þt þ b3 logðRÞt þ b4 logðYUSÞt þ b5 logðRERÞt þ eit logðLi Þt ¼ b0 trend þ b1 logðW i Þt þ b2 logðW j Þt þ b3 logðRÞt þ b4 logðYUSÞt þ b5 logðRERÞt þ eit Dependent variable skilled labor log Lst b0 b1 b2 b3 b4 b5 Lags Wald test on lag excl. LM-test on k = 6 [p-value] Coint. tests versus critical values

ECM in VECM (std. error) Adj. R2 in VECM

9.007*** (2.086) 14.018*** (3.326) 1.473 (1.726) 2.684 (2.807) 1.378*** (0.314) 2.501*** (0.778) 7.623*** (1.157) 12.966*** (3.932) 0.471(0.448) 1.246 (0.781) 4 4 On k = 5 lags: 30.506 [0.727] 30.518 [0.727] 33.070 [0.609] 33.252 [0.600] Trace 136.8 > 117.7** 123.6 > 95.8** ** 83.0 > 69.8 95.9 > 88.8** 64.0 > 63.9** 53.2 > 47.9** 25.3 < 29.8 35.4 < 42.9 Max. Eig. 40.9 < 44.5 40.6 > 40.1** 29.8 < 33.9 31.8 < 38.3 0.010 (0.006) 0.007* (0.004) 0.258 0.261

Dependent variable unskilled Labor log Lst 0.007 (0.010) 0.001 (0.002) 1.156*** (0.493) 1.207*** (0.532) 0.982*** (0.425) 1.099*** (0.468) 0.169*** (0.075) 0.112 (0.123) 2.804*** (0.269) 2.396*** (0.621) 0.529*** (0.103) 0.597*** (0.119) 2 2 On k = 3 lags: 47.067 [0.103] 48.313* [0.082] 36.712 [0.436] 36.583 [0.442] Trace 118.7 > 95.8** 131.1 > 117.7** 66.3 < 69.8 78.2 < 88.8

Max. Eig. 52.4 > 40.1** 27.1 < 33.9 0.010 (0.010) 0.508

52.9 > 44.5** 27.9 < 38.3 0.009 (0.009) 0.508

Notes: Data are of monthly frequency from 1990:1 to 2006:3. The total number of observations is 190. Standard errors are reported next to the coefficients. The method of estimation is the vector error–correction model with 11 seasonal dummies. See notes to Table 4 for details. * Rejection of the hypothesis of zero coefficients at 10% level, respectively. ** Rejection of the hypothesis of zero coefficients at 5% level, respectively. *** Rejection of the hypothesis of zero coefficients at 1% level, respectively.

(b5-coefficients of 0.529 and 0.597). This result is not statistically significant for the skilled labor specification. Combined with the own-price elasticity, this implies that less skilled workers are more affected than skilled workers when the peso weakens and when the price of skilled labor rises. It is consistent with the notion that the adjustment of labor at the maquiladoras is felt more severely by the less skilled. Away from the steady-state, the ECM term is negative and statistically significant only for one of the skilled labor cases in Table 6: the responses are slightly negative, implying that close to 0.7% of the deviations are adjusted in the following month. When current employment is higher than forecasted employment, there is a small reduction in the hiring of subsequent skilled workers. Despite the dynamic adjustment in the ECM specifications, the steady-state values of the long-run cointegrating vectors reported in Table 6 are very strong and in line with the previous tables and expected signs. 5. FINAL REMARKS Since globalization has changed the nature of trade in Mexico (Kose et al., 2005), we introduce this fact into Mexican maquiladoras’ labor demand. Using monthly data from 1990 to 2006, new results emerge. We find that the augmented mod-

