## wpiea2020170-print-pdf

## Source details

**Canonical URL:** [wpiea2020170-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020170-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020170-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020170-print-pdf.pdf.json)

---

### Introduction and contributions
- Research question: What is the socio-economic impact of Special Economic Zones (SEZs) on local labor markets in a low-income country setting (Cambodia)?
- Novel contributions:
  - First causal evidence on socio-economic impact of a place-based policy on local communities in a low-income country.
  - First study to evaluate SEZs’ effects on composition of local employment including female employment and effects on income inequality.
- Methodology overview:
  - Constructed a novel geo-tagged SEZ database matched to Cambodia’s household survey data at the district level between 2007 and 2017.
  - Event-study analysis examining changes in employment, wages, income levels, income inequality, price levels, and high school drop-out rates following SEZ entry.
  - Two identification strategies to address non-random SEZ location selection:
    - Inverse propensity scores based on districts’ initial characteristics.
    - Alternative control group of non-SEZ districts closely mimicking treatment districts (including future participants and districts bordering SEZ locations).

### Stylized facts on Cambodia’s SEZs and data
- Legal and program features:
  - SEZ legal framework: “Sub-Decree No.147 on the Organization and Functioning of the CDC” (2005).
  - SEZs defined as geographically bounded areas larger than 50 hectares.
  - SEZs must provide basic infrastructure and operate management/zone administration offices as one-stop service providers.
  - SEZs within 20km from the border benefit from expedited customs clearing within SEZs.
- SEZ counts and characteristics (as of 2019):
  - 23 operating SEZs and 7 authorized to begin operation.
  - At least 13 additional locations considered as potential zones but not officially authorized as of 2019.
  - Zone administrations set up in 18 out of 23 operational SEZs.
  - SEZs accounted for approximately 18 percent of total goods exports in 2018.
  - Between 2005 and 2019, SEZs received over US$2 billion in initial investment.
  - By 2019, SEZs employed more than 131,000 Cambodians, accounting for over 22 percent of formal employment in Cambodia; 64 percent of these workers are Cambodian women.
- Data sources and sample:
  - Household data: Cambodia Socio-Economic Survey (CSES) annual since 2007; district-level panel covering 180 of 202 districts and municipalities between 2007 and 2017.
  - SEZ data: Council for Development of Cambodia (CDC) and Open Development Cambodia.
  - SEZ entry defined as start of operation of first firm within a given zone.
  - SEZ locations concentrated in capital, border, and coastal regions (e.g., Sihanoukville and Svay Rieng host 15 of 23 operating SEZs).

### Estimation strategy and identification
- Baseline event-study specification:
  - y_dt = α + β D_dt + λ T_dt + δ_d + γ_pt + ε_dt.
  - D_dt = 1 if SEZ present in district d at time t; T_dt = (current year) − (year of establishment of the first SEZ in the district); δ_d = district fixed effects; γ_pt = province-by-year fixed effects.
  - Standard errors clustered at the province level.
- Dependent variables include:
  - Employment: paid employment share; manufacturing employment share; female employment rate.
  - Income and inequality: average real wages (log), size-adjusted household income (log), Gini coefficient.
  - Prices: self-reported residential land resale values (log).
  - Education: adult high-school drop-out rate; average years of education (log).
- Bias adjustments:
  - Propensity score weighting (inverse predicted probability from logistic regression).
  - Alternative control group of adjacent districts and “future SEZ districts.”
- Key sample counts and balance facts:
  - SEZ sample: seven districts.
  - Full non-SEZ sample: 173 districts.
  - Alternative control group (adjacent & future SEZ districts): 36 districts.
  - Selected pre-treatment mean differences (SEZ vs Non‑SEZ): Vietnam Border 0.14 vs 0.12 (Diff. 0.02); Capital Region 0.43 vs 0.08 (Diff. 0.35); Female employment 0.77 vs 0.85 (Diff. -0.08); Log(Wage) 16.15 vs 16.42 (Diff. -0.27); Log(Household income) 15.48 vs 15.05 (Diff. 0.43); Log(Years of education) 1.76 vs 1.45 (Diff. 0.31).
  - Propensity score weighting renders initial mean differences statistically indistinguishable from zero.

### Main results — Employment composition and labor outcomes
- Paid employment share:
  - SEZ coefficients (Panel A/B/C): 0.004 (s.e. 0.011); -0.011 (s.e. 0.011); -0.003 (s.e. 0.014).
  - Interpretation: little effect on district paid employment share across specifications.
- Manufacturing employment share:
  - SEZ coefficients (Panel A/B/C): 0.009 (s.e. 0.005); 0.009 (s.e. 0.009); -0.005 (s.e. 0.005).
  - Interpretation: negligible effect on manufacturing employment share.
- Female employment rate:
  - SEZ coefficients (Panel A/B/C): 0.060 *** (s.e. 0.009); 0.053 *** (s.e. 0.014); 0.050 *** (s.e. 0.013).
  - Interpretation: SEZ entry associated with about five percent increase in female employment rate.
  - Mechanism: SEZ firms concentrated in garments and light manufacturing; female workers account for more than 60 percent of employment in SEZs.

