The Impact of Special Economic Zones on Economic Development: Evidence from Nightlight Analysis in the Lao People’s Democratic Republic

I. Introduction

Special economic zones (SEZs) are regarded as one of the most popular industrial policy tools adopted in recent decades due to their transformative impact on economies. When properly executed, the distinct policy and regulatory framework provided by SEZs has proven instrumental in attracting foreign direct investment, creating jobs, and generating exports and foreign exchange. This model has shown to be particularly successful in some countries and economies in East Asia and Latin America (Zeng 2021), where SEZs have played a crucial role in promoting remarkable economic growth and development. The proven success of SEZs in these emerging economies has inevitably generated significant interest among developing countries in adopting SEZ strategies, albeit with varying degrees of success.

In recent decades, SEZ adaptation has experienced exponential growth: from 79 SEZs in 29 economies in 1975 to approximately 5,400 in 147 economies in 2018, of which 4,000 are located in Asia (United Nations Conference on Trade and Development 2019). Despite prominent growth, the evidence of their impact on the economy is still inconclusive, especially in developing countries. This is partly due to the challenges in obtaining data for such analyses. This study aims to identify the causal impact of SEZ establishment on economic activity using the newly established counterfactual estimators for causal inference and nighttime lights (NTL) data for districts in the Lao People’s Democratic Republic (Lao PDR).

SEZs are delimited areas within a country where governments incentivize industrial activity through fiscal, regulatory, and infrastructure support (United Nations Conference on Trade and Development 2019). The definitions and objectives of SEZs vary among countries, with developed economies focusing on innovation and logistics. In contrast, developing economies use them to attract investment and stimulate employment, exports, and government revenue. Transferring the experience of developed countries to developing countries is often challenging due to differences in institutional quality, purpose, and structure.

In the Lao PDR, there are currently 12 SEZs located in 10 districts. Despite the government’s plan to establish up to 40 additional SEZs in the next decade, there has yet to be any empirical evidence of their impact on the economy. Existing studies (Phonvisay, Thipphavong, and Manolom 2017; Vixathep and Phonvisay 2020) have mainly examined the effects of SEZs on changes in employment, exports, investments, and government revenue since their establishment.

This study uses new counterfactual estimators for causal inference with remote-sensing, time-series, cross-sectional data to examine the effects of establishing SEZs in the Lao PDR (Liu, Wang, and Xu 2024). There are precisely estimated positive effects on a proxy for economic activity in the districts hosting SEZs, albeit with a lag, perhaps due to the lack of a unified legal framework and strategy during the early stages of SEZ policy adoption. Heterogeneity analyses in terms of subperiods and types of zones show that SEZs after 2010 and industrial SEZs, rather than tourism ones, appear to have larger effects on economic activity. Notably, 2010 marked the beginning of a more coordinated approach to SEZ development in the Lao PDR.

This study makes novel contributions to the study of the impact of SEZs on developing countries’ economies using NTL data. We contribute positively to the literature by providing evidence from the unique perspective of a small, landlocked developing country surrounded by fast-growing, export-oriented economies. Heterogeneity analysis further highlights the importance of supporting institutions and policies in reaping the benefits of SEZs.

Our study stands out as the only analysis to empirically examine the impact of SEZs using the counterfactual estimators for causal inference with time-series, cross-sectional data. Unlike other studies that primarily employ the difference-in-differences (DID) framework (Wu, Hong, and Muhammad 2020; Gallé et al. 2022; Wang, Yang, and Wei 2022; Otchia and Wiryawan 2024), we recognize potential biases in DID results when dealing with a small treatment group. The proposed framework accommodates a relatively small treatment group and multiple treatment timings, as observed in our study (Liu, Wang, and Xu 2024). Finally, this study offers empirical evidence of the overall impact of SEZs in the Lao PDR. These findings can offer valuable insights to policymakers as they consider establishing additional SEZs in the near future.

