BIS Working Papers No 931 The fintech gender gap by Sharon Chen, Sebastian Doerr, Jon Frost, Leonardo Gambacorta, Hyun Song Shin Monetary and Economic Department March 2021 JEL classification: E51, J16, O32. Keywords: fintech, gender, financial inclusion, personal data, privacy. BIS Working Papers are written by members of the Monetary and Economic Department of the Bank for International Settlements, and from time to time by other economists, and are published by the Bank. The papers are on subjects of topical interest and are technical in character. The views expressed in them are those of their authors and not necessarily the views of the BIS. This publication is available on the BIS website (bis). Bank for International Settlements 2021. All rights reserved. Brief excerpts may be reproduced or translated provided the source is stated. ISSN 1020-0959 (print) ISSN 1682-7678 (online) The ntech gender gap S. Chen S. Doerr J. Frost L. Gambacorta H. S. Shin March 2021 Abstract Fintech promises to spur nancial inclusion and close the gender gap in access to nancial services. Using novel survey data for 28 countries, this paper nds a large ntech gender gap: while 29% of men use ntech products and services, only 21% of women do. The gap is present in almost every country in our sample. Country characteristics and several individual-level controls explain about a third of the unconditional gap. Gender di erences in the willingness to use new nancial technology or ntech entrants if they o er cheaper services account for over half of the remaining gap. The paper concludes by suggesting potential explanations for the gender gap and implications for challenges in fostering nancial inclusion with new technology. JEL Codes: E51, J16, O32 Keywords: ntech, gender, nancial inclusion, personal data, privacy Sharon Chen () is at Ernst Brown et al., 2019; C elerier and Matray, 2019). And yet, women all over the world remain unbanked or underbanked relative to men: they have lower rates of bank account ownership than men (Demirg u c-Kunt et al., 2017), are less likely to manage household nances (Guiso and Zaccaria, 2021) or to participate in the stock market (Ke, 2020). Hopes are high that new nancial technology or ntech can enhance nancial inclusion and close the gender gap in the access to nancial services (Demirg u c-Kunt et al., 2018; Breza et al., 2020). By leveraging new technology and non-traditional data, both traditional nancial institutions (incumbents) and new ntech rms ( ntech entrants) promise to o er novel products better tailored to consumers needs at a lower cost (Arner et al., 2020; Boot et al., 2020; Philippon, 2020; Thakor, 2020). These technological advances could bene t disadvantaged groups disproportionately (Suri and Jack, 2016; Bachas et al., 2017; Ouma et al., 2017; Lee et al., 2021). However, evidence on whether ntech helps to close the gender gap in the access to and use of nancial services is scarce. This paper uses data from a large survey of over 27,000 adults from 28 major economies to investigate gender di erences in the adoption of new nancial technol- ogy. The dataset draws on a survey used to construct the 2019 EY Global Fintech Adoption Index (2019). The survey sample is representative along the age and gender distributions and includes details on individuals use of ntech products, as well as at- titudes towards ntech entrants and incumbents. It also contains detailed background information along several key dimensions. Our key nding is the presence of a large and ubiquitous ntech gender gap: namely, that women are signi cantly less likely to use ntech products or services o ered by ntech entrants than men. On average, 29% of men report having used ntech entrants over the previous six months. The respective gure for women is 21%. The gap is present in almost all countries in our sample and not fully explained by a large set of individual 2 characteristics, such as age, income, education, marital or employment status, or a proxy for nancial literacy. Nor is it explained by country-speci c characteristics. Accounting for individual characteristics reduces the gap from 8.4 percentage points (pp) to 5.9 pp, or by 30% relative to its unconditional mean. Further including country xed e ects reduces the gender gap to 5.2 pp, but it remains statistically highly signi cant. How does the ntech gender gap compare with the gap in bank account ownership? Demirg u c-Kunt et al. (2018) report that 72% of men and 65% of women have a bank account globally. The unconditional gap in bank account ownership (7 pp) is thus smaller than the ntech gender gap (8 pp). Scaled by mean adoption rates, which equal 69% for traditional bank accounts and 25% for ntechs, the di erence equals (7=69 =) 10% vs (8=25 =) 32%. These ndings suggest that ntech entrants have so far not closed the gender gap in the access to nancial services. Fintech products di er greatly in scope. For example, some products facilitate cross- border