The Impact of Fintech Startups on Financial Institutions' Performance and Default Risk

We examine the impact of fintech start-ups on the performance and default risk of traditional financial institutions. We find a positive relationship between fintech start-up formations and incumbent institutions’ performance for the period 2005–2018 and a large sample of financial institutions from 87 countries. We further analyze the link between fintech start-up formations and the default risk of traditional financial institutions. Fintech start-up formations decrease stock return volatility of incumbent institutions and decrease the systemic risk exposure of financial institutions. The findings indicate that legislators and financial supervisory authorities should closely monitor the development of fintech start-ups, because fintechs not only have a positive effect on the financial sector’s performance but also can improve financial stability relative to the status quo.


Introduction
The rise of financial technology (fintech) has received considerable attention from academics, practitioners, and regulators. The recent hype about fintech is due to the development and deployment of novel technologies such as artificial intelligence, big data, cloud computing, machine learning, blockchain, and other technologies that have the potential to revolutionize the financial sector, which was historically considered among the most traditional and conservative sectors in the economy. The Financial Stability Board of the Bank for International Settlements defines fintech as "as technology-enabled innovation in financial services that could result in new business models, applications, processes or products with an associated material effect on the provision of financial services" (European Banking Authority, 2017, p. 7). Fintech innovations have emerged in many of the traditional value-adding sections of a universal bank, including financing, asset management, payment services, and others (Dorfleitner et al., 2017). Fintech startups not only challenge traditional financial institutions by providing cheaper, faster, and easier access to financial services but also potentially foster the transformation and innovation activities of incumbent institutions (Milian et al., 2019;Di et al., 2021;Panos and Wilson, 2020;An and Rau, 2021). However, little is known about how fintech start-ups affect the traditional financial sector.
A core function of financial institutions is the intermediation of financial resources (Merton, 1992).
Yet the financial crisis of 2007-2008 created a credit crunch (Campello et al., 2010;Campello et al., 2011;Cowling et al., 2012), which resulted in financial constraints of many small and mediumsized firms (Mc Cahery et al., 2015). Economic output declined sharply, and unemployment rates increased worldwide (Daly et al., 2012;Bruno et al., 2014). Moreover, bank customers lost their confidence in many of the traditional financial institutions, which in some regions resulted in bank runs, such as that of British Northern Rock in 2008. This fragile environment provided the ground for novel products and services of fintech start-ups, which started with a clean slate and did not need to overcome a history of failure and excessive risk-taking. As a result of consumers' distrust in banks, marketplace lending has become more prevalent (Saiedi et al., 2020). In particular, banks' misconduct is related to the emergence of the United States (U.S.) online lending market (Bertsch et al., 2020). Moreover, research indicate that fintech start-ups have the potential to better address information asymmetries (Lin et al., 2013;Ge et al., 2017;Xu and Chau, 2018), because they leverage additional information about borrowers from the Internet, thereby enabling them to receive credit for the first time and, in some cases, at cheaper rates (Serrano-Cinca et al., 2015;Iyer et al., 2016;Jagtiani and Lemieux, 2019). Services provided by fintech start-ups also include algorithm-based investment advice, mobile banking services, instant online and mobile payment infrastructure, innovative risk management systems, and cost-efficient foreign exchange services (Haddad and Hornuf, 2019). The increasing prevalence of fintech start-ups and the potential pressure they put on incumbent firms raise the question of how fintech start-ups affect the performance and risk-taking of traditional financial institutions.
Research has often argued that fintech start-ups do not fully comply with financial regulation and engage in regulatory arbitrage using existing legal exemptions (Hornuf and Schwienbacher, 2017;Buchak et al., 2018;Cumming and Schwienbacher, 2018), which might consequently undermine financial stability (Fung et al., 2020;Li et al., 2020;Vučinić, 2020). Because financial stability has a direct impact on economic growth, scholars have investigated financial institutions' default risk and the conditions under which this risk led to subsequent financial turmoil (Beck et al., 2006;Acharya and Naqvi, 2012;Diamond and Rajan, 2012). Fintechs can, for example, affect systemic risk through their increasing interconnectedness with traditional financial institutions and a lax supervision by authorities (BIS, 2017). By contrast, new business models that are uncorrelated with traditional financial institutions can also reduce systemic risk in the financial industry. To the best of our knowledge, empirical research has not yet examined the systemic risk of traditional financial institutions resulting from fintech start-up formations. In this article, we therefore raise the question: Does the emergence of fintech start-ups affect systemic risk of traditional financial institutions?