The rest of this paper is organized as follows. Section II outlines the relevant literature on key policies in SEZ development in the Lao PDR, the economic impact of SEZs, and the use of NTL as a proxy for economic activity. Section III details the data and the counterfactual estimators for causal inference with time-series, cross-sectional data. Section IV discusses the main results, including robustness and additional heterogeneity tests. Sections V–VII provide the discussion, limitations, and conclusions, respectively.

II. Related Literature

A. Special Economic Zone Development and Key Policies in the Lao People’s Democratic Republic

Since the initiation of market-oriented economic reforms in 1986, the Government of the Lao PDR has prioritized the development and expansion of the private sector. SEZs have emerged as a key focus to attract foreign investment, diversify the economy, and achieve rapid socioeconomic development and regional integration (Secretariat Office to the National Committee for Special Economic Zones [S-NCSEZ] 2012; Phonvisay, Thipphavong, and Manolom 2017). The first SEZ in the Lao PDR was established in 2003 as an experimental site. The decree on management regulations and incentive policies for the Savan-Seno SEZ (Decree No. 177/PM) would later be legislated and become the pioneering legislation for subsequent SEZs.

Despite government efforts, the absence of a unified legal framework and strategy has hindered the effectiveness of SEZs, with different agencies managing them under individually specific legal frameworks. Challenges include resource constraints, inefficient land acquisition processes, compensation disputes, and failure to meet targets for local communities. The lack of tailored lessons for the country’s socioeconomic conditions has contributed to poor administration (Kuaycharoen et al. 2020). The turning point occurred in 2010 with the enactment of Decree No. 443/PM, providing a crucial legal framework for SEZ development in the Lao PDR.

Decree No. 443/PM outlined principles, regulations, organization, and management, including procedures for establishing SEZs and incentives for developers and investors, such as tax exemptions and land use rights. The decree also specified the rights and obligations of stakeholders, addressing community relations, land-use compensation, and cultural and environmental preservation (National Committee for Special Economic Zones [NCSEZ] 2010).

To oversee the decree’s implementation, the NCSEZ and the S-NCSEZ were established. These government units were tasked with (i) developing overall guidelines and strategies for SEZ development based on a market-driven approach; (ii) ensuring competitiveness; (iii) streamlining investment and business processes through a one-stop service; and (iv) addressing issues of equity, accountability, sustainability, and environmental conservation (Vixathep and Phonvisay 2020). In 2016, these responsibilities were transferred to the Ministry of Planning and Investment.

The Lao PDR currently has 12 SEZs (Figure 1), with the government planning to establish up to 40 additional SEZs in the next decade. SEZs fall into two categories: special and specific economic zones. SEZs cover at least 1,000 hectares and are designed to evolve into modern towns with residential areas, whereas specific economic zones serve specific purposes. As of 2021, all zones had been upgraded to SEZ status (Vixathep and Phonvisay 2020). SEZs can be further classified into three sectors: (i) industrial zones, (ii) tourism and new urban centers, and (iii) trade and logistics zones. However, many zones have adopted an integrated development approach that combines manufacturing, commercial, residential, and tourism areas (Kuaycharoen et al. 2020).

B. Impact of Special Economic Zones

Despite the widespread popularity of SEZs as a policy tool, their impact on the economy remains unclear, particularly in developing countries. The existing literature on developing countries has primarily focused on case studies in the People’s Republic of China and India, leaving a significant gap in the understanding of other developing countries (Luo et al. 2015; Alkon 2018; Lu, Wang, and Zhu 2019; Wu, Hong, and Muhammad 2020; Gallé et al. 2022; Wang, Yang, and Wei 2022).

While the majority of studies suggest a positive impact of SEZs on economic growth, capital investment, employment, productivity, wages, innovation, and entrepreneurial activities (Luo et al. 2015; Frick and Rodríguez-Pose 2019; Lu, Wang, and Zhu 2019; Wang, Yang, and Wei 2022; Brussevich 2024), others find evidence of structural change and industrialization (Luo et al. 2015; Xi, Sun, and Mei 2021; Gallé et al. 2022; Otchia and Wiryawan 2024). Conversely, some studies reported no significant or even negative impact and spillover (Asian Development Bank 2015, Brussevich 2024), citing factors such as high initial costs (Shenoy 2018), local capture (Alkon 2018, Liu and Ma 2019), and poor implementation (Vu and Yamada 2022). Evidence from Cambodia also indicates negative spillover effects, such as increased school dropout rates in districts neighboring SEZ districts (Brussevich 2024).

