Draft – please do not cite or circulate

  

Can Economic Growth be Inclusive and Immiserising?

The Kuznets Hypothesis and ‘Premature Deindustrialisation’

  Arief Anshory Yusuf *, Andy Sumner** & Ekki Syamsulhakim*

• Padjadjaran University, Indonesia and ** King’s College London, UK

  Draft

• – please do not cite or circulate

  In this paper, we consider the impact of economic growth on poverty and inequality

  Abstract:

  in Indonesia. We revisit the ‘Kuznets hypothesis’ and review contemporary theorists writing in the Kuznets tradition. We then argue that there is a new relevance for Kuznets in terms of developing countries experiencing ‘premature deindustrialisation’.

  1. Introduction

  In this paper, we use district level data from Indonesia to consider how economic growth has impacted poverty and inequality during a period of ‘premature deindustrialisation’. To put our findings in context and to build new theory, we revisit the

  ‘Kuznets hypothesis’ and review writing in the tradition of Kuznets. We then hypothesise a new relevance of Kuznets to developing countries experiencing ‘premature deindustrialisation’. The paper is structured as follows. Section 2 discusses the literature on poverty, inequality and growth. Section 3 considers the case of Indonesia. Section 4 revisits Kuznets. Section 5 concludes.

  2. Poverty, inequality and growth

  Numerous scholars have, without doubt, given considerable attention to the poverty, inequality

• – and growth relationship in developing countries. Historically, the interest in the broad area defined as who benefits from growth and how much
• – grew from debates in the early 1970s that were critical of the distribution of the benefits of growth at that time (e.g., Adelman and

  1 became the umbrella term for considering who benefits from growth. .

  The body of literature provides the empirical basis for the accepted notion that economic growth is inclusive in a general sense: on average, the poverty headcount falls and the incomes of the poorest rise in line with average income growth (see Dollar and Kraay, 2002; Kraay, 2006; Dollar et al., 2013). However, two recent contributions have reopened this debate. First, Shaffer (2015) notes that in 10-15 per cent of episodes, poverty actually rises with growth and he connects this with historical debates on ‘immiserising growth’. Second, Sen (2013) concurs with this finding in that there are a surprising number of growth episodes that are not inclusive (i.e., episodes are without falling poverty and/or with rising inequality).

  Sen separates types of growth episo des between ‘growth acceleration’ and ‘growth maintenance’ and finds that the former is much less likely to benefit the poor than the latter.

  Sen argues that this is because the institutional factors that lead to growth accelerations are

  2

  different from those that lead to growth maintenance. This points towards a well noted argument that the average inclusivity of growth can be misleading as it is subject to enormous ₁ variation across countries and highly sensitive to where the poverty line is set (see also the

  

The various labels differ from each other in nuances. For example, ‘growth with equity’ was typically defined discussions of Edward and Sumner, 2015). Much of the debate turns on whether inequality is high or rising, as high and rising inequality can hamper not only poverty reduction but also

  3 future growth prospects, which can impact future poverty reduction.

  What variables are thought in the literature to explain such cross-country variations? The role of initial inequality is most often cited in determining the responsiveness of poverty to growth. Specifically, that the extent of poverty reduction depends on prevailing inequality levels and that a higher level of initial inequality leads to less poverty reduction at any given level of growth which is a mathematical identity (see Adams, 2003; Deininger and Squire, 1998; Fosu, 2011; Hanmer and Naschold, 2001; Kalwij and Verschoor, 2007; Misselhorn and Klasen, 2006; Ravallion, 1995, 1997, 2001, 2004, 2005a, 2005b; Son and Kakwani, 2003; Stewart, 2000). A recent observation is – perhaps Kuznets revenge – that rapid growth with structural change is associated with rising inequality in many developing countries such as Indonesia (see Sumner, 2016). Although this is not a very large set of countries experiencing growth with structural change data on decile shares to the richest decile corroborate rising inequality with fast growth. That said of course governments are not impotent and the range of policies to manage inequality is well known (see UNDP, 2010). Growth incidence curves (GICs) provide one way to visualise the poverty, inequality and growth relatonship beyond simply presenting elasticities. Figure 1 shows the growth incidence curves for different administrations in Indonesia from 1994-2014. Figure 1 shows that, with the exception of the period of 1994-1999, the Soeharto-Habibie (ST-HB) era, the Indonesian growth incidence curve for all administrations before Jokowi has always been upward-sloping. Comparing the growth-incidence-curves (GIC) for four different adminstrations, one general observation is that the period of Yudhoyono (SBY) presidency (2004-2014) is the period of the most rapidly rising prosperity. However, as can be seen from the slope of the GICs, these periods are also the most unequalising. The later SBY term (2009-2014) is notable in that during this period the expenditure per capita of the top 5% grew annually at 10% (see the second panel of Figure 1). The period is the period where Indonesia experienced the greatest increase in consumption inequality (see below).

