A Lab-in-the Field Experiment on Incentivized Group Leaders in Joint-Liability Group Loans

We design two versions of a Microfinance Lab Game: The Investment Game and the Repayment
Game. The Investment Game corresponds to a lending scenario with ex-ante moral hazard, the
Repayment Game corresponds to a lending scenario with ex-post moral hazard. These games imitate
the investment, repayment and monitoring decisions that microfinance clients face in reality. For both
games, we design a control setting, where individuals are subject to joint-liability under symmetric contracts, and a treatment setting,
in which individuals are subject to joint-liability under asymmetric
contract terms. The asymmetry refers to the interest rate that individuals are expected to pay.
Specifically, the asymmetric contract terms specify that the group leader repays the loan with a lower
interest in case of all group members repaying the loan. The lower interest rate is based on the
argument that one member thereby becomes an optimally incentivized lead-borrower, who is expected
to increase monitoring efforts and thereby reduce moral hazard of all group members, which also has
long-run implications for the bank’s profits and overall social welfare.
Together with the Bolivian microfinance institution IDEPRO IFD, we will invite their clients to
participate in the game sessions, which will last about 2 hours.

2024-11-12

2024-11-27

Final Moral Hazard Action
○ Ex-ante moral hazard in the Investment game
○ Ex-post moral hazard in the Repayment game

Mechanisms:
○ Monitoring decision
○ Project choice decision before being monitored (Investment game)
○ Repayment decision before being monitored (Repayment game)

Investment Game:
In the Investment Game, two (randomly matched) players form a joint-liability group and receive a
loan from the bank (MFI) which they have to repay with an interest. Each player has to decide
whether to invest the loan in a safe project or a risky project with a private benefit. In case the project
is successful, the earnings are such that the loan with interest can be repaid. There is asymmetric
information in that the MFI cannot observe behavior of the borrowers. However, repayment can be
enforced, conditional on the project being successful. That is, the MFI can observe whether the
project was successful. In case only one player’s project is successful, the successful player is
enforced to partially cover the group members’ outstanding loan. In case both projects are not
successful, the group defaults. Within joint liability groups there is also no perfect information, i.e.
members do not observe each other’s investment decisions. However, players can monitor each other,
which is costly, but ensures that the monitored player can only invest in the safe project.
In the groups assigned to the control setting, both players face identical interest rates. In the groups
assigned to the treatment setting, one player is randomly assigned as the group leader, who faces a
lower interest rate, conditional on both projects being successful.
Each group will play four independent rounds of the investment game. Decisions will be indicated by
the players through game cards (an investment card and a monitoring card), which are unobservable
for other group members, but will be recorded on tablets by trained enumerators. In terms of
decisions, we focus on the investment choice and the monitoring decision. These result in a final
investment action that may or may not constitute ex-ante moral hazard.

Repayment Game:
In the Repayment Game, two (randomly matched) players form a joint-liability group and receive a
loan from the bank which they have to repay with an interest. The loan is automatically invested in a
project that can be successful (high outcome) or fail (low outcome). In case the project is successful,
the players can repay their loan and interest to the bank. In case the project is not successful, they
cannot repay. Before learning about the project outcome, players individually decide whether to
reveal an eventual positive outcome, and thus repay their loan or not in case their investment is
successful. In case both players repay, both pay only their share of the loan. In case only one repays,
the repaying player automatically pays the full outstanding loan, including the share of the group
member. In case none repays, the group defaults. There is asymmetric information in the sense that
the MFI cannot verify project outcomes. However, the loan has a dynamic incentive, i.e. whenever the
full loan is repaid (either by two players jointly or by one player), both have access to another loan in
the next round. In case of default, both will not receive another loan in the next round. Within joint
liability groups there is also no perfect information, i.e. members do not observe each other’s project
outcomes. However, players can monitor each other, which is costly, but ensures that players cannot
“lie”, i.e. repayment is enforced, conditional on the player’s project being successful.

In the groups assigned to the control setting, both players face identical interest rates. In the groups
assigned to the treatment setting, one player is randomly assigned as the group leader, who faces a
lower interest rate, conditional on both players repaying their loan.