Inconclusive results in the literature may stem from challenges in obtaining reliable data for analysis, especially in developing countries, where data may be scarce or unreliable. Scholars have addressed these limitations by using proxy measurements, such as employing NTL data from satellite observations to represent SEZs or economic activity (Frick and Rodríguez-Pose 2019, Gallé et al. 2022). For instance, Frick and Rodríguez-Pose (2019) used NTL data to proxy for the performance of 346 SEZs in 22 emerging countries, revealing a positive impact with a strong distance-decay effect in the surrounding areas.

Studies on the impact of SEZs in the Lao PDR are scarce. Phonvisay, Thipphavong, and Manolom (2017) conducted semistructured interviews with relevant SEZ stakeholders in the Lao PDR. Their findings revealed that some SEZ firms outsourced production inputs from local firms and collaborated with local polytechnic schools to train workers. This finding hints at forward and backward linkages and local spillovers, potentially enhancing long-term productivity and performance. However, research on this topic is still limited and requires more empirical evidence. This study aims to fill this gap by employing empirical methods to examine the impact of SEZs on the economy of the Lao PDR.

C. Using Nighttime Lights as a Proxy for Economic Activity

NTL data have gained widespread use as a proxy for economic activity (Henderson, Storeygard, and Weil 2012; Frick and Rodríguez-Pose 2019; Mendez and Santos-Marquez 2021; Otchia and Asongu 2021) due to availability, comparability across countries, and ability to capture informal sector activity, which is often overlooked in official gross domestic product (GDP) statistics. In the Lao PDR, where subnational GDP data are unavailable, NTL data offer a valuable alternative for assessing economic development.

Researchers commonly use two NTL datasets: (i) Defense Meteorological Satellite Program (DMSP) data, available from 1992 to 2013; and (ii) newer Visible Infrared Imaging Radiometer Suite (VIIRS) data, available from 2012 to 2021. The DMSP, initiated in 1962 with the Operational Line-Scan System, was originally designed by the United States Air Force for cloud detection in short-term weather forecasts. The satellites scan the Earth with a spatial resolution of 30 arc seconds, approximately $0.9\times 0.9$ kilometer (km) at the equator, and produce output pixels with a ground footprint of 25 square kilometers (km²) at nadir (Zhang and Gibson 2022). This ground footprint can be extended by 240% at the half scan and 400% at the edge of the scan (Gibson et al. 2021; Puttanapong, Prasertsoong, and Peechapat 2023). The data were stored in digital number (DN) values ranging from 0 to 63, where lower values indicate less brightly lit areas. Due to the absence of built-in calibrations to address sensor amplification changes on board each satellite (F10, F12, F14, F15, F16, and F18), the DN values lack temporal consistency (Kim, Gibson, and Boe-Gibson 2024). Satellites initially have an overpass time starting at 7:30 p.m., but the orbits gradually shift from a day–night orbit to a dawn–dusk orbit (Nechaev et al. 2021).

Using DMSP data as a proxy for economic activity poses inherent challenges because of the original purpose of observing clouds rather than lights. Key issues include the blurring effect and top-coding problems, resulting in mean-reverting errors (Zhang and Gibson 2022; Kim, Gibson, and Boe-Gibson 2024). Blurring occurs when NTL are attributed to locations without emitted light. Meanwhile, top-coding arises in brightly lit areas, such as urban centers experiencing light saturation due to additional unrecorded amplification performed by DMSP satellites so cloud tops can be viewed with similar brightness across the light and dark parts of the lunar cycle. The top-coding problem is further exacerbated by pixel aggregation to conserve the limited data storage (Zhang and Gibson 2022). DMSP’s limited ability to detect low light also contributes to a less discussed bottom-coding issue, resulting in false zeros, particularly in dimly lit areas that are common in many developing countries. The spatially mean-reverting errors affect econometric estimators with lights on the left-hand side, attenuating coefficients and making apparent impacts appear smaller and less precise (Kim, Gibson, and Boe-Gibson 2024).

