COVID-19 containment policies, digitalization and sustainable development goals: evidence from Alibaba's administrative data

The study investigates how digital platforms, specifically Alibaba Group’s Ele.me on-demand delivery platform, affected MSME sustainability and progress toward the UN SDGs under varying COVID-19 containment policies in China. Motivated by rapid pandemic-driven digitalization, heterogeneous policy stringency (complete, partial, and checkpoint lockdowns), and limited quantitative evidence at high frequency, the research asks: (1) How did different containment policies impact MSME entry and performance on digital platforms during and after lockdown? (2) How did impacts differ across sectors (cooked food, uncooked food, pharmacy)? (3) Which business strategies (chain affiliation, multiapp presence) enhanced resilience and sustainability? The study’s importance lies in quantifying platform-enabled contributions to SDGs (notably 2.c, 3.8, 8.3) and informing policy on containment measures and digital ecosystem support.

The paper reviews work linking digitalization and the SDGs, noting digital technologies’ promise for localization, governance, and delivery of services (e.g., digital health, smart cities), while acknowledging risks (privacy, surveillance, environmental impacts, potential inequality). Most prior SDG studies are qualitative or country-level quantitative analyses; few assess digital food delivery platforms or use merchant-level data. The paper also surveys research on COVID-19’s effects on MSME sustainability and resilience across environmental, social, and economic dimensions, highlighting that prior quantitative work often uses surveys or low-frequency financial data; detailed high-frequency administrative data are rare. It underscores heterogeneity by location and firm characteristics (e.g., chain status) and identifies a gap in measuring the long-term impact of specific containment policies on platform merchants, users, riders, product variety, and pricing.

Setting and policy context: The study examines China’s COVID-19 timeline from late 2019 through early 2021, focusing on cities with heterogeneous containment policies: complete lockdown (Wuhan, Yichang), partial lockdown (Hangzhou, Harbin), checkpoint restrictions (Suzhou, Chongqing), and no formal policy (Xi’an). Lockdown stringency definitions: complete (city immobilization; residents confined; public and private transport banned), partial (residents can leave; public transit halted; private travel allowed), and checkpoints (localized movement restrictions). Data: High-frequency administrative data from Alibaba’s Ele.me and related Koubei platform covering 104 weeks from 2019 week 5 to 2021 week 4 (one year pre- and post-outbreak). Samples include shops with at least one order over the full period. Data levels: (1) city-district-week aggregates (numbers of shops, entrants, active users, full-/part-time riders) and (2) merchant-week operational data (orders, gross merchandise value (GMV), net GMV, subsidies by shop/platform, number of SKUs, rating scores 0–5, categories, chain status and chain brand ID, multiapp (Koubei) status, shop creation date, location). Proxies: price discounts = subsidies/GMV; product variety = number of SKUs. Categories: 204 narrow categories grouped into four broad types—cooked food; uncooked food (e.g., supermarkets, fruit stores); pharmacy; others (e.g., beverages, desserts, flowers). Additional data: city-day COVID-19 cases aggregated to city-week and policy start/end dates by city. Market geography: district within city (most orders occur within ~3 km). Empirical strategy: The authors estimate dynamic policy effects using interactions of policy indicators with linear and quadratic time trends rather than static dummies to capture nonlinearity and adaptation over time. At the district-week level, regressions assess impacts on entries, agent counts (shops, users, riders), and overall performance (orders, net GMV, GMV per order; category net GMV). At the shop-week level, regressions assess impacts on orders, net GMV, GMV per order, subsidy ratio, and SKUs, including shop fixed effects, city-district/week fixed effects, and controls for shop characteristics (rating, age, chain status and network size), same-side network effects (log number of shops in the district), and log new COVID-19 cases. Xi’an (no policy) serves as a control for general time patterns. Models: (1) District-week regressions of outcomes on policy dummies interacted with time and time-squared during and post-lockdown, plus controls and fixed effects; (2) Shop-week regressions of outcomes on policy-time interactions and shop characteristics with shop and week fixed effects; (3) Pooled shop-week regressions with district-week fixed effects to assess business strategies (chain affiliation and chain network size; multiapp presence) during Wuhan’s lockdown and post-lockdown. Robustness includes alternative time trend specifications (linear, cubic), finding quadratic adequately parsimonious and similar to cubic. Outcomes are visualized via dynamic effect plots and summarized in tables for strategy heterogeneity and by product category.

