Discretion, Information, and Credit Screening: Evidence from Italian Loan Officers

2026-06-23

2026-07-31

For each vignette, the primary outcomes are measured twice: first after the initial case description, and then again after additional hard information is provided. The primary outcomes are:

1. The respondent’s perceived probability that the credit becomes problematic, defined as recording a payment delay of more than 90 days within 12 months.
2. The respondent’s willingness to let the application proceed beyond initial screening, measured on a Likert scale.
3. The share of the requested amount that the respondent would recommend for approval.
4. The time the respondent would invest in further assessing the request and supporting the credit evaluation, measured on a Likert scale relative to the time usually dedicated to a similar application.

For each outcome, we will analyze both the initial assessment and the updated assessment after additional information is provided.

Secondary outcomes include:

1. The additional information reported by the respondent as most important for any change in evaluation: credit-registry or credit-bureau information, financial-burden information, guarantee status, no relevant change, or other.
2. Heterogeneity in treatment effects by respondent characteristics, including professional experience, main area of activity, type of institution, typical client segment, geographic area, and self-reported approval and default experience.
3. Heterogeneity in treatment effects by vignette type, comparing SME loan applications and residential mortgage applications, and by specific case profile.
4. Differences between initial assessments and post-information assessments across randomized applicant characteristics and randomized hard-information treatments.

The study is an online survey experiment with randomized credit-application vignettes. Respondents are bank employees or credit professionals. After collecting information on respondents’ demographic profile, professional background, credit-evaluation experience, and institutional context, the survey presents three hypothetical credit applications randomly selected from four possible cases. The four cases consist of two SME working-capital loan applications and two residential mortgage applications.

For each vignette, the respondent first sees an initial case description and a summary table describing the application. The respondent is instructed to evaluate the request as they would in the initial screening or first review of a credit application, assuming that only the information shown at that stage is available. The initial vignette information includes both fixed case attributes and randomized applicant characteristics. The randomized initial characteristics are the applicant’s name, which jointly signals gender and geographic origin, age, firm age or job tenure, and the applicant’s relationship with the bank.

After this initial assessment, the respondent receives additional information about the same application. This second-stage information is randomized and includes credit exposure or outstanding debt position, financial burden, and the presence or absence of an accessory personal guarantee. Respondents then answer the same assessment questions again, allowing us to measure both initial credit evaluations and within-vignette updating after additional hard information is provided. A central part of the analysis will examine whether the effects of randomized applicant identity differ depending on the additional hard information provided. In particular, we will test whether information on credit exposure or outstanding debt, financial burden, and guarantee status attenuates, amplifies, or leaves unchanged differences in credit assessments associated with the applicant’s gender and geographic origin. Finally, respondents report which piece of additional information was most important, if any, in changing their evaluation.

Not available

Randomization is implemented by the online survey platform. The platform randomizes which three of the four vignettes each respondent sees, the order in which these vignettes are presented, the applicant characteristics shown in the initial vignette, and the additional hard information shown after the initial assessment.

The unit of randomization is the respondent-vignette. Each respondent evaluates multiple vignettes. Vignette assignment and order are randomized at the respondent level, while applicant characteristics and additional hard information are randomized within each respondent-vignette.

Yes

Around 300 respondents

Around 900 hypothetical credit applications

All respondents will evaluate three vignettes with randomized applicant characteristics and additional hard information