  The faster growth of the richest during the second SBY period may be related to the commodity boom. Yusuf and Sumner (2013) discuss this specifically in an attempt to suggest reasons behind rising inequality in Indonesia during the 2000s. The commodity boom hypothesis is that because natural resources sectors are typically capital intensive, the proceeds boom. Of course this is insufficient to establish causation though it does provide a clear correlation between rising commodity prices and inequality in Indonesia.

  Looking carefully the GIC under the second SBY term for the poorest 30% of the population, we can make two observations: First, up-to approximate the 30th percentile the slope is downward sloping. This means that the expenditure of the very poor increased more than the less-poor. Second, the rural poor experienced higher expenditure growth than the urban

  4

  poor. One of the important milestones in the development of social protection programs in Indonesia during the second SBY term was the introduction of a unified database for beneficiaries to support the implementation of targeted social assistance. TNP2K (Tim

  

Nasional Percepatan Penangulangan Kemiskinan) , a new body replacing the earlier

  coordinating agency, TKPK (which did not have authority, capacity and resources to the same extent) was behind the effort of the database unification. TNP2K was also given responsibility to oversee the coordination of household-based social assistance programs, community empowerment programs, and programs to expand economic opportunities for low-income households (Sumarto and Suryadarma, 2011).

  Figure 1. Growth incidence curves under various administrations (Percentage change in real expenditure per person by percentile of expenditure per person), 1994-2014

  2009-2014 (SBY II)

  10 2004-2009 (SBY I)

  8

  6

  4 1994-1999 (SHT-HB)

  2 1999-2004 (GD-MGW)

  4 8 12 16 20 24 28 32 36 40 44 48 52 56 60 64 68 72 76 80 84 88 92 96 100 Percentile of exp./person

  Source: Authors’ calculation based on SUSENAS The major social protection programs include the rice for the poor (Raskin), conditional cash transfers (CCT), and national community health insurance (Jamkesnas). The negative but the richer less so (proportionally to their initial level) than the poor. This may relate to the Asian Financial Crisis of 1997/98. The deterioration of Indonesian currency, the collapse of the banking and financial sectors that happened during the crisis would have tended to hurt the rich more than the poor though all of society was hit by the crisis. The period of the recovery from the crisis (1999-2004) under the presidency of Abdurrahman Wahid (or Gus Dur) and later on Megawati was a period of the lowest growth in expenditure. The flatness of the GIC of this period indicates equal benefits of growth during that period.

  Next, we link changes in poverty to growth elasticities. Table 1 thus seeks to compare administrations. The period of 1994-1999 (Soeharto-Habibie) was successful in reducing multidimensional poverty despite the crisis. As noted above the fact that the elasticity of multidimensional poverty to growth is high in the period interrupted by the economic crisis may suggest the lower correlation of multidimensional poverty to economic business cycles. We conclude that the SBY period was the least successful in term of generating poverty reduction for every 1% economic growth. This applies to both the poverty incidence by the national poverty line or multidimensional poverty incidence. In contrast, the Megawati and Gus-Dur period had made a 2% reduction in multidimensional poverty for every 1% economic Figure 2. Growth incidence curve for Jokowi’s presidency, Sept. 2014 to Sept 2015 Source: Authors’ calculation

  

Percentage change of real expenditure per person (Sept'14 - Sept'15)

  20

  15

  10

  5

1 4 7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64 67 70 73 76 79 82 85 88 91 94 97

• 5 Urban Rural Urban+Rural

  Source: Authors’ calculation based on SUSENAS Jokowi’s administration could be judged against the baseline of his predecessors. We concluded above that economic growth between 1994 and 2014 has been accompanied by an impressive decline in multidimensional poverty and moderate progress in reducing monetary poverty incidence. Inequality, however, has risen to an unprecedented magnitude, particularly people. The increase in poverty is most likely due to: (i) the slowing down of economic growth

  5

  and hence weaker employment growth; (ii) an increase in the price of rice ; and (iii) The delayed disbursement of cash compensation intended to protect the poor from the impact of the

  6 fuel price increase in November 2015 (See Yusuf and Sumner, 2015).

  The growth incidence curve (for the first year of the Jokowi administration) shows that the upper middle income group, particularly those close to the top 10% saw their expenditure fall, particularly in urban area. Some caution is required though in interpretation of this. This dynamic is most likely cyclical and related to the nature of the slowing down of the economy which is characterized by resource and capital-intensive (and most likely skill-intensive) sector being the hardest hit. This is supported by the figure which shows that urban upper middle class suffered more than rural upper middle class.