### Main results — Wages, incomes, inequality, and land prices
- Real wages (log):
  - SEZ coefficients (Panel A/B/C): -0.035 (s.e. 0.223); -0.018 (s.e. 0.144); -0.137 (s.e. 0.177).
  - Interpretation: SEZ presence does not raise district average real wage levels.
- Household income (log):
  - SEZ coefficients (Panel A/B/C): 0.031 (s.e. 0.072); 0.011 (s.e. 0.076); 0.042 (s.e. 0.073).
  - Interpretation: negligible effects on average household incomes at the district level.
- Gini coefficient:
  - SEZ coefficients (Panel A/B/C): -0.049 *** (s.e. 0.009); -0.046 *** (s.e. 0.012); -0.047 *** (s.e. 0.007).
  - Post-SEZ Trend coefficients (Panel A and C): 0.008 *** (s.e. 0.003); 0.009 *** (s.e. 0.002).
  - Interpretation: SEZ entry associated with a decline in district-level Gini; combined effects correspond to about 1.5 percentage point decline in income inequality (text); Panel B implies about 4.6 percent decline (text).
- Residential land values (log):
  - SEZ coefficients (Panel A/B/C): 0.034 (s.e. 0.101); 0.105 ** (s.e. 0.042); -0.043 (s.e. 0.142).
  - Interpretation: Panel B suggests possible increase in average land prices with SEZ entry; overall effects mixed across specifications.

### Main results — Education
- High-school drop-out rate (share, older than 18, not in school, without diploma):
  - SEZ coefficients (Panel A/B/C): 0.008 (s.e. 0.036); -0.011 (s.e. 0.035); 0.004 (s.e. 0.042).
  - Interpretation: SEZ entry has negligible direct impact on high-school drop-out rates in the baseline; neighboring-district analyses indicate increased drop-outs in neighbors.

### Intensive margin — Effect of multiple SEZs
- Treated districts hosted, on average, 2.4 SEZs at the end of the sample period.
- SEZ intensity indicator M_d = 1 if district had more than one SEZ at end of sample.
- Key findings:
  - Core employment and income results robust to controlling for number of SEZs; no additional impact of multiple SEZs on paid employment, manufacturing employment, or female employment.
  - Wages: interaction estimates marginally significant in some panels (Panel B SEZ × Multiple SEZs = 0.327 * (s.e. 0.168)); interpret cautiously.
  - Land values:
    - SEZ × Multiple SEZs (Panel A/B): -0.341 *** (s.e. 0.096); -0.562 *** (s.e. 0.135).
    - Interpretation: Subsequent SEZs in the same district associated with large declines in residential land values; joint effect cited as 20 to 44 percent decline in residential land values in districts with multiple SEZs (text).
    - Suggested explanation: governance shortcomings (e.g., land grabbing) may negatively affect residential or agricultural land prices when multiple SEZs locate in a district.

### Spillover effects on neighboring districts
- Specification includes SEZ presence in neighboring districts DNt and post-entry trend TNt.
- Findings (selected coefficients):
  - Direct (treated district) effects replicate baseline: Female empl. 0.058*** (0.014); Gini coef. -0.044*** (0.008); log(Land value) 0.149** (0.054).
  - SEZ in neighboring district:
    - Female empl.: 0.018* (0.010) — small positive spillover on female employment in neighboring districts.
    - School Drop-out rate: -0.045*** (0.014) for SEZ in neighboring district; Post-SEZ Trend in Neighboring District for School Drop-out rate = 0.015*** (0.002).
  - Interpretation:
    - Spillovers on paid and manufacturing employment negligible.
    - Small positive spillover on female employment—consistent with commuting into SEZ jobs from neighboring districts.
    - No evidence of spillovers on wages, household incomes, or income inequality in neighboring districts.
    - Land valuations in treated districts rose by about 15 percent after SEZ entry (statistically significant).
    - Neighboring districts show slight increases in high-school drop-out rates over time (joint coefficients imply about 1.5 percent increase given average SEZ age ~4 years), suggesting SEZ employment is an attractive outside option for youth and migration effects may influence measured schooling outcomes.

