Home›Research›AI Exposure, Credit & Employment in Brazil
Working Paper · Forthcoming

AI Exposure, Credit Markets, and Employment in Brazil

Leila Aghabarari  ·  Bernardus Van Doornik
Leila Aghabarari — International Finance Corporation, World Bank Group · Bernardus Van Doornik — Banco Central do Brasil
Forthcoming 2026
Abstract. Does AI reshape labor markets not only directly, but through the financial system? We study Brazil, linking occupational AI exposure (GAIA-E) to the Central Bank's credit registry and matched employer–employee records (RAIS). The hypothesis is a transmission chain: as generative AI raises the perceived riskiness or revised prospects of AI-exposed activities, banks adjust credit, and firms in turn adjust employment and wages. Using a difference-in-differences design around the late-2022 generative-AI shock, with rich firm and bank fixed effects, we trace how exposure propagates from technology to credit to jobs. Results forthcoming.
Figure 1 · Transmission chain
How a technology shock reaches local labor markets through the banking system.
🤖
AI Exposure
GAIA-E by occupation, aggregated to firms & sectors
→
🏦
Credit Tightening
Banks re-price / ration lending to exposed borrowers
→
👥
Employment & Wages
Firms adjust hiring, separations, and pay
Conceptual framework · GAIA-E → BCB credit registry → RAIS employment
Figure 2 · Brazil in GAIA — use-case mix vs. peers
Share of AI conversations by use case, Brazil vs. United States, India, and Germany.
Source: data/gaia_countries.csv · uc_work, uc_personal, uc_coursework
Figure 3 · AI exposure by major occupation group
Mean GAIA-E by group. Highlighted groups are heavily represented in Brazilian formal employment (RAIS).
Source: data/gaia_occupations.csv · GAIA-E by group · RAIS-weight highlighting illustrative
Figure 4 · Identification strategy timeline
Pre-period, the generative-AI shock, and the estimated bank response. Treatment intensity = GAIA-E.
Pre-period (2006–2022) 2006 2012 2018 2022 2024 ChatGPT launch · Nov 2022 Estimated bank response (2024) Treatment variable: occupational AI exposure (GAIA-E)
Design schematic · difference-in-differences around the late-2022 shock
Figure 5 · Expected results preview (event study)
Stylized DiD event-study: coefficient on GAIA-E × post by year, with a vertical line at the 2022 shock. Illustrative — results forthcoming.
Placeholder figure · not estimated · shows the expected shape of treatment effects

Data

BCB credit registry (SCR). Loan-level records from the Central Bank of Brazil's credit information system, providing borrower–bank exposures, volumes, rates, and risk classifications at high frequency.

RAIS. The annual matched employer–employee census of Brazil's formal labor market, with occupation (CBO), wages, hires, and separations at the worker–firm level.

GAIA crosswalk. A pipeline mapping GAIA-E from O*NET-SOC occupations to Brazil's CBO occupational classification, then aggregating to the firm and sector level to construct AI-exposure intensity for each borrower.

Methodology

Difference-in-differences. We compare more- vs. less-AI-exposed firms (by GAIA-E intensity) before and after the late-2022 generative-AI shock, in both credit and employment outcomes.

Fixed effects. Bank, firm, sector × time, and region × time fixed effects absorb confounders, isolating differential responses along the exposure gradient. A bank-firm relationship structure lets us separate credit-supply from credit-demand shifts.

Falsification. Placebo event dates in the pre-period and pre-trend tests check that exposed and non-exposed firms moved in parallel before the shock, supporting a causal reading.