  Table 2. Changes in poverty incidence and inequality under Jokowi’s administration

  Poverty incidence Gini coefficient Palma Ratio Urban Rural All Urban Rural All Urban Rural All September 2014

  8.16

  13.76

  10.96

  0.43

  0.34

  0.41

  2.20

  1.33

  2.00 March 2015

  8.29

  14.21

  11.22

  0.43

  0.33

  0.41

  2.16

  1.34

  1.95 September 2015

  8.22

  14.09

  11.13

  0.42

  0.33

  0.40

  2.03

  1.30

  1.80 Sept '14-March '15

  0.13

  0.45

0.26 -0.01 0.00 -0.01 -0.04 0.01 -0.05 The nature of the distribution of economic growth (as shown by the GIC curves) implies a slight decline in inequality nationally and in urban areas. However, the magnitude of the change is not that significant and discussion of a change in inequality over such as short period of time is questionable.

  The official BPS poverty estimates based on that Susenas suggest that the poverty incidence still increased over the one year period. However, there is small improvement on the March 2015 position, but the poverty incidence in September 2015 is still higher compared to the previous one year. It suggest that during the first year of Jokowi’s administration, the poverty incidence rose, nationally by 0.17% (0.06% in urban area and 0.33% in rural areas). As a result of this, and in contrast to the previous 4 administrations, Jokowi’s first year has a positive growth elasticity of poverty. This specifically meaning that for every one per cent of economic growth generated, poverty increase by 0.04% percentage point.

  In sum, although the aspirations and the budgetary reform undertaken, the first year of Jokowi’s presidency has been less successful in term of the inclusiveness of growth compared to previous administrations. That said, the outlook for the remaining period of the current administration is likely to be more position given changes in social spending. The inclusivity

  

3b. Urban and rural poverty, inequality and growth in Indonesia: An econometric

analysis

  The figure below plots the annualised change in the poverty headcounts (by the national poverty line which is very close to the global poverty line of $1.90) of around 300 districts in Indonesia for the period of 2002-2012 against the economic growth during the same period. It suggests that most districts that have positive economic growth per capita experienced a decline in poverty incidence. However, we can observe from the plot that districts that are classified as urban areas are clustered around a smaller change in poverty incidence even when they have high economic growth. There are quite a number of cities with rises in poverty incidence despite positive or even substantially above-national-average economic growth during the period. One plausible explanation for this is that the rural poor migrate and become the urban poor though other factors are likely to also play a role.

  Figure 3. The relationship between growth and poverty reduction in Indonesian districts

  1

• 1
<
• 20 -10

  10

  20 Annual change in total GDP per person curban crural Econometric analysis

  We model changes in poverty incidence as a function of economic growth, initial poverty incidence, initial inequality and cities dummies, and model change in the Gini coefficient as a use and downloadable from the World Bank’s micro-data website. INDODAPOER covers information related to the economy, infrastructure, education, health, poverty, labor and social protection, and financial sectors at the district- and province-level from 1976 to 2014. There are 219 variables from 546 districts covered in the data to date. For our analysis, we choose the period of 2002

• – 2012, which as discussed earlier, an era of democratization and decentralization in Indonesia. The included districts in our analysis is 295, consisting of 64 urban and 231 rural areas. We ended up with much less districts partly due to the grouping of expanded districts. As mentioned above, we basically develop two econometric models. The first model is intended to investigate factors affecting changes in poverty incidence, whereas the second is formed to examine factors affecting changes in inequality. Included as the independent variables in the first model are economic growth, initial poverty incidence, and some other covariates such as city dummies, literacy rate for population above 15 years old, different level of education participation (net enrolment ratio of school aged children in primary, junior secondary and senior secondary school in each districts), and infrastructure conditions (household access to electricity, safe water, and sanitation). Similar to the first model, the

As growth may potentially be endogenous, we use rainfall data for each district to act as the instrument for growth, and employ an Instrumental Variable (IV) regression to estimate the coefficients of the models. Rainfall is chosen as the instrument because it is correlated with growth and not correlated with the unobserved heterogeneity that may affect changes in poverty incidents or change in inequality (Bhattacharyya and Resosudarmo, 2013). There are different ways of how rainfall should enter the first stage regression. Miguel, et.al (2004, page 737), for example, use the growth of rainfall as the instruments for economic growth. Meanwhile, Sarsons (2015) uses the level-data of rainfall as instruments, arguing that the growth of rainfall may mistakenly identify rain shocks. In our case, we follow Bhattacharyya and Resosudarmo (2013) and Sarsons (2015) method in treating our rainfall variable.

  Specifically, the reduced form for growth in equation (1) and (2) can be written as

• ℎ = + ∑ , = 2, … , (3)

  1

  where denotes districts, and

• ℎ = + ∑ , = 2, … , (4)

  1

  where denotes sectors. the bottom of the tables). Those sectors may not be directly impacted by the changes in precipitation. The second approach statistically provides more consistent evidence that rainfall is partially correlated with growth, albeit showing a lower value of F-statistics than the first method. Finally, the third approach can be seen as the weakest instrument for growth, as it rarely provides a statistically significant first-stage regression statistics. Having said that, the magnitude of the coefficients of the variables in our structural equations do not alter too much using the three instruments. Table 3-11 report the result of the various regressions analysis.