### Robustness checks
- Methods:
  - Inverse propensity score-weighted regression with adjacent & future SEZ controls (Panel A, Observations: 354).
  - Specification using lags of the dependent variables (Panel B, Observations: 1,084).
- Key robustness results (selected):
  - Propensity-weighted (Panel A):
    - Paid empl.: -0.024*** (0.004).
    - Female empl.: 0.050** (0.017).
    - Gini coef.: -0.039** (0.014).
    - Land value: 0.082* (0.041).
    - Interpretation: Female employment increase and Gini reduction persist; paid employment may fall, consistent with growth in informal activities; land values rise in treated districts, suggesting gains to landowners.
  - Lagged specification (Panel B):
    - Female empl.: 0.033* (0.019).
    - Gini coef.: -0.036** (0.013).
    - Post-SEZ Trend for manufacturing employment = -0.013** (0.006) indicating a small negative post-treatment trend in manufacturing share.
  - Overall inference: Robust support for increases in female employment and reductions in income inequality; some sensitivity in paid employment and manufacturing trends pointing to sectoral shifts and possible informal activity expansion.

### Policy implications and recommendations
- Core recommendations:
  - Invest in infrastructure (energy, transportation, water and sanitation) to:
    - Encourage SEZ entry into disadvantaged areas.
    - Promote formation of input-output linkages with domestic firms.
    - Attract foreign investment in more technologically sophisticated industries and support export diversification.
  - Invest in human capital and encourage worker training by firms to:
    - Address limited evidence that existing SEZs create higher-paid jobs or attract higher-skilled labor.
    - Mitigate incentives for youth to leave school for low-skilled SEZ jobs; foster knowledge transfer and quality upgrading.
  - Fine-tune tax incentives:
    - Move away from tax holidays toward incentives linked directly to the size of the investment to attract long-term, higher-quality investment and reduce rent-seeking.
    - Recognize that tax incentives are less effective in poor investment climates; broader reforms (infrastructure, rule of law, transparency, governance) yield greater gains.
  - Support domestic entrepreneurial activity and productivity:
    - Improve governance and transparency of tax and business registration systems.
    - Promote productivity of domestic firms via human capital and infrastructure to strengthen backward linkages and generate positive productivity spillovers.
- Rationale linking evidence to policy:
  - SEZs increased female employment and reduced income inequality in host districts but did not raise real wages, and land values rose—implying distributional gains favor landowners.
  - SEZ firms purchase only 12 percent of inputs domestically compared to 62 percent by non‑SEZ firms (World Bank and ADB, 2014), consistent with assembly/re-export specialization and limited backward linkages.
  - Policy focus: upgrade sectoral composition of SEZ investors, improve domestic firm productivity, and widen geographic reach of SEZ benefits.

### Conclusion — key takeaways
- Identification: variation in SEZ location and timing with propensity score weights, alternative control groups, and lagged-variable robustness checks.
- Between 2007 and 2017, Cambodia’s SEZ program:
  - Increased female employment rates and reduced income inequality in SEZ host districts.
  - Did not increase real wages, implying distributional benefits favor landowners (land values rose).
  - Was concentrated in areas with better infrastructure and trade access, potentially widening urban-rural divides.
  - Associated with rising high school drop‑out rates in districts with higher SEZ concentration and in neighboring control districts.
- Policy priorities to enhance socio-economic spillovers:
  - Invest in energy, transportation, water and sanitation infrastructure.
  - Invest in human capital and encourage firm-provided worker training.
  - Shift from tax holidays to investment-size-linked tax incentives to attract longer-term, higher-value investment and limit rent-seeking.

*Source: IMF working paper — Sections 3, 5, 6, 7, 8, and 9 (Tables and accompanying text) from wpiea2020170-print-pdf.*

### References22

### References22

### Introduction
- Research question: What is the socio-economic impact of Special Economic Zones (SEZs) on local labor markets in a low-income country setting (Cambodia)?
- Novel contributions:
  - First causal evidence on socio-economic impact of a place-based policy on local communities in a low-income country.
  - First study to evaluate SEZs’ effects on composition of local employment including female employment and effects on income inequality.
- Contextual differences:
  - SEZ programs in low-income countries (including Cambodia) often target locations with more developed infrastructure and access to transportation networks rather than economically-distressed areas.
- Methodology overview:
  - Constructed a novel geo-tagged SEZ database matched to Cambodia’s household survey data at the district level between 2007 and 2017.
  - Event-study analysis examining changes in employment, wages, income levels, income inequality, price levels, and high school drop-out rates following SEZ entry.
  - Two identification strategies to address non-random SEZ location selection:
    - Inverse propensity scores based on districts’ initial characteristics.
    - Alternative control group of non-SEZ districts closely mimicking treatment districts (including future participants and districts bordering SEZ locations).