- Digital platform contribution to SDGs and MSME sustainability: The platform supported food security, wellbeing, employment, and MSME business continuity during and after COVID-19, with notable gains in uncooked food and pharmacy transactions. - Extensive margin scarring from strict lockdowns: Under complete lockdown (Wuhan), entrants initially spiked but then fell by 58% by late February 2020, indicating long-term scarring in entry, especially in cooked food. Cities with milder restrictions (partial/checkpoint) saw early surges in entry that gradually declined, suggesting acceleration of digitalization under moderate policies. - Platform agents contracted then recovered: Numbers of shops, active users, and riders declined during lockdown, with longer recovery in complete-lockdown cities. - Overall market performance dynamics: Orders and net GMV in partial/checkpoint cities recovered quickly post-lockdown; Wuhan’s net GMV recovered to prepandemic levels only by early May 2020. Average net GMV per order rose sharply—195% of prepandemic level under complete lockdown—consistent with order consolidation, stockpiling, and reduced logistics capacity. - Category shifts: During lockdown, cooked food net GMV fell while uncooked food and pharmacy net GMV rose (pharmacy especially in Wuhan). After lockdown, uncooked food and pharmacy continued to exceed prepandemic levels; cooked food recovered more slowly in stricter-policy cities. - Intensive margin resilience for open shops: For shops that remained open, orders, net GMV, and GMV per order spiked during lockdown, reflecting higher average demand per seller, while the ratio of subsidies to GMV (discounts) fell in Wuhan, consistent with higher costs and weaker competition. - Product variety (SKUs): SKUs dropped at the pandemic’s onset (particularly under strict containment) due to supply chain disruptions, then gradually increased post-lockdown; open shops in Wuhan still maintained higher-than-average SKUs compared to closed shops. - Business strategies enhancing resilience and sustainability: • Chain stores: Outperformed independents during and after lockdown; larger chain networks conferred greater gains. Joining a chain increased net GMV by about 7.6% during lockdown and 13.1% after. • Multiapp stores (Koubei + Ele.me): Outperformed single-app stores; multiapp presence raised net GMV by about 10.2% during lockdown and 13.8% after. - Heterogeneity by product category: Chain strategy effects varied—e.g., limited gains for cooked food during Wuhan’s lockdown when demand collapsed, and weaker effects for uncooked food post-lockdown when demand normalized; multiapp advantages generally positive except for uncooked food post-lockdown (possible shift toward in-person dining/revenge consumption).

The findings show that digital food delivery platforms can advance SDG 2.c by sustaining food markets and stabilizing access/prices (notably through uncooked food growth), SDG 3.8 by facilitating access to medicines and reducing contagion via digital payments, and SDG 8.3 by fostering MSME digitalization, job creation (including delivery riders), and formalization. Policy implications include: (1) containment policy design matters—partial/checkpoint measures can catalyze MSME digitalization, while complete lockdowns risk long-lasting scarring on entry and service sectors; (2) pandemic disruptions propagate across regions via supply chains and labor/logistics networks, affecting even cities without formal policies, implying a role for governments in internalizing spillovers; (3) on the intensive margin, business strategies such as chain affiliation (especially larger networks) and multiapp participation enhance resilience and sustainability; (4) governments should strengthen digital infrastructure, promote standards for business continuity (e.g., ISO 22301), and support logistics coordination to maximize digital transformation benefits.

Merchants responded heterogeneously to COVID-19 containment policies. Complete lockdowns depressed entry—particularly for cooked food—leaving persistent scarring at the extensive margin. Conversely, on the intensive margin, chain stores and multiapp stores demonstrated stronger resilience during and after lockdown, highlighting the importance of specialization and broader network coverage for business sustainability under shocks. Digital delivery platforms in middle-income contexts can help achieve SDGs by sustaining food transactions, enabling access to medicines, and creating jobs. Policymakers should be cautious with the strictest containment measures due to lasting and contagious economic effects and should foster supportive digital ecosystems and resilient business practices.

The study lacks comparable offline administrative data, preventing direct online-versus-offline comparisons. It also does not observe full multihoming behavior beyond Ele.me due to lack of access to other platforms’ data, limiting analysis of cross-platform dynamics. Findings are based on selected cities and one platform, which may affect generalizability. Future work should integrate multi-platform administrative data and matched offline datasets under appropriate privacy and confidentiality controls.