Innovation Across Borders: Experimental Evidence on Artificial Intelligence

How to cite

Faia, Ester; Ferrando, Annalisa; Rariga, Judit; Cullen, Zoe; Baumann, Ursel; Perez-Truglia, Ricardo. 2025. "Innovation Across Borders: Experimental Evidence on Artificial Intelligence." AEA RCT Registry. AEARCTR-0017121. https://doi.org/10.1257/rct.17121-3.1

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Ester Faia* , Annalisa Ferrando , Judit Rariga , Zoe Cullen , Ursel Baumann , Ricardo Perez-Truglia

* Primary investigator

Registry ID
AEARCTR-0017121
DOI
https://doi.org/10.1257/rct.17121-3.1
First registered
Last updated
Status
Completed
Trial dates
2025-11-19 – 2025-12-15
Countries
Germany (Europe)
JEL codes
O3, C83
Registry record
http://www.socialscienceregistry.org/trials/17121

Abstract

We designed an information-provision experiment on the adoption of Artificial Intelligence (AI) technologies. Our survey module will be embedded in the Survey on the Access to Finance of Enterprises (SAFE), conducted by the European Central Bank (ECB). We expect to collect responses from roughly 5,000 firms across 12 European Union (EU) countries. The survey elicits firms’ beliefs about the adoption rates of AI among competitors within their own country and, separately, among competitors in the largest EU economies. We then randomize the provision of information about these two adoption rates. Finally, we measure how this information affects firms’ beliefs about future AI adoption by competitors, domestically and abroad, as well as their own intentions to adopt AI in the future. Additionally, we plan to explore heterogeneity by access to credit. Intuitively, a firm that learns it is lagging behind competitors in technology adoption may be able to respond by investing in adoption if it has sufficient liquidity or credit access, but may be unable to react otherwise.

Intervention

The survey elicits firms’ beliefs about the adoption rates of AI among competitors within their own country and, separately, among competitors in the largest EU economies. We then randomize the provision of information about these two adoption rates. Finally, we measure how this information affects firms’ beliefs about future AI adoption by competitors, domestically and abroad, as well as their own intentions to adopt AI in the future.

Experimental design

The subjects will be randomized between treatment and control groups. The information treatment provides firms with data on the share of firms that have invested in AI among: (1) domestic competitors (defined as firms in the same sector and size class within the same country); and (2) international competitors (defined as firms in the same sector and size class but located in the largest EU economies: Germany, France, and Italy).

Following the standard design of information-provision experiments, we seek to exploit heterogeneity in prior beliefs. Firms that underestimated the share of adopters are expected to revise their beliefs upward, those with accurate priors should show little or no updating, and those that overestimated the share are expected to revise their beliefs downward. The design leverages heterogeneity in both prior beliefs (one for within-country adoption and another for cross-country adoption) to identify the relative strength of within-country versus cross-country diffusion of innovation.

We also plan to explore heterogeneity by access to credit. Intuitively, a firm that learns it is lagging behind competitors in technology adoption may be able to respond by investing in adoption if it has sufficient liquidity or credit access, but may be unable to react otherwise. To measure access to credit, we can rely on survey questions from other modules of the SAFE survey and, where possible, complement these with administrative records, such as (i) balance sheet data from Orbis and (ii) credit data from the European credit registry (AnaCredit).

Design details

Randomization method
Randomization done in office by computer
Randomization unit
The unit of randomization is the firm
Clusters
Around 5,000 firms
Observations
Around 5,000 firms
Arms
50% control and 50% treatment
Primary outcomes
(i) Expectations about competitors' investment in AI (domestically an abroad). (ii) The firm's own intention to invest in AI
Secondary outcomes
In the future, we may be able to explore additional data. First, we may be able to conduct a follow-up survey to assess the persistence of the effects, if any. Second, we might be able to merge the survey data with various administrative sources (e.g., balance sheet data from Orbis). At this stage, it is unclear whether either option will be feasible, and we need to seek for approval in advance. If they are, we intend to update this RCT registration before analyzing any additional datasets.

Cited by

No citation data available for this registration yet.

Keywords