Information Disclosure and Demand Elasticity of Financial Products: Evidence from a Multi-Country Study

This study tests the effectiveness of behavioral-based disclosure formats. Around 1,700 individuals from Mexico and Peru chose among loans and savings accounts presented in different formats, including a simplified key facts statement (KFS) and current marketing brochures. The study finds that the price elasticity of loans is -1.04 using brochures and -3.19 using the simplified KFS, with smaller effects for savings products. Finally, while financial literacy is correlated with better decision-making, the effect of the disclosure format for loans is about three times as large as that of financial literacy. More importantly, the KFS helps financially illiterate individuals relatively more.


Introduction
Many consumers in financial markets find out about the characteristics of products exclusively from providers, even when their staff has better information about the cost of the product and the incentives between customers and staff are misaligned. It is thus important to ask whether firms are successful at providing information and ultimately, whether consumers are well-informed.
Two crucial pieces of evidence, however, should give us pause. First, financial markets are characterized by substantial price dispersion (see, for example, Hortaçsu and Syverson, 2004, Stango and Zinman, 2016, Giné and Mazer, 2017and Zinman, 2015 for a review). 1 While the variation in prices may be driven by riskiness or the transaction costs involved in servicing different customers, Stango and Zinman (2016) and Giné and Mazer (2017) find that the same individuals in the same months are offered credit cards and credit or savings products, respectively, with substantially different interest rates.
Second, there is little comparison shopping, even when customers face substantial price dispersion. Woodward and Hall (2012), for example, show that mortgage borrowers overpay at least USD 1,000 by shopping from too few brokers.
In rational search models, consumers expand their choice set up to the point where the benefit of doing so is equal to its cost. But even when choices are provided at no additional cost, comparison frictions may still cause consumers to place more weight on non-financial factors like brand loyalty or non-professional advice from friends and family to the detriment of the cost of the product (Bertrand et al., 2010). In addition, the decision of which product to contract may involve unfamiliar concepts especially to individuals with limited financial capabilities Mitchell, 2011, 2014) and as a result, financial consumers may not necessarily choose the most cost-effective product or the one most suitable to their needs (see for example, Gross and Souleles, 2002; Choi et 1 Table 1 in Giné and Mazer (2017) reports summary statistics for estimated costs and yields of credit and savings products, respectively, offered by a sample of financial institutions in Mexico and Peru. The total annual cost of credit including usage fees ranges from 22.1 percent in Peru to over 225 percent in Mexico. The total annual yield of a transaction account also including usage fees ranges from -14.5 percent in Mexico to 0.8 percent in Peru.

4
OA4 contains two examples of market-designed key facts statements in Peru. 2 The APR, for example, appears around the center in the 10 th row for one institution but in the top left corner for the other. Other terms also appear in different places, making the comparison between both products difficult.
This study seeks to understand the role of disclosure formats in facilitating comparison shopping for savings and credit products by low-income consumers in Peru and Mexico. These countries were chosen because they both have similar levels of financial access but have de jure regulations with different transparency requirements.
According to the market conduct index published by the Economist Intelligence Unit's 2014 Global Microscope Index and Report, Peru is ranked second while Mexico is ranked 25th (middle of the sample of countries). 3 We implement a laboratory experiment in which low-income consumers were assigned a profile and then incentivized to choose the product that best fit their needs from among 5 or 10 products. In each round of decision-making, information about the products was presented in a different format, including current marketing materials gathered from financial institutions during sales visits by actual consumers and a simplified key facts statement (KFS) designed using behavioral insights to facilitate comparison shopping.
Our results show that the simplified KFS with its standardized format significantly improves consumer decision-making compared to the marketing materials currently provided by financial institutions. The effects are however much stronger for credit than savings products. The probability of choosing the cheapest loan increases from 42 percent using the marketing materials to 65 percent using the simplified KFS but it only increases from 32 percent to 34 percent for savings accounts. One reason is perhaps that individuals may not care about the total yield of the savings account and focus instead on other characteristics. After all, a consistent finding from the literature on savings is that the behavioral response to changes in the price of saving is not large (see Hastings et al., 2013 for a review). Alternatively, subjects evaluating savings products may not have had enough information as the lack of impact is concentrated in Mexico where the simplified KFS did not contain the total yield earned by the savings account.
We also find that transparency increases price elasticity. The price elasticity of credit products is -1.04 using brochures and -3.19 using the simplified KFS, that is, about three times as large. For savings products, the price elasticity is 0.02 using brochures and 0.03 using the simplified KFS, and the difference is not statistically significant. In addition, non-price factors such as the (random) order in which savings products are presented to consumers matter more when brochures are used rather than simplified KFS, consistent with the idea that transparency allows individuals to focus on the price. Finally, we show that financial education is correlated with better financial decision-making, but for credit products, the effect of the disclosure format is about three times as large as that of the effect of financial literacy. More importantly, the simplified KFS is particularly useful to financially illiterate individuals as financial literacy increases the price-elasticity of credit products by 58 percent when using brochures but only by 7 percent when using the simplified KFS.
These results, therefore have relevant implications for how government-mandated information should be presented and its potential to influence choices and competition in consumer finance markets. This paper contributes to the literature in household finance and other fields that study the impact of disclosure regulations in various consumer markets (see Ben-Shahar, 2011 and Dranove and Jin, 2010 for a review). The evidence on the effectiveness of disclosure regulations is mixed, and because there may be factors other than the disclosure regime that change at the same time, it is hard to attribute changes in behavior to the disclosure regulation alone. In contrast, in our controlled setting the same subjects make decisions using the same information presented with a different format, allowing us to make causal statements about the effectiveness of the disclosure treatment.
Our evidence comes from the laboratory rather than the field. The advantage is that subjects are rewarded to pay attention when making decisions. This increases our 6 statistical power considerably and may explain the magnitude of the price elasticities found. 4 The disadvantage is that the environment may be artificial as individuals typically face competing demands on their time and attention when making decisions. In the field, complexity can lead to inaction, as shown by Bettinger et al. (2012) in the context of college financial aid applications, by Hastings and Weinstein (2008) that study school choice and by Barghava and Manoli (2015) that study benefits of simplification in the take-up of the Earned Income Tax Credit. More related to our context, in Adams et al.
(2016), for example, holders of low-yielding savings accounts were given information about higher-rate paying products, a form that enabled simplified switching and a reminder about their low rate. About 90 percent of study participants failed to take any action when it was in their interest to do so. Perhaps more troubling, they did not even recall receiving or reading such information. Similarly, Ponce et al. (2017) find muted effects from information disclosures in the Mexican credit card market.
Another advantage of field experiments is that they allow researchers to observe not only consumer responses, but also how firms respond to greater disclosure requirements.
Duarte and Hastings (2012) evaluate a change in government disclosure in Mexico's privatized social security system and find strong evidence that firms find ways to undermine the effects of disclosure reform by altering their fee structures. Anagol and Kim (2012) also find evidence that firms respond to disclosure policy by altering products to maintain lack of clarity in pricing. 5 The paper finally contributes to the literature documenting consumer financial mistakes and the role of disclosures in preventing them. Hastings and Tejada (2008) show that presenting the cost of a financial product in amounts instead of percentages allows people to choose better products and to focus on other characteristics like fees. Thus, echoing the findings here, minor changes on how information is presented can have significant effects on decision-making. Related, Bertrand and Morse (2011) show that 4 Beshears et al. (2013) andChoi, Laibson, andMadrian (2010) also conduct laboratory experiments that vary the presentation of investment fees while holding other fund characteristics constant to test whether making fees less shrouded changes fund choice; both studies find however, little evidence to suggest that changing the framing of fees has a large impact on investor decisions. 5 For evidence of responses to disclosure from another industry see Newell et al. (1999) for purchases of appliances. 7 disclosing the cumulative costs of payday loans in amounts (rather than percentages) significantly reduces the demand of such loans.
The remainder of the paper is organized as follows. Section 2 presents the experimental design and the different treatments; Section 3 reports the empirical strategy, Section 4 presents the results and Section 5 concludes.