In contrast, the research-oriented VIIRS on the Suomi National Polar Partnership satellite, launched in 2011, was specifically designed for NTL detection, addressing many of the limitations of DMSP satellites. The VIIRS scans the earth at a 15 arc-second grid, maintaining a consistent ground footprint of 0.55km² and an overpass time of 1:30 a.m. Unlike DMSP, VIIRS sensors measure light in radiometric units of nanowatts per square centimeter per steradian at a 14-bit resolution, providing up to 65,536 possible values compared to DMSP’s 64 (Zhang and Gibson 2022). Moreover, VIIRS incorporates in-flight calibration for data comparability across time and space. These enhancements make VIIRS data significantly superior to DMSP in terms of spatial, temporal, and radiometric resolutions (Nechaev et al. 2021).

In cases where the analysis period spans both DMSP and VIIRS datasets, efforts have been made to cross-calibrate the sensors (Li et al. 2020, Chen et al. 2021, Nechaev et al. 2021). Li et al. (2020) globally harmonized NTL data from 1992 to 2018 by inter calibrating DMSP data between each year and satellite. They established a sigmoid regression between VIIRS and the inter calibrated DMSP NTL, transforming VIIRS data into a DMSP-like format. In contrast, Chen et al. (2021) enhanced DMSP data by converting it into a VIIRS projection and radiance range, utilizing a vegetation index and auto-encoder neural network model to construct VIIRS-like NTL data from 2000 to 2018. However, inherent differences persist when joining DMSP and VIIRS data, including the changing ground footprint (25km² for DMSP to 0.55km² for VIIRS) and overpass time transition (from 7:30 p.m. to 1:30 a.m. local time). These disparities may still impact analyses using NTL as the primary dependent variable (Kim, Gibson, and Boe-Gibson 2024).

This study utilizes the harmonized NTL data from Li et al. (2020). The derived DMSP-like data (2014–2018) show a temporal trend and spatial pattern that is consistent with the DMSP data (1990–2013).¹ Additionally, the selection of these data is driven by its availability for at least five periods before the 2003 initiation of SEZs, which is a critical requirement for implementing the counterfactual framework.

III. Data and Methodology

A. Data

This study uses NTL data as a proxy for economic activity to estimate the impact of SEZs on the economy of the Lao PDR. Specifically, we use the harmonized NTL data constructed by Li et al. (2020) for 1992–2021. Similar to Yamada and Yamada (2021), the mean value of NTL for each of the 148 districts was calculated with a value ranging from 0 to 63. The main treatment variable of the analysis is the establishment of an SEZ in the district. Information regarding each SEZ was obtained from the Special Economic Zone Promotion and Management Office reports, which contain information on the year of establishment and location of each SEZ (Table 1).

In addition to the primary variable of interest, the analysis considered control variables such as population, temperature, precipitation, and vegetation. Population data obtained from WorldPop offer an estimated count of people per 1km pixel. To meet the counterfactual framework’s requirement of at least five periods before treatment, population data for 1998 and 1999 were interpolated. Precipitation and temperature data were sourced from the University of Delaware’s Terrestrial Precipitation and Terrestrial Air Temperature Gridded Time Series database (version 5.01). These variables were included in the analysis because of the significant impact of weather changes on the overall economy (Yamada and Yamada 2021), especially in countries reliant on weather-induced agricultural income, such as the Lao PDR. The vegetation index used in this study is the Yearly Normalized Difference Vegetation Index derived from NASA’s Long-Term Data Record (v5) Advanced Very-High-Resolution Radiometer data. This index assesses vegetation health by measuring the difference between near-infrared light reflected by vegetation and red light absorbed by vegetation. GeoQuery compiled and computed all control variables (Goodman et al. 2019) and Table 2 presents the list of variables along with descriptive statistics.

