Revisiting Seguino's Gender Accounting

Revisiting Seguino’s Gender Accounting ________________________________________ Benjamin H. Mitra-Kahn* ABSTRACT In November 2000, Feminist Economics published a paper by Dr. Stephanie Seguino, entitled ‘Accounting for gender in Asian economic growth’. This paper caused a stir within feminist economics, as it implied a distinct positive relationship between the wage discrimination of women and the high growth rates of both GDP and investment experienced in the South East Asian Miracle. Further, this relation it was argued, could be generalized for any semi industrial country across the world, and was a representative sample for ‘Asian’ growth. This paper re-visits those conclusions and argues that Seguino was correct to point out that gender inequality played a role in the economic growth of the Newly Industrialised Countries, as opposed to the official World Bank view. Evidence is presented to argue that the early industrializes in South East Asia are distinctly different from other countries and any results from this sample should not be generalized. The countries in question used government micro policy and existing social structures to create a hyperbolic relationship between gender wage inequality and growth, meaning that inequality did lead to growth, but not in the sense that wider gaps leads to greater growth, in fact the opposite is true for the sample, in contrast with Seguino’s original argument. KEYWORDS Economic growth, gender, inequality, Asia, semi-industrialized economies, exchange rates, Seguino, gender wage gap I. INTRODUCTION In 2000 Feminist Economics published a controversial article by Stephanie Seguino entitled ‘Accounting for Gender in Asian Economic Growth’. This paper was controversial because it argued that gender inequality had played a significant part in the South-East Asian growth miracle, and that inequality had been a positive factor for growth. * I am thankful for the correspondence and original data sets supplied by Professor Seguino both of which were extremely helpful and illuminating. Extended Data sources can be found in the Bibliography. In South East Asia, governments and firms used the social structure in place to financially exploit women, to various degrees, in order to cut costs and keep prices competitive. Focusing on the labor intensive export sectors, countries acquired a cheap and flexible labor force which through different institutions was restricted to accept its place in life and on the factory floor. This created a set of circumstances where the manufacturing sector became relatively calm, encouraging foreign investment and with their low costs, boosting profits through bigger margins. This is the very convincing micro story told, and it was a big contribution to the literature on the South-East Asian Newly Industrialized Countries (NICs), as it emphasized a hitherto gap in the growth story of those countries. What this paper is going to do is to take issue with the other half of Seguino’s argument. The general and global macro thesis that gender wage inequality in can be good for growth and exchange rates. More than that, it will present evidence that the growthinequality connection is not as simple as Seguino presents it, with wider gaps leading to greater growth. The fact that gender inequality played a role in the South-East Asian miracle should be recognized, but evidence will be presented here, in opposition to the fact that “in general, these results provide consistent evidence that gender wage inequality is a stimulus to growth.” (Seguino 2000: 44, my emphasis). II. THE METHOD At the base of Seguino’s argument is a growth accounting exercise, along the lines of an ‘augmented’ Solow growth function where output is driven by changes in the labor and capital inputs as well as a human capital variable. It is recognized that this might not be the optimal way of approaching gender in the growth process1, but it is used “for consistency with the methodology of many recent studies that have considered the determinants of Asian growth” (Seguino 2000: 40): Yit = Ait F ( K , LF , HK ) it (1) This is the basic growth accounting relationship where Y is output, A is technological change, K is capital, LF is the labor force, HK is a proxy of human capital, i is the country index and t is time. 2 Gender wage gaps are introduced not in the general growth accounting with a suitable WGAP j variable, where j is the wage gap specification used. Rather, wage gaps are introduced through the technological change A. Here the wage gap is supposed to drive technological change (and not growth explicitly) through a leap-frogging method, as lower wages encourages both exports and a higher exchange rate allowing industry to earn foreign exchange and import foreign technology. (2) Ait = Ci (1 + tφ )eσ ⋅WGAP j it Where Ci is the country-specific time-invariant effect, φ measures the effect of external factors on growth otherwise not captured in the model, differentials on growth and WGAP is the gender wage gap, where j denotes the type of wage gap being tested. To follow the standard procedure (2) is substituted into (1), the natural logs are taken and the equation is differentiated with respect to time in order to allow the growth accounting to be undertaken using regression analysis. d log Yit = φ + Σλi + α 1WGAP j it + α 2 d log K it + α 3 d