  7 Table 3 Dependent Variable : Change in Poverty 2002

• – 2012 Independe Urban and Rural Urban Rural nt Variables

  Growth -0.111 -0.023 -0.116 0.374 -0.025 0.159 -0.129 -0.024 -0.142

  (0.052)** (0.017) (0.096) (2.269) (0.026) (0.485) (0.052)** (0.017) (0.111)

  Initial -0.035 -0.040 -0.035 -0.087 -0.040 -0.062 -0.034 -0.039 -0.034

  Poverty

  (0.004)*** (0.002)*** (0.006)*** (0.266) (0.008)*** (0.057) (0.005)*** (0.002)*** (0.006)***

Initial -0.207 -0.211 -0.207 0.380 -1.661 -0.718 0.411 0.251 0.430

Inequality (0.569) (0.391) (0.583) (12.345) (0.648)** (2.670) (0.725) (0.459) (0.785) Urban 0.148 0.176 0.146 (0.075)* (0.048)*** (0.083)*

Constant 0.531 0.187 0.551 -0.991 0.786 -0.035 0.453 0.068 0.498

(0.265)** (0.131) (0.416) (10.247) (0.219)*** (2.153) (0.285) (0.149) (0.471) Other NO NO NO NO NO NO NO NO NO Covariate s 295 295 295

  64

  64 64 231 231 231 N First Stage Regression: Instrumen (a) (b) (c) (a) (b) (c) (a) (b) (c) ts 9.84 *** 3.88***

  2.67 0.03 3.72*** 0.18 10.28*** 2.99 ***

  2.36 F-stat

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  7

(a) intentionally choose the rainfall data on one specific month in a year which yields the highest relevant statistical values; (b) monthly rainfall data (linear combination

of 12 monthly rainfalls; (c) annual average rainfall data. All of the standard errors in both first and second stage regressions are robust to heteroskedasticty.

  

Table 4

Dependent Variable : Change in Poverty 2002

• – 2012 Independent Urban and Rural Urban

  Rural Variables Growth -0.148 -0.030 -0.152 0.010 -0.014 0.108 -0.215 -0.025 -0.280 (0.084)* (0.018)* (0.172) (0.118) (0.020) (0.290) (0.118)* (0.018) (0.393) Initial Poverty -0.038 -0.043 -0.038 -0.038 -0.036 -0.047 -0.036 -0.044 -0.034 (0.006)*** (0.003)*** (0.008)*** (0.013)*** (0.010)*** (0.024)* (0.008)*** (0.003)*** (0.016)** Initial 0.217 -0.161 0.230 -1.913 -2.089 -1.200 1.238 0.393 1.528 Inequality

  (0.764) (0.395) (0.919) (1.067)* (0.661)*** (2.242) (1.272) (0.446) (2.252) Urban 0.233 0.233 0.233 (0.105)** (0.058)*** (0.107)** Constant 2.164 1.820 2.176 -3.369 -4.730 2.138 2.282 1.935 2.400

  (1.286)* (0.623)*** (1.496) (7.565) (4.506) (15.645) (2.028) (0.686)*** (2.889) Other YES YES YES YES YES YES YES YES YES Covariates N 295 295 295

  64

  64 64 231 231 231 First Stage Regression: Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c) 2.86 * 2.78***

  0.91 0.05 2.72*** 0.26 4.63** 2.30***

  0.65 F-stat

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 5

Dependent Variable : Change in Inequality 2002

• – 2012

  Growth

• 0.004 0.003 0.030

  (all sectors)

  (0.006) (0.002) (0.031)

  Initial

• 0.685 -0.679 -0.655 -0.687 -0.677 -0.644 -0.691 -0.662 -0.641

  Inequality

  (0.071)*** (0.073)*** (0.148)*** (0.070)*** (0.074)*** (0.129)*** (0.079)*** (0.078)*** (0.100)***

  Urban 0.016 0.022 0.047 0.013 0.025 0.071 0.009 0.040 0.062

  (0.008)* (0.007)*** (0.028)* (0.012) (0.010)*** (0.041)* (0.015) (0.012)*** (0.031)**

  Growth

• 0.003 0.002 0.017

  (Agriculture)

  (0.003) (0.002) (0.014)

  Growth

• 0.002 0.003 0.006

  (Mining)

  (0.002) (0.001)*** (0.004) 0.280 0.245 0.101 0.270 0.251 0.186 0.271 0.238 0.214

  Constant

  (0.035)*** (0.021)*** (0.161) (0.023)*** (0.020)*** (0.063)*** (0.025)*** (0.021)*** (0.037)***

  Other NO NO NO NO NO NO NO NO NO

  Covariates 295 295 295 295 295 295 279 279 279

  N First Stage Regression: Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c)

  9.84*** 3.88*** 2.67 16.37*** 2.77*** 3.65* 9.20*** 4.08*** 4.37** F-stat

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 6

Dependent Variable : Change in Inequality 2002

• – 2012
• 0.006 0.003 0.023

  Growth (manufacture)

