GSCU Delinquency Reduction Field Experiment

The research flow divides GSCU members into three sub-populations labeled A, B, and C, which correspond. At the start of
the study, GSCU will randomly assign all members into these three populations, and members within each population will then be randomly assigned to either a “control” or “treatment” sub-group. If the latter, the member will receive the appropriate treatment if pre-determined (e.g. experiencing delinquency) criteria are met.

“A” Population — Members in the “A” population will receive treatments designed to help them successfully resolve their delinquency if they enter delinquency during the study. They will be randomly assigned into four sub-groups: CA, A1, A2, and A3. “CA”
is the control group. “A1” will receive “resolution emails” designed by SPI to encourage successful resolution. “A2” will receive “credit score emails” designed to help them understand the impacts of delinquency on their credit score. “A3” will receive a hybrid treatment including both the “resolution emails” and the “credit score emails.” These treatments will take place outside of GSCU's traditional Account Resolution workflow.

“B” Population — Members in the “B” population will receive treatments designed to help prevent them from entering delinquency during the study. After being randomly assigned into the “B” population, GSCU will identify the 15,000 members of the “B” Population most likely to experience delinquency over the next three months. These 15,000 members will be the “B Target Population” and will be randomly assigned into five sub-groups: CB, B1, B2, B3, and B4. “CB” is our control group. “B1” will receive “prevention emails” designed by SPI to encourage members to prevent delinquency. “B2” will receive letters with a magnet inside with the due date of their GSCU loan and encouraging them to sign up for auto-pay. “B3” will receive marketing emails encouraging them to consider enrolling in GreenState’s pilot “Rainy Day Savings” program, which is designed to help them build their liquid savings balances. “B4” is an omnibus group that will receive the B1, B2, and B3 treatments.

“C” Population, approximately 70,000 members — Members in the “C” population will be encouraged to go through holistic financial counseling offered by Horizons, an Iowa-based financial coaching non profit organization. Similar to the “B Target Population,” after being randomly assigned into the “C” population, GSCU will identify the 15,000 members of the “C” Population most likely to experience delinquency over the next three months, and then sub-divide them into CC, C1, and C2.

2020-09-28

2021-09-27

Incidence of 30 day delinquency
Incidence of 60 day delinquency
Incidence of 90 day delinquency
Amount of GSCU debt liability
Number of late debt payments
Number of skipped debt payments
Ownership of GSCU savings account

Amount of liquid savings in GSCU account

All data will be taken directly from GSCU account data.

GSCU members who held some debt with GSCU were randomized into one of three groups -- A, B, and C.

Members who were randomized into the A group were then randomized into one of four subgroups -- CA, A1, A2, and A3. Each of the roughly 27,000 members randomized into this group will be included in the analytical sample for the "A" interventions.

Members who were randomized into the B group were then randomized into one of give subgroups -- CB, B1, B2, B3, and B4. In each of these subgroups, the 3,000 members at highest risk of delinquency (as determined by the CURise model) will be included in the analytical sample. In groups B1, B2, B3, and B4, only the 3,000 highest risk members will receive the intervention.

Members who were randomized into the C group were then randomized into one of three subgroups -- CC, C1, and C2. In each of these subgroups, the 5,000 members at highest risk of delinquency (as determined by the CURise model) will be included in the analytical sample. In groups C1 and C2, only the 5,000 highest risk members will receive the intervention.

Randomization was conducted in an office using Stata 14.

Randomization occurred at the household level.
Some members lived in the same household as another member whose "home branch" is going to included in a quasi-experiment that tests the efficacy of an automatic savings account program. If a household has any member whose home branch was in the quasi-experiment, then no member of the household would be randomized into one the A, B, or C groups.

No

Roughly 54,000 households.

57,000 members will be included in this study.

AC: ~7,200 members
A1: ~7,200 members
A2: ~7,200 members
A3: ~7,200 members

BC: 3,000 members
B1: 3,000 members
B2: 3,000 members
B3: 3,000 members
B4: 3,000 members

CC: 5,000 members
C1: 5,000 members
C2: 5,000 members