### Key Findings (Main Results)
- Employment and composition:
  - Entry of SEZs boosts female employment.
  - Limited effect on aggregate formal employment share.
  - Firms operating in SEZs are mostly foreign-owned but hire predominantly local labor; female workers account for more than 60 percent of employment in SEZs.
  - Majority of jobs are low-skilled and concentrated in garments and other light manufacturing industries.
- Income inequality and wages:
  - Entry of SEZs contributes to declining income inequality within a district.
  - Local wage levels remain unchanged following SEZ entry.
- Land and prices:
  - Land values tend to rise in treated districts after SEZ entry.
  - Result aligns with spatial equilibrium model predictions where property price increases can offset wage gains.
- Agglomeration and program intensity:
  - Limited agglomeration effects; SEZ program intensity has little additional effect on labor market outcomes.
- Spillovers to neighboring districts:
  - Small spillovers on female employment, potentially due to commuting across district borders.
  - Increase in high school drop-out rates in neighboring districts, suggesting SEZ employment is an attractive outside option for youth.

### Theoretical and Empirical Context
- Theoretical mechanisms highlighted:
  - Agglomeration effects and knowledge spillovers (higher productivity in denser areas; attracting highly educated people).
  - Countervailing force of labor mobility: with perfect mobility, landowners capture benefits and local labor gains can be nullified (Glaeser and Gottlieb (2008)).
  - Place-based policies can generate local welfare effects that differ from aggregate effects; targeted gains can come at expense of non-participating areas (Neumark and Simpson (2015)).
- Empirical literature summary:
  - Evidence on enterprise zone employment effects is mixed (Neumark and Simpson (2015)): some studies find no effects (e.g., Neumark and Kolko, 2010), others find substantial gains (e.g., Busso et al., 2013).
  - Studies on China (Wang, 2013; Lu et al., 2019) and India (Alkon, 2018) show heterogeneous results.
  - Previous firm-level studies in Cambodia (World Bank and Asian Development Bank, 2014; Warr and Menon, 2016) find SEZs attracted significant FDI and boosted exports but generated limited knowledge and technology spillovers.

### Stylized Facts: Cambodia’s SEZs and District Characteristics
- SEZ program overview:
  - Legal framework established in 2005 by “Sub-Decree No.147 on the Organization and Functioning of the CDC.”
  - SEZs defined as geographically bounded areas larger than 50 hectares designated for industrial production and support activities.
  - Construction and operation left to the private sector; SEZs must provide basic infrastructure (energy supply, sewage and waste water treatment networks, storage facilities, solid waste management) and operate management/zone administration offices as one-stop service providers.
  - SEZs located within 20km from the border benefit from expedited customs clearing within SEZs.
- SEZ counts and characteristics (as of 2019):
  - 23 operating SEZs and 7 authorized to begin operation (at various stages of construction and development).
  - At least 13 additional locations considered as potential zones but not officially authorized as of 2019.
  - Zone administrations set up in 18 out of 23 operational SEZs.
  - SEZs mainly specialize in manufacturing of garments, footwear, travel goods, electronics, vehicle parts, plastics, and other consumer products.
  - SEZs accounted for approximately 18 percent of total goods exports in 2018.
  - Number of registered firms per SEZ ranges from one to over a hundred.
  - Between 2005 and 2019, SEZs received over US$2 billion in initial investment.
  - By 2019, SEZs employed more than 131,000 Cambodians, accounting for over 22 percent of formal employment in Cambodia; 64 percent of these workers are Cambodian women.
- Incentive framework:
  - Cambodia’s Law on Investment provides a two-tier incentive structure:
    - SEZ developers: initial exemptions from profit tax, import duties and other taxes on machinery and construction equipment, and other temporary and permanent concessions.
    - Firms within SEZs: eligible to generous tax incentives (exemptions, in whole or in part, from VAT, import and export duties, excise taxes, corporate and dividend taxes), access to infrastructure, and streamlined administrative/regulatory treatment.
    - Firms must pay a set fee to an SEZ developer and provide relevant training to Cambodian workers they employ.

### Data and Geographic Distribution
- Data sources:
  - Household data from Cambodia Socio-Economic Survey (CSES) conducted annually since 2007; district-level panel constructed from repeated cross-sections covering 180 of 202 districts and municipalities between 2007 and 2017.
  - SEZ data (location, number of firms, initial investment amount, age, and employment in 2019) from the Council for Development of Cambodia (CDC); information on planned SEZs from CDC and Open Development Cambodia.
- Definition of SEZ entry:
  - Entry defined by the start of operation of the first firm within a given zone.
- Location patterns:
  - SEZ locations are non-random and concentrated in capital, border, and coastal regions.
  - Two provinces—Sihanoukville and Svay Rieng—host 15 of the 23 currently operating SEZs.
  - Banteay Meanchey hosts three zones; provinces in central and northeastern Cambodia (Kampong Thom, Mondul Kiri, Stung Treng, Ratanak Kiri, Preah Vihear) currently have no operational SEZs, though some have been considered.
- Province-level changes 2007–2017 (CSES-based):
  - Paid employment:
    - Provinces with highest SEZ concentration (Svay Rieng, Preah Sihanouk, Kandal, Phnom Penh, Bantey Meanchey) experienced at least 20 percent growth in paid employment.
    - Some surrounding provinces without SEZs (Kampong Chhang, Kampong Speu, Kampong Cham, Oddar Meanchey) also saw significant growth in paid employment.
    - Slowest growth in paid employment documented in northeastern provinces.
  - Manufacturing employment share:
    - Manufacturing employment share rose by more than 10 percent in provinces with highest SEZ concentration; dropped or remained flat in areas with no SEZs.
    - Noted that manufacturing sector growth cannot be attributed to SEZs alone, as many factories operate outside SEZs.
  - Wages:
    - Real wages, on average, have risen by more than 40 percent across the country since 2007.
    - Variation across SEZ-present areas: about 10 percent wage growth in Phnom Penh area to 62 percent growth in Kandal province.
    - Real wages declined in Ratanak Kiri and Mondul Kiri.
  - Educational attainment:
    - National average increase in average years of education completed is 25 percent; Phnom Penh rose by only 10 percent.
    - Relationship between educational attainment changes and SEZ entry appears weak at province level.

