Frictions in Mortgage Refinancing

Last registered on July 26, 2021

Pre-Trial

Trial Information

General Information

Title
Frictions in Mortgage Refinancing
RCT ID
AEARCTR-0007990
Initial registration date
July 21, 2021
Last updated
July 26, 2021, 11:20 AM EDT

Locations

Region

Primary Investigator

Affiliation
Universidad de Chile

Other Primary Investigator(s)

PI Affiliation
Northwestern University
PI Affiliation
Stanford University
PI Affiliation
Northwestern University
PI Affiliation
Universidad de Chile

Additional Trial Information

Status
In development
Start date
2020-11-19
End date
2022-07-21
Secondary IDs
Abstract
Mortgage rates have decreased considerably in Chile over the past decade, but mortgage holders rarely refinance their loans. The Chilean National Consumer Service (SERNAC) is the agency of the State of Chile in charge of ensuring the protection of the rights of consumers. SERNAC has been considering using informational treatments to encourage consumers to refinance. However, the type of informational treatment that would be most effective would depend on the nature of the friction preventing consumers from refinancing. First, search may be costly: it may take substantial effort for a consumer to go from bank to bank soliciting offers. Second, switching may be costly: not only might it take mental effort to even start the process of searching for a loan (perhaps because the consumer is unaware even how to) but changing banks from one's home bank could require substantial effort. Third, consumers may have incorrect beliefs over the benefits of refinancing: they may be unaware of the prevailing interest rate in the market, for instance. Broadly, this project asks whether informational treatments can increase the prevalence of refinancing, and if so, what the behavioral frictions are that led to the low likelihood of refinancing.

To answer this question, we partner with a major national bank in Chile. We conduct three pieces of analysis. First, we conduct a survey of bank clients to solicit information about clients' understanding of the refinancing process, how they evaluate loans, and their beliefs over market rates. Second, we conduct a field experiment in which we distribute information about their current loan, the refinancing process, and the market rates to all mortgage holders with the Bank. This consists of approximately 250,000 clients. We randomize the clients into five treatment arms, each giving information about a different part of the process. Finally, we analyze the behavior post-treatment through the lens of a structural model to quantify the behavioral frictions.
External Link(s)

Registration Citation

Citation
Bhattacharya, Vivek et al. 2021. "Frictions in Mortgage Refinancing." AEA RCT Registry. July 26. https://doi.org/10.1257/rct.7990-1.0
Sponsors & Partners

Sponsors

Partner

Type
government
Experimental Details

Interventions

Intervention(s)
We randomize the population into a control or one of five information treatments. The population will receive an email from the Bank with a PDF file attached.

Control Group. A group of people will not get any additional letter through this experiment. They will however receive a standard letter describing the terms of their contract, although at a time pre-set by regulation and independent of the experiment. (All Chileans with mortgages receive this letter.)

Treatment to Change Beliefs of the Possibility of Refinancing. This letter will contain information about the recipient's current mortgage, a statement that alerts the recipient that it is possible to refinance one's mortgage and reduce the monthly payment (and defining what refinancing means), and a link where the recipient could find more information about the terms one can receive. It is on a regular-sized sheet and contains both the logo and the stamp of SERNAC.

Treatment to Alter Beliefs. This treatment arm provides information that alters the bias in the beliefs individuals have about the rates they will obtain in the market. Here, we include the information in the T1 treatment and also add graphical illustrations of where the individual stands in the distribution of prevailing interest rates. This documents (1) a short explanation that different financial institutions will offer different interest rates, (2) an interest rate ``thermometer'' that includes the mortgage holder's current interests and the average prevailing interest rates of refinanced loans that have similar characteristics, and (3) a written explanation of the information in the interest rate thermometer together with a computation that includes annual and total savings for a reduction in the interest rate to the market average. For those who are above the market rate, we show no savings (i.e., we do not show negative savings).

The prevailing market rates are computed using predicted values of a regression of interest rate on various characteristics. In particular, we run a regression of interest rate at the mortgage level on variables like loan amount, term, income, neighborhood (``comuna''), loan type, and whether the loan was a subsidy. Continuous variables are typically binned, and categorical variables are included as separate fixed effects. We then predict the rate using the borrower's current personal and mortgage characteristics, using the month FE for May 2021. The market average will be presented as what a customer would have received on average had they refinanced recently (with fine print outlining the idea behind the procedure).

Treatment to Alter Search Costs. This treatment arm is designed to facilitate search across banks. Since search could be difficult because either (i) customers are unable to compare contracts or because (ii) customers do not even know the process to search, we design a treatment that affects both avenues. First, the letter for this treatment includes the baseline information from above. Second, we include a diagram designed by SERNAC that aims to teach people how to compare different credit offers. Finally, we include a step-by-step checklist for how someone can solicit offers from different banks.

Treatment to Alter Switching Costs. In this treatment arm, we inform the recipient of the Portability Law, which makes it easier to port information between banks. We include the baseline information included in the control and then add a step-by-step checklist for how to change banks.