Experimental Design
Individuals from around the capital city of Peru and Mexico were invited to participate in the experiment to test different disclosure forms. During recruitment, individuals were told that they would earn money making decisions but no details were provided about the nature of the decisions. Experimental sessions were conducted in 2013 in Mexico and 2017 in Peru. They took place in a room set up in a way to ensure that communication between subjects was not possible. A total of 600 subjects in Peru and 1,071 in Mexico participated in 57 sessions (10 sessions for each product in Peru and 20 and 17 for credit and savings, respectively, in Mexico), with around 30 subjects per session (see Table 1 for details). Prior to the sessions, a subset of participants in Mexico received SMS and live calls with tips about the terms that were important for financial decision-making. In particular, prospective participants in credit sessions were told to verify the total amount to be paid, including interest payments, commissions and insurance premia. Participants of savings sessions were told to choose the accounts offering the highest yield. Online Appendix OA2 contains the scripts to the live calls and the text of the SMS. While there is an extensive literature on messages as reminders (see, for example, Karlan et al., 2016) here we test messages as a way to disclose information.
Subjects only participated in one experimental session that lasted between 1.5 and 2 hours. Each session started with a 20-minute survey, then three rounds of decisionmaking in Peru and 5 in Mexico where subjects were instructed to choose the product that best fit their needs, followed by an end-of-session survey. The initial survey included questions on demographic characteristics, knowledge and preferences of financial institutions, factors that affect subjects' financial decisions, and financial literacy. 8 Table 2 presents the characteristics of participants. Participants were stratified by gender and they vary by education and occupation. Although Mexico is a richer country, the Peruvian sample has a lower proportion of low-income participants (NSE C-or D).
For this reason, monthly household income is slightly higher in Peru (USD 641 in Mexico compared to USD 783 in Peru) and participants seem on average more educated.
Participants in Peru are also more familiar with financial institutions and report higher ownership of savings accounts and credit cards. In addition, less than one-third of participants in Peru report comparing more than one product when they last contracted a savings account or loan. When we correlate a dummy for comparing different products against individual characteristics we find that richer individuals (as per their socioeconomic status), those with internet at home and those that are familiar with banks are more likely to engage in comparison shopping. 6 While more than half tend to view the staff and marketing materials as the primary source of information about financial products, less than 15 percent of participants in either country is familiar with a key facts statement. These individuals appear to be poorly equipped when deciding among financial products (Lusardi and Mitchell, 2011) and thus are good candidates for the focus of this study. We note that there is within country variation in the levels of education and financial literacy, a feature that we will exploit later when comparing the difference in the probability of choosing the right financial product for a given participant facing different disclosures designs to that of participants with different levels of financial literacy facing a given format. Online Appendix Table OA1 regresses our   preferred proxy for financial literacy, which takes value 1 if the participant answered a question about interest rate correctly, against other individual characteristics. 7 Not surprisingly, for the pooled sample in column 3 financial literacy is correlated with household income, education and usage of financial products. It is also correlated with 6 Data on product comparisons were only collected in Peru and therefore are not reported in Table 2. 7 The interest rate question used is a simplified version of the one from Lusardi and Mitchell (2011): "If you deposit 100 pesos / soles in a bank account that charges you nothing and guarantees you a yield of 2% per year, how much would there be in the account by the end of the year, if no deposits or withdrawals are made?" Possible answers are: (a) Over 102. (b) Exactly 102. (c) Less than 102. (d) I don't know. (e) I prefer not to answer. Lusardi and Mitchell (2011) use the timeframe of 5 years instead of one year. being a male, although it is not always him who makes financial decisions in the household.
Following the survey, the experimenter explained the rules for decision-making to all participants in Spanish. In each decision-making round, subjects were first provided with a sheet to mark their decisions. They then were given 10 minutes to record the three best products on the sheet. 8 Sheets were then collected after each round by an assistant and inputted into a computer to calculate payouts for the round. After the end-of-session survey, subjects were paid a show-up fee of 200 pesos (USD 16) in Mexico and could win 100 pesos (USD 8) more depending on the number of correct answers. In Peru, they were paid similar amounts using a voucher for a family meal in a popular fast food restaurant.

