Implementing Helicopter Money: A Randomized Policy Evaluation

This experiment will consist in performing monetary transfers to randomly selected French households. The purpose is to understand if monetary transfers can be used as an economic stimulus. This experiment also helps better understand the role of the banking sector as an agent of intermediation and support for individuals in the context of economic recovery policies.

We conduct a randomized controlled trial to study the impact of an unconditional money transfer on consumption among French households. Participants receive a monetary transfer in the form of pre-paid debit cards; the treatment and control groups vary in whether (and how fast) the unspent value on the card decreases over time, as well as whether participants are subject to a framing treatment. We study the differences in the consumption response to these treatments, and their heterogeneity with observed characteristics.
There are six different groups of participants. The groups differ based on whether and how the unspent value on the card decreases of time, which we call “interest rate treatment groups” (G1, G2, G3). The characteristics of each interest rate treatment group are described below. Each interest rate treatment group contains two subgroups that receive either no framing treatment (F1) or a framing treatment (F2). Participants are drawn uniformly from a specific subset of clients of the French retail bank CIC. This subset of clients has been selected based on characteristics observed by the bank and by the researchers.

G1 participants receive a payment card linked to a newly created account with an initial account balance of EUR 300. The card expires six months after the date when card and instructions have been mailed, at which point any unspent value on the account is transferred to the participants’ main checking account.

G2 participants receive a payment card linked to a newly created account with an initial account balance of EUR 300. The card expires three weeks after the date when card and instructions have been mailed, at which point any unspent value on the account is transferred back to the researchers and is lost to the participants.

G3 participants receive a payment card linked to a newly created account with an initial account balance of EUR 300. The card has a 6-month expiry date, but every week the remaining balance decreases by the following amounts: by 30 euros if the balance is in the interval (200, 300]; by 20 euros if the balance is in the interval [100, 200], by 10 euros if the remaining balance is below 100. Thus, treatment in group G3 approximates a 10 % negative interest rate, at a weekly frequency.

All treatment participants receive a letter with instructions and explanations at the start of the trial, and are contacted by their bank advisor.
Within each of the treatment groups G1, G2, and G3, there are two framing groups, F1 and F2. In the baseline group (F1), the participants are told in the instructions letter that they are free to use the money as the please. In the framing treatment groups (F2), the letter contains an additional paragraph with the framing treatment. This framing treatment explains that, although the participants are free to spend the money as they pleased, they are encouraged to (i) spend quickly, (ii) spend on French-made goods or services, and (iii) spend on items they would not have purchased otherwise, so that the overall increase of their spending and their impact on the French economy is maximized.

During this round, some money will be returned to the experimenters (the weekly deductions in G3, as well as the unspent balance in G2). The entirety of these funds will be redistributed in an additional round several months after the main round; the treatment takes the same form as those in G1, with no framing treatment. The number of treated individuals in this round will depend on the amount of funds returned from the first round.

2022-04-25

2023-05-01

- Total spending on the treatment cards/accounts
- Total spending on all cards including the treatment card, including cash withdrawals
- Total spending on all cards including the treatment card, including cash withdrawals, plus checks and direct debits (“prélèvements”)
- Total spending on all cards including the treatment card, including cash withdrawals, plus checks, direct debits, and wire transfers.
- Total spending on treatment card, by merchant category code (MCC)
- Total spending on all cards incl. treatment card, by merchant category code (MCC)
- Local content spending: average spending share on domestic products across MCCs, weighted by the card spending shares of the above two measures
All variables will be constructed from the available transaction microdata. Baseline specifications will use weekly aggregates, additional results will use daily or monthly aggregates.

Outcome variables are constructed from confidential transaction account data collected and held by the implementation partner (CIC/Euro-Information). Weekly and monthly variables will be aggregated from credit card transaction microdata stored in the table OPECRT. For aggregates that contain all account transactions, these will be merged in from the table OPE, which contains all banking movement in and out of the household accounts. For aggregates that contain more information on direct debits, we will merge in these transactions from the table PLV. Data on accounts balances at the monthly level come from the table GRPCLI. Finally individual information (age, profession …) will come from the TIE table.

