创新无国界？技术扩散的地理学

Innovation without borders?The geography of technologicaldiffusionUrsel Baumann, Zoe Cullen, Ester Faia,Annalisa Ferrando, Ricardo Perez-Truglia,Judit Rariga Abstract How well does innovation diffuse across geographic boundaries?To shed light on thisquestion, we present a large-scale field experiment involving 3,300 firms across twelve EuropeanUnion countries.We elicit firms’ perceptions of the share of similar firms in their owncountry that had invested in artificial intelligence (AI), as well as the corresponding shareamong similar firms in Germany, France, and Italy. We randomly provide half of the samplewith accurate information about both domestic and foreign AI investment.We show thatfirms substantially underestimate competitors’ current AI investment, both domestically andabroad, and that they update their expectations about competitors’ future AI investmentin response to the information treatment. The treatment also causes a statistically significantincrease in firms’ own expected AI investment rate. We find strong strategic complementaritieswithin borders: a 1 pp increase in the expected share ofdomesticpeers investing in AI raisesa firm’s own expected AI investment rate by 0.570 pp. These complementarities are absentacross borders: the effect of an increase in the expected share offoreignpeers investing inAI on a firm’s own expected AI investment rate is statistically insignificant.Overall, ourevidence shows that innovation diffusion and strategic complementarities in AI investmentare much stronger domestically than internationally. JEL Classification:O33, D22, C93, L21.Keywords:Innovation diffusion, artificial intelligence, field experiment, survey data. Non-technical summary The diffusion of new technologies across firms is a central driver of productivity growth.Despite the rapid expansion of artificial intelligence (AI) as a general purpose technology, itsadoption remains uneven across firms and countries, including within integrated economic areassuch as the euro area. Understanding why AI investment remains concentrated in some firmsand locations while lagging in others is essential for both economic research and policy design. This paper examines whether competitive pressure from peers influences firms’ decisions toinvest in AI, and whether this pressure differs depending on whether competing firms are locateddomestically or abroad. To address these questions, we designed a large-scale field experimentintegrated into the Survey on the Access to Finance of Enterprises (SAFE), a regular firm-level survey jointly conducted by the European Central Bank and the European Commission.The experiment covers approximately 3,300 firms across twelve euro area countries and wasimplemented in the fourth quarter of 2025. The experimental design leverages the fact that firms form beliefs about the extent of AIadoption among their competitors, and that these beliefs often deviate substantially from reality.Initially, the survey elicited each firm’s perception of the share of comparable domestic firms- defined as firms of similar size and operating in the same sector - that had invested in AIby mid-2025.It also collected estimates of the corresponding investment rates among firmsin the three largest euro area economies (Germany, France, and Italy).Subsequently, half ofthe surveyed firms were randomly assigned to receive accurate, factual information about theseinvestment rates, drawn from the previous SAFE survey round. After receiving the information,firms were asked to update their expectations about future AI investment among competitorsand to report the share of their own planned investment they intended to allocate to AI overthe coming twelve months. The findings highlight three key results. First, firms systematically underestimate the extentto which their competitors have invested in AI, both domestically and abroad. On average, thegap between actual investment rates and firms’ prior beliefs exceeded twenty percentage pointsand was skewed towards underestimation across nearly all countries.Moreover, firms tendedto misperceive their country’s relative position within Europe, systematically perceiving AIinvestment as more concentrated in the three largest economies than the data suggest.Thisdistortion persisted even when their own country’s AI adoption rate was above the euro areaaverage.Second, providing accurate information meaningfully shifted firms’ beliefs.Firmsthat received factual signals revised their expectations regarding competitor AI investment ina manner consistent with Bayesian learning. The magnitude of these belief revisions correlated directly with the gap between the information provided and firms’ prior beliefs. The estimatedweight that firms placed on the new information aligns with findings from similar information-provision experiments conducted in other contexts.Third, and most importantly, the paperidentifies how updated beliefs translate into changes in investment plans.