B. Methodology

1. Counterfactual for Causal Inference

For the main analysis of estimating the impact of SEZs on the economy, we adopt the counterfactual estimators for causal inference with time-series, cross-sectional data introduced by Liu, Wang, and Xu (2024). This framework provides several novel estimators for determining the counterfactual when conducting an inference analysis. In other words, it compares the observed outcomes with those expected in the absence of intervention. Many studies on this topic have commonly used the DID or event study framework (Lu, Wang, and Zhu 2019; Gallé et al. 2022; Wang, Yang, and Wei 2022; Brussevich 2024; Otchia and Wiryawan 2024). However, the DID could present a biased result if a small number of treatment units or unobserved time-varying confounders exist, making the parallel assumption unlikely to hold.

The counterfactual estimators provided by the framework account for unobservable time-varying confounders and offer more reliable causal estimations. Additionally, it accommodates heterogeneous treatment timing, addresses issues such as the negative weight problem in the traditional two-way fixed effects (TWFE) model, and allows flexibility in incorporating time-varying covariates. Additionally, it provides diagnostic tests to assess the validity of identifying assumptions (Liu, Wang, and Xu 2024).

The proposed framework includes three sets of counterfactual estimators: fixed effects, interactive fixed effects (IFE) (Xu 2017), and matrix completion (MC) (Athey et al. 2021). For the main analysis, we chose the IFE and MC models due to their ability to address time-varying confounders. While these estimators differ in their assumptions (Pan and Qiu 2022), their general approach to constructing the counterfactual can be summarized in three steps:

Step 1:	Consider the observation under treatment as a missing value and build a predictive model with only untreated observations.
Step 2:	Predict the counterfactual outcome for the treated observations using the predictive model obtained from Step 1.
Step 3:	Obtain the average treatment effect on the treated (ATT) by taking the average difference between the observed outcomes and the outcome from the counterfactual for treated observations.

The counterfactual estimation of the IFE model is represented as follows :

where $Y_{it}$ is the outcome dependent variable (mean NTL) for district i at time t, ${\delta }_{it}$ is the heterogeneous treatment effect on district i at time t, $D_{it}$ is a binary treatment indicator that takes the value of 1 if district i at time t has an SEZ. $X_{it}$ is a vector of observed covariates, $\beta$ represents the vector of the unknown parameters, and ${\epsilon }_{it}$ is the error term. $f_t$ is a vector of unobserved common time-varying factors, and ${\lambda }_i$ is a factor of unknown factor loading.

Factors refer to the time-varying trends that impact each district differently, whereas factor loading represents the heterogeneous impacts resulting from the observed characteristics of each district. The interaction term ${\lambda }_i{f}_t$ implicitly accounts for the effects of unobserved time-varying confounders and the effect of other policies, through which the influence is eliminated or controlled when performing the estimation of ${\delta }_{it}$ (Sun et al. 2022). ${\alpha }_i$ and ${\xi }_t$ are district and time fixed effects, respectively, and ${\epsilon }_{it}$ represents the idiosyncratic shocks for district i at time t.

In contrast to the IFE model mentioned above, which decomposes unobserved factors into ${\lambda }_i$ and $f_t$, the MC model does not explicitly estimate ${\lambda }_i$ and $f_t$, but instead directly estimates the unobserved factor based on the nuclear norm regularization (Mader and Rüttenauer 2022). The unobserved factors can, therefore, be expressed as $L={\lambda }_i{f}_t$ (Maamoun 2019).

Our main interest is the ATT of the intervention on the outcome after the policy implementation over time. The ATT can be estimated as the average differences between the observed outcome $Y_{it}(1)$ and the estimated counterfactual ${Ŷ}_{it}(0)$. The confidence interval presented in the results was based on block bootstraps of 1,000 runs.