Full Treatment. This treatment arm includes all information from the previous treatments, aggregated together.
Intervention Start Date
2021-07-21
Intervention End Date
2021-07-22

Primary Outcomes

Primary Outcomes (end points)
The primary outcomes of interest are refinancing behavior, searching behavior, and beliefs about searching and refinancing process.

Primary Outcomes (explanation)
Refinancing Activity. These include outcomes related to refinancing offers and its associated conditions.

- Refinancing Offer: A dummy that equals 1 if the individual received an offer to refinance her mortgage (either in own or other bank), and zero otherwise (measured both through surveys and admin data).
- Number of Refinancing Offers Received.
- Refinancing Conditions Offered: Conditions offered, including interest rate, term, dividend, among others (measured both through surveys and admin data).
- Refinancing Acceptance: A dummy that equals 1 if the individual refinanced her mortgage (either in own or other bank), and zero otherwise (measured both through surveys and admin data).
- Refinancing Conditions Accepted: Conditions accepted, including interest rate, term, dividend, among others (measured both through surveys and admin data).
- Request-Refinance Offer Window: Time elapsed between refinancing request and refinancing offer from the bank (measured through admin data).
- Request-Refinance Acceptance Window: Time elapsed between refinancing request and refinancing approval from the bank (measured through admin data).

Searching Activity. These include outcomes related with the searching process of refinancing opportunities.

- Refinancing Inquiry: A dummy that equals 1 if the individual solicited refinancing her mortgage (either in own or other bank), and zero otherwise (measured both through endline surveys and admin data).
- Number of Refinancing Inquiries.
- Refinancing Rejected: A dummy that equals 1 if the individual solicited refinancing her mortgage (either in own or other bank) but request was not attended, and zero otherwise (measured both through Endline surveys and admin data).
- Number of Rejections.
- Intermediate Search (Hard): A dummy that equals 1 if the individual accessed the CMF Simulator to get estimates of rates online, and zero otherwise (measured through CMF monitoring system).
- Intermediate Search (Soft): A dummy that equals 1 if the individual reports she searched for opportunities of mortgage refinancing or portability during the last 2 months, and zero otherwise (measured through Endline surveys).
- Request to First Response Window: Time elapsed between refinancing request and first response from the bank (measured through admin data)
- Intensity of Intermediate Search: Times the individual entered the CMF Simulator.
- Banks to which individuals requested refinancing or portability
- Banks from which individuals obtained offers for refinancing or portability
- Banks from which individuals accepted offers for refinancing or portability
- Types of criteria used to search for refinancing opportunities in a given bank: Dummies for type of reasons leading individuals to search, including information about lower interest rate in competing banks, recommendation from relatives and friends, marketing, proximity to home, among others.


Beliefs. These include outcomes related to individuals' beliefs about the process of refinancing mortgages.

- Beliefs about Knowledge: A dummy that equals 1 if the individual believes she is well informed about the credit conditions of his mortgage (measured through a 4-point Likert scale).
- Beliefs about own Interest Rate: Dummies that equal 1 if the individual believes the interest rate he pays is larger/equal/lower than the market interest rate for credits with similar characteristics.
- Beliefs about Refinancing: A dummy variable that equals 1 if the individual believes she will solicit refinancing her mortgage in the near future with high probability (measured through a 5-point Likert scale)
- Beliefs about Refinancing Approval: A dummy variable that equals 1 if the individual believes her refinancing request would be accepted with high probability (measured through a 7-point Likert scale)
- Beliefs about potential gains of refinancing: The value of reductions in monthly dividend that the individual would obtain had she refinanced her mortgage (measured in a scale of \$0 to \$100,000 Chilean pesos)
- Beliefs about potential interest rate obtained from refinancing: the interest rate that the individual would obtain had she refinanced her mortgage (measured in a scale ranging 0-10\%).
- Beliefs about timing of refinancing: number of months she believes it takes to refinance/port a mortgage (from searching to approval)
- Beliefs about other mortgage conditions relative to own: proportion of individuals that pay a lower interest rate than her considering individuals with similar mortgages


Secondary Outcomes

Secondary Outcomes (end points)
Secondary outcomes of interest includes knowledge about own credit conditions and satisfaction regarding the information services provided by the bank.
Secondary Outcomes (explanation)
Knowledge. These include outcomes related to the knowledge that the individual has regarding the credit conditions of his own mortgage.

- Knowledge about Interest Rate: a dummy that equals 1 if the individual knows the correct interest rate he pays in his mortgage, and zero otherwise. We can use various tolerances to construct this variable.

Satisfaction. These include outcomes related to the level of satisfaction that clients have with respect to the services provided by own bank.