log LFit + α 4 d log HK it + ε it (3) Where d is the difference operator, φ is now the growth rate of technological change residual. It is worth noting that by assuming the wage gap to be in the technological change variable, one avoids logging it in the growth accounting exercise, and a subtle assumption of an explicit relationship between the wage gap and growth has been worked into the growth accounting, without any explicit evidence being presented about the link from the wage gap to higher exchange rates. a. Expanding the Basic growth accounting relationship Seguino takes the basic growth accounting relationship, and sets out to test a second specification (1b) where the total labor force is substituted for the female and male labor force to see if changes in the gendered labor force have any effect on the growth accounting. ¤ when variables measured at the mean ( it) are fixed effects and   is the effect of gender wage ¢ £¡ is the ubiquitous 3 (1b) Yit = Ait F ( K , LFF , LFM , HK ) it Where LFF and LFF is the percentage of the labor force that is female and male respectively. While only one result using this specification is reported, it is relevant to note that no matter if LFF and LFM were statistically relevant or irrelevant at the 5% significance level, the coefficient of LFM and LMM was always -0.001 and 0.001 in the results making them effectively irrelevant no matter the statistical significance. The reason for this result, I would argue, is that the total labor force is not a well chosen variable, as it includes all women and men in the labor force, no matter if they are factory workers, policy makers or economists. More than that, even the gendered changes in the labor force are small over the years and it is not clear why they should have any pronounced effect on growth under the hypothesis being tested. In addition to using the (1b) specification, I propose including a different variable to account for the gendered labor force, namely the percentage of female and male employees as a part of the labor force (EMF and EMM respectively), as provided in the World Bank’s (2007) World Development Indicators. These statistics give the share of the labor force which consists of employees, as opposed to employers and public servants. It may not be an ideal statistic, but it is focused on the part of the labor force that Seguino is interested in, and the changes within this sub-sample are significantly bigger than in the general labor force, and it is the group of labor (unskilled, temporary) which policy makers in South East Asia has focused on in their gendered labor policies. Yit = Ait F ( K , EMF , EMM , HK ) it (1c) b. The Exchange Rate Transmission from micro to macro The micro level wage gap is meant to transfer into the macro growth experience as the “lower wages substitute for currency devaluations, making exports more competitive” (Seguino 2000: 37), than just by lowering the labor costs of production. This happens as export demand depends solely on the real exchange rate for exports, in a positive relationship. 4 * Where EX is export demand, Z is constant, e is the nominal exchange rate, PX is the foreign currency price of the goods that competing countries are offering, PX is the local So as the sales price of domestic goods fall, or the exchange rate rises, the demand for exports also rises. To tie gender wage gaps in with this, one assumes a mark up pricing rule for producers, combined with the labor costs of employing women. * PX = (1 + µ )(W f a X + ePm n X )   currency price of the country’s export goods and is the price elasticity of exports. (5) * coefficient in the export sector, Pm is the foreign currency price of imported intermediaries, and nX is the import coefficient in the export sector. Seguino argues that “a decline in the female wage causes the domestic price of exports to fall” which is only partially correct. In fact, depending on the propensity of producers to cut prices in accordance with costs, or gain larger margins depends on the elasticity with which exports are demanded. This relation ignores the gender mirror image as male wages (Wm) are not included. A more comprehensive function would be: * PX = (1 + µ )(W f a X + Wm [1 − a X ] + ePm n X ) Again a higher exchange rate would improve the PX and thus exports, but the average labor coefficient of whether men or women are employed in the export sector is highly significant in the sense that on a practical level, it may be easier to exchange labor than to change wages. If this relation is correct across the sample of nations under investigation, one would expect that larger shares of women would be employed in the industrial sector during the period, where data is available. The data does to some extent support this idea, but the transmission mechanism requires further investigation: ¢ Where is a flexible mark-up over unit costs, Wf is female wages, aX is the average labor (5b) Hong Kong, China Indonesia 1980 56.2 13.1 1985 47.0 12.2 1990 33.0 12.0 ¡   (4) * E X = Z (ePX / PX )ψ Z > 0, 0 < < 1995 18.7 15.5 5 Korea, Rep. Malaysia Philippines Singapore Thailand 23.8 20.0 14.9 40.3 7.8 24.4 20.5 12.8 33.4 10.7 30.2 28.0 12.8 -12.3 23.7 31.4 13.2 25.0 17.0 Table 1: Female Employment in Industry Source: World Bank (2007) c. The startling results Seguino presents a very graphic example of the relationship between economic growth