Task
In each round, subjects received information about 5 or 10 products, each offered by a different institution. The terms of each product were simulated using the dispersion of values in the market. No pair of participants received the same combination of products. Participants were instructed to choose the best product in accordance to a profile randomly assigned to them at the beginning of the session. 9 Half of the participants were randomized into one profile and the other half into the other. In credit sessions, all participants were told that they were going to acquire a 12-month loan with monthly installments of 10,000 pesos (USD 800) in Mexico and 1,500 Peruvian soles (USD 450) in Peru. In Mexico there was only one profile that mentioned that every monthly installment was made on time. In Peru, half of the participants were randomized into another profile where all monthly installments but one were paid on time. Put differently, there was one missed installment but paid in full in the next installment. In savings sessions, participants were told that they had a fictional endowment to be deposited into a savings account of 1,000 Peruvian soles (USD 300) in Peru and 5,000 pesos (USD 400) in Mexico. Savings Profile 1 mentioned that each month participants would make two deposits and two withdrawals of 100 Peruvian soles (USD 30) in Peru and 250 pesos (USD 20) in Mexico each and two balance inquiries at a teller of the financial institution. Savings Profile 2 in Mexico had no monthly activity, while in Peru it was similar to Savings Profile 1 in that one transaction (withdrawals, deposits and balance inquiries) would be made instead of two. The balance inquiry would be made at an ATM instead of at a teller.

2 Treatments
In recent years, both Mexico and Peru have developed a regulatory framework to supervise and promote the use of financial services. Mexico enacted a law similar to the U.S. Truth in Lending Act of 1968 in 2009 that also requires financial providers to disclose the APR and APY. Peru enacted disclosure regulation in 2005 and in 2012, which, similar to the regulatory financial transparency regime in Mexico, also defines the criteria for the determination and definition of interest rates, fees, charges and yieldsincluding methods for calculating the total effective costs and rates for credit and savings products. Current regulation requires financial institutions to disclose information to consumers through brochures, key facts statements, webpages, ATMs, and verbally at the branches.
While disclosure regulation of most countries dictates what terms should be disclosed and how they should be calculated, the actual design of the forms is typically left to the financial providers. The goal of the experiment is to test alternative disclosure formats to the ones developed by the financial industry.
Each session in Peru had three rounds and in Mexico 5 rounds, each with a different disclosure format. In Peru, the first disclosure treatment used marketing materials such as brochures, amortization tables, and simulations that were offered to 11 prospective clients when shopping for financial products at the time of the experiment. 10 These materials combined pictures with information about the terms, but each institution had its own design, making comparisons across similar products difficult. The second disclosure treatment used key facts statements (KFS) that institutions were required to give customers after contracting a product. The SBS regulated the minimum number of terms that had to be disclosed, but the design and whether to show the terms in fine print was left to the financial institution. As a result, these market-designed KFS had different layout of information, again making comparisons across products difficult. The third and final disclosure format used a standardized key facts statement designed jointly by SBS and us. This format presents the more relevant information in the top right corner using a large font and because the information is standardized, a given term will always be in the exact same place for every institution thus facilitating comparability. In Mexico, the first disclosure treatment also used brochures. The second disclosure treatment used a standardized key facts statement designed jointly by CONDUSEF and us that is similar to the standardized KFS used in Peru. The remaining treatments used comparative tables with either 5 or 10 products that varied the number of financial terms presented. The complex treatment presented information for 5 products with 8 terms for credit and 12 terms for savings (Complex 5). The simple treatment also presented information for 5 products, but with 4 terms for credit and 3 terms for savings (Simple 5). Finally, the long, simple treatment presented the same terms as the Simple 5 treatment just described but provided information about 10 different products (Simple 10). Online Appendix OA4 contains examples of all the disclosure formats used in the experiment.
The order in which formats were presented to participants was randomized in each session to avoid learning effects. Given the objective of comparing the performance of different formats, all materials had to have key information about the APR / YPR and user fees to make the informational content comparable across formats. This meant that terms of the product had to be added to the materials if these were missing in the original one, which was typically the case for brochures.