payments, while others o er peer-to-peer loans. Potentially, the gap is more pronounced in some categories than others, which could explain the aggregate gap. To test this possibility, we estimate regressions at the respondent-product level for 19 narrowly de ned product categories. We nd that including product xed e ects and comparing the use of ntech services and products by men and women within the same product category does not a ect our estimates of the gender gap in any statistical or economically meaningful way. Respondent-product level regressions also allow us to exploit variation across genders within the same product. For example, we nd that the gender gap is around 50% smaller among products that complement traditional banking services, relative to those that are substitutes. These ndings indicate that women might be more likely to adopt ntech products that complement familiar services. Including granular xed e ects at the individual level does not materially change our coe cients, despite the fact that the R-squared more than quadruples. Individual observable and unobservable characteristics are hence unlikely to explain the product-speci c gap, alleviating concerns about self- selection and gender di erences in unobserved characteristics (Altonji et al., 2005; Oster, 2019). 3 Does it matter who o ers ntech products? We nd that 49% of respondents use novel nancial products and services that are o ered by traditional nancial institutions, compared with 25% for ntech entrants. Moreover, if individuals use ntech products provided by incumbents, they also report to use ntech entrants signi cantly more than respondents who do not use incumbents (35% vs 15%). These ndings suggest that ntech entrants are a complement to rather than a substitute for traditional banks (Fuster et al., 2019; Tang, 2019a). Still, men are more likely to use ntech products irrespective of the provider. The gender gap equals 6.4 pp (25% of the mean adoption rate) among services provided by entrants and 7.1 pp (14% of the mean) among those provided by incumbents. The di erence across providers is statistically insigni cant, which implies that the gender gap is not speci c to who provides ntech products or services, but rather to the products themselves.1 To investigate potential determinants of the gender gap, we document di erences between women and men in their reported attitudes towards privacy and technology. Women report more than men that they worry more about their security when dealing with companies online. They also report being signi cantly less willing to adopt new nancial technology, for example digital banks. Results further suggest that men are more price-sensitive: they are more willing to use a ntech entrant or share their personal data with ntechs for cheaper o ers.2 Finally, women report being less willing to use a ntech even if it o ers better products or products that are better-suited to the respondents lifestyle. To shed further light on our ndings, we examine whether di erences in attitudes can explain the gender gap. Controlling for whether an individual worries about his or her security does not materially a ect results; neither does controlling for di erences in the suitability of products. However, controlling for attitudes towards technology and price sensitivity reduces the gap from 5.2 pp to 2.3 pp. Thus, while individual and country characteristics reduce the unconditional gender gap by around one-third, 1The di erence remains insigni cant when we estimate regressions at the respondent-provider or respondent-product-provider level and include individual and/or product xed e ects. 2This nding is consistent with evidence from outside nancial services. For instance, Farrelly et al. (2001) nd that men are more responsive to changes in the price of cigarettes than women. 4 further accounting for di erences in attitudes reduces the gap by another 40%.3 Nevertheless, we are unable to fully explain the gender gap, in spite of introducing additional exercises. For example, while it could be that men are more likely to make nancial decisions among couples, we nd that the gender gap is also present among respondents who live alone. This result suggests that arguments that try to tie the gap to traditional gender roles within households fall short. Similarly, we nd a signi cant gap among the groups that are employed, have multiple accounts at nancial institutions, or are nancially literate. One important caveat is that our survey does not contain direct measures of risk aversion, which limits our options to directly investigate one plausible explanation for the ntech gender gap. We do nd, however, that including individual xed e ects to account for unobservable characteristics leaves our ndings una ected. Our results suggest that the gap in the use of ntech is closely linked to di erences in attitudes towards technology and price sensitivity. What determines di erences in these factors, however, remains