and one measure of gender inequality – the log difference between male and female wages (WGAP1) – where “Asian data provide some support for the hypothesis that gender wage inequality is a stimulus to growth and investment… [and] wider gender wage gaps are associated with faster growth” (2000: 38). The graph presented by Seguino has been re-calculated (with some updated values). It still looks the same, and could be described as the graph that launched a thousand regressions. Average annual growth rate GDP (%)/Fitted values 3 4 5 6 7 8 Taiwan Thailand Hong Kong Singapore Malaysia Indonesia South Korea Sri Lanka Philippines .2 .3 .4 .5 Log (Wm) - Log (Wf) .6 .7 Figure 1: Growth rate of GDP, 1975-95 and Gender Wage Gap There are some immediate things to note on this graph: The slope is suggestive of a large effect from wage gaps to economic growth, which was later examined using the growth accounting. The sample consists of only eight observations, counting gender wage gaps and growth averages over a 21 year period, which is highly restrictive. Seguino later ran 6 the growth accounting exercises with a slightly larger cross sample, using 5 year averages, leading to regressions with four variables per country, both of these are very restricted samples. Finally the sample may not be internally consistent as two outliers (Sri Lanka and the Philippines) appear to be dragging the slope of the curve. All these issues need to be addressed. When it comes to the growth accounting results, there are some contradictions with the coefficients on the wage gap and the one pictured in figure 1. In the three reported results in Seguino (2000), the coefficients on WGAP1 are all found to be statistically significant (to the 1% or 5% level), but the coefficients are 0.016 and 0.017 for specification (1) with 5 year averages, 0.041 and 0.34 for specification (1b) with panel data and a whopping 0.138 and 0.147 for a growth accounting specification looking at investment determinants. What this means for the sample is that a 0.10 point increase in WGAP1 should lead to a 0.16, 0.17 point increase in average GDP growth. The slope on that fitted line would be approximately four times shallower than the one in figure 1, and considering they are both average cross sectional data they should correspond somehow. The slope of the curve fits better with the results for panel data, but even so, the relationship in figure 1 implies a relationship of 1:6.1881, which outstrips even the panel data. d. The wage gap specifications The various growth accounting regressions were tested with four different wage gap specifications (j) in Seguino (2000), and the same wage gaps are used in this paper. The wage gaps are introduced into the A equation (2), and when logged and differentiated they are substituted into WGAP j in (3): d log Yit = φ + Σλi + α 1WGAP j it + α 2 d log K it + α 3 d log LFit + α 4 d log HK it + ε it (3) j=0 j=1 j=2 j=3 WGAP0 = 0 WGAP1 = WGAP1 WGAP2 = WGAP2 WGAP3 = WGAP1 + EDGAP 7 j = 0 is a regression run where no wage gap is used, to see what the results look like without an explicit gender wage gap. j = 1 is a basic gender wage gap: log(Wm) – log(Wf) where Wm and Wf are male and female earnings respectively. j = 2, WGAP2 attempts to control for productivity differences by taking account of the educational attainment of male and female labor, by dividing earnings (conceived as marginal productivity) by the average years of secondary education per male and female 15 and over (SEm and SEf respectively). WGAP2 is also referred to as an ‘efficiency’ gender wage gap. ¢     ¡ £ £ ¤ ¥ Finally, to try and control for educational attainment when using WGAP1, a control variable EDGAP is introduced as the log difference of female and male secondary educational attainment. More details on the gaps are available in Seguino (2000), but it is worth pausing to consider the data which these indicators are based on, and their availability. There are no easily available data on secondary educational attainment along gender lines. In fact, the education data have to be approximated from five year population data and school leaving data corresponding to population growth. Data on this and the Wage gaps are not available on a regular basis, so regressions not based on five year, or twenty year averages have to exclude a sizeable portion of data points (61 out of 180 data points were excluded on this basis). That being said, averages have to choose a way to generate nonexisting variables, generally by linear approximation, which given the smaller data-sets might give less reliable estimates. For human capital (which is also based on educational attainment in the population), data is sometimes available once in a decade, and a lot of linear approximation, averaging or simply assuming one value across the dataset has to be done. For all the tests in this paper, when more than one data point has been available a linear approximation was made, otherwise single data points were taken as full period averages.     ¡ £ ¤ (6) £ WGAP 2 = log Wf Wm − log SE m SE f ¢ ¥ 8 III. WHEN, WHERE AND WHO? As is sometimes the case in big empirical papers, a great number of interesting statistical overviews were presented to introduce the reader to the region and issues under consideration. However, to draw on an example, in the process of going through comparisons of East Asia, Latin America, Sub-Saharan Africa and Middle-Eastern growth averages for 1965-91 and growth indicators for ‘Asia’ in 1990, 1979-93, 1970-93 and 1975-95, all in one page (Seguino 2000: 29) some of the perspective on what countries were under consideration, and who and when they were representatives for could be lost on the reader. As the case was, the abstract which talked about the period 1975-1990, was to be corrected in all the econometric exercises which worked only with data in the 21 year period 1975-1995. When the title refers to such a wide geographic area as ‘Asia’, the subtopic is ‘semi-industrialized economies’, and the issue is gender inequality, the generality of the sample is of great importance, and the question arises: Exactly who is being investigated, and why should the results be globally applicable? a. Defining Asia The claim to account for “Gender in Asian Economic Growth” is substantiated by testing a ‘sample’ population of Asia consisting of Hong Kong, Singapore, Taiwan, South Korea, Thailand, Malaysia, Indonesia, Sri Lanka and the Philippines. Geographically this area spans from the southern tip of India in ‘South Asia’ through the ‘South-East Asian’ area east of Bangladesh, North of Australia and south of China, through ‘East Asia’. Already excluded are central and north Asia which both house considerable populations. As a sample of South, East and South-East Asia, the countries include some 14% of the population in the region today. 9 Figure 2: Population figures, Source: CIA Factbook 2007 The sample is far from homogenous, nor geographically balanced. In fact Seguino identifies a group of “earlier industrializers (South Korea, Singapore, Hong Kong and Taiwan)” (2000: 29) and another group “Thailand, Indonesia and Malaysia… [as] second-tier NICs” (2000: 51). The first of these (NIC1) accounts for a mere 15% of the sample population, while ‘second-tier’ NIC2 accounts for some 63% mainly through the inclusion of Indonesia. The possible economic and geographic outliers then become Sri Lanka and the Philippines, NIC3. On the whole, the sample is not representative of a wider Asian grouping; rather it is made up of two waves of NICs and two possible anomalies. Seguino performed a number of growth accounting exercises, and when repeating these for bigger samples of raw data, I ran them with dummy variables to see if there was a statistically significant difference between the two waves of NICs and the two potential outliers. 10 NIC 1 8 Taiwan Hong Kong South Korea Singapore NIC 2 Thailand NIC 3 Average annual growth rate GDP (%) Malaysia Indonesia 6 Sri Lanka 4 Philippines 2 .2 .4 .6 .8 .2 .4 .6 .8 .2 .4 .6 .8 Log (Wm) - Log (Wf) Graphs by var13 Figure 3: Separating the sample nations For the basic growth accounting specification (1) used by Seguino in her first rounds of tests (2000: 43), I found a statistically significant difference between NIC1+NIC2 and NIC3, with a 1% significance level between these two groups, for all specifications of the wage gap j. Considering this, it might be more appropriate to consider the NIC3 countries with a different group of countries than the original sample. Considering that the am of the paper was to address the ‘semi-industrial’ parts of Asia, which effectively is the sample, Seguino suggests an international control group of semiindustrial countries. b. What it means to be ‘Semi-Industrial’ There is no clear definition of what makes a country ‘semi-industrial’ but a benchmark could be the contribution to GDP of non-agricultural sectors in the economy (i.e. industry and services) based on the notion that Darity (1995) made an explicit separation between rural economies and the rest. Based on this, there should be a group of countries which falls between the rurally focused economy of Darity and the industrialized nations in the EU and North America. Seguino suggested a group of countries which could be semi- 11 industrial, and proceeded to use them in cross country regressions with the original sample group. Service & Industrial Contribution to GDP Old Sample Hong Kong Singapore South Korea Thailand Indonesia Malaysia Sri Lanka Philippines 1975 *99.2 97.7 72.9 73.1 69.8 71.1 69.6 69.7 1985 99.5 99.0 86.5 84.2 76.8 80.1 72.3 75.4 1995 99.9 99.8 93.7 90.5 82.9 87.1 77.0 78.4 New Sample Brazil Chile Colombia Costa Rica Cyprus El Salvador Greece Mexico Paraguay Portugal Turkey 1975 87.9 93.4 75.6 76.7 84.3 -84.9 88.2 63.1 74.9 64.2 1985 88.5 92.4 82.5 78.2 92.5 **82.6 86.8 89.9 71.1 85.6 79.6 1995 91.0 90.8 84.7 86.3 ***94.9 85.5 89.8 94.3 75.2 94.6 83.6 Table 2: Service and Industrial Contribution to GDP Source: World Bank (2007) * 1980 Value, **1990 Value, ***1994 Value, no values available for Taiwan. On general inspection one might suspect that these two groups do not easily correspond and knowing that NIC3 is significantly different from NIC1 & 2 some further tests should be warranted before using them as similar samples. Plotting a predicted regression line with a 95% confidence interval for the new sample group, and seeing how many of the old sample falls within this, would be a good first exercise. 