  (0.011) (0.003) (0.024)

• 0.734 -0.656 -0.494 -0.708 -0.643 -0.599 -0.709 -0.596 -0.596

  Initial Inequality

  (0.133)*** (0.076)*** (0.324) (0.089)*** (0.079)*** (0.110)*** (0.083)*** (0.081)*** (0.085)*** 0.007 0.026 0.066 0.012 0.031 0.043 0.019 0.024 0.024

  Urban

  (0.023) (0.009)*** (0.044) (0.013) (0.008)*** (0.014)*** (0.007)*** (0.007)*** (0.007)***

• 0.003 0.005 0.010

  Growth (utility)

  (0.005) (0.002)** (0.006)*

• 0.002 0.005 0.005

  Growth (Construction)

  (0.002) (0.001)*** (0.002)** 0.303 0.238 0.103 0.288 0.218 0.172 0.277 0.205 0.205

  Constant

  (0.082)*** (0.024)*** (0.182) (0.048)*** (0.026)*** (0.052)*** (0.032)*** (0.024)*** (0.030)***

  Other Covariates NO NO NO NO NO NO NO NO NO

  295 295 295 295 295 295 295 295 295 N

  First Stage Regression:

  Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c) F-stat 1.05 2.15** 1.98 6.53** 2.03** 6.56** 13.63*** 3.47*** 12.81*** p

  

&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 7

Dependent Variable : Change in Inequality 2002

• – 2012 Growth -0.002 0.007 0.010 (Comm) (0.003) (0.002)*** (0.006)

  Initial -0.692 -0.653 -0.640 -0.659 -0.706 -0.797 -0.765 -0.637 -0.573 -0.676 -0.676 -0.738

  Inequality

  (0.073) (0.079)*** (0.091) (0.078) (0.080) (0.151) (0.207) (0.080) (0.136) (0.073) (0.074) (0.222)

• *** *** ***

  Urban 0.017 0.029 0.033 0.014 0.027 0.052 0.022 0.019 0.018 0.015 0.015 0.072

  (0.008) (0.008)*** (0.011) (0.011) (0.008) (0.030) (0.011) (0.008) (0.012) (0.010) (0.008) (0.066)
• Growth
• 0.004 0.004 0.020 (trades)

  (0.006) (0.002)* (0.020) Growth

• 0.012 0.006 0.016 (Financial)

  (0.026) (0.002) (0.011)

• Growth (Othr
• 0.003 -0.004 0.036 Services)

  (0.005) (0.002) (0.045)

Constant 0.277 0.209 0.186 0.278 0.239 0.162 0.360 0.206 0.129 0.278 0.279 0.067

(0.031) (0.024)*** (0.050) (0.034) (0.023) (0.114) (0.218) (0.024) (0.097) (0.032) (0.022) (0.245)

• Other NO NO NO NO NO NO NO NO NO Covariates

• *** *** *** *** *** *** ***

  N 295 295 295 295 295 295 295 295 295 295 295 295 First Stage Regression: Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c) (a) (b) (c)

  14.14* 3.31*** 6.98** 5.53** 3.30***

  2.43 0.49 6.95*** 2.84* 7.14*** 1.68***

  1.29 F-stat IV

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 8

Dependent Variable : Change in Inequality 2002

• – 2012
• 0.012 0.002 0.055

  Growth (all sectors)

  (0.010) (0.002) (0.085)

• 0.640 -0.680 -0.836 -0.663 -0.677 -0.746 -0.671 -0.679 -0.690

  Initial Inequality

  (0.084)*** (0.074)*** (0.356)** (0.076)*** (0.075)*** (0.322)** (0.088)*** (0.075)*** (0.120)*** 0.001 0.004 0.016 -0.011 0.007 0.097 -0.017 0.011 0.054

  Urban

  (0.010) (0.008) (0.036) (0.015) (0.009) (0.152) (0.019) (0.011) (0.042)

• 0.008 0.001 0.048

  Growth (Agriculture)

  (0.006) (0.002) (0.076)

• 0.003 0.001 0.008

  Growth (Mining)

  (0.002) (0.001) (0.006) 0.663 0.568 0.194 0.591 0.579 0.520 0.670 0.559 0.392

  Constant

  (0.127)*** (0.097)*** (0.773) (0.100)*** (0.097)*** (0.377) (0.116)*** (0.096)*** (0.231)*

  Other YES YES YES YES YES YES YES YES YES

  Covariates 295 295 295 295 295 295 279 279 279

  N First Stage Regression: Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c)

  2.86* 2.78*** 0.91 8.92*** 1.98** 2.03 3.85* 2.39*

  1.53 F-stat

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 9

Dependent Variable : Change in Inequality 2002

• – 2012
• 0.010 0.003 0.020

  Growth (manufacture)

  (0.009) (0.002) (0.015)