2. Two-Way Fixed Effects

In addition to using the counterfactual estimators for causal inference with time-series, cross-sectional data under different estimation models, we used the traditional TWFE model to compare the results with the following specification :

where $y_{it}$ is the output variable (mean NTL). ${SEZ}_{it}$ is the treatment variable, SEZs in district i at time t. $X_{it}$ denotes the time-varying control variables, while ${\delta }_i$ and ${\pi }_t$ denote district and time fixed effects, respectively. At the same time, ${\epsilon }_{it}$ represents the error term.

IV. Results

A. Main Result

Figure 2 illustrates the outcomes of the main analysis. Panel A displays the dynamic treatment effect of SEZs over time, where period 0 marks the initial treatment phase when SEZs were established. Each point signifies the difference between the counterfactual and observed outcomes or ATT. Before period 0, the ATT remained relatively constant at about 0. However, following the introduction of SEZs, the ATT began to increase, indicating a positive impact on the economic activity of the host districts. This effect became statistically significant after the fourth period.

Panel B presents a result similar to that in panel A, but the period is expressed in calendar years. Notably, the first SEZ was established in 2003, yet its impact did not become statistically significant until around 2010.² Interestingly, this period aligns with the establishment of the NCSEZ and S-NCSEZ, potentially hinting at the positive impact associated with institutional changes. These findings are consistent with those of Frick and Rodríguez-Pose (2019), who observed a positive impact of SEZs on economic activity in selected emerging economies.

While Figure 2 illustrates the dynamic effects, Table 3 depicts the average treatment effect across all periods under various specifications. Columns 1 and 2 present the outcomes for the counterfactual framework using the IFE and MC specifications, respectively. Column 3 presents the results obtained using the TWFE method.

In panel A, the results indicate that SEZs have a positive and statistically significant impact on the economic activity of the host district at the 5% significance level for all model specifications. The IFE and MC models exhibited similar magnitudes, whereas TWFE tended to overestimate this effect. In panel B, where the control variables are included, the effect remains consistent in terms of sign and significance level. However, the magnitude decreased for all specifications. This reduction may be attributed to the loss of observations when control variables are included, given their availability from 2000, whereas the dependent variables are accessible from 1992.

B. Heterogeneity Test

Building on the findings in Figure 2, which reveal a notable increase in the effect of SEZs after 2010, we delved deeper into the data by conducting a heterogeneity test. To accomplish this, we segmented the data into two groups, pre-2010 and post-2010, which align with the time when the NCSEZ and S-NCSEZ were established. Prior to their establishment, SEZs in the Lao PDR lacked a unified government agency to offer guidance and strategic direction for their development. The analysis was conducted using the TWFE method and the results are presented in Table 4.

Recognizing that SEZs in distinct sectors may exhibit structural differences and emit varying intensities of NTL, potentially influencing our results differently, we conducted an additional analysis that specifically focuses on different SEZ types. In this analysis, we categorized districts into two main groups: (i) those hosting industrial zones, and (ii) those hosting tourism and new urban zones. Trade and logistics zones were excluded due to their limited representation, resulting in only one district remaining after removing districts with multiple types of SEZs. This approach ensures a more refined and coherent comparison between districts hosting industrial zones and those hosting tourism and new urban zones. The results of the sector-specific analyses are presented in Table 5.

Table 4 shows the results of the effects of the SEZs after 2010. Panel A represents the results without control variables, and panel B shows the results with the control variables included. The empirical evidence suggests that having a unifying legal framework, guidelines, and competitive strategy is associated with a positive and significant impact of SEZs on economic activity. These results are consistent when control variables are included and after controlling for province-specific and region-specific time trends. This finding is similar to that in India, where SEZs experienced exponential growth after a similar SEZ policy was implemented in 2005 (Gallé et al. 2022). These findings highlight the importance of having supporting institutions and policies for SEZs to be effective.

