Big data reveals who really travels to Shanghai for business

A new study has now put a number on this hidden flow: business travelers make up 36.9 percent of all short-term visitors to the city, roughly 654,000 people in a single month, and their destinations stretch far beyond the famous towers of the central business district into suburban industrial parks and new towns. The finding comes from a research team led by Junnan Chen and De Wang, published as an open-access research article in Frontiers of Urban and Rural Planning on 28 October 2026, and it offers one of the most direct behavioral pictures yet of how a global city is wired to the rest of its country.

The difficulty with studying business travel has always been identification. Surveys capture trip purposes accurately but cover only small samples, while the mobile phone signaling data that track millions of people say nothing about why anyone is traveling. The researchers built what they call a big–small data integration framework to bridge this gap. They first surveyed 652 short-term visitors to Shanghai, combining 602 online questionnaire responses with 50 in-person interviews at railway stations, all targeting people who had visited the city within the previous seven days. From this survey they estimated a binary logistic regression model that distinguishes business travelers from other visitor types using socioeconomic attributes, mobile usage patterns, and activity characteristics.

The model performed strongly in stratified ten-fold cross-validation, achieving an accuracy of 88.65 percent, a sensitivity of 79.44 percent, a specificity of 94.31 percent, and a cross-validated ROC–AUC of 0.9417, with a Brier score of 0.085 indicating well-calibrated probabilities. Notably, an ablation analysis showed that the model’s discriminatory power depended almost entirely on travelers’ observed activity patterns in Shanghai: removing those variables dropped the cross-validated AUC from 0.9417 to 0.7825, while removing socioeconomic or mobile-usage variables produced only small decreases. The researchers then transferred this survey-calibrated model to large-scale mobile signaling data from China Unicom’s SmartSteps platform covering November 2019, a period chosen to avoid distortions from the COVID-19 pandemic and subsequent 5G deployment changes.

Because privacy protections prevented any direct check of individual classifications, the team relied on three forms of indirect validation. The aggregate share of business travelers in the signaling data, 36.9 percent, closely matched the 38.0 percent observed in the survey, a difference of just 1.1 percentage points. Behavioral patterns also aligned with expectations: identified business travelers concentrated in companies, office buildings, industrial parks, and government agencies on weekdays, while non-business travelers gravitated toward scenic areas, shopping centers, and museums, particularly on weekends. Finally, during the China International Import Expo, the number of visitors identified as business travelers at the National Exhibition and Convention Center rose by 127.8 percent relative to the reference period, an event-consistent response the authors describe as additional external evidence that the transferred model captures business-related mobility.

Mapping where these business travelers went revealed a strikingly polycentric geography. The highest densities fell within Shanghai’s inner ring, in established centers of finance, consulting, and information technology, but substantial secondary concentrations appeared in suburban new towns and along major transport corridors, frequently corresponding to industrial parks, development zones, and town-level employment centers. Non-business activities, by contrast, tracked tourism, shopping, and cultural destinations more closely. The outer suburbs showed smaller, scattered clusters near major transport facilities and administrative boundaries. The authors argue that this pattern provides behavioral evidence that Shanghai’s polycentric structure is expressed not only in the distribution of firms and employment but in the actual movements of external business travelers, and they suggest that planning for external connectivity should extend beyond the traditional central business district to suburban employment centers and their transport links.

The origins of these flows showed a dual structure. Most source cities lay in the Yangtze River Delta, forming a continuous zone of intensive interaction around Shanghai, where short distances, dense transport networks, and frequent connections make repeated same-day or short-duration trips feasible. Yet several economically advanced but geographically distant cities also generated substantial flows, concentrated in a limited number of major urban centers, while many less developed and less accessible cities showed weak connections. The authors interpret this as a spatial hierarchy combining an intensive regional hinterland, selected long-distance source centers, and a broad set of weakly connected peripheral areas, with Shanghai functioning both as a regional center and as a node in the national urban network.