Econometric Analysis
To examine the effectiveness of different disclosure formats for loans and savings products, we first look at the impact of the different formats on the probability of choosing the best product. We run logit regressions using data from Peru and Mexico separately and then combined. For Peru we use the following specification: where ℎ is a dummy variable that takes value 1 if participant i in round j and session k chose the cheapest credit product or the savings product with highest yield given the profile assigned. and are dummy variables that take value 1 if participants in round j and session k were given the standardized KFS or the marketdesigned KFS, respectively. The omitted treatment is the one that uses marketing materials such as brochures. takes the value of 1 if participant i in session k was assigned to Profile 1 described in the previous section. The vector of characteristics includes whether participant i in session k is a male, whether he or she has postsecondary education, age and age squared (divided by 100) and our proxy for financial literacy. All specifications include round fixed effects and standard errors are clustered at the participant-session level.
In Mexico, selected participants received either a phone call or a series of SMS messages on behalf of CONDUSEF one or two days prior to participating in the experiment. These phone calls and text messages contained simple information on key financial terms that were used in the disclosure formats and would help participants select the best product. Online Appendix OA2 contains the scripts of the calls and messages. In addition, we follow the literature (Hasting and Tejada, 2008;Gigerenzer et al., 2003 andShu andTownsend, 2010) and varied whether the total amount to be paid for the loan or the total amount earned in the savings accounts was displayed in peso amounts or in percentages. In particular, the marketing brochures and the simple table with 5 products had the total amount in peso amounts for participants with an even-numbered ID and in 13 percentage values for participants with odd-numbered IDs. In contrast, the complex table and simple table with 10 institutions were presented in peso amounts for participants with odd-numbered IDs and in percentage terms for participants with even-numbered IDs. The simplified KFS designed by us with CONDUSEF always displayed the total amount to be paid or earned in peso amount. This way, the same individual was presented with the total amount to be paid (credit products) or earned (savings products) in peso amounts or percentages depending on the format. Finally, a glossary explaining key financial terms was distributed in half of the sessions. Table 1 reports the number of participants who received the glossary and the messages or calls. Given these interventions, we use the following specification for Mexico: where , , 5 and 10 are dummies that take value 1 if participants received the respective treatment in round j of session k. The omitted treatment is again the one that uses marketing materials such as brochures.
is a dummy variable that takes the value 1 if participant i in round j of session k saw the total amount to be paid (credit product) or earned (savings product) displayed in pesos and 0 if in percentages. is a dummy indicating that participant i was provided with a glossary of terms during session k and 0 otherwise and / takes the value of 1 if participant i received a call / SMS prior to session k and 0 otherwise. We finally pool the data from both Peru and Mexico and run the following regression: In this pooled specification, we include country fixed effects and keep only the treatments that are common in both countries, namely the simplified format and brochures.
14 Because individuals ranked the choice of three products, we can also run a rankorder logit specification that includes the total cost of the credit product or the total yield in case of a savings product. This specification is ideal to assess the price sensitivity across treatments by interacting the price variable (cost of loan or yield of savings account) with the treatment dummies. In addition, we can also assess how financial literacy affects price sensitivity across treatments. Using data from Peru we run the following specification:  = + 1 + 2 + 3 * + 4 * + 5 * * + 6 + 7 * + 8 * * + 9 5 + 10 5 * + 11 5 * * + 12 10 + 13 10 * + 14 10 * * where , , 5 and 10 are treatment dummies defined before. Finally, we pool data from both countries and run: Standard errors for all specifications are calculated using bootstrap and clustered at the participant level. 15 We can also use a ranked-order logit specification to investigate whether certain treatments make individuals more likely to rely on non-financial factors like brand loyalty or the (random) order in which products were received in detriment to the cost of the product (Bertrand et al., 2010). We run the following specification using pooled data where takes value 1 if product chosen in order c by participant i in round j of session k was among the first half of the products given in a round or was shown in the upper half of a comparative table in Mexico. is another dummy that takes value 1 if the participant had or had ever had a financial product from the institution of the product chosen in order c. Table 3 reports the coefficients from regressions in (1) (in columns 1 and 4) in (2) (in columns 2 and 5) and in (3) in columns 3 and 6. The dependent variable is a dummy that takes value 1 if the individual chose the best product. Columns 1 to 3 (4 to 6) report the results for credit (savings) sessions. In all regressions, the first rows show the coefficients associated with the disclosure treatments followed by the coefficient for the profile dummy and the participant characteristics. The table also reports the mean of the dependent variable for the omitted disclosure treatment (marketing materials) and the pvalue of a t-test that two different disclosure treatments are equal.