Table 5 presents the findings of the heterogeneity test based on the type of SEZ. Across various specifications, the results consistently demonstrate that industrial SEZs have a more substantial impact on economic activity than tourism SEZs. Both the IFE and MC estimations yielded similar results in terms of magnitude and statistical significance. However, the TWFE estimation tends to overstate this effect, similar to our main result. When the control variables are introduced in panel B, the effect diminishes. Nonetheless, the results still indicate a positive trend.

C. Robustness Test

The counterfactual estimators for causal inference with time-series, cross-sectional data include diagnostic tools, such as placebo and pretrend tests, to assess the validity of the underlying identifying assumptions. The placebo test assumes that the treatment starts at a specific period before the actual treatment and runs a counterfactual estimation to obtain the estimated ATT for the period between the placebo and the actual treatment. If the underlying functional form and strict exogeneity assumptions are valid, the estimated ATT is expected not to be statistically different from zero (Liu, Wang, and Xu 2024).

The pretrend test is similar to the placebo test, albeit a leave-one-out approach was used, where instead of hiding a few periods before the actual treatment, we consecutively hid one pretreatment period (relative to the timing of the treatment) and repeatedly conducted a placebo test on observations in that period. Consequently, we have a more holistic view of whether the identifying assumptions are likely to hold (Liu, Wang, and Xu 2024).

Additionally, the pretrend test uses the equivalence approach, which reverses the null hypothesis of the placebo test based on the difference-in-mean approach. By doing so, it allows us to set the null hypothesis as ${\mathrm{\text{ATT}}}_s<{-\theta }_2$ or ${\mathrm{\text{ATT}}}_s>{\theta }_1$, where ${\theta }_1$ and ${\theta }_2$ are a prespecified range, and we set ${\theta }_1={\theta }_2=0.36{\widehat{\delta }}_{\epsilon }$, following Liu, Wang, and Xu (2024), where ${\widehat{\delta }}_{\epsilon }$ is the standard deviation of the residualized untreated outcome. This essentially provides us with an equivalence range. The test then checks whether the 90% confidence intervals for the estimated ATTs during the pretreatment period exceed these thresholds. The identifying assumption is assumed to hold if the estimated ATTs fall within the prespecified range.

Figure 3 shows the results of the placebo and pretrend tests. The results of the placebo test in panel A provide evidence supporting the validity of the underlying assumptions, as seen in the nonsignificant p-value of 0.39. The result of the pretrend test further confirms this finding, as the trend leading toward the onset of the treatment falls within the equivalence range.

V. Discussion

This study contributes to the ongoing debate on the impact of SEZs on the economic activity of developing countries. By implementing the newly developed counterfactual estimators for causal inference with time-series, cross-sectional NTL data and variations in SEZ establishment across districts in the Lao PDR, our results show that SEZ establishment leads to a positive and statistically significant increase in the economic activity of districts hosting SEZs. Our results offer unique insights from a small, landlocked developing country’s perspective and echo similar conclusions found in the literature (Frick and Rodríguez-Pose 2019, Gallé et al. 2022). However, the result indicates a 4-year lag before the effect becomes statistically significant. This is relatively long compared to a study by Otchia and Wiryawan (2024), who found a significant effect on industry share to GDP after only 1 year of implementing the SEZ policy. Still, the effect started to grow faster after the fifth year.

The observed 4-year lag in our study may be attributed to several challenges faced by the government during the initial stages of SEZ development. Before 2010, there was a fragmented legal framework and strategy for SEZ development, with various SEZs established and administered by different government agencies (Vixathep and Phonvisay 2020). The lack of experience in SEZ management and supervision exacerbated these issues. Furthermore, SEZ developers have encountered many challenges—such as a shortage of financial resources for infrastructure development, disputes over land ownership with local communities, and difficulties in meeting local employment quotas (Kuaycharoen et al. 2020). These factors likely contributed to the gradual impact of the SEZ policy in the Lao PDR.