el complements the baseline model and show that real exchange rate depreciations contribute to a drop in unskilled labor employment (coefficients of 0.727 and 0.822). Combined with the cross-price elasticity, these coefficients become 0.529 and 0.597, but still suggest that unskilled workers are more affected than skilled workers when the peso weakens and when the price of skilled labor rises. The degree of complementarity is also found to be much higher between capital and skilled labor than between capital and unskilled labor. Ramı´rez (2006) has argued that foreign capital has had a key role in long-run Mexican wages. Given the substantial increase in FDI inflows into Mexico for this sample, it is likely that foreign capital has amplified the channel between cost of capital and labor. The time series approach implemented herein comes with a methodological gain. Several articles under panel data methods initiated by Milner and Wright (1998) and Greenaway et al. (1999) have let the user cost of capital appear as time dummies under the assumption of perfect capital markets. We relax this assumption and find strong capital-labor complementarities when making labor demand sensitive to the interest rate, as in Neumeyer and Perri (2005). This approach paid off, in particular, for the case of skilled employment. Further work in understanding why unskilled labor responds differently is left for further research.

NOTES 1. The maquiladora program allows the inputs and their operating machinery to enter Mexico free of tariffs. The mechanism can perhaps be best summarized in this way: ‘‘Maquiladora firms are those that import nearly 100% of their inputs and then export nearly 100% of their output” (Robertson, 2003, p. 40).

2. A far from exhaustive list on Mexico starts with Feenstra and Hanson (1997) in exploring FDI inflows into Mexico and the role of outsourcing by Northern multinationals. See also Hanson and Harrison (1999) and Revenga (1997), who find that changes in trade policy have negative effects on employment. Technology effects on the skill premium

EMPLOYMENT RESPONSES OF SKILLED AND UNSKILLED WORKERS AT MEXICAN MAQUILADORAS: THE EFFECTS OF EXTERNAL FACTORS

are studied by Esquivel and Rodrı´guez-Lo´pez (2003) and Mollick (2008). 3. Data for Mexican manufacturing (excluding the maquiladoras) start from 1994 onwards and for maquiladoras start from 1990 onwards. The contrasting dynamics of these two sectors are clear. The Wall Street Journal (2005) states that ‘‘while growth in a number of economic sectors in Mexico has been disappointing lately, one important segment – maquiladoras – is firing on all cylinders. Employment is up, fresh investment is streaming in and exports from the assembly-for-export industry are showing double-digit growth. The expansion pales compared with strides made by manufacturers in China, but industry leaders are still optimistic.” Along the same lines, Hanson (2002) contrasts the real value added by the maquiladoras at an annual average rate of 10% between 1990 and 2002 against only 3% for real Mexican GDP. 4. Calculations with value added of the maquiladoras do not change qualitatively the basic results but are subject to the endogeneity criticism. We represent foreign demand by US real GDP (YUS) since the US is the major destination of Mexican maquiladora production. Earlier works with YUS in employment growth equations include Fullerton and Schauer (2001), Mollick (2003), Coronado, Fullerton, and Clark (2004), and Can˜as, Fullerton, and Smith (2007), while US real output is also included in Gruben (2001)’s export growth equations. 5. Robertson (2003) lists that the effect of an appreciation of RER on labor demand varies with the following: the share of exported output (almost 100% in maquiladoras), the amount of imported intermediate inputs (also very high in the sector), and the degree of complementarity between imported inputs and labor. The net effect will depend on the relative impact of these factors. 6. While the decision to use the US real cost of capital was based on economic reasons, two statistical factors are also present. First and foremost, since the depreciation rate (d) is unknown, we estimate the user cost of capital with varying depreciation rates. When calculating the user cost for Mexico, the nominal interest rate is the 1-month rate on Mexican government bills (CETES) from Mexico’s Central Bank (‘‘Banxico”, data available at: http://www.banxico.org.mx) and deflated by the consumer