Results
Column 1 of Table 3 shows the results for credit sessions in Peru. The simplified KFS increases the probability that individuals chose the right loan product by 12.6 percentage points relative to the brochures (and other marketing materials) and it is significant at the 1 percent level. In contrast, KFS designed by financial institutions do not significantly improve decision-making relative to brochures. The p-value associated with the t-test that the coefficients on both disclosure treatments are equal is 0. The coefficient associated to the profile is also not significant, suggesting that individuals are equally able to choose the best loan product regardless of whether they expect to make all 16 payments on time or with one missed payment. We note that for 80 percent of all products (including loans and savings accounts), the best product according to Profile 1 would not be chosen under the other profile so that choosing according to the profile was important. Finally, none of the participant characteristics in Peru affects financial decision-making, including proxies for education and financial literacy.
Column 2 of Table 3 shows that in Mexico, the simplified KFS, the complex table and simple table with 5 products were equally effective at increasing the probability of choosing the best credit product by around 25 percentage points or by 64 percent (from a base of 38.6 percentage points) relative to credit choices using brochures. Interestingly, doubling the number of products from 5 to 10 eviscerates the positive impact on decisionmaking as participants do no better with the 10-product comparative table than with brochures. This result is consistent with the concept of choice overload coined by Toffler (1970) and described in Iyengar and Lepper (2000) and Schwartz (2004).
We also find that showing the total cost of the credit in pesos (instead of in percentages) increases the probability of choosing the cheapest credit product by 8 percentage points, confirming the findings of Hasting and Tejada (2008), Gigerenzer et al. (2003) and Shu and Townsend (2010). In contrast, the live calls, SMS or the glossary did not improve decision-making. About half of the participants had heard about CONDUSEF, and indeed receiving an SMS prior to the experiment increased the odds of knowing about CONDUSEF by 25 percent. Participants received an average of 2.4 SMS from CONDUSEF in the 5 days prior (and they recalled receiving around 3 SMS per day). They received about 1.2 live calls from CONDUSEF. Despite the number of SMS and calls, the message did not register, as it did not improve the ability to choose the best financial product. Perhaps since they were not facing a teachable moment at the time they received the SMS or call, they ignored the content. The lack of impact of the glossary may be explained by the fact that it was difficult to understand. Online Appendix OA3 reports the glossary that was handed out.
Among the participant characteristics, financial literacy is the only one that contributed to better decision-making by 10 percentage points, but the impact of the simplified KFS is 2.5 times larger than that of financial literacy. Column 3 of Table 3 presents the pooled regression comparing the simplified KFS to the brochures. These were the disclosure treatments common to both countries. We find that the simplified KFS improves the probability of choosing the best credit product by 23 percentage points (p-value is 0). This result is remarkable because it indicates that the same individual can improve his or her decision-making simply by using a different format. When we compare individuals with and without financial literacy, we find that financially literate individuals are 6.6 percentage points more likely to choose the right credit product. This comparison is somewhat problematic because financially literate individuals may differ in other characteristics to those that are financially illiterate and therefore differences in decision-making cannot be solely attributed to financial literacy. By comparing the coefficients, however, what is remarkable is that the simplified KFS is almost 3.5 times more effective than financial literacy.
While the simplified format significantly improves decision-making for credit products, columns 4 to 6 show that this is not the case for savings products. The coefficient on simplified KFS is positive in columns 4 and 6 but negative in column 5 and never statistically significant. In column 4, the market-designed KFS does not improve the choice of savings products (relative to marketing materials) either. In column 5, the simple table with 5 products and to a lesser extent the complex table and the simple table   with 10 products increase the probability of choosing the highest-yielding savings account according to the profile assigned. The simple table increases the probability by 17.7 percent or by 51 percentage points (from a base of 34.2 percent among those offered brochures). Similar to column 2, when the yield is presented in pesos rather than percentage terms, the probability of choosing the right savings product increases by 4.1 percent. Among participant characteristics, being a male, having post-secondary education and correctly answering the financial literacy question improve financial decision-making. The impact of financial literacy is again about half that of using the simple table with 5 products to compare across savings products. Table 4 presents the results of the rank-ordered logit. In Peru and in the pooled regression (columns 1 and 3, respectively), the higher the cost of the loan, the lower the probability that that loan will be selected as the first choice. This suggests that individuals are price sensitive. More interestingly, in all of the credit-related specifications (columns 1 to 3), the interaction of the total cost with a dummy for the simplified format is also negative and significant, suggesting that price sensitivity is enhanced by the simplified KFS. Put differently, comparison shopping is enhanced with more transparent disclosure.
Related, financial literacy also increases price sensitivity in Mexico and using the pooled sample (columns 2 and 3, respectively). Echoing the results of Table 3, the impact of financial literacy is however more muted than that of the disclosure format.
Individuals also seem price sensitive when evaluating savings products (columns 4 to 6). In Peru, using the simplified KFS helps, but not in Mexico. Online Appendix OA4 shows an example of the simplified KFS used in Peru (E) and in Mexico (H). While the format in Peru includes the total yield earned in a month with two usage profiles, the format in Mexico does not contain this information. This is perhaps the reason why the simplified KFS is effective in Peru but not in Mexico. As a result, in the pooled regression only financial literacy matters. Table 5 computes the price-elasticity from the estimates of Table 4. In practice, we use data from one disclosure treatment and run a simple rank-ordered logit without interactions. 11 Columns 1 and 4 use the sample of all respondents. In columns 2 and 5, the sample is restricted to individuals that are not financially literate, that is, individuals that did not answer the question on interest rate correctly ( = 0), while in columns 3 and 5 only individuals who correctly answered that question are included ( = 1).
Panel A of Table 5 contains data from Peru, Panel B from Mexico and Panel C pools data from both Peru and Mexico. The price-elasticities reflect the results discussed in Table 4. For example, according to the elasticities reported in column 1 of Panel C, when individuals compare credit products using brochures, an increase of 1 percent in the cost of the loan leads to a decline in the probability of that loan being chosen first of 1 percent (p-value is 0). In contrast, when individuals use the simplified KFS the decline in 11 In particular, we run the following specification: = + + . See Online Appendix 5 for details on how the elasticities were calculated.

19
the probability is roughly three times as large at 3.2 percent. When comparing columns 2 and 3, financial literacy leads, as previously discussed, to higher elasticities in magnitude.
When individuals use brochures to compare credit products, the probability that a loan will be chosen first when its costs increases by 1 percent declines by 0.8 percent for a financially illiterate individual to 1.2 percent for a financially literate one. Similarly, when individuals use a simplified KFS the probability declines by 3 and by 3.2 percent, respectively for a financially illiterate and literate individual. Also notably, the simplified KFS is able to correct differences in financial literacy by not only increasing the magnitude of everyone's price elasticity but also by making the difference in price elasticity between financially literate and illiterate individuals smaller, in percentage terms. Related, Online Appendix Table OA2 reports individual perceptions about the different formats and their knowledge of the interest rate for the chosen product. As shown in column 3, financially illiterate individuals are 17 percent more likely to perceive the simplified KFS as being clear compared to the marketing brochures.
Similarly, in column 6 these individuals are 13.2 percent more likely to correctly state the interest rate of the loan chosen.
For savings products, the results are far more muted in magnitude and statistical significance. In column 1 of Table 5, an increase in the yield of a savings account by 1 percent increases the probability of choosing that product first by 0.02 percent when individuals use brochures to compare across savings products. When individuals use simplified KFS, the probability of choosing the account first increases by 0.03 and the difference in elasticities is not statistically significant. Columns 9 and 12 of Online Appendix Table OA2 show neither differences in clarity nor correct recall of interest rates between formats. Why are the results for savings so muted? There are a couple of reasons. First, unlike Peru, the simplified KFS in Mexico did not contain the net yield from the account to be earned in a year. Thus, while the simplified KFS in Peru is the only disclosure format with a positive and significant elasticity, in Mexico the simplified KFS is not at all effective. In fact, the price elasticity is lower than that of the brochures and is not significant for financially illiterate individuals. Second, when choosing a savings product, the net yield from the account may not be as relevant as other factors 20 such as convenience, speed, familiarity and trust of the institution offering the product, etc.
In the context of our lab experiment we cannot explore convenience and speed of service, but we can study other non-price factors such as the order in which products are presented or appear in a comparative table in Mexico, and whether familiarity with the institution offering the product, defined as a dummy that takes value 1 if the individual has ever contracted a product from that institution. In the context of elections, Ho and Imai (2008) use the naturally occurring variation in ballot order and find that being listed first on the ballot increases the likelihood of being voted especially in primary elections.
Similarly, Luca and Smith (2013) study a change in the way universities were ranked in the U.S. News and World Report and find that the ranking matters.
Columns 1 and 4 of Table 6 show that individuals using the market-designed KFS in Peru are more likely to choose products that are presented among the first three of the five or that come from institutions that are familiar to the individual. When using the simplified KFS, individuals are less influenced by these factors. Thus, in Peru transparency increases price elasticity while familiarity and the order in which products are presented matter less. In Mexico, neither the order nor familiarity affects the probability of choosing a credit or savings product in columns 2 and 5. As a result, the pooled regression reflect, to some extent, the results in Peru. Individuals are more likely to choose products presented first when using brochures than when using the simplified KFS.

