Can Low-altitude Logistics Drive County-level Green Transformation? Synthetic Control Evidence from Nankang District, China

Technical Summary: Low-altitude Logistics and County-level Green Transformation in Nankang District

Problem Statement
While low-altitude logistics, facilitated by unmanned aerial vehicles (UAVs), is increasingly recognized as a new form of intelligent transportation infrastructure with potential low-carbon attributes, credible empirical evidence regarding its impact on composite green-transformation outcomes at the county level remains scarce. Existing literature largely focuses on provincial or city-level analyses or theoretical lifecycle assessments of drone delivery. There is a significant identification gap concerning whether real-world low-altitude logistics pilots are causally associated with improvements in a multi-dimensional county-level economic green transformation index. This study addresses this gap by evaluating the 2017 low-altitude logistics delivery pilot in Nankang District, Jiangxi Province, China.

Methodology
The study employs a Synthetic Control Method (SCM) to construct a data-driven counterfactual for Nankang District.

• Data: An annual panel dataset covering 80 counties in Jiangxi Province from 2007 to 2022. Nankang is the single treated unit; the remaining 79 counties form the donor pool.
• Outcome Variable: A composite County-level Economic Green Transformation (CEGT) index. This index aggregates three dimensions: economic development foundation (GDP, fiscal revenue, industrial structure), low-carbon development level (total electricity consumption and sector-specific CO₂ emissions), and green sustainable development capacity (PM₂.₅ concentration, agricultural machinery power, and patent grants). The index was pre-computed using a time-space range entropy-weighting procedure.
• Predictors: Six standardized pre-treatment predictors (2007–2016) were used to construct the synthetic control: log retail sales, fixed-asset investment to GDP ratio, per capita loan balance, log number of secondary-school teachers, mobile-phone users per capita, and log number of healthcare/welfare beds.
• Inference Strategy:
  • Full Permutation Inference: The 79 donor counties are sequentially assigned as pseudo-treated units to generate a distribution of placebo effects.
  • Comparable-Fit Restricted Permutation: To address sensitivity to poor pre-treatment fits, a restricted inference is performed retaining only placebo units with a pre-treatment Root Mean Squared Prediction Error (RMSPE) no greater than twice that of Nankang.
  • Robustness Checks: The study includes leave-one-out re-estimation, an in-time placebo test (assigning 2014 as the policy year), and multi-threshold restricted permutation exercises.

Key Results

• Synthetic Construction: Synthetic Nankang is composed of five donor counties, with Fengcheng City (55.93%), Shicheng County (15.50%), Wannian County (14.29%), Shanggao County (13.89%), and Yudu County (0.39%) receiving positive weights. The pre-treatment fit is tight, with a Pre-treatment RMSPE of 0.00742 and an average pre-treatment gap of +0.0061 index units.
• Treatment Effect: Following the 2017 pilot implementation, the observed CEGT index for Nankang diverges positively from the synthetic counterfactual.
  • The average post-treatment gap (2017–2022) is +0.0317 CEGT index units.
  • The divergence becomes more pronounced from 2019 onward, peaking at +0.0541 in 2021.
  • The Post-to-Pre RMSPE ratio is 4.84, indicating a significant increase in prediction error post-treatment relative to the pre-treatment period.
• Statistical Significance:
  • In the full permutation distribution, Nankang ranks 1st out of 80 units, with an exact p-value of 0.0125.
  • In the comparable-fit restricted permutation (31 retained placebo units), Nankang still ranks 1st out of 32, with a restricted exact p-value of 0.03125.
  • Robustness checks (leave-one-out, in-time placebo, and varying fit thresholds) consistently support the main pattern, showing no comparable positive divergence in placebo scenarios.

Significance and Claims
The paper claims three marginal contributions:

1. Contextual: It extends policy evaluation of low-altitude logistics to the county-level economic green transformation literature, complementing existing provincial and city-level studies.
2. Methodological: It leverages a single real-world UAV pilot to construct an explicit synthetic counterfactual, combining strict full permutation inference with comparable-fit restricted permutation and a suite of falsification tests. This adheres to recent methodological guidance for single-unit inference.
3. Policy: It provides synthetic control evidence to inform the design of pilot evaluation and monitoring schemes for low-altitude logistics.

Modesty and Limitations
The authors explicitly frame their findings with caution:

• Causality vs. Association: The results indicate a policy-related improvement relative to the counterfactual but do not identify specific causal mechanisms or transmission channels. The divergence accelerating after 2019 is reported descriptively due to a lack of verified operational records linking specific milestones to the outcome.
• Scope of Impact: The findings should not be interpreted as direct evidence of UAV-related carbon savings or reductions in specific sub-components (e.g., emissions alone). The outcome is a composite index reflecting broader structural changes.
• External Validity: As a single-treated-unit design, the results are best interpreted as evidence specific to the Nankang pilot rather than a generalizable estimate for all county-level low-altitude logistics.
• Unobserved Confounders: While the SCM absorbs common shocks, unobserved local initiatives or potential within-prefecture spillovers (given that some donor counties are in the same prefecture as Nankang) cannot be completely ruled out.

The study concludes that the low-altitude logistics pilot in Nankang is associated with an improvement in the county-level economic green transformation index, but emphasizes that future research requires granular operational data and multi-unit identification strategies to further validate and generalize these findings.