The empirical literature on financial innovation in general and the interaction between traditional financial institutions and fintechs in particular is still scarce. Lerner (2002) and Miller (1986) measure financial innovation by the filing of financial patents and show that it has been increasing since the late 1970s. The quality of financial patents and financial innovations, however, was often low (Lerner et al., 2016). Scott et al. (2017) show that the financial industry traditionally invested a large share of expenses in information technology (IT), reaching around one-third of all expenses in the early 1990s. In particular, early on, the financial industry employed computers. However, only a few financial innovations (e.g., automated teller machines) have led to considerable changes in financial institutions and their business models (Merton, 1995). Whether fintechs affect incumbents' ability to innovate and consequently perform is still an open question.
A related article to ours focusing on bank-fintech alliances is that of Brandl and Hornuf (2020), who run a bank-fintech network analysis for Germany and find that bank-fintech relationships are often product-related. They argue that this form of alliance is due to fintechs' development of an algorithm or software, the value of which can only be determined when the software has been adapted more thoroughly to customer needs. Hornuf et al. (2020) refine these findings by analyzing bank characteristics associated with bank-fintech alliances. They hand-collect data for the largest banks from Canada, France, Germany, and the United Kingdom and provide detailed evidence that banks are more likely to form alliances with fintechs when they pursue a well-defined digital strategy and/or employ a chief digital officer. Furthermore, they find that banks more often invest in small fintechs but often build product-related collaborations with larger fintechs, which is in line with predictions from incomplete contract theory (Grossman and Hart 1986;Aghion and Bolton 1992). Phan et al. (2020) investigate a sample of 41 Indonesian banks. They find that fintechs negatively predict bank performance and argue that fintechs substitute for traditional banks by providing less expensive and more efficient services In this article, we collected data for 8,092 financial institutions and 12,549 fintech start-ups from 87 countries to assess the effect of fintech start-ups on the performance and default risk of traditional financial institutions. Our results indicate a positive and significant impact of fintech formations on financial institutions' performance. An increase of fintech start-up formations is associated with an increase of incumbent institutions' performance. Our findings also suggest that fintech formations decrease stock return volatility and financial institutions' exposure to systemic risk. These findings might be of interest to academics, practitioners, and regulators alike, especially as the fintech sector is steadily growing and becoming increasingly integrated with the traditional economy and incumbent financial institutions (Li et al., 2020).
The remainder of the article proceeds as follows. Section 2 summarizes the literature and introduces our hypotheses. Section 3 describes the data and introduces the variables used in the quantitative analysis. Section 4 presents the descriptive and multivariate results. Section 5 provides a discussion and conclusion of our study.

Literature and hypotheses
A wealth of literature has investigated the performance of financial institutions. In the past decade, research has examined the determinants of financial institutions' performance, analyzing how firms address corporate governance issues (Aebi et al., 2012;Peni and Vähämaa, 2012;Zheng and Das, 2018), master the diversification of their business activities (Berger et al., 2010;Brahmana et al., 2018;Chen et al., 2018;Kim et al., 2020), deal with external regulation (Naceur and Omran, 2011;Psillaki and Mamatzakis, 2017), react to monetary policies (Mamatzakis and Bermpei, 2016;Gambacorta and Shin, 2018), deal with the legal and institutional framework (Kalyvas and Mamatzakis, 2017;Bitar and Tarazi, 2019;El Ghoul et al., 2021), generate intellectual capital (Talavera et al., 2018;Nawaz, 2019;Adesina, 2021), and engage in shadow banking activities (Tan, 2017;Lin et al., 2018). Given the all-embracing and massive development of the fintech sector in the past decade, it seems worthwhile also to investigate how fintech start-up formations affect financial institutions' performance.
Consumer theory stipulates that new products or services, such as those developed by fintech startups, act as either complements to or substitutes for existing products or services (Aaker and Keller, 1990;Frank, 2009). The products and services that fintech start-ups offer are more likely to benefit traditional financial institutions if they are complements but threaten incumbent institutions' performance if they are substitutes (Kaul, 2012). While fintechs have the potential to develop revolutionary business models, collaborations between banks and fintechs have most often been evolutionary in nature (Bhalla, 2019). Thus, existing products or services have merely been enhanced, with innovations rarely replacing existing ones (Merton, 1995). For example, invoice trading and factoring always existed, but the innovation of fintechs was to scale these services down and offer them to small and medium-sized enterprises (Dorfleitner et al., 2017).
Most research concludes that IT is beneficial for incumbent institutions because it helps reduce transaction costs, thereby improving service quality, optimizing business structure, and promoting business transformation and upgrading (Shu and Strassmann, 2005;Lapavitsas and Dos Santos, 2008;Martín-Oliver and Salas-Fumás, 2008). Moreover, empirical evidence shows that many incumbent institutions acknowledge the superiority of fintech start-ups and have incorporated these start-ups and/or their products and services into their own business models .