• 0.728 -0.656 -0.563 -0.725 -0.645 -0.602 -0.735 -0.602 -0.590

  Initial Inequality

  (0.112)*** (0.078)*** (0.195) (0.123)*** (0.080)*** (0.129)*** (0.098)*** (0.081)*** (0.087)***

• 0.006 0.007 0.025 -0.011 0.012 0.025 0.001 0.008 0.008

  Urban

  (0.013) (0.009) (0.024) (0.017) (0.009) (0.018) (0.010) (0.009) (0.010)

• 0.009 0.005 0.014

  Growth (utility)

  (0.009) (0.002)** (0.008)

• 0.005 0.005 0.006

  Growth (Construction)

  (0.004) (0.002)*** (0.003)** 0.662 0.554 0.413 0.571 0.586 0.595 0.604 0.552 0.548

  Constant

  (0.128)*** (0.099)*** (0.218) (0.119)*** (0.103)*** (0.138)*** (0.110)*** (0.100)*** (0.105)***

• Other Covariates YES YES YES YES YES YES YES YES YES

  295 295 295 295 295 295 295 295 295 N

  First Stage Regression:

  Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c) F-stat

  0.37 1.87**

  0.88 0.70 1.62* 3.71* 4.19** 1.98** 8.95***

  

p &lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  

Table 10

Dependent Variable : Change in Inequality 2002

• – 2012 Growth -0.007 0.007 0.013 (Comm) (0.006) (0.002)* (0.008)

• Initial -0.676 -0.672 -0.671 -0.572 -0.719 -0.924 -0.520 -0.652 -0.610 -0.616 -0.666 -1.299

  Inequality (0.078) (0.080)* (0.097) (0.121)* (0.081)* (0.255)** (0.549) (0.079)** (0.137)** (0.119)** (0.073)** (2.985)
• Urban -0.003 0.010 0.016 0.003 0.004 0.006 -0.024 -0.000 -0.008 -0.016 0.001 0.213

  (0.010) (0.009) (0.013) (0.010) (0.009) (0.019) (0.086) (0.009) (0.021) (0.025) (0.008) (0.977) Growth
• 0.010 0.005 0.025 (trades)

  (0.009) (0.002)* (0.026)

• Growth

  0.045 0.006 0.019 (Financial) (0.151) (0.002)** (0.016)

• Growth (Othr
• 0.014 -0.002 0.151 Services)

  (0.015) (0.002) (0.694)

Constant 0.696 0.472 0.370 0.690 0.533 0.316 -0.500 0.426 0.130 0.545 0.576 0.971

(0.128) (0.109)* (0.187) (0.143)* (0.104)* (0.326) (3.602) (0.114)** (0.422) (0.128)** (0.093)** (2.030)

• Other

  YES YES YES YES YES YES YES YES YES YES YES YES Covariates 295 295 295 295 295 295 295 295 295 295 295 295 N First Stage Regression: Instruments (a) (b) (c) (a) (b) (c) (a) (b) (c) (a) (b) (c) F-stat IV 6.53** 2.89*** 4.06** 2.97* 3.07***

  1.03 0.09 5.14***

  1.64 1.19 2.53***

  0.05

• * p&lt;0.1; ** p&lt;0.05; *** p&lt;0.01

  Discussion

  Table 3-10 report various regression results. Table 3 and 4 reports the estimation of the poverty model, while table 5-10 reports the estimation of the inequality model. For poverty model, the benchmark specification is that change in poverty is a function of growth, initial poverty incidence, initial inequality and areas (urban-rural). The first table report the benchmark model, the second table add other relevan correlates of poverty incidences to account for more control variable. What we will be discussed further here is the later tables which report the estimation results with all covariates. The benchmark is for comparison purpose only. As also mentioned previously, the table also report results with different variations of how we measure instrumental variables. As can be seen from Table 2, the effect of growth on poverty reduction is rather mixed. As we include initial poverty incidence (which is actually strong and statistically significant), we are actually testing whether districts with similar initial level of poverty incidence will experience faster poverty reduction episodes when their economic growth is higher. The result suggest yes if we look at nationally as economic growth is negative and significant (despite only marginally at 10% level). However, when we divide sample into urban (cities) and rural (non-cities) inequality level with is associated with faster absolute change in inequality and vice versa. In terms of the impact of growth, the regression results suggest that at aggregate level economic growth does not have any effect on inequality. That is said it does not increase or decrease inequality. The growth variable is not significant at standard level of significance.

  At sectoral level, the results provide no evidence that economic growth can reduce inequality. In all specification, we found that the coefficient of the economic growth variable in any of all the sector we consider is not found to have negative sign, not to mention statistically significant.

  This result is actually consistent with the discussion on GIC earlier which suggest that in the period of 2002-2012, the GIC curve has always been positively sloped.

  Moreover, we found an indication that economic growth of certain sector exhibit negative impact on equality (or inequality-increasing). The economic growth variable of those sectors are positive and statistically significant. Those economic sectors are construction, communication and transportation, trades, and financial sectors.