12 Figure 4: Predicted fit of New Sample with 95% Confidence interval, and Sample Data Immediately one notes that the vast majority of the data points for Seguino’s original sample fall far outside the confidence interval for the new sample nations. To verify this observation, dummies were run on all the forms of growth accounting used in the later regression analysis (1, 1b and 1c) to test whether the new sample (or ‘Rest’) was significantly different from NIC1,2,3 and whether NIC1 & 2 were significantly different from the joint NIC 3 & Rest Samples. Running the very simple average growth values and average WGAP1 values for the whole time-period one finds a stark difference between Seguino’s sample group and the ‘rest’ of the semi-industrial countries: Average annual growth rate GDP (%)/Fitted values Rest 8 Seguino Taiwan Thailand Hong Kong Cyprus Indonesia South Korea Singapore Malaysia 6 Chile Paraguay Colombia Turkey Costa Rica Portugal Mexico Greece Brazil Philippines Sri Lanka 2 El Salvador 4 .2 .4 .6 .8 .2 .4 .6 .8 Log (Wm) - Log (Wf) Average annual growth rate GDP (%) Graphs by Dummy Seguino Sample against Rest Fitted values Figure 5: Comparing Seguino’s sample and the ‘rest’ of semi industrial countries. When testing across the growth accounting, all the dummy variables are significant to a 1% significance level across all j, implying that both Seguino’s original sample countries and the suggested alternative (NIC1 and NIC2 alone) do not correspond to a more widely 13 defined category considered as ‘semi-industrial’. More than that, it implies that it may be fruitful to consider NIC3 as part of the ‘Rest’ Sample. Based on this I argue that Seguino (2000) can not justifiably claim to address ‘semiindustrial’ countries, nor ‘Asian’ countries. Rather, her paper was trying to account for the impact of gender discrimination in the Newly Industrialised South East Asian Countries only. All the results from this exercise cannot be generalised from this very specialised and statistically different sub-sample of nations. Taking this argument to the next stage, we can review figure 1, and reconsider the steepness of the WGAP1 fitted curve, on the basis that NIC3 should not be included in the original sample of nations. Average annual growth rate GDP (%) 8 Taiwan Thailand Hong Kong Indonesia Chile Paraguay Lanka Sri Turkey Colombia Costa Rica Philippines Portugal Mexico Greece El Salvador South Korea Singapore Malaysia Cyprus 6 4 Brazil 2 0 0 .2 .4 Log (Wm) - Log (Wf) .6 .8 NIC 3 & ‘Rest’ NIC 1 & NIC 2 Figure 6: Re-assessing Seguino by new groupings Figure 6 re-affirms Seguino’s conclusion that there is ‘some’ relation between gender wage gaps and growth, but the slope variables are much more in sync with the growth accounting results achieved with the 5 year averages in the 2000 paper. 14 IV. GROWTH ACCOUNTING AND GENDER WAGE GAPS The next step is to expand the data set in the growth accounting to include all the data for the period 1975-95. This will hopefully throw some new light on the conclusions reached in the growth accounting exercise by Seguino (2000). I present the results in the same tabular form as the one given in the original paper for easy comparison3. The first set of results is for the same sample of nations that is used in Seguino’s original paper (2000: 43), with the same gender wage gaps and the same growth accounting specification (1). The only difference is that the data has been updated where relevant, and I am using the full sample of data as opposed to five year averages. Table 3: Determinants of GDP growth (1) Variable Constant d log K d log LF d log HK WGAP 1 WGAP 2 EDGAP Observations: 189 106 106 j=0 0.751 (0.001)* j=1 1.049 (0.000)* j=2 1.082 (0.000)* j=3 1.052 (0.000)* 0.984 (0.000)* 0.996 (0.000)* 0.987 (0.000)* 0.995 (0.000)* 0.058 (0.000)* 0.027 (0.065)*** 0.0275 (0.029)** 0.023 (0.063)*** 0.008 (0.805) 0.154 (0.000) 0.149 (0.000)* 0.153 (0.000)* -0.450 (0.000)* -0.459 (0.000)* -0.409 (0.000)* 0.054 (0.756) 106 F-Statistic accepted at the 1% significance level for all j. Note: Numbers in parentheses are P>|t| statistics; Asterix is short-hand: *p<1%, **p<5%, ***p<10% What becomes immediately apparent is that the wage gaps rather than being positively correlated, as in Seguino’s average sample, are now negatively correlated with growth. The statistical significance has also improved for all the wage gaps, while the education gap (EDGAP) is insignificant both in this result and in Seguino’s original. The coefficients on the other variables are as one would expect, although HK is found to be statistically insignificant for the samples where education is not part of the wage gap analysis. This may be associated with the fact that the data for the education variables 15 were averaged across each respective country series, and so they generally have relatively little movement in relation to changes in growth. b. Using a new sample of countries Using the fact that NIC1 & 2 are statistically different from NIC 3 and the ‘rest’ of the nations, we can run a new set of growth accounting regressions, to see if the relation between growth and the wage gaps are the same for the two top tier NICs. Table 4: Determinants of GDP growth (1c), NIC1 & NIC 2 Variable Constant d log K d log EMM d log EMF d log HK WGAP 1 WGAP 2 EDGAP Observations: 76 50 50 j=0 0.851 (0.121) j=1 1.381 (0.117) j=2 1.430 (0.088)*** j=3 1.052 (0.000)* 1.009 (0.000)* 0.991 (0.000)* 0.986 (0.000)* 0.980 (0.000)* 0.339 (0.031)** 0.544 (0.030)** 0.504 (0.031)** 0.424 (0.079)*** -0.376 (0.010)* 0.033 (0.556) -0.580 (0.001)* 0.164 (0.074)*** -0.532 (0.001)* 0.156 (0.076)*** -0.460 (0.007)* 0.151 (0.083)*** -0.535 (0.001)* -0.706 (0.000)* -0.677 (0.000)* 1.264 (0.023)** 50 F-Statistic accepted at the 1% significance level for all j. Note: Numbers in parentheses are P>|t| statistics; Asterix is short-hand: *p<1%, **p<5%, ***p<10% Again one finds that the wage gaps are negatively correlated with the growth of GDP, and the education gap is also positive. Furthermore, the employment of women and men in the labor force as ‘employees’ is statistically significant, and the coefficients for employing women are negative as opposed to the positive coefficients of male participation. The labor coefficients for both labor force participation and employee percentages are consistent with table 4’s coefficients across all possible specifications. The coefficient for LFF and EMF in both Seguino’s Sample and the top-two-tier NIC sample remains negative, and the opposite is true for LFM and EFM. Running the same growth 16 accounting exercises for the NIC3+Rest and Rest samples do not correspond to this observation however. The implication is that employing more women in low-skill low-wage jobs can be detrimental for the countries in the Seguino or ‘Top Two’ sample. This might signify that the countries in question have overdone gender wage inequality, and have hit some sort of tipping point, conceptualized in some higher power specification of the relationship between wage gaps and growth. 15 Annual GDP Growth % 10 5 0 -5 -10 -.5 WGAP1: Ln (Wm) - Ln (Wf) Annual GDP Growth % Seguino (NIC1, 2, 3) 0 .5 NIC 1 & NIC 2 NIC 3 & Rest 1 predicted g Figure 7: Whiplash effect When running a fractal polynomial regression, on the four sub-samples considered so far, and plot them against the wage gaps and the growth, a curious picture emerges. The South East Asian NIC’s, are statistically different from the other groups, but not only in the slope, but also the general shape and curvature of the relationship between growth and the wage gap. A similar relationship is exists for WGAP2 and growth. The NICs have a higher annual growth rate at the 0.4-0.5 wage gap level than the other sample group. More than that, it seems that movements away from this top point has a 17 negative effect on growth, which explains some of the negative coefficients on both wage gaps and labor participation in the growth accounting. The ‘Top Two’ sample curve does appear to be flattening out as the WGAP approaches zero. The reason for the higher GDP growth at a higher WGAP is most probably the effect of the government policies instituted by the NICs over the period in question as described by Seguino. However the end result has been a sustainable status quo, which cannot be improved by reducing or increasing the gender wage gap. I would suggest that further research, which is beyond the scope of this critique, should be attempted, to discover whether it was the economic policies of the NICs which created this hyperbola relationship between growth and gender inequality, or if they were social structures in place prior to the policies. The parabola relationship seen for the outside samples, are reminiscent of the Kuznets curve. If the minima had been where WGAP1 = 0, we could tell a story about a lack of exploitation, and re-affirm Seguino’s argument that gender wage gaps can improve growth. That being said, the gender wage gap can be in either direction, and it would still improve growth, meaning that the gender wage gap argument could simply be a question of cheaper labor defined by one social group (gender) as opposed to another. As it stands, there is some evidence here that all the countries in the data have some structural gender bias against female wages, and that removing it would be beneficial for growth to some extent, but worsening it, could also lead to higher growth up to a point. This graph does something rather remarkable, in that it agrees with Seguino’s analysis that gender inequality was a positive factor in the growth experience of the NICs, but it was not through increasing the gender wage gap that this was achieved, rather a set of micro social structures and policies inverted the ‘growth wage-gap’ parabolic relationship, and put the NICs on a higher trajectory. 18 V. EXCHANGE RATES AND GENDER WAGE GAPS I have argued that the micro-macro relation from wage gaps to growth are not really significant as part of the story on how gender wage gaps influenced the growth experience of the NICs. For completeness, I investigate the possible link from wage gaps to maintaining a higher exchange rate, by using data for the percentage change in the price of one U.S. dollar in domestic currency, and regressing it against the growth rate. It might be more suitable to consider the relationship in a timely sense, so that a higher wage gap in year one, leads to a stronger currency in year two. Further to that, the sample for currency prices include some extreme years, with the Latin American liquidity crisis, and the associated hyperinflation. In an attempt to control for this effect, the sample is restricted to annual swings in currency prices of 15% up or down. Immediately I consider Seguino’s sample of countries first, regressing the lagged and current exchange rate against WGAP1, and also testing whether Seguino’s sample is statistically different from the ‘rest’ of the sample. Table 5: Exchange Rates and Wage Gap 1: Variable Exchange Rate Exchange