1295

price index (INPC). Under a higher d = 12%, the user cost of capital would always be positive. See Neumeyer and Perri (2005) for implausibly negative real interest rates when inflation swings are very high. Since the real cost of capital can be negative when d = 6%, a few observations were lost when applying logarithms under the user cost for Mexico. Second, the Mexican user cost turned out to be stationary in levels, in contrast to all the other series. 7. Further details are provided in the notes to Table 2. On the Ng and Perron (2001) tests, two modifications were followed in this paper. First, we adopt the recommendation by Perron and Qu (2007) to use OLSdetrended data when constructing the modified Akaike information criterion. Second, while we first used a standard default criterion to select the maximum number of lags (k) at 14 lags for ADF (k), we chose the maximum number of lags at 6 since the series were not subject to large moving average terms. In fact, a ‘‘large k is generally necessary for noise functions with a moving-average root that is large”, according to Ng and Perron (2001, p. 1520), which was not observed in our data. Remaking the ADF tests with kmax = 6 yielded very similar results to those reported in Table 2. We omit them due to space constraints. 8. For skilled employment, the Trace tests suggest more than one cointegrating vector. The coefficients for the other vector (not shown) are little changed, however. If one assumes two cointegrating vectors, another possible vector would imply – without the trend – a positive coefficient for RER (0.292 with standard deviation of 0.119) and a positive coefficient for US output (2.420 with standard deviation of 0.099) but not statistically coefficient for user cost. With the trend, a positive coefficient is again found for RER (1.938 with standard deviation of 0.432) and a positive coefficient for US output (6.487 with standard deviation of 2.302) but not statistically significant for user cost. These additional results support the findings of Table 5, in which skilled employment increases with RER depreciations. 9. Other explanations are theoretically possible. Faria and Leo´nLedesma (2005), for example, show that an RER depreciation increases the present value of financial wealth in foreign bonds. This positive income effect can make workers keep the same level of utility by working less (negative effect) or deciding to increase their labor supply (positive effect).

REFERENCES Campbell, J., & Perron, P. (1991). Pitfalls and opportunities: What macroeconomists should know about unit roots. NBER Macroeconomics Annual, 6, 141–201. Can˜as, J., Fullerton, T., Jr., & Smith, W. D. (2007). Maquiladora employment dynamics in nuevo laredo. Growth and Change, 38(1), 23–38. Coronado, R., Fullerton, T., Jr., & Clark, D. (2004). Short-run maquiladora employment dynamics in tijuana. The Annals of Regional Science, 38, 751–763. Draper, D. A. G., & Manders, A. J. G. (1997). Structural changes in the demand for labor. De Economist, 145(4), 521–546. Esquivel, G., & Rodrı´guez-Lo´pez, J. A. (2003). Technology, trade, and wage inequality in Mexico before and after NAFTA. Journal of Development Economics, 72, 543–565. Fajnzylber, P., & Maloney, W. (2005). Labor demand and trade reform in Latin America. Journal of International Economics, 66, 423–446. Faria, J. R., & Leo´n-Ledesma, M. (2005). Real exchange rate and employment performance in an open economy. Research in Economics, 59, 67–80. Faria, J. R., & Carneiro, F. G. (2003). Devaluation, output and wages. International Economic Journal, 17(4), 15–27. Feenstra, R., & Hanson, G. (1997). Foreign direct investment and relative wages: Evidence from Mexico’s maquiladoras. Journal of International Economics, 42, 371–393. Frenkel, R., & Ros, J. (2006). Unemployment and the real exchange rate in Latin America. World Development, 34(4), 631–646.

Fu, X., & Balasubramanyam, V. N. (2005). Exports, foreign direct investment and employment: The case of China. The World Economy, 28(4), 607–625. Fullerton, T., Jr., & Sprinkle, R. (2005). An error–correction analysis of US–Mexico trade flows. The International Trade Journal, 19(2), 179–192. Fullerton, T., Jr., & Schauer, D. (2001). Short-run maquiladora employment dynamics. International Advances in Economic Research, 7, 471–478. Gale, W. G., & Orszag, P. R. (2005). Deficits, interest rates, and the user cost of capital: A reconsideration of the effects of tax policy on investment. Brookings Institution, July. . Greenaway, D., Hine, R., & Wright, P. (1999). An empirical assessment of the impact of trade on employment in the United Kingdom. European Journal of Political Economy, 15, 485–500. Gruben, W. (2001). Was NAFTA behind Mexico’s high maquiladora growth? Economic and Financial Review, FRB of Dallas, Third Quarter. pp. 11–21. Hanson, G. (2002). The role of maquiladoras in Mexico’s export boom. University of California, Davis. http://migration.ucdavis.edu/rs/more.php?id=8_0_2_0. Hanson, G., & Harrison, A. (1999). Trade liberalization and wage inequality in Mexico. Industrial and Labor Relations Review, 52(2), 271–288. Haouas, I., Yagoubi, M., & Heshmati, A. (2003). Labour-use efficiency in tunisian manufacturing industries. Review of Middle East Economics and Finance, 1(3), 195–214.