The study’s most provocative result concerns industrial similarity. Using enterprise counts from the Nationwide Industrial and Commercial Enterprise Registration Database, the researchers computed cosine similarity between the industrial profiles of each source county and each Shanghai subdistrict, then modeled inbound business flows with negative binomial regression. A linear model suggested a straightforward positive association, with an incidence rate ratio exceeding 2.6. But when a squared term was added, the pattern shifted: the linear coefficient turned negative and the quadratic term became positive and significant, revealing a U-shaped association. Predicted flows reached their minimum at a cosine similarity of approximately 0.24 and rose at both higher and lower similarity levels. The authors are careful to stress that this is a statistical association, not proof that intermediate similarity suppresses exchanges or that dissimilarity promotes them, and they decline to attribute the pattern to a specific industrial mechanism given measurement uncertainty and sample heterogeneity.

To probe which industrial configurations mattered most, the team turned to a LightGBM machine-learning model interpreted through SHAP and SHAP interaction values. The results were unambiguous on one point: inter-industry pairs accounted for approximately 77 percent of total interaction intensity, against 23 percent for intra-industry pairs. Shanghai’s Finance sector showed positive interactions with IT services, Entertainment, and Manufacturing in source counties, while Shanghai’s Manufacturing sector interacted positively with origin-side Finance and Business services. Finance, Real estate, and Business services made the largest cross-industry contributions on the Shanghai side, suggesting that advanced producer services frequently appear in the configurations associated with inbound business flows. Among within-industry pairs, Finance, IT services, and Real estate showed positive interactions, while Manufacturing displayed the strongest negative intra-industry interaction. The authors repeatedly caution that these are modeled statistical associations, not evidence of cooperation, competition, complementarity, or spillovers, and that direct verification would require transaction, investment, supply-chain, or collaborative-project data.

The SHAP analysis also surfaced nonlinear patterns among the control factors. Spatial distance showed a strong decay, with shorter distances associated with more intensive flows especially within 500 kilometers, leveling off beyond that range. Larger source populations were associated with stronger connections but with diminishing marginal gains, while employment size in Shanghai showed no comparable diminishing pattern. Tertiary-industry output had the strongest and most consistent positive association among the economic variables, and exhibition-center provision showed positive SHAP values increasing up to a moderate range before leveling off. In the regression models, high-speed rail connectivity significantly increased business exchange volumes, whereas the airline dummy variable did not reach statistical significance, which the authors suggest may reflect higher costs or less regular flight frequency.

The authors are candid about the limits of their framework. The signaling data cover only domestic mobile users within mainland China, excluding international business travel that may matter greatly in global-city contexts, and the absence of individual-level travel-purpose labels prevents direct measurement of post-transfer classification accuracy. The enterprise-count-based similarity measure may not fully capture functional relationships between regions, and transferability to other cities, periods, or operators would require local recalibration and independent validation. The study also does not quantify cost savings relative to repeated population-wide surveys. Even so, the work demonstrates that a modest behaviorally detailed survey can unlock the purpose hidden inside massive mobility datasets, and that when it does, the picture that emerges is more varied than firm-network studies imply: business ties to a global city thrive both where economies mirror each other and where they differ, and they land in places that formal plans never designated for them.

Subject of Research: Identification of inbound business travelers to Shanghai using integrated survey and mobile phone data and analysis of industrial-structure associations with intercity business flows

Article Title: Inbound business activities and industrial linkages: insights from Shanghai using a big–small data integration framework

Article References: Chen, J., Yan, L., Mu, L., Yu, W., Hu, Y., Cao, Z., & Wang, D. (2026). Inbound business activities and industrial linkages: insights from Shanghai using a big–small data integration framework. Frontiers of Urban and Rural Planning, 4(1), Article 28. https://doi.org/10.1007/s44243-026-00098-1

Image Credits: AI Generated

DOI: 10.1007/s44243-026-00098-1

Keywords: Shanghai, business travel, mobile signaling data, industrial similarity, intercity linkages, big–small data, LightGBM, SHAP, polycentric structure, advanced producer services, negative binomial regression, Yangtze River Delta