Conclusion
This study conducts a laboratory experiment to test the effectiveness of behavioralbased financial disclosure by focusing on how comparability and the standardization of formats can improve financial decision-making for low-to-middle income consumers in Peru and Mexico.
We find that a standardized key facts statement significantly improves the ability of consumers to make financial decisions and suggests that regulators should not only 21 mandate certain key terms but also the format in which these terms are presented to encourage comparison shopping and improve financial wellbeing. It is encouraging that, in recent years, some regulators have begun mandating standardized formats. 12 Interestingly, the standardized format tested increased the probability that individuals choose the best credit product by a factor of three, relative to the increase in probability between financially literate and illiterate individuals. More importantly, the standardized format seems to "democratize" financial decision-making as it is particularly effective for individuals that are financially illiterate. The effects are however concentrated in credit products rather than savings. The lack of impact in savings is concentrated in Mexico, where the simplified format failed to disclose the total yield of the savings account in pesos, which may indicate that total yield calculations can be a useful requirement for disclosure rules to improve consumers' ability to understand the cost or return of the savings account.
The laboratory setting approach taken also suggests an effective mechanism to test the design of financial disclosure initiatives. This approach is not new. For example, the Consumer Financial Protection Bureau and the Federal Reserve of the U.S. constantly survey and test financial consumers on how they understand information, which information they think is useful, and finally how the information can be more effectively conveyed (Kroszner, 2007). Interestingly, Online Appendix Table AO3 shows that among individuals who showed up for a session in Mexico, those that participated in the laboratory experiment (compared to those that were randomly turned away due to lack of space) were more likely to report contracting the session's financial product in the following 6 months. Owning a business and being a male were also positively correlated with the self-reported likelihood of contracting the financial product.
In addition, regulators in Mexico and elsewhere are requiring lenders to send detailed product information in a machine-readable format so they can be downloaded by startups like ComparaBien, ComparaGuru, and rocket.la which provide timely comparative information to individuals looking for financial products. These channels also have the advantage of being fully digital, removing the time and travel burdens to shopping around, and making it easier to review and compare competing key facts statements on the same screen at the same time. In the U.S., a similar initiative called "Smart Disclosure" was undertaken by the Obama Administration (Sunstein, 2013) to provide sharable information so that intermediaries could help consumers make, for example, better informed decisions about energy and health care.
A final important consideration for the effectiveness of KFS is the timing of the provision of information. The KFS is most useful early on during the sales visit, so that consumers can quickly receive competing offers and compare across products. However, sales staff may be incentivized to only disclose the KFS late, after the product has been contracted. This practice will undermine the effectiveness of disclosure and the consumer's propensity to shop around. Policy makers should therefore take care to develop rules regarding the timing of disclosures and monitor compliance with timely disclosure of KFS through mystery shopping (see Giné and Mazer, 2017)    Notes: This table reports the estimation of the following specifications: 1) For Peru, BestChoice ijk = α j + β 1 Simplified jk + β 2 Mkt jk + β 3 LowProfile ik + X ik 'γ+ ε ijk . Simplified jk and Mkt jk denote the different disclosure formats. LowProfile ik takes the value of 1 if participant i in session k is assigned to Profile 1. 2) In columns 2 and 5, for Mexico, BestChoice ijk = α j + β 1 Simplified jk + β 2 Complex jk + β 3 Simple 5 jk + β 4 Simple 10 jk + β 5 AmountPesos ijk + β 6 Glossary ik +β 7 LiveCall ik +β 8 SMS ik + β 9 LowProfile ik + X ik 'γ + ε ij . Simplified jk , Complex 5 jk , Simple 5 jk and Simple 10 jk denote the different disclosure formats. AmountPesos ijk is a dummy variable that takes the value 1 if participant i in round j of session k saw the total amount to be paid (credit product) or earned (savings product) displayed in pesos and 0 if in percentages. Glossary ik is a dummy indicating if participant was provided with a glossary of terms during session k and 0 otherwise. LiveCall ik / SMS ik take the value of 1 if the participant i received a call/SMS prior to session k and 0 otherwise. LowProfile ik is only used in regressions for savings sessions and takes the value of 1 if participant i in session k is assigned to Profile 1. 