Rate Exchange Rate Exchange Rate Exchange Rate Sample Seguino Rest NIC1 & NIC 2 NIC 3 & Rest Whole Sample Dummy Significant X X V V n/a Coefficient -6.48 (0.017)** -6.98 (0.152)† +0.78 (0.808) -7.03 (0.107)† -8.54 (0.00)* Lagged E.R. (t+1) Lagged E.R. (t+1) Lagged E.R. (t+1) Lagged E.R. (t+1) Seguino Rest NIC 1 & 2 NIC 3 & Rest V V V V -6.14 (0.022)** -5.65 (0.275) +0.03 (0.970) -4.73 (0.253) Note: Numbers in parentheses are P>|t| statistics; *p<1%, **p<5%, ***p<10% †p<20% For the current exchange rate against the wage gap there is no significant difference between Seguino’s sample and the ‘Rest’ of the data available, so a regressions for the 19 whole sample was run. The dummy was however significant for the lagged exchange rate effect. The overall picture (which is generally mirrored for WGAP2) is that there is a large negative relationship between the WGAP and the cost of a dollar in domestic currency. This means that this dataset agrees with Seguino’s link from the wage gap to exchange rates, and their potentially positive effect on export viability. That being said, the relation does not extend to actually maintaining the exchange rate, but simply to slow down its depreciation. Ironically, the only sub sample which do not exhibit this pattern is the NIC 1+2 sample, where a positive coefficient is found for both the lagged and current exchange rate, and the current rate is significant inside a 10.7% significance level. This signifies that there is a significant difference between the early industrializers (NIC1 & NIC2) and all the other sub-samples, and ironically they are the only group for which exchange rates have a positive (and thus detrimental) relationship with the wage gap. % Change in Exchange Rate for $/Fitted values NIC 1 & 2 20 NIC 3 & Rest -10 0 0 10 .2 .4 .6 .8 0 .2 .4 .6 .8 WGAP1: Ln (Wm) - Ln (Wf) % Change in Exchange Rate for $ Graphs by Dummy between NIC 1+2 and NIC 3 + Rest Fitted values Figure 8: Exchange rates and WGAP1. 20 VI. CONCLUDING REMARKS Seguino argued that wage inequality had a positive effect on economic growth in the Newly Industrialized South East Asian Countries. This had happened through a number of micro economic policies and social structures in place in those countries during the growth years, and through the exchange rate this had manifested on the macro-level, and could be used as a form of policy lever, where wider inequality gaps could lead to higher growth. This observation it was argued could be applied to other similar semi-industrial countries across the world This paper has tried to argue that the micro economic analysis performed was correct, and that gender inequality did have an effect on the growth experience during the period 1975-95 in the NICs. However, the micro-macro link for the sample was not accurately reflected, and in fact the relation from the micro to the macro is a lot more complex than a simple linear regression might suggest. I have argued that the findings for the NICs cannot be generalized across the geographic region, and in fact there is a structural difference inside the geographic sample between the earlier industrializers, and those who are still trying to catch up. Furthermore neither Seguino’s sample nor the smaller (more highly industrialized) sub-sample I propose have any structural similarities to other ‘semi-industrial’ countries, and as such results can not be generalized across this group either. The early industrializers in South East Asia, I have argued are distinctly different from other countries, as government micro policy and social structures created countries with a hyperbolic relationship between gender wage inequality and growth, and this has meant that inequality has been structurally beneficial in the sample, but widening the wage gaps will not improve the growth performance of these countries. More than that, the possible relation from the wage gap to economic growth, the micro to macro link, was not conducive to growth in the sample nations. This paper may have opened up more questions than it answered, but its original intention was to critique the idea that widening gender inequality can be used as a policy tool for boosting growth as part of the development process, and I think this has been 21 shown to be false in the case of the South East Asian growth miracle. This paper agrees with Seguino’s notion that the ‘orthodox’ equity-growth story for South East Asia is not sufficient for understanding the growth process, but it is not a question of wider gaps, bigger growth. Some directions for future research would be to expand the data-sets and country samples to see how wage gaps and growth interacts in other regions and economies, and to understand how the slope and curvature of the inequality-growth relation can be shaped and changed by government and social institutions. 22 END-NOTES: 1: based on regression result for the growth averages and average wage gap 1 2. Particularly if one considers that the technical meaning of the log derivative results for ‘contributions’ to growth are a measure of income shares between the coefficients, and are meant as a proxy for the input-output relation between the independent and dependent variable in the accounting. See JeonYongbok Chen (2007) for a more thorough review. 