Financial Education, Health Scoring and Downstream Behaviour: Evidence from FinTech

How to cite

Federle, Jonathan; Loos, Benjamin; Renkl, Emanuel; Wichert, Sebastian. 2023. "Financial Education, Health Scoring and Downstream Behaviour: Evidence from FinTech." AEA RCT Registry. AEARCTR-0009594. https://doi.org/10.1257/rct.9594-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Jonathan Federle* , Benjamin Loos , Emanuel Renkl , Sebastian Wichert

* Primary investigator

Registry ID
AEARCTR-0009594
DOI
https://doi.org/10.1257/rct.9594-1.0
First registered
Last updated
Status
In development
Trial dates
2023-05-01 – 2023-09-01
Countries
Private
JEL codes
D1, D14
Registry record
http://www.socialscienceregistry.org/trials/9594

Abstract

We conduct a field experiment within the chatbot of a mobile app of a fintech. The goal is to examine the effect of financial education and a financial health score on consumer financial behavior. Experiment participants will be randomly assigned to three groups on an individual level.

Participants in the control group neither receive financial education nor a financial health score. The first treatment group receives financial education. More specifically participants are educated about a standard budgeting rule and how to use the apps' personal financial management features to implement it. The second treatment group receives identical financial education but is further provided with their individual subjective financial health score. Thereby, the financial health score is calculated based on the participant's answers to five questions on their finances. With this experiment we want to answer two research questions: First, is financial education within a chatbot effective? Second, does a financial health score amplify the effect of financial education?

To investigate the research questions, we examine several outcomes and compare them among the different groups. More specifically, we measure the adoption of personal financial management features within the app as well as the spending and saving behavior of the participants. Our first hypothesis is that participants in the first treatment group have a higher adoption of personal financial management features and spend less/ save more as compared to the control group. Our second hypothesis is that the provision of an individual financial health score in the second treatment group further amplifies these effects.

Experimental design

The fintech app used for the field experiment contains a chatbot which can be used by the fintech's clients to ask service related questions. For instance, if a client wants to change his/ her address and does not know how, the client can ask the chatbot to assist him/ her with this task. The recruitment mechanism now works as follows: If a client consults the chatbot and his/ her question is resolved, he/ she will subsequently be asked by the chatbot whether he/ she would like to learn more about financial health:

"One last thing - I'm learning a lot about financial health these days. They say money can't buy happiness... But financial health can definitely improve one's life and prevent hardships. Can I share my insights with you? It just takes 3-5 minutes." (The exact wording might be subject to change depending on editorial requests of the fintech)

Upon answering the question with yes, the participant will be randomly assigned to one of the two treatment groups. In general, the participant is at no point obliged to participate in the experiemnt as he/ she can exit the chatbot any time.
As stated, the experiment will be launched in a staggered release. The control group is composed of clients that consult the chabot and have their question resolved, but are not asked if they would like to learn more about financial health.

Design details

Randomization method
Each user of the FinTech app is assigned a random user id according to the UUID-4 standard upon registration. The treatment/control assignment within our experiment is based on this user id. In particular, we use the Python NumPy RandomState random number generator and initialize the seed function determining the randomization with the aforementioned UUID-4. Based on this initialized random number generator, we assign the users to the Control, Treatment A, or Treatment B group, respectively. A code excerpt outlining the exact mechanism of the randomization is attached in the rubric "Docs & Materials".
Randomization unit
Users are randomized at a user level. See "Randomization Method" for further details.
Clusters
As many user which we are able collect in the experiment time period or 30.000 users, whichever comes first.
Observations
As many user which we are able collect in the experiment time period or 30.000 users, whichever comes first. If the participation rate of app users in the experiment is significantly lower than anticipated, we might fall back to a maximum observation period of 4-6 months and reside with the number of observations that we have collected within this time frame. The exact specification of the time frame depends on the preferences of the project partner.
Arms
The experiment will be introduced as a staggered release by the project partner. We cannot yet determine how many users will finally end up in the control/treatment groups. We, however, plan to assign the Control:Treatment A:Treatment B according to the following ratio: 50:25:25
Primary outcomes
- Actual spending behavior of users in the period following the presentation of a financial health score. Separate evaluation horizons ranging from several days to several months are planned. We expect spendings to decrease following the presentation of a financial health score. - Actual saving behavior of users in the period following the presentation of a financial health score. Separate evaluation horizons ranging from several days to several months are planned. We expect savings to increase following the presentation of a financial health score. - Adoption of personal financial management features as measured by the usage of the features by the respective users in the groups. We expect the adoption of personal financial management features to increase in both Treatment arms.

Cited by

No citation data available for this registration yet.

Keywords