- General Satisfaction: A dummy that equals 1 if the individual is satisfied with the information provided by the bank regarding her mortgage, and zero otherwise (measured through a 7-point Likert scale)
- Readiness: A dummy that equals 1 if the individual agrees in that the information contained in the cartilla is easy to read, and zero otherwise (measured through a 5-point Likert scale)

Experimental Design

Experimental Design
Experimental Sample. Our experimental sample consists of 254,515 bank clients with active mortgages by May 31, 2021. Of them, 247,311 were not surveyed at baseline (``Non-surveyed Sample (NS)"), while 7,204 were surveyed (``Surveyed Sample (S)").

Stratified Randomization: By June 26, 2021, we conducted a stratified-randomization considering the following set of dummies as stratification variables: gender (1 if the client is female); education (1 if the client has higher education degree); age (1 if the client is below the median age); degree of banking competition (1 if the client resides in a municipality (comuna) where the number of bank branches is below the median number of bank branches across municipalities); 5 quantile dummies of predicted savings (calculated based on the difference between the mortgage holder’s current interests and the average prevailing interest rates of refinanced loans that have similar characteristics) plus a zero predicted savings dummy (1 if predicted savings equals zero); and client's performance (1 if the client is cataloged by the bank as a ``good payer"). We form a total of 192 strata. Then, within each stratum, we randomly assigned each non-surveyed client to the control or one of the 5 treatment groups, with the number of observations per group almost evenly distributed across groups (around 41,220 observations per group). Likewise, within each stratum we randomly assigned surveyed clients to either the control or the ``Full Treatment" group (3,601 and 3,603 observations, respectively).

Implementation. The cartillas are officially delivered on July 21, 2021. A pilot round with 3,000 cases was implemented a week before (week of July 12, 2021). The objective of pilot round was to test whether the technological aspects of the experiments were functioning well (e.g., email system sent out by clients, etc.) and to bring forward potential concerns on the part of receiving clients. No changes were made to the design after the pilot was initiated.
Experimental Design Details
Not available
Randomization Method
Randomization done in office by a computer
Randomization Unit
Randomization is at the mortgage origination level.
Was the treatment clustered?
No

Experiment Characteristics

Sample size: planned number of clusters
We do not cluster the observations, but stratify them across 192 stratas (defined above), and randomize within strata.
Sample size: planned number of observations
Our experimental sample consists of 254,515 bank clients with active mortgages by May 31, 2021. Of them, 247,311 were not surveyed at baseline (``Non-surveyed Sample (NS)"), while 7,204 were surveyed (``Surveyed Sample (S)").
Sample size (or number of clusters) by treatment arms
For the Non-surveyed Sample (NS), we have:
Control Group = 41,220
T1 Group = 41,221
T2 Group = 41,216
T3 Group = 41,207
T4 Group = 41,221
T5 Group = 41,226

For the ``Surveyed Sample (S)", we have
Control Group = 3,601
T5 Group (Full Treatment) = 3,603
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)
Given our sample size, we estimate the statistical power of the experiment, i.e. the probability of rejecting the null hypothesis of no treatment effect. We do it separately for S and NS subsamples. Since the probability of being assigned to each treatment is equal across observations, and these are assumed to be uncorrelated across and within groups, the exercise is common for all treatment-control comparisons regardless to the specific treatment group we evaluate in each subsample. Importantly, we do not have access to baseline data on the distribution of mortgage refinancing for our target population, meaning we cannot simulate the statistical power of the experiment under known mean and variance of our outcome of interest. We thus proceed by assuming - a standardized outcome with mean 0 and variance 1, - a significance level of $\alpha = 0.05$, - P=0.5 (portion of assigned-to-treatment units in each treatment-control comparison), - a take-up rate among assigned-to-treatment units of c=0.3 (i.e. 30\% of those receiving the email with the attached cartilla actually open the email), - a take-up rate among assigned-to-control units of $s = 0$ (no one in the control group receives the email with the attached cartilla), and - a statistical power of 80\%. Given this, our sample size allows us to identify an effect size (Minimum Detectable Effect) of 0.065 standard deviations for the case of treatment-control comparisons in the ``Non-surveyed" sample (i.e, any comparison between the control group and the 5 treatment groups), and an effect size of 0.22 standard deviations for the case of treatment-control comparisons in the ``Surveyed" sample (i.e., control group versus ``Full" treatment group). Finally, it is important to note that our power calculation exercise is based on plain treatment-control comparisons that do not consider regression adjustments for baseline covariates and/or dummies by strata, both of which could potentially increase the statistical power of the experiment. Lastly, our exercise is also sensitive to the take-up rate assumed, which may also vary.
Supporting Documents and Materials

There are documents in this trial unavailable to the public. Use the button below to request access to this information.

Request Information
IRB

Institutional Review Boards (IRBs)

IRB Name
Comite de Etica, Facultad de Ciencias Fisicas y Matematicas, Universidad de Chile
IRB Approval Date
2020-11-23
IRB Approval Number
038
Analysis Plan

Analysis Plan Documents

Pre-analysis Plan

MD5: 9f4ba690053a6d27065ed261d443d947

SHA1: fad7df9b3392325fa817676d29634101c485d092

Uploaded At: July 21, 2021