- Income, as measured by incoming salary transfers, at monthly level (see primary outcomes)
- Total sum of incoming transfers, at monthly level (see primary outcomes)
- Fraction of spending on online purchases (treatment card only / all cards; see primary outcomes)
- Employment status
- If the response rate on the post-trial survey is sufficiently high, we will also use some of the survey responses as outcome variables. At the time of writing, the exact format of the survey has not yet been approved by the implementation partner, but we aim to include the following outcomes: (1) time to first spending after receiving the card; (2) self-reported spending on domestic produce, internet purchases, regularly purchased items; (3) answers related to the decision process and self-reported impact.

See primary outcomes. Income will be constructed from incoming wire transfers. We may also use supplementary information that the bank has on individuals’ self-reported socio-professional categories. The location of survey response data in the database is not yet known at the time of writing. The data provider is currently in the process of tagging transaction, and if such information is available to us, we will try to use it too.

The experiment will consist of three groups that vary according on the interest rate conditions on the card. Each of the three groups will be further divided into two subgroups, one of which receives a framing treatment.

Potential participants will be drawn from a pool of about 90k households (corresponding to about 106k individuals on which data is available) that are in CIC’s panel. To get to this sample, we have excluded from the initial sample (around 300k household) prior to the draw:
- Individuals that were not easily reachable (permanent address and email not on record; not agreed to online banking)
- Individuals without a debit card, or individuals who had not used their debit card or online banking within the last 20 days
- Individuals outside the age bracket [26,75], individuals who do not have legal capacity or are heavily indebted
- Individuals in families that have a family-owned company account
- Individuals that according to CIC are likely to use another bank for transactions.

We uniformly draw participants from this restricted set. The total number of participants that will be drawn initially is 1,500. 40 of those have been treated in a pilot that was conducted in Oct/Nov 2021 to check the operational implementability (10 in G1, 10 in G2, 20 in G3; half of each group with framing F2). 960 will be treated in the main round of the experiment (40% of them in G1, 30% in G2, 30% in G3; each group equally split between F1 and F2).

The remaining list of 500 may be used in an additional round: 19 weeks after the main round, any money that is being returned to the researchers (from G3 because of the weekly “interest payments”, or unspent balances from G2) will be distributed to another group “G1+” that has the same treatment as G1 (and do not receive a framing treatment). The sample size in this treatment will be chosen to use up the returned money.

We plan to evaluate the trial and present results after the main round, unless take-up is very low. We reserve the right to include data from the additional round in the final results.

The draw was done by CIC, on a computer, under supervision of a notary (“huissier” in French).

Individual

No

-

40 households in the pilot. For the main trial, 960 households have been drawn. The eligibility filter will be again applied to treatment and control participants shortly before the start of the full experiment. As a result, the sample size in the experiment will be slightly smaller than the number of drawn participants.

The number of drawn households are as follows:
Pilot
• 5 households receiving treatment G1 with framing F1
• 5 households receiving treatment G1 with framing F2
• 5 households receiving treatment G2 with framing F1
• 5 households receiving treatment G2 with framing F2
• 10 households receiving treatment G3 with framing F1
• 10 households receiving treatment G3 with framing F2
Main round: treatment groups:
• 20% -> 192 households receiving treatment G1 with framing F1
• 20% -> 192 households receiving treatment G1 with framing F2
• 15% -> 144 households receiving treatment G2 with framing F1
• 15% -> 144 households receiving treatment G2 with framing F2
• 15% -> 144 households receiving treatment G3 with framing F1
• 15% -> 144 households receiving treatment G3 with framing F2
Additional round G1+: number of treated individuals is the largest integer less or equal to the amount of money returned in the main round (in eur) divided by 300. All households receive treatment G1, half with F1, the other half with F2.

The eligibility filter will be again applied to treatment and control participants shortly before the start of the full experiment. As a result, the sample size will be slightly smaller than the number of drawn participants.

Control group:
All households in the sample from which we have uniformly drawn the treated households, minus those that have become ineligible in the meantime.

Power calculations were conducted in 2019 prior to applying for the ANR grant.

Pre-Analysis Plan (20 April 2022)