For these financial institutions, the emergence of fintech start-ups results in a beneficial partnership rather than a threat (PwC, 2016). For example, the verification of customers' identity through account or video verification supports the customer onboarding process, without cannibalizing existing business from incumbents.
Historically, some scholars have claimed that the opposite is true and that IT could bring enormous challenges to commercial banks (Holland et al., 1997), because IT, globalization, and deregulation allow for new market entrants, disintermediation, innovation, and customer changes on a massive scale. Accordingly, fintech start-ups would take over several key functions of traditional financial institutions (Li et al., 2017). New market entrants benefit from their lack of legacy infrastructure and low levels of organizational complexity, which allows them to be more agile, innovate faster, and be more radical in their approach to innovation (Brandl and Hornuf, 2020). In other words, fintech start-ups are likely to absorb the inefficient operation of traditional financial institutions' existing business. This substitution effect is also in line with disruptive theory (Christensen, 2013), which claims that new entrants effectively compete with traditional players by providing accessible and cost-effective goods and services to customers. As a result, start-ups eventually replace incumbents. Fintech have already sparked such a disruptive evolution when offering financial products and services to customers in novel and more cost-efficient ways (Ferrari, 2016). The efficiency increase due to fintechs results, for example, from disintermediation that significantly lowers transaction costs for consumers (KMPG, 2016;PwC, 2016). Blockchain technology is one of the most prominent inventions that can accomplish such efficiency increases (Wood and Buchanen, 2015;Peters and Panayi, 2016), for example, by making the clearing and settlement of securities and many other services of the financial sector more cost-effective.
Moreover, fintechs have developed applications to improve efficiency in financial services across a range of other services, including mobile and instant payment services, automated asset management, and digital information and data management (Villeroy de Galhau, 2016). These innovations take advantage of traditional financial institutions because many incumbents still rely on an outdated IT infrastructure (Laven and Bruggink, 2016;Brandl and Hornuf, 2020) and have difficulties in adopting new financial products and services or in the same quality as fintechs.
Furthermore, traditional financial institutions are often less likely to adopt new technologies quickly because of restrictions stemming from the regulatory environment that applies to fully regulated institutions (Hannan and McDowell, 1984).
It might also be argued that fintechs have no effect on banks performance, because fintechs attract customers who traditional financial institutions do not serve. One of the most prominent examples is the implementation of mobile payment and banking services in Kenya (Jack and Suri, 2014;Suri and Jack, 2016). Moreover, Jagtiani and Lemieux (2018) find that consumer-lending activities on the platform LendingClub have penetrated areas that may be underserved by traditional banks, mostly in highly concentrated markets and areas that have had fewer bank branches. For example, risky start-up firms and consumers who lack credit history often do not obtain access to credit, especially if the desired loan amounts are small and associated with high transaction costs (Demos, 2016;Hayashi, 2016). Fintech start-up often use novel, sometimes algorithm-driven technology to assess borrowers' creditworthiness at lower costs, which has been an advantage over traditional banks that operate physical branches and employ human loan officers (Hayashi, 2016).
Finally, existing financial institutions can acquire fintech start-ups perceived as "too" innovative and cost-effective. In this way, incumbent institutions gain access to new technology and can adapt it to their own specific needs. For example, Capital One, one of the largest banks in the U.S. in total assets and market capitalization, acquired the fintech start-up Level Money in 2015. Level Money was a San Francisco-based digital banking technology firm that provided customers with a simple overview of their finances. With more than 800,000 downloads, the Level Money app connects to 250 U.S. financial institutions (Li et al., 2017). After its acquisition, Level Money became part of Capital One's Digital Innovation Team, which enables the bank to strengthen its capabilities in digital banking technologies (High, 2016).
With the acquisition of fintech start-ups, financial institutions might not only obtain new retail customers but also extend their existing business through fintech corporate clients. Some of the more traditional financial institutions have realized the potential that stems from the emergence of fintech start-ups and have specialized in what is called "banking as a service" (BaaS) or "banking as a platform" (BaaP). In the BaaS business models, financial institutions operate a licensed and regulated banking back end and offer BaaS middleware to fintech start-ups that cannot or do not want to incur the costs of being fully regulated themselves. In other cases, financial institutions might offer regulatory advice or technology to fintechs that have not yet acquired the respective knowledge or find doing so not cost-efficient. In either case, the division of value creation between fintechs and banks might ultimately benefit both. Overall, we therefore conjecture that financial institutions will not go down without a fight or without any attempt to improve their business models after the emergence of fintech start-ups. We therefore hypothesize the following: H1. Fintech start-up formations are positively related to traditional financial institutions' performance.