  To summarize, we find that during the period of 2002-2012, growth has been both inclusive and immiserising in Indonesia. We find that, in this particular period, economic growth has reduced poverty in rural areas. However, no such association between growth and

  We do however find no overall association between aggregate economic growth and changing inequality which suggests that growth-inequality impacts in Indonesia are geographically (urban/rural) specific. Indeed, we find that it is sectoral growth that matters. Growth in such sectors as constructions, tranport and communication and financial sector is associated with rising inequality, while agriculture and manufacturing are not associated with rising inequality.

  Figure 4. Relationship between initial poverty (2002) and change in poverty (2002-2012)

  1

• 1

Figure 5. Relationship between initial inequality (2002) and change in inequality (2002-2012)

thesis was based on time-series data for three countries (US, UK and two states in Germany) plus point estimates for inequality in India, Puerto Rico and Ceylon. He argued that inequality would rise in an ‘upswing’ and then fall later in the ‘downswing’ of what became known as the inverted-U or Kuznets Curve. Kuznets (1955, pp. 7-8) argued that inequality would rise as the inter-sectoral shift away from agriculture lead to income differences between rural/agriculture and other sectors. He further argued that the only way to offset this was for the share of lower non-agriculture/urban income groups to rise. He contended that, in democracies, urban migrants would become politically organized leading to redistribution.

  Kuznets argued that the early benefits of growth go to those who have capital and education but that, as more people move out of the traditional sector, real wages rise in the modern sector and inequality falls. He further argued that the poorest lost out more rapidly than other groups as income expanding opportunities arose outside agriculture. Inequality, Kuznets noted, is composed of inequality between and within segments (urban and rural) and inequality tends to

  8

  be lower in the rural segment (relative to the urban segment). Thus as the size of the more unequal urban segment increases this will further add upward pressure on inequality. And, given that, during economic growth, productivity in urban areas is likely to increase faster than about the Kuznets hypothesis. Such scholars have focused on open economies and agrarian liberalisation, the role of technology, and aspects of national political economy and land distribution. It is worth reviewing these new theories.

  Galbraith (2010) argues that changes in national inequality since 1970 are driven by global forces. He shows that inequality patterns are similar across countries, suggesting that the key drivers of changes in national inequality are world interest rates and commodity prices (and between-sector terms of trade). He argues that a commodity boom reduces inequality in countries with a dominant agricultural sector as it raises the relative income of framers. OPEC oil rises are more complicated as they reduce inequality in oil exporters but drive inequality up in net oil-importing countries. Finally, high rates of interest are bad for debtor countries and this increases inequality. Galbraith presents an ‘augmented Kuznets curve’ or S-curve where

  11 by the curve rises, then falls, and then rises again.

  In a somewhat similar vein, at least in the sense of a focus on open economies, Lindert and Williamson (2001) argue that it is the shift towards market orientation (domestic to export) of agriculture and not the shift from agriculture to manufacturing and services that causes inequality to rise. Lindert and Williamson predict an initial rise in inequality. However, while

  In contrast, Roine and Waldenström (2014) suggest a new Kuznets curve based on technological developments starting not a sectoral shift of agriculture to industry, but a shift from traditional industry to technologically intensive industry. If a given technology makes skilled workers more productive and there is an increase in the relative demand for those workers, the rewards accrue to a small proportion of the population who are skilled workers. Based on Tinbergen’s (1974, 1975) hypothesis that the returns to skills are a competition between education and technology, the supply of skilled workers then determines whether their wages rise or not. Roine and Waldenström (2014) argue that the drivers of the Kuznets

  

12

downturn were political and exogenous shocks.

  Focusing more on domestic economic structure as deterministic, Oyvat (2016) argues that it is agrarian structures

• – land inequality – that frame the relationship. Consistent with Kuznets (and Lewis potentially), he argues that migration is driven by higher urban incomes and this suppresses wages in the urban sector. If land inequality is higher, more people will

  13

  migrate for lower wages and this will further depress urban wages. Empirically, Oyvat argues that the level of land inequality has a significant impact on urbanisation, intra-urban inequality and overall inequality. The results suggest that land reforms or subsidies to rural small holders

  Similarly, though with a different focus, Acemoglu and Robinson (2002) discuss the political economy of the Kuznets curve in two models of ‘late capitalism’. The first is a high- inequality, low- output model which they call ‘autocratic disaster’. In this model, inequality does not rise and political mobilisation is too limited to address existing inequality. A second model is the ‘East Asian miracle’ of low inequality and high output where inequality does not rise which ensures political instability and avoids democracy being forced on elites. They argue that when the process of industrialisation does increase inequality, this leads to the political mobilisation of the masses that are concentrated in urban areas and factories. Political elites thus undertake reform to ensure their continued position at the top. The extension of the franchise is the best option for elites as it acts as a commitment to future redistribution and thus

  15 prevents unrest.