### Policy-relevant Observations (from findings)
- Bottlenecks limiting SEZ spillovers:
  - Lack of basic infrastructure (stable electricity, transport links, water/sewage).
  - Labor skills shortages and low productivity of domestic suppliers.
  - Governance and transparency of tax and business regulations affecting domestic firm productivity and attractiveness of high value-added FDI.
- Implication:
  - Investing in infrastructure and education, and improving governance and transparency of tax and regulatory institutions are necessary conditions for generating greater socio-economic benefits of the SEZ program.

*wpiea2020170-print-pdf - References22*

### 3.  Estimation Strategy

### 3. Estimation Strategy

### Baseline specification and identification
- Estimation approach: event-study comparing districts with an SEZ (treatment) to districts without an SEZ (control) using the specification
  y_dt = α + β D_dt + λ T_dt + δ_d + γ_pt + ε_dt.
- Definitions:
  - D_dt = 1 if an SEZ is present in district d at time t, 0 otherwise.
  - T_dt = (current year) − (year of establishment of the first SEZ in the district); T_dt = 0 for all non‑SEZ districts.
  - δ_d = district fixed effects; γ_pt = province‑by‑year fixed effects.
- Standard errors clustered at the province level.
- Dependent variables (y_dt) include:
  - Employment: paid employment share; manufacturing employment share; female employment rate.
  - Income and inequality: average real wages (log), size‑adjusted household income (log), Gini coefficient.
  - Prices: self‑reported residential land resale values (log).
  - Education: adult high‑school drop‑out rate; average years of education (log).

### Bias concerns and adjustments
- Non‑random SEZ placement may bias baseline estimates because SEZ locations differ systematically (e.g., infrastructure, skills, cost of living).
- Three estimation strategies / robustness checks adopted:
  - Baseline event‑study (district and province‑year fixed effects).
  - Propensity score weighting to balance pre‑treatment means (weights = inverse predicted probability from a logistic regression of SEZ status on location indicators and initial socio‑economic characteristics).
  - Alternative control group: adjacent districts (bordering treated districts) and “future SEZ districts” (authorized for SEZ after 2017 or considered for SEZ status).

### Key sample and balance facts (from Table 1)
- SEZ sample: seven districts.
- Full non‑SEZ sample: 173 districts.
- Alternative control group (adjacent & future SEZ districts): 36 districts.
- Pre‑treatment mean differences (selected):
  - Vietnam Border: SEZ 0.14; Non‑SEZ 0.12; Diff. 0.02.
  - Capital Region: SEZ 0.43; Non‑SEZ 0.08; Diff. 0.35.
  - Female employment: SEZ 0.77; Non‑SEZ 0.85; Diff. -0.08 (statistically significant).
  - Log(Wage): SEZ 16.15; Non‑SEZ 16.42; Diff. -0.27 (statistically significant).
  - Log(Household income): SEZ 15.48; Non‑SEZ 15.05; Diff. 0.43 (statistically significant).
  - Log(Years of education): SEZ 1.76; Non‑SEZ 1.45; Diff. 0.31 (statistically significant).
- Propensity score weighting (Column 3 of Table 1) renders initial mean differences statistically indistinguishable from zero.