an open question. They could be explained by di erences in preferences across genders, for example di erences in risk aversion (Croson and Gneezy, 2009; Dohmen et al., 2011), or di erences in the costs and bene ts that consumers attach to the use of these new products. They could also result from gender-based discrimination (Bartlett et al., 2019), for example from bad previous experiences by women with nancial institutions (Brock and De Haas, 2021). Finally, the gap could arise from social norms or laws that a ect the cost-bene t trade-o di erently across genders (Burda et al., 2013; Falk and Hermle, 2018; Hyland et al., 2020). For instance, if women have reason to worry more about a leak of personal data, then it may be rational to avoid services that require collection and processing of personal data.4 As factors related to attitudes towards technology and price sensitivity explain a sizeable part of the overall gap, future research focusing on the determinants of these factors could be particularly promising in understanding the ntech gender gap. Several important questions are opened up by our ndings on policies to foster 3As we show in the Appendix, these factors explain the gender gap in ntech products o ered by traditional FIs to a signi cantly smaller extent. 4Okat et al. (2020) argue that trust in traditional nancial institutions is not a signi cant driver of ntech adoption, while Yang (2020) shows that a scandal in the US banking sector has led to an increase in ntech adoption. 5 nancial inclusion. For one, our results suggest that improvements in technology alone may fall short of the objective of closing the gender gap in access to nancial services. The ntech revolution may need to be complemented by targeted policy initiatives that take account of di erences in attitudes across demographic groups. Depending on the cause of the ntech gender gap, the speci c policy response may di er. If di erences in adoption rates are based on di erences in hard-wired preferences, then the scope for interventions through policy is limited. Should, however, the observed outcome be the result of discrimination or social norms and laws that disadvantage women, then policy that addresses and remedies these factors could help to promote nancial inclusion through nancial innovation. These policy options also raise deeper conceptual questions on where to draw the line between hard-wired preference di erences and attitudes that are susceptible to changes in prevailing norms. Our paper contributes to the current literature on the e ects of nancial technology on nancial inclusion and the gender gap in access to nancial services.5 Fuster et al. (2019) and Tang (2019a) show that ntech often serves as a complement, rather than a substitute, to traditional banking services. Jagtiani and Lemieux (2018), Hau et al. (2018), Agarwal et al. (2019) and Frost et al. (2019) instead argue that ntech and big tech lenders serve borrowers that are traditionally underserved by banks. Other papers also highlight how ntechs could spur nancial inclusion, for example by reducing the costs of nancial intermediation (Philippon, 2020; Sahay et al., 2020) or changing con- sumer behaviour (Breza et al., 2020). Our results are, to the best of our knowledge, among the rst that use individual-level information to investigate the adoption of n- tech products from the consumer side across genders for a large sample of countries.6 We establish a persistent gender gap in the use of ntech that could pose an obstacle to nancial inclusion through nancial innovation. The results further show that the willingness to share data and concerns about privacy di er across subgroups in the population. In particular, we nd that women appear less willing to share personal data than men and that they worry more about their 5See Demirg u c-Kunt et al. (2017) for a survey on nancial inclusion. 6Carlin et al. (2019) show for Iceland that younger generations adopt nancial technology more readily than older generations. 6 security when dealing with companies online. If users value privacy di erentially, then e ective privacy regulation needs to take these di erences into account, for example when assigning control rights (Acquisti et al., 2016; Tang, 2019b; Jones and Tonetti, 2020). In light of the debate on algorithmic fairness and bias in data (Kleinberg et al., 2015; Corbett-Davies and Goel, 2018; Kleinberg et al., 2018), algorithms trained on non- representative data that are then used to derive conclusions about the general population could lead to an ine cient outcome (Bergemann et al., 2020).7 Our ndings hence highlight the need to better understand the causes of di erences in the willingness to share data across demographic groups. 2 Data Our main source of data is the EY Global Fintech Adoption Index (2019). The consumer survey is based on 27,103 online interviews with digitally active adults between February and March 2019 in 28 countries.8 The countries i