In sum, there have been various iterations of a new Kuznets relationship within a Lewis model dual economy with different ways of segmenting the ‘sectors’ not only by production

  (e.g., agriculture or non-agriculture) but as modern/capitalist or traditional/pre-capitalist (as per Lewis), as urban and rural (as per Kuznets originally and Oyvat, 2016), by technology (e.g., Roine and Waldenström, 2014), by the extent of global integration or internal/external market

  The period in question for Indonesia is one where manufacturing shares are consistent with what Palma (2005) originally and later, Rodrik (2016) labelled as ‘premature deindustrialisation’, in that developing countries have reached ‘peak manufacturing’ in

  16

  employment and value added shares at a much earlier point than the advanced nations. In contrast, service shares of GDP and employment are on an upward trend in general.

  Premature deindustrialisation has two components. The inverted-U of manufacturing shares versus GDP per capita is shifting down and leftwards over time, making it harder for late developers to attain the benefits of industrialization that earlier developers saw. The first component is that ‘peak manufacturing’, in employment or GDP shares (or export shares) has been reached and the inverted-U curve is now on the plateau or even downswing of the curve. The second component is that the inverted-U curve moves leftward over time. This means that the point at which the inverted-U turns is, on average, lower in per capita income terms now than in the 1990s which was already lower than in the 1980s (noted originally by Palma, 2005).

  Indonesia's economic growth since the Asian financial crisis has been accompanied ‘premature deindustrialisation’. Figure 6 shows the development of the share of the industrialisation’ as he defines it and triggered by trade liberalisation and China’s entry into global manufacturing on a larger scale in the late 1990s and early 2000s. Aswicahyono and Hill (2015, p. 11-

  13) however argue that relative to per capita income, Indonesia’s share of manufacturing value added in GDP is larger than predicted by standard cross-country regressions over the 1960-2012 period but Indonesia is below average with respect to employment shares. The authors argue that this simply shows higher relative capital intensity of manufacturing in Indonesia compared to other countries and that there is no desirable share for manufacturing. They go on to argue that the industrial slow down since the Asian Financial Crisis is due to historically high terms of trade since the early 2000s leading to a larger natural resource sector and real exchange rate appreciation hence squeezing non-commodity tradable sectors; a policy regime that has hindered competitiveness of manufacturing exports (such as rises in minimum wages and poor infrastructure) and the rise of China in a wide range of manufactures has depressed returns to labour intensive manufacturing. Figure 6. The trend of manufacturing share in value added and employment Source: World Bank (2016) and ADB Key Indicators

  1961 1971 1980 1990

  2000 2010 2014 1960

  1970 1980 1990 2000

  2010 2014 1975 1980

  1985 1990

1995

2000

  2005 2010 2014

  5

  10

  15

  20

  25

  30

  35

  40 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 MANVA/GDP (%)-INDONESIA

  MANEMP/EMP (%)-INDONESIA MANEMP/EMP (%)-THAILAND MANVA/GDP (%)-THAILAND MANEMP/EMP (%) What are the causes of the ‘premature deindustrialisation’ phenomenon? There are differing views. Rodrik (2015) links the phenomenon to trade liberalization over time and the impact of entry of China into manufacturing. Felipe and Mehta (2016) argue that premature deindustrialisation is caused by the fact that large national increases in labour productivity were counteracted by a shift of manufacturing jobs to lower productivity

  17

  economies. Further, Palma (2005) argues that there several other potential hypotheses (which are not mutually exclusive) that could explain the phenomenon observed: (i) it is due to a statistical illusion caused by contracting out of manufacturing jobs to services (e.g., cleaning or catering); (ii) it is due to a fall in the income elasticity; (iii) it is due higher productivity growth in manufacturing; or (iv), it is due to outsourcing globally whereby manufacturing employment has fallen in OECD countries; (v) it is due to the change in policy regimes in OECD countries away from Keynesianism; or (vi), it is due to

  18 technological progress.

  A further or accompanying issue for Indonesia has been what some have called ₁₇ ‘jobless growth’. More specifically, the employment absorption capacity of Indonesia’s manufacturing sector changed significantly after the Asian financial crisis. Before the crisis, the manufacturing sector was the primary driver of Indonesia’s economic growth and job creation. The manufacturing sector value added grew 11.2% per annum during 1990-1996 (while the average economic growth was 7.9%) and its employment grew 6% per annum (while the average national employment growth was only 2.3%). Aswicahyono et al. (2010) estimated that the implied output elasticity (percentage change in employment with respect to percentage change in output growth) of the manufacturing sector was 0.53 in 1990-1996. However, the authors highlighted that the elasticity declined to 0.18 in 2000- 2008 and analysed this as a period of (virtually)

  ‘jobless growth.’ This

  ‘jobless growth’ phenomenon was most visible during the period of 1998- 2005. Figure 7 plots value added and employment of the manufacturing sector. Between 1990 and 1996,