---

### Main Results — Employment (Table 2, Columns 1–3)
- Sample sizes:
  - Panels A and B: Observations = 1,555.
  - Panel C: Observations = 354.
- SEZ presence (SEZ coefficient) on employment indicators:
  - Paid employment (Column 1):
    - Panel A SEZ = 0.004 (s.e. 0.011).
    - Panel B SEZ = -0.011 (s.e. 0.011).
    - Panel C SEZ = -0.003 (s.e. 0.014).
    - Interpretation: little effect on paid employment share across specifications.
  - Manufacturing employment (Column 2):
    - Panel A SEZ = 0.009 (s.e. 0.005).
    - Panel B SEZ = 0.009 (s.e. 0.009).
    - Panel C SEZ = -0.005 (s.e. 0.005).
    - Interpretation: negligible effect on manufacturing employment share.
  - Female employment rate (Column 3):
    - Panel A SEZ = 0.060 *** (s.e. 0.009).
    - Panel B SEZ = 0.053 *** (s.e. 0.014).
    - Panel C SEZ = 0.050 *** (s.e. 0.013).
    - Average reported effect: SEZ entry associated with about five percent increase in female employment rate.
    - Note: In Panel B the Post‑SEZ Trend coefficient is negative; with an average post‑treatment period of about four years this attenuates to about 2 percentage points (see footnote).

- Mechanism: SEZ sectoral profile (garments, light manufacturing) concentrates demand for female labor; >60 percent of jobs created by SEZ firms are taken by women (textual evidence).

---

### Main Results — Incomes, Inequality, and Land Prices (Table 2, Columns 4–7)
- Wages (Column 4):
  - SEZ coefficients:
    - Panel A SEZ = -0.035 (s.e. 0.223).
    - Panel B SEZ = -0.018 (s.e. 0.144).
    - Panel C SEZ = -0.137 (s.e. 0.177).
  - Interpretation: SEZ presence does not raise district average real wage levels or accelerate wage growth.
  - Explanation provided: firms in garments/light manufacturing tend to set wages at national minimum; SEZ workers are a small share of household income; non‑monetary benefits may induce substitution from informal to formal work without raising aggregate wages.

- Household income (Column 5):
  - SEZ coefficients:
    - Panel A SEZ = 0.031 (s.e. 0.072).
    - Panel B SEZ = 0.011 (s.e. 0.076).
    - Panel C SEZ = 0.042 (s.e. 0.073).
  - Interpretation: negligible effects on average household incomes at the district level.

- Gini coefficient (Column 6):
  - SEZ coefficients:
    - Panel A SEZ = -0.049 *** (s.e. 0.009).
    - Panel B SEZ = -0.046 *** (s.e. 0.012).
    - Panel C SEZ = -0.047 *** (s.e. 0.007).
  - Post‑SEZ Trend coefficients (Panel A and C) are positive on Gini:
    - Panel A Post‑SEZ Trend = 0.008 *** (s.e. 0.003).
    - Panel C Post‑SEZ Trend = 0.009 *** (s.e. 0.002).
  - Interpretation and magnitudes:
    - Entry of SEZs is associated with a decline in district‑level Gini. Combined effects in Panels A and C correspond to about 1.5 percentage point decline in income inequality (text).
    - Panel B specification implies about 4.6 percent decline in income inequality from SEZ entry, though the dynamic effect is not statistically significant (text).

- Land values (Column 7):
  - SEZ coefficients:
    - Panel A SEZ = 0.034 (s.e. 0.101).
    - Panel B SEZ = 0.105 ** (s.e. 0.042).
    - Panel C SEZ = -0.043 (s.e. 0.142).
  - Interpretation:
    - Overall effect on residential land prices is negligible in Panels A and C.
    - Panel B suggests possible increase in average land prices with SEZ entry; this is explored further in Sections 5 and 6.

---

### Main Results — Education (Table 2, Column 8)
- High‑school drop‑out rate (share of population older than 18, not in school, without diploma):
  - SEZ coefficients:
    - Panel A SEZ = 0.008 (s.e. 0.036).
    - Panel B SEZ = -0.011 (s.e. 0.035).
    - Panel C SEZ = 0.004 (s.e. 0.042).
  - Interpretation: SEZ entry has negligible direct impact on high‑school drop‑out rates in the baseline.

---

### 5. Effect of Multiple SEZs (intensive margin; Table 3)
- Context and variable construction:
  - Treated districts hosted, on average, 2.4 SEZs at the end of the sample period.
  - SEZ intensity indicator M_d = 1 if district had more than one SEZ at end of sample; M_d = 0 if only one SEZ present.
  - Estimation augments baseline with interaction: β_M (M_d × D_dt).
  - Interaction of post‑SEZ trend and multiple SEZs excluded because coefficients were not statistically significant.