1296

WORLD DEVELOPMENT

Hijzen, A., Go¨rg, H., & Hine, R. (2005). International outsourcing and the skill structure of labour demand in the UK. Economic Journal, 115, 860–878. Kamin, S., & Rogers, J. (2000). Output and the real exchange in developing countries: An application to Mexico. Journal of Development Economics, 61, 85–109. Kose, M. A., Meredith, G., & Towe, C. (2005). How has NAFTA affected the Mexican economy? Review and evidence. In R. Langhammer, L. Vinhas, & de Souza (Eds.), Monetary policy and macroeconomic stabilization in Latin America (pp. 35–81). Berlin-Heidelberg: Springer. Kwiatkowski, D., Phillips, P., Schmidt, P., & Shin, Y. (1992). Testing the null hypothesis of stationarity against the alternative of a unit root: How sure are we that economic series have a unit root?. Journal of Econometrics, 54, 159–178. Miller, J., Hom, P., & Gomez-Mejia, L. (2001). The high cost of low wages: Does maquiladora compensation reduce turnover? Journal of International Business Studies, 32(3), 585–595. Milner, C., & Wright, P. (1998). Modeling labour market adjustment to trade liberalisation in an industrializing economy. Economic Journal, 108, 509–528. Mollick, A. V. (2008). The rise of the skill premium in Mexican maquiladoras. Journal of Development Studies, 44(9), 1382–1404. Mollick, A. V. (2003). Employment determination at Mexican maquiladoras: Does location matter?. Journal of Borderlands Studies, 18(2), 45–67. Neumeyer, P., & Perri, F. (2005). Business cycles in emerging economies: The role of interest rates. Journal of Monetary Economics, 52, 345–380. Ng, S., & Perron, P. (2001). Lag length selection and the construction of unit root tests with good size and power. Econometrica, 69(6), 1519–1554. Ng, S., & Perron, P. (1995). Unit root test in ARMA models with data dependent methods for the selection of the truncation lag. Journal of the American Statistical Association, 90, 268–281.

Perron, P., & Qu, Z. (2007). A simple modification to improve the finite sample properties of Ng and Perron’s unit root tests. Economics Letters, 94, 12–19. Ramı´rez, M. (2006). Is foreign direct investment beneficial for Mexico?. World Development, 34(5), 802–817. Revenga, A. (1997). Employment and wage effects of trade liberalization: The case of Mexican manufacturing. Journal of Labor Economics, 15(3), S20–S43. Robertson, R. (2007). Are Mexican and US workers complements or substitutes? Macalester College. . Robertson, R. (2003). Exchange rates and relative wages: Evidence from Mexico. The North American Journal of Economics and Finance, 14, 25–48. Rodrı´guez, F., & Rodrik, D. (2000). Trade policy and economic growth: A skeptic’s guide to the cross-national evidence. In B. Bernanke, & K. Rogoff (Eds.), Macroeconomics annual 2000. Cambridge, MA: MIT Press for NBER., 2001. Rodrik, D. (1997). Has globalization gone too far? Washington DC: Insititute for International Economics. Sargent, J., & Matthews, L. (2004). What happens when relative costs increase in export processing zones? Technology, regional production networks, and Mexico’s maquiladoras. World Development, 32(12), 2015–2030. Slaughter, M. (2001). International trade and labor-demand elasticities. Journal of International Economics, 54, 27–56. The Wall Street Journal (2005). Maquiladoras offset Mexico’s sluggish sectors. August 18. Villarreal, M. A. (2005). US–Mexico economic relations: Trends, issues, and implications. Congressional Research Service, The Library of Congress. .

Available online at www.sciencedirect.com