Each group will play a minimum of 2 and a maximum of 4 rounds of the Repayment Game,
conditional on repayment in the previous round (dynamic incentive). Decisions will be indicated by
the players through game cards (a repayment card and a monitoring card), which are unobservable for
other group members, but will be recorded on tablets by trained enumerators. In terms of decisions,
we focus on the repayment choice and the monitoring decision. These result in a final repayment
action that may or may not constitute ex-post moral hazard.

Hypotheses
The hypotheses to be tested are as follows:
Investment Game:
H1) The rate of risky/low effort investments (i.e. ex-ante moral hazard), after investment changes due
to monitoring, is lower in groups with asymmetric contract terms than in groups with symmetric
contract terms.

Repayment Game:
H2) The rate of hidden positive outcomes (ex-post moral hazard), after revelation changes due to
monitoring, is higher in groups with asymmetric contract terms than in groups with symmetric
contract terms.

Econometric Specification
We will use the following econometric specification (logit model) to test the hypotheses at the
individual level. where the outcome is specified as the moral hazard action observed for individual i of group g in
round r. Standard errors will be clustered on the individual level. is a binary variable equal to one for
individuals that were assigned to a group with asymmetric contracts.
We expect to be significantly smaller than 0 (<0) under asymmetric contracts.

Assignment as group leader in the treatment settings will be done through the choice of chairs in the
lab, which in turn will be randomly assigned. For the Investment and the Repayment Game, two
groups of two will play simultaneously and sit on chairs (1-4). Chairs 1 and 2 will be the group
leaders in the Investment Game; chairs 1 and 4 in the Repayment Game.

The games will be played over a period of a maximum of 3 weeks. 4 sessions will be running in
parallel with 2 sessions under the control setting and 2 sessions under the treatment sessions (in
different rooms). Two groups of two players will play simultaneously the Investment first and the
Repayment game second within each session. The unit of randomization for the treatment condition is
the group of four, i.e. both groups of two players will play the identical game version. Playing in
groups of 2x2 guarantees anonymity of the players as well as plausible deniability for moral hazard
actions.

The unit of randomization for the leader in the asymmetric treatments is the individual player.
Matching of the two players is done randomly and depends on show up date and time, i.e.
participation in the games is conditional on participants’ availability and assignment to groups
(treatment or control) is done on a rolling basis conditional on arrival time.

No

We plan to conduct the experiment with a sample of 400 participants in total and all participants play
both game versions. This means that the Investment Game and the Repayment Game in each
treatment condition will be played by 100 groups (“clusters”) in each treatment condition (see Table
1).

Investment game: 1600 Repayment game: min 800, max 1600

Investment game:
Asymmetric treatment: 100 groups à 2 players
Symmetric treatment: 100 groups à 2 players

Repayment game:
Asymmetric treatment: 100 groups à 2 players
Symmetric treatment: 100 groups à 2 players

The number of individuals was chosen to be able to detect effect sizes of about 0.25 SD (small effect sizes) for the individual level outcomes. We use the following formula to calculate MDEs: where is the MDE, is the critical value in the t-distribution for the significance level , is the critical value in the t-distribution for power (i.e. is the chance of a type II error), is the share of participants allocated to the treatment condition, is the variance of the outcome variable (of the control group), is the total sample size, is the cluster size (i.e. in the case of individual outcomes the number of observations per individual; in the case of group-level outcomes, the number of observations per group), and is the intra-cluster correlation coefficient. Since all outcome variables are binary, the variance is simply This value is naturally restricted to a maximum of 0.25. Table 2 in Pre-Analysis-Plan displays the minimum detectable effect sizes for binary outcomes at the individual level (i.e., moral hazard (yes/no) assuming an intra-cluster (i.e. intra-individual) correlation coefficient of 0.6, as calculated from the experimental data from Carli et al. (2024). With a minimum of 800 round#individual observations, we are able to detect effects as small as 12.5 percentage points corresponding to 0.25 SD. If we can recruit 500 participants (and hence have a minimum of 1000 round#individual observations per game version), the MDE decreases to 0.22 SD, if we can only recruit 300 (600 round#individual obs.) or 200 (400 round#individual obs.) participants, the MDEs increase to 0.29 and 0.35 SD respectively.