- Main findings (selected, from Table 3)
  - Core results remain robust when controlling for number of SEZs—primary effects documented in Table 2 are not driven by a few districts with multiple SEZs.
  - No additional impact of multiple SEZs on employment indicators (paid employment, manufacturing employment, female employment).
  - Wages (Column 4, log(Wages)):
    - Panel A(SEZ × Multiple SEZs) = 0.325 (s.e. 0.419) — marginal statistical significance noted; coefficients on SEZ and interaction not jointly significant (interpret cautiously).
    - Panel B(SEZ × Multiple SEZs) = 0.327 * (s.e. 0.168).
  - Incomes, Gini, school drop‑out: no additional heterogeneity by SEZ intensity (Columns 5, 6, 8).
  - Land values (Column 7) — heterogeneous and notable effect:
    - Panel A SEZ × Multiple SEZs = -0.341 *** (s.e. 0.096).
    - Panel B SEZ × Multiple SEZs = -0.562 *** (s.e. 0.135).
    - Interpretation: while land values initially rise with the entry of the first SEZ, subsequent SEZs in the same district are associated with large declines in residential land values.
    - Joint effect cited in text: 20 to 44 percent decline in residential land values in districts with multiple SEZs.
    - Possible explanation: in the low‑income country context, governance shortcomings (e.g., land grabbing for industrial production) could negatively affect residential or agricultural land prices when multiple SEZs establish operations in a location.

---

*Source: IMF working paper — Section 3 (Estimation Strategy), Tables 1–3, and accompanying text.*

### 6.  Spillover Effects on Neighboring Districts

### 6. Spillover Effects on Neighboring Districts

### Spillover analysis framework
- Extended specification includes indicators for SEZ presence in neighboring (bordering) control districts, DNt, and corresponding post-entry time trend, TNt:
  ydt = α + βDdt + λTdt + βN D N dt + λN T N dt + δd + γpt + εdt.
- DNt equals one when a control district shares a border with a treated district. The coefficient on TNt measures dynamic spillover effects in neighboring control districts.

### Empirical findings on neighboring-district spillovers (summary of Table 4)
- General pattern:
  - Spillover effects on paid employment share and manufacturing employment share are negligible.
  - Small positive spillover on female employment in neighboring districts, interpreted as commuting of female workers from neighboring districts to treated districts.
  - No evidence of spillover effects on wages, household incomes, and income inequality (Gini) in neighboring districts.
  - Limited evidence of remittance-linked income effects toward neighboring districts.
- Selected coefficients and standard errors (Table 4):
  - SEZ (treated district direct effect):
    - Paid empl.: -0.001 (0.007)
    - Mnf. empl.: 0.011 (0.010)
    - Female empl.: 0.058*** (0.014)
    - log(Wages): -0.051 (0.133)
    - log(HH Income): -0.003 (0.069)
    - Gini coef.: -0.044*** (0.008)
    - log(Land value): 0.149** (0.054)
    - School Drop-out rate: -0.037* (0.021)
  - Post-SEZ Trend (treated district):
    - Paid empl.: -0.001 (0.003)
    - Mnf. empl.: 0.006 (0.011)
    - Female empl.: -0.011*** (0.004)
    - log(Wages): 0.069 (0.048)
    - log(HH Income): 0.064 (0.042)
    - Gini coef.: 0.007 (0.007)
    - log(Land value): 0.020 (0.026)
    - School Drop-out rate: 0.009 (0.008)
  - SEZ in Neighboring District:
    - Paid empl.: 0.043 (0.030)
    - Mnf. empl.: 0.016 (0.023)
    - Female empl.: 0.018* (0.010)
    - log(Wages): -0.075 (0.130)
    - log(HH Income): 0.002 (0.082)
    - Gini coef.: -0.003 (0.017)
    - log(Land value): 0.066 (0.087)
    - School Drop-out rate: -0.045*** (0.014)
  - Post-SEZ Trend in Neighboring District:
    - Paid empl.: 0.001 (0.005)
    - Mnf. empl.: 0.003 (0.005)
    - Female empl.: 0.000 (0.002)
    - log(Wages): 0.014 (0.018)
    - log(HH Income): 0.017 (0.012)
    - Gini coef.: -0.003 (0.002)
    - log(Land value): -0.028 (0.019)
    - School Drop-out rate: 0.015*** (0.002)
- Interpretation of schooling and land results:
  - Land valuations in treated districts rose after SEZ entry by about 15 percent (statistically and economically significant).
  - Adding neighboring-district spillover variables changes direct effects on land values and school drop-out rates:
    - Joint coefficients imply drop-out rates increase slightly in neighboring districts by 1.5 percent, assuming an average age of an SEZ is about 4 years.
    - Direct effect on high school drop-out rates in treated districts is negative, implying educational attainment in treated districts rises relative to neighbors — possibly reflecting migration of more educated workers into SEZ districts rather than behavioral improvements in education.

### Inference
- Spillovers are present for female employment (likely commuting) but limited for wages, household income, and income inequality.
- Land price increases and diverging school drop-out dynamics highlight potential distributional and migration-driven effects of SEZ placement on neighboring districts.

---

### 7. Robustness Checks

### Methods
- Two additional checks:
  1. Inverse propensity score-weighted regression with alternative control group (adjacent & future SEZ controls).
  2. Specification using lags of the dependent variables as controls.