Our main results also hint at the heterogeneous effects of SEZs. Hence, we performed additional heterogeneity tests by subperiod and type of SEZ. The results suggest that after 2010, SEZs had a more substantial impact on economic activity than prior to 2010. This could be due to the institutional shift that occurred in 2010 when the NCSEZ and the S-NCSEZ were created. This finding is similar to that of India, where SEZs experienced exponential growth after a similar SEZ policy was implemented in 2005 (Gallé et al. 2022). The results based on different types of zones suggest that industrial zones have a greater impact on the economy than tourism zones. More importantly, our results highlight the importance of supporting institutions and effective policies to reap the benefits of SEZs.

VI. Limitations

Although this study found a positive causal impact of SEZs on the economic activity of host districts in the Lao PDR, caution is warranted in interpreting the results because of the inherent weaknesses of the harmonized NTL data. First, after shifting from DMSP data to VIIRS, there was an improvement in low-light detection and a change in the ground footprint, leading to fewer top- and bottom-coding issues. Because of these changes, the periods of observation using DMSP data have a blurring problem, which causes light to extend to unlit areas, leading to a mean-reverting error. Kim, Gibson, and Boe-Gibson (2024) investigated the impact of closing the Kaesong Industrial Zone in the Democratic People’s Republic of Korea and found that the presence of spatially mean-reverting errors in econometric estimators using DMSP NTL on the left-hand side leads to coefficients being attenuated, resulting in the impacts appearing smaller and less precise. Therefore, the larger and more precise impact we found post-2010 could be attributed to this shift from the DMSP to the VIIRS NTL dataset.

Second, the shift in the overpass (local) time from 7:30 p.m. for DMSP to 1:30 a.m. for VIIRS may introduce bias in the assessment of SEZ types. This change could potentially lead to an underestimation of the impact of tourism SEZs, as these zones are more likely to experience limited activities at 1:30 a.m. than industrial zones that can operate around the clock.

Finally, Keola, Andersson, and Hall (2015) noted that NTL are primarily observed in urban areas, and not all economic activity emit NTL as they expand. Consequently, NTL may not effectively capture all activities, particularly those in agricultural and rural areas, which many developing countries depend upon. As a result, this analysis may not comprehensively capture economic activity in the agriculture sector.

Despite these data limitations, this study offers promising insights into the overall impact of SEZs in the Lao PDR and represents one of the first attempts to empirically examine their effects. Future research should explore methods that would allow the use of more precise data sources such as VIIRS NTL or NASA Black Marble. Additionally, integrating additional datasets, such as land cover data, can enhance the understanding of agricultural activities, offering a more comprehensive view of the effects of SEZs on economic activity.

VII. Conclusion

In the past 3 decades, SEZ policies have been widely adopted by a host of countries around the world. Despite the prevalence of SEZs, their impact on the economy remains inconclusive, especially in developing countries. This study contributes to the ongoing discussion on the impact of SEZs on the economic activity of developing countries by utilizing the newly developed counterfactual estimators for causal inference with time-series, cross-sectional NTL data in the Lao PDR. The results indicate a positive and statistically significant increase in economic activity in districts hosting SEZs. Our findings are robust for several estimation models and specifications. However, this study also highlights a 4-year lag before the effect becomes statistically significant. Additionally, the results suggest a heterogeneous impact, with SEZs established after 2010 and industrial zones (versus tourism zones) having a more substantial impact on economic activity. These findings emphasize the importance of having supportive institutions and policies, and of focusing on establishing industrial SEZs to maximize the benefits of SEZ policies.

ORCID

Nilaphy Phommachanh  https://orcid.org/0009-0001-2783-7922

Notes

¹ The derived DMSP-like data exhibited high fluctuation between 2015 and 2018 for pixels with DN values greater than seven. However, this fluctuation notably diminishes for pixels greater than 20 and 30 in the 2015 and 2018 data, respectively. This implies that the dataset is more reliable for pixels with a DN value (Li et al. 2020).

² The year 2010 also marks the start of the F18 series of DMSP data, which is a satellite whose sensor records substantially more light than found in any prior or subsequent years (the F18 series was replaced after 2013), even for settings that should not have had any on-the-ground change (Gibson, Olivia, and Boe-Gibson 2020).