How sharing is the “sharing economy”? Evidence from 97 Airbnb markets

TL;DR

Analyzing 97 Airbnb markets, this paper finds the platform is highly unequal and increasingly centralized, contradicting “sharing economy” claims. It reports an average Gini coefficient of 0.68, with Black hosts receiving 22% less revenue and women 12% less on average.

Key findings

• Across 97 cities and regions, the average Gini coefficient is 0.68 (median 0.68, SD 0.069); about 10% of hosts receive roughly 50% of market revenue, and in the most unequal cities (Columbus, Austin, Prague) a large share goes to just 1% of hosts.
• Inequality rose over time in all nine cities with data since 2015; regulation did not significantly alter the long-term trend, though San Francisco’s 2018 registration deal caused a temporary spike and drop before returning to growing inequality.
• Black hosts receive on average 22% less revenue, and women 12% less; gaps are larger in some cities, e.g., Black hosts in Cape Town receive 57.9% less and in Rio de Janeiro 47.1% less, while women in Rio receive 32.4% less.
• Black hosts are underrepresented relative to city demographics—e.g., in Cape Town 4.8% of Airbnb hosts are classified as Black versus 38.6% of the population, while 74.4% of hosts are classified as White versus 1.4% of the population, an over 5,300% overrepresentation of White hosts.

Methods

• Uses InsideAirbnb scraped data for 97 cities/regions; the abstract reports 834,722 listings, 513,785 hosts, and 13,466,854 reviews, with 2019 data for current inequality and historic data since 2015 for nine cities.
• Relative revenue is proxied as number of reviews × listing price, summed per host; hosts with no reviews are excluded; inequality is measured via Lorenz curves, Gini coefficients, and power-law fitting.
• Race and gender are classified from host profile pictures using Clarifai (discarding 26% with no face and 22% with multiple faces); OLS regressions use revenue normalized to city average, baseline White male, validated with Genderize (91.5% classified by first name; 92% agreement for women).

Limitations

• The paper states the relative-revenue proxy may bias estimates—e.g., if small/personal hosts receive more reviews, inequality may be underestimated—and assumes profile photos show the actual host; hosts using logos or other photos may be missed.
• The paper states race and gender operationalization is limited: race is socially constructed and not merely skin color, gender is treated as binary, and the analysis is descriptive rather than explanatory, not disentangling direct and indirect discrimination.
• Evident from the text: historic analysis covers only nine cities, and race/gender gap analysis focuses on the 14 cities with the most hosts due to limited classification resources.