### Main robustness results (summary of Table 5)
- Panel A: Propensity Score Weights and Adjacent & Future SEZ Controls (Observations (A): 354 for each column)
  - SEZ coefficient estimates:
    - Paid empl.: -0.024*** (0.004)
    - Mnf. empl.: -0.004 (0.008)
    - Female empl.: 0.050** (0.017)
    - Wages: -0.072 (0.104)
    - HH Income: 0.018 (0.072)
    - Gini coef.: -0.039** (0.014)
    - Land value: 0.082* (0.041)
    - Drop-out rate: -0.030 (0.029)
  - Notable points:
    - Under this restrictive specification, direct SEZ effects on female employment and district-level income inequality remain largely unchanged.
    - Presence of SEZs may have a negative impact on paid employment (Column 1), consistent with growth in total employment including informal activities.
    - Land values rise in treated districts, suggesting disproportionate gains to landowners.
- Panel B: Lagged Specification (Observations (B): 1,084 for each column)
  - SEZ coefficient estimates:
    - Paid empl.: 0.012 (0.018)
    - Mnf. empl.: -0.015 (0.015)
    - Female empl.: 0.033* (0.019)
    - Wages: 0.014 (0.231)
    - HH Income: 0.034 (0.099)
    - Gini coef.: -0.036** (0.013)
    - Land value: 0.067 (0.089)
    - Drop-out rate: -0.012 (0.047)
  - Post-SEZ Trend (Panel B) highlights:
    - A small negative post-treatment trend in manufacturing employment share: Post-SEZ Trend for Mnf. empl. = -0.013** (0.006).
    - This is consistent with slowdown in total paid employment growth found in Panel A and suggests auxiliary growth may occur outside manufacturing.

### Inference
- Robustness checks largely support baseline findings: increases in female employment and reductions in income inequality in treated districts persist.
- Some sensitivity in paid employment and manufacturing trends suggests sectoral shifts and possible growth in informal or non-manufacturing activities following SEZ entry.

---

### 8. Policy Implications

### Main policy recommendations
- Promote foreign investment in higher value-added industries and across wider geographic areas; strengthen input-output linkages with local economy through:
  - Investment in infrastructure (energy, transportation, water and sanitation) to:
    - Encourage SEZ entry into disadvantaged areas.
    - Promote formation of input-output linkages with domestic firms.
    - Attract foreign investment in more technologically sophisticated industries and support export diversification.
  - Investment in human capital:
    - Address limited evidence that existing SEZs create higher-paid jobs or attract higher-skilled labor.
    - Mitigate incentives for youth to leave school for low-skilled SEZ jobs; encourage foreign firms to provide worker training to foster knowledge transfer and quality upgrading.
  - Fine-tune tax incentives:
    - Move away from tax holidays toward incentives linked directly to the size of the investment to attract long-term, higher-quality investment and reduce rent-seeking.
    - Recognize that tax incentives are less effective in poor investment climates; broader reforms (infrastructure, rule of law, transparency, governance) yield greater gains.
  - Support domestic entrepreneurial activity and productivity:
    - Improve governance and transparency of tax and business registration systems.
    - Promote productivity of domestic firms through investment in human capital and infrastructure to strengthen backward linkages and generate positive productivity spillovers.

### Rationale linking evidence to policy
- SEZs in Cambodia between 2007 and 2017:
  - Increased female employment and reduced income inequality in SEZ host districts.
  - Did not raise real wages, suggesting gains accrue more to landowners (land values rose) than to workers.
  - Showed limited backward linkages: firms within SEZs purchase only 12 percent of inputs domestically compared to 62 percent by non-SEZ firms (World Bank and ADB, 2014), consistent with assembly/re-export specialization.
- Therefore, policy focus should be on upgrading the sectoral composition of SEZ investors, improving domestic firm productivity, and widening geographic reach of SEZ benefits.

---

### 9. Conclusion (key takeaways)
- Identification strategy uses variation in SEZ location and timing, with propensity score weights and alternative control groups, plus robustness checks using lagged dependent variables.
- Between 2007 and 2017, Cambodia’s SEZ program:
  - Increased female employment rates and reduced income inequality in SEZ host districts.
  - Did not increase real wages, implying distributional benefits favor landowners.
  - Was concentrated in areas with better infrastructure and trade access, potentially widening urban-rural divides.
  - Is associated with rising high school drop-out rates in districts with higher SEZ concentration and in neighboring control districts.
- Policy priorities to enhance socio-economic spillovers:
  - Invest in energy, transportation, water and sanitation infrastructure.
  - Invest in human capital and encourage worker training by firms.
  - Shift from tax holidays to investment-size-linked tax incentives to attract longer-term, higher-value investment and limit rent-seeking.

*Source: IMF Working Paper chapter "6. Spillover Effects on Neighboring Districts" (from the provided content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020170-print-pdf.pdf_