Price transparency and remittance provider choice in the GCC-to-Pakistan corridor

The main intervention is the introduction of a high transparency provider into the respondents choice set, making a choice between two hypothetical remittance providers. A standard-transparency formal-provider display shows the provider’s fee, offered exchange rate, and service attributes (arrival time, how the receiver collects the money; how the sender initiates the transaction. A high-transparency formal-provider display also shows the official market-rate benchmark and the calculated final amount received in PKR.

2026-08-23

2026-10-30

The main outcome of interest is the choice of provider.

The two pre-specified primary estimands are:
- In Part 1, the transparency-by-exchange-rate-margin interaction in formal-provider utility; and
- In Part 2, the interaction between formal-option transparency and the formal-minus-informal exchange-rate-margin difference.

Hypothesis 1: Transparency makes choice more responsive to the exchange-rate margin when comparing formal providers.
Hypothesis 2: Transparency makes formal-versus-informal choice more responsive to the formal option's relative exchange-rate margin.

Both are tested two-sided. Holm-adjusted and unadjusted p-values and 95% confidence intervals will be reported across the two primary tests.

Secondary outcomes and estimands include fee and margin sensitivity; preferences and willingness to pay for arrival time, payout method, and send channel; the transparency main effect; transparency-by-fee; average marginal effects; formal-choice probabilities; and indifference thresholds. Pre-specified exploratory heterogeneity analyses cover age, monthly income, actual remittance amount, education, employment, and remittance frequency. Gender and GCC-country or corridor differences are descriptive or exploratory subject to estimability. Other respondent characteristics are post-registration exploratory analyses.

Within-respondent blocked discrete choice experiment using a fixed Bayesian D-efficient design. Six design blocks distribute 48 selected task templates across the sample while each respondent completes eight estimation choice sets. The main intervention is the introduction of high transparency providers within the choice set.

Not available

Within each assigned block, the order of the eight estimation tasks, the eligible repeat-task source position and content, and the A/B placement of alternatives are assigned using reproducible pseudorandom draws, and the realized assignments are stored. In Part 2, the formal option uses the high-transparency display in two of four tasks and the standard display in two. The same respondent-specific task order and A/B placement are used in the web-link and image-card versions.

The respondent is the unit for pseudorandom task sequencing and repeat-task selection; the respondent-task choice set is the unit for the joint pseudorandom A/B permutation of the two transparency conditions

No

Approximately 1,000 completed remittance sender respondents, contributing up to 8,000 real estimation choice sets (each respondent completes 8 real choice tasks). Approximately 1,000 additional repeated choices are collected only for quality control.

9,000 estimation choice sets completed, including 8,000 real choice sets and 1,000 additional repeated choices collected for quality control

All respondents get 50% control (standard transparency providers) and 50% treatment (high transparency providers) in both Part 1 and Part 2.

Power is based on 1,000 joint Monte Carlo replications of the current selected six-block templates. The same 1,000 respondents complete four part 1 and four part 2 real choice tasks in each replication; the ninth displayed repeat is excluded. Choices are generated at the design-prior means. The assumed transparency-by-margin coefficient, TM = -0.2, is used for design search and simulation. It represents an additional utility change of -0.2 when the exchange-rate margin worsens by one percentage point under high transparency and does not constrain estimation. Part 1 is estimated by the fixed conditional-logit equivalent. Part 2 is estimated as a formal choice with the hawala-like alternative-specific constant and block fixed effects. Both use respondent-clustered sandwich standard errors. The registered Holm correction is applied to the two primary transparency-by-margin p-values within each replication. At the prior mean of -0.200, Holm-adjusted empirical power is 0.142 in part 1 and 0.065 in part 2 (Monte Carlo standard errors 0.011 and 0.008); unadjusted power is 0.218 and 0.117. The MDEs were calculated from 1,000 joint simulations over a grid of effect magnitudes, setting both primary interactions to the same negative value at each point. Interpolation at 80% power gives unadjusted MDEs of 0.487 and 0.763 and Holm-adjusted MDEs of 0.522 and 0.764 for parts 1 and 2, respectively. Thus, the current sample is not conventionally powered for either interaction at the design-prior mean. The estimands remain pre-specified and will be reported with confidence intervals and an explicit statement of limited precision. The repeat contributes no power.