3: In Seguino (2000) the headers Eq. (1), Eq (2), Eq(3) and Eq(4) is equivalent to j = 0, j = 1, j = 2 and j = 3 respectively BIBLIOGRAPHY: Adelman, Irma and Sherman Robinson; 1989; Income Distribution and Development; in Chenery, Hollis and T.N. Srinivasen (editors); Handbook of Development Economics: Vol II; Amsterdam; North Holland. Amsden, Alice; 1989; Asia’s next giant: South Korea and late industrialization; Oxford; Oxford University Press. Bourguignon, F.; 1981; Pareto-Superiority of Unegalitarian Equilibria in Stiglitz’s Model of Wealth Distribution with Convex Savings Function; Econometrica; 49; pages 1469-1475. Bowman, Kirk S.; 1997; Should the Kuznets effect be relied on to induce equalizing growth: Evidence from post-1950 development; World Development, Volume 25, Issue 1, January, Pages 127-143 Braunstein, Elissa and Nancy Folbre; 1998; To honor and obey: The patriarch as Residual Claimant; Working Paper, Economics Department, University of Massachusetts at Amherst. Carter, Michael and Elizabeth Katz; 1997; Separate spheres and the conjugal contract: Understanding the impact of gender biased development; in Haddad, Lawrence, John Hoddinott and Harold Elderman, editors; 1997; Intrahousehold resource Allocation in Developing Countries: Models Methods and Policy; Baltimore; Johns Hopkins University Press. Darity jr, William; 1995; The formal structure of a gender-segregated low-income economy; World Development; Volume 23; Issue 11, Nov.; Pages 1963-1968. Gustaffson, Björn and Li Shi; 2001; Economic Transformation and the Gender Earnings Gap in Urban China; in Riskin, Carl, Zhao Renwei and Li Shi (Editors); 2001; China's Retreat from Equality: Income Distribution and Economic Transition; New York; An East Gate Book. Hsiung, Ping-Chun; 1996; Living Rooms as Factories: Class, Gender and the SatelliteFactor System in Taiwan; Philadelphia; Temple University Press. 23 Jeon, Yongbok, 2007, “Total Factor Productivity, Income Distribution, and Growth Accounting in China: A Methodological Critique”, presented at ICAPE conference 2007, University of Utah Kuznets, Simon; 1955; Economic Growth and Income Inequality; American Economic Review; 45; pages 1-28. Larrain, Felipe and Rodrigo Vergara; 1998; Income Distribution, Investment and Growth; in Solimano, Andres (editor); 1998; Social Inequality: Values, Growth and the State; Ann Arbor; University of Michigan Press. Oaxaca, Ronald; 1973; Male-female wage differentials in urban labor markets; International Economic Review; Vol.14, No.3, pages 693-709. Riskin, Carl, Zhao Renwei and Li Shi (Editors); 2001; China's Retreat from Equality: Income Distribution and Economic Transition; New York; An East Gate Book. Sadeghi, J.M.; 1995; The Relationship of Gender Differences in Education to Economic Growth: A Cross-country Analysis; Cairo; Working Paper No. 9521, Economic Research Forum for the Arab Countries, Iran and Turkey, (ERF). Seguino, Stephanie; 1997; Gender Inequality and Export-Led Growth in South Korea; Journal of Development Studies; 34(2); pages 102-32 Seguino, Stephanie; 2000; Accounting for Gender in Asian Economic Growth; Feminist Economist, 6(3), pages 27-58. Seguino, Stephanie; 2000b; Gender Inequality and Economic Growth: A Cross-Country Analysis; World Development, Volume 28, Issue 7, 1 July, Pages 1211-1230 Tien-Tung, Hsueh and Li Qiang; 1999; China's National Income 1952-1995; Oxford, Westview Press. World Bank; 1993; The East Asian Miracle; Oxford; Oxford University Press Data Sources: Available as separate Excel Sheet, tabs: Other World Bank, 2007, World Development Indicators for Agricultural, Service and Industry contribution to GDP and Employment in Agricultural, Service and Industry. CIA, 2007, CIA Factbook Data Sources, as received by e-mail from Dr. Seguino 24/12-04: Available as separate Excel sheet, tabs: Main and Sources. The Penn World Tables 5.6 are online at:http://www.nber.org/pub/pwt56/dos/ 24 Accessed 6/12-05. They contain data from 1950 to 1992 and include all of the countries in the sample. The World Bank World Development Indicators 1997 CD-ROM. It contains data from 1970 to 1995 for each country and includes each country in the sample except Taiwan. The disk contains 500 different variables. The Barro Lee dataset was downloaded from the Internet at http://www.worldbank.org/html/prdmg/grthweb/ddbarlee.htm. The data is available for each of the sample countries in the years 1960, 1965, 1970, 1975, 1980, 1985 and 1990. The Nehru and Dhareshwar data set was downloaded from the web site "http://www.worldbank.org/html/prdmg/grthweb/ddnehda.htm". This data is not available after 1990. ILO Wage data is available from 1969 to 1995 at most from the data extract sent in the file "Laborsta.exe." GENMAC formatted data may be found in the file "T5A.xls". Received 24/12-04. Additional Social Indicators from the World Bank Social Indicators may be found in the file "sid2.xls". This information was downloaded from the internet "http://www.ciesin.org" received 24/12-04. The Gini Coefficient from Deininger and Squires can be found in the file "Ndata.xls". The data was downloaded from the internet at (received 24/12-04) "http://www.worldbank.org/html/prdmg/grthweb/dddeisqu.htm" ILO Unemployment data is mainly from a data extract file now called "laborsta.exe" which contained the file "T3A.dat". This information was collected and formatted in the file "T3A.xls" where it was supplemented with data from various years of the ILO's – also received 24/12-04 25