3) Pooling data from both Peru and Mexico, BestChoice ijk = α j + β 1 Simplified jk + β 2 LowProfile ik + X ik 'γ+ ε ijk . In this specification we include country fixed effects. In all specifications, BestChoice ijk takes the value of 1 if participant i in round j and session k has chosen the best product and 0 otherwise. Vector of characteristics X ik includes the following variables: male, whether the individual has a post secondary education, age and age squared (divided by 100) and a proxy for financial literacy that takes value 1 if the individual correctly answered the question on interest rate. In all specifications we use round fixed effects. Control treatment includes promotional materials (brochures, amortization tables etc.) collected from financial institutions. Standard errors are clustered at the participant level and are reported in parenthesis under coefficient estimates. Levels of significance * p<0.10 ** p<0.05 *** p<0.01. Notes: This table reports the following rank-ordered logit specification: 1) For Peru, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 Simplified jk *Price cijk + β 4 Price cijk *FinLit ik + β 5 Simplified jk *Price cijk *FinLit ik + β 6 Mkt jk + β 7 Mkt jk *Price cijk + β 8 Mkt jk *Price cijk *FinLit ik + ε cijk ,. Simplified jk , and Mkt ij denote the Simplified KFS and the Market designed KFS. 2) For Mexico, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 Simplified jk *Price cijk + β 4 Price cijk *FinLit ik + β 5 Simplified jk *Price cijk *FinLit ik + β 6 Complex5 jk + β 7 Simple5 jk + β 8 Simple10 jk + β 9 Complex5 jk *Price cijk + β 10 Simple5 jk *Price cijk + β 11 Simple10 jk *Price cijk + β 12 Complex jk *Price cijk *FinLit ik + β 13 Simple5 jk *Price cijk *FinLit jk + β 14 Simple10 jk *Price cijk + ε cijk . Simplified jk , Complex jk , Simple5 jk and Simple10 jk denote the different main treatments. 3) Pooling data from both Peru and Mexico, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 Simplified jk *Price cijk + β 4 Price cijk *FinLit ik + β 5 Simplified jk *Price cijk *FinLit ik + ε cijk . An observation is a choice c made by each individual participant i in round j and session k. In all specifications, Price cijk is either the total loan cost or the savings yield of the each product. Order cijk takes value 3 for the first choice of individual i in round j in session k, 2 for the second choice and 1 for the third choice. FinLit ik is a proxy for financial literacy that takes value 1 if the individual correctly answered the question on interest rate. Control treatment includes promotional materials (brochures, amortization tables etc.) collected from financial institutions. Standard errors are clustered at the participant level and are reported in parenthesis under coefficient estimates. Levels of significance * p<0.10 ** p<0.05 *** p<0.01.  Notes: This table reports the price-elasticities of the probability that a specific prouct is chosen first by an individual. These elasticities are estimated based on the following rank-ordered logit specifications: Order cijk = α j + β 1 Price cijk + ε cijk . Each elasticity is calculated with correspondent data samples. An observation is a choice c made by each individual participant i in round j and session k. Price ijc is either the total loan cost or the savings yield of the each product. y ijc is a variable that takes the value of 1 if the individual choses product first. Standard errors are clustered at the participant level and are reported in parenthesis under coefficient estimates. Levels of significance * p<0.10 ** p<0.05 *** p<0.01.  Notes: This table reports the following rank-ordered logit specification: 1) For Peru, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 TopTable cijk + β 4 Familiarity cijk + β 5 Mkt jk + β 6 Simplified jk *TopTable cijk + β 7 Mkt jk *TopTable cijk + β 8 Simplified jk *Familiarity cijk + β 9 Mkt jk *Familiarity cijk + ε cijk ,. Simplified jk , and Mkt jk denote the Simplified KFS and the Market designed KFS. 2) For Mexico, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 TopTable cijk + β 4 Familiarity cijk + β 5 Complex5 jk + β 6 Simple5 jk + β 7 Simple10 jk + β 8 Simplified jk *TopTable cijk + β 9 Complex5 jk *TopTable cijk + β 10 Simple5 jk *TopTable cijk + β 11 Simple10 jk *TopTable cijk + β 12 Simplified jk *Familiarity cijk + β 13 Complex ij *TopTable cijk + β 14 Simple5 jk *TopTable cijk + β 15 Simple10 jk *TopTable cijk + ε cijk . Simplified jk , Complex jk , Simple5 jk and Simple10 jk denote the different main treatments. 3) Pooling data from both Peru and Mexico, Order cijk = α j + β 1 Price cijk + β 2 Simplified jk + β 3 TopTable cijk + β 4 Familiarity cijk + β 5 Simplified jk *TopTable cijk + β 6 Simplified jk *Familiarity cijk + ε cijk . An observation is a choice c made by each individual participant i in rounda j and session k. In all specifications, Price cijk is either the total loan cost or the savings yield of the each product. Order cijk is a variable that takes value 3 for the first choice of individual i in round j in session k, 2 for the second choice and 1 for the third choice. TopTablecijk is a variable that takes the value of 1 if option c is the first on the table presented to individual i in round j and session k . Familiarity cijk dummy that takes value 1 if the participant had or had ever had a financial product from the institution of the product chosen in order c. Control treatment includes promotional materials (brochures, amortization tables etc.) collected from financial institutions. Standard errors are clustered at the participant level and are reported in parenthesis under coefficient estimates. Levels of significance * p<0.10 ** p<0.05 *** p<0.01.

