Transfer learning for credit across countries

Transfer learning, domain adaptation and domain generalization appear on nearly 700 recent papers, almost all in AI journals, and only 10 of them concern credit, lending, banking or markets. Lenders entering new countries, and credit bureaus in countries with little data, want to reuse models trained elsewhere. How much transfers, and what breaks?

What prompted this#

Since October 2025, 679 papers carry a transfer-learning, domain-adaptation or domain-generalization keyword (664 in AI journals): "domain adaptation" on 182, "domain generalization" on 178, "transfer learning" on 162, "unsupervised domain adaptation" on 76. Only 10 of the 679 also carry a credit, loan, bank, finance, default, microfinance, emerging-market, fraud or stock keyword.

Questions#

Does a credit model trained in one country, adapted with a few thousand local loans, beat a model trained from scratch on the local data, and how does the answer depend on how different the countries' borrowers and institutions are?
Which features transfer (payment behaviour, mobile phone data) and which do not (income, employment categories)?
Does transferred scoring import the source country's biases against groups that do not exist in the target?

Data that could help#

Loan datasets from several countries (public lending datasets, microfinance data, partner lenders), and borrower features that are available in both.

Next step#

Adapt a credit model trained on one public lending dataset to another country's dataset with small local samples and compare it with local-only training at each sample size.

Related research activity#

The keywords this note follows, in the tracked finance, accounting, economics, management, operations research and AI journals as of 2026-10-26
Keyword	Papers, last 30 days	Rank	Papers, last 36 months
transfer learning	9	#149	721
domain adaptation	14	#67	556
domain generalization	9	#149	476

@misc{scholar4ai-ideas-transfer-learning-for-credit-across-countries, author = {{Scholar4AI}}, title = {Transfer learning for credit across countries}, year = {2026}, howpublished = {\url{https://scholar4ai.com/ideas/transfer-learning-for-credit-across-countries}}, urldate = {2026-10-23}, note = {Journal Research Rankings; data from OpenAlex (CC0); licensed CC BY-NC 4.0} }