A. Script for live calls (Savings)
Good morning/afternoon, my name is Guadalupe Gomez and we are calling you as a part of a pilot study on financial literacy in the name of Condusef. The goal of this call is to provide you some advice on how to choose an account to save your money. This call will take up to 5 minutes.

YES
Condusef is a public institution that protects and defends the rights of the users of financial services. As part of this mission, we are calling you today to provide some advice on how to choose an account to save your money.
Have you ever signed up for an account to save your money at any financial institution?
Before opening an account to save your money, you should not only be aware of the offered interest rate (i.e. yield), but also be aware of the fees incurred. The total yield minus the monthly fees determine if the account will provide gains or losses to your savings. For example, if your total yield is 5 pesos but you pay 10 pesos per month in fees, your savings will decrease by 5 pesos every month.
Additionally, we always recommend comparing at least 3 institutions when opening an account and choosing the one that offers the highest yield and charges the lowest fees.

Was the information I just gave you clear?
Thank you for your time. We hope the information provided was useful.

B. Script for live calls (Credit)
Good morning/afternoon, my name is Guadalupe Gomez and we are calling from Condusef. The goal of this call is to provide you some advice on how to choose a credit product.

START call
Condusef is a public institution that protects and defends the rights of the users of financial services. As part of this mission, we are calling you today to provide some advice on how to choose a credit product.
Have you ever signed up for a credit product at any financial institution?

When signing up for a credit product or personal loan is very important to be aware of the total loan cost (i.e. the total amount to be paid). This information is not usually provided by the financial institution, but it is easily calculated by adding up all credit payments, fees and insurance costs.
For example, if you ask for 5,000 in credit and the total cost is 15,000, you will end up paying three times the borrowed amount. You should use the total amount to be paid for a loan to make comparisons among different financial institutions.
We recommend comparing at least 3 institutions when signing up for a loan, asking for the same amount and term conditions. You should always choose the one that offers the lowest total cost. Origination fee: One-time fee. Fixed amount or percentage that a financial institution charges the client for giving him/her credit.

CREDIT
Disbursement fee: Fixed amount charged by the financial institution when the client uses the money lent to him/her.

Credit check fee:
Fixed amount charged by the financial institution for checking a client's credit history.
Annual maintenance fee: Fixed amount charged annually by the financial institution for maintenance of credit product.
Credit history: Report that contains the summary of all credit products taken by a client. Can be checked with special entities, such as the Buró de Crédito or the Circulo de Crédito.
Advance payment: Payment of partial or total amount the credit borrowed by the client before the payment due date set by the financial institution.
Principal or capital: Total amount deposited by the financial institution to the client when giving him/her a loan.
Interest rate: Percentage charged by a financial institution when lending money to a client.
Interest rate for late payment: Percentage charged by a financial institution when the clients delays the payment.

Insurance products
Life: Financial instrument that, in case of death of the insured, ensures the payment of an amount corresponding to the deficit balance, according to contract terms. (1) Se encuentra expresada en un año de 360 días.
La garantía respalda las obligaciones que usted tenga o pueda tener de forma directa o indirecta con el BCP.
De no encontrarse conforme con las condiciones contractuales, EL CLIENTE podrá solicitar unilateralmente la resolución del contrato suscrito ingresando una comunicación por escrito en la red de Agencias de LA FINANCIERA a nivel nacional.
El CLIENTE declara haber recibido la presente Cartilla y el Contrato para su lectura y que la Financiera ha absuelto todas sus preguntas, suscribiendo el presente documento y el Contrato con absoluto conocimiento de sus alcances en cuanto a derechos, obligaciones y responsabilidades contenidas. El presente documento carece de valor si no está acompañado del respectivo contrato firmado por los representantes de la Financiera.
De acuerdo a lo señalado en el Contrato, el cliente otorga autorización a la FINANCIERA a cargar en cualquier cuenta, depósito y/o valor que mantuviere en la FINANCIERA las sumas que pudieren resultar de cualquier obligación exigible que mantiene o pudiera mantener en la FINANCIERA.

Section OA5: Computation of price elasticities
The elasticities are calculated based on the following rank-ordered logit specification: Following Beggs et al. (1981) 1 , the probability that a product is chosen first among 3 options is given by , where ℎ = ℎ + ℎ where ℎ has distribution extreme value. The reported elasticities in Table 5 are estimated as follows assuming individuals choose Section OA6. Tables   (1) (2)

APY APY
Notes: This table reports the estimation of the following specifications: 1) For Peru, y ijk = α j + β 1 Simplified jk + β 2 Simplified jk *FinLit ik + β 3 Mkt jk + β 4 LowProfile ik + X ik 'γ + ε ijk, , where i indexes each individual participant and in session j . y ijk is the outcome variable that takes the value of 1 if participant i reports clear presentation of the product or if he/she is able to recall correctly the APR /APY of the product given in the last round. Simplified jk , and Mkt jk denote the Simplified KFS and the Market designed KFS. LowProfile ik takes the value 1 if participant is assigned to profile 1. 2) For Mexico, y ijk = α j + β 1 Simplified jk + β 2 Simplified jk *FinLit ik + β 3 Complex jk + β 4 Simple 5 jk + β 5 Simple 10 jk + β 6 Complex jk *FinLit ik + β 7 Simple 5 jk *FinLit ik + β 8 Simple 10 jk *FinLit ik + β 9 LowProfile ik + X ik 'γ + ε ijk . Simplified jk , Complex jk , Simple 5 jk and Simple 10 jk denote the different main treatments. LowProfile ik is only used in regressions for savings sessions and takes the value of 1 if participant i in session k was assigned to Profile 1. 3) Pooling data from both Peru and Mexico, y ijk = αj + β 1 Simplified jk + β 2 Simplified jk *FinLit ik + β 3Low Profile ik + X ik 'γ + ε ijk . In savings sessions regressions, we assign all individuals in Mexico to profile 1. In this specification we include country fixed effects. An observation is an individual. Vector of characteristics X ik includes the following variables: male, whether the individual has a post secondary education, age and age squared (divided by 100) and a proxy for financial literacy (FinLit ij ) that takes value 1 if the individual correctly answered the question on interest rate. In all specifications we use round fixed effects. Control treatment includes promotional materials (brochures, amortization tables etc.) collected from financial institutions. Standard errors are clustered at the participant level and are reported in parenthesis under coefficient estimates. Levels of significance * p<0.10 ** p<0.05 *** p<0.01.

Clarity
Credit Clarity (1)  Notes: This table reports the estimates of the regression of a dummy variable that takes value of one of the respondent considers getting a credit or savings product in the next 6 months on another dummy variable that takes value of one if the individual participated in the lab experiment and individual characteristics. Data comes from survey taken in Mexico only. Data come from the initial survey. *In Mexico, C-D socioeconomic groups identify low-to-middle income household. ** Knowledge of interest rate is tested with the following multiple choice question: "If you deposit 100 soles in a bank account that charges you nothing and guarantees you a yield of 2%