Extensive theoretical and empirical research has investigated the determinants of the default risk of financial institutions, because financial stability is of utmost importance for the economy and financial supervisory authorities. Finance scholars have examined the default risk of financial institutions mostly from two perspectives. The first stream of literature focuses on financial institutions' characteristics, including their size (Saunders et al., 1990;Laeven and Levine, 2009;Afonso et al., 2014), liquidity (Diamond and Dybvig, 2000;Diamond and Rajan, 2012), diversification of funding activities (Demirgüç-Kunt and Huizinga, 2009), bank capital as a share of risk-weighted credit exposures (Furlong and Keely, 1989), and corporate governance (Agoraki et al., 2010;Chen et al., 2017). The second stream of literature focuses on the determinants of risktaking that results from external sources, such the degree of bank competition (Boyd and De Nicolò, 2005;Beck et al., 2006;Beck et al., 2013), monetary policy (Borio and Zhu, 2012;Chen et al., 2017), deposit insurance schemes (Demirgüç-Kunt and Detragiache, 2002;Angkinand and Wihlborg, 2010), external regulation (Barth et al., 2004;Klomp and De Haan, 2012) such as creditor and minority shareholder protection (La Porta et al., 2000;Houston et al., 2010), and political institutions (Chen et al., 2015;Ashraf, 2017;Wang and Sui, 2019).
In this study, we investigate the default risk of financial institutions following the emergence of fintech start-ups. Fintechs' impact on the default risk of financial institutions is not clear per se.
Several factors could lead to an increase in the default risk in the financial industry. Fintech startups often provide similar financial products and services to those of incumbents (Dorfleitner et al., 2017;Yao et al., 2018;Kommel et al., 2019), and in some cases, their business models are inherently linked to traditional financial institutions. For example, in many jurisdictions commercial loans can only be extended by institutions that possess a banking license. Marketplace lending platforms, for example, often do not possess a banking license, and a bank in the background ultimately extends the loan between the borrower and the lenders (Cumming and Hornuf, 2020). Thus, banks are often an integral part of fintech business models. However, startups generally fail more often than established firms (Evans, 1987;Dunne et al., 1989;Cressy, 2006), which could increase the risk of firms that collaborate with them. Conversely, fintechs could also lower the default risk of financial institutions. The digitalization of lending activities likely lowers transaction costs and improves the efficiency of the loan origination and maintenance processes (BIS, 2017). This could reduce the costs of capital for borrowers and improve the risk-adjusted returns for fintechs and traditional financial institutions. Moreover, because fintech start-ups employ modern technology and use big data, at least theoretically, they can better address information asymmetries (Lin et al., 2013;Ge et al., 2017;Xu and Chau, 2018).
Ecosystems that promote the sharing of data can further enable the development of novel products and services. The European Banking Authority (2019) expects a positive effect of application programming interfaces, which allow for a more direct exchange of data, leading to increased competitive pressure and improved customer experiences. We therefore hypothesize the following: H2a. Fintech start-up formations decrease financial institutions' default risk.
Traditional financial institutions invest in fintech start-ups, which allows them to better access their knowledge Shin, 2018, Hornuf et al., 2020). As fintechs grow larger and become more integrated and interconnected with the financial sector, they may also affect systemic risk. A prominent example is the German payment acquirer Wirecard, which in 2020 collapsed and subsequently filed for default because of a series of fraudulent accounting activities and inflated profits. Although Wirecard had been part of the Prime Standard, the market segment of the Frankfurt Stock Exchange with the highest transparency standards, it was itself considered a fintech company. Wirecard not only collaborated with other fintech start-ups, such as Holvi, Lendico, Number 26 (now N26), Rate Pay, and Zencap, 2 but also engaged in alliances with large financial conglomerates such as the insurance company Allianz (Reuters, 2020). To offer lending services, Wirecard operated the subsidiary Wirecard Bank, which had a banking license and was fully regulated and monitored by the Federal Financial Supervisory Authority (Bundesanstalt für Finanzdienstleistungsaufsicht [BaFin]). Wirecard was not classified as a financial holding company, and only the subsidiary had been classified as a financial company by BaFin, which implied that the holding company's activities were supervised only loosely, and accounting fraud remained undetected (Navaretti et al., 2020).
Although bank-fintech collaborations have been rapidly growing in many economies, related supervision has developed only slowly, as the Wirecard case evidences. After the collapse of Wirecard, the German legislator proposed a draft law to strengthen financial market integrity (Finanzmarktintegritätsstärkungsgesetz), targeting a wide range of financial market regulations.
While the Wirecard accounting scandal did not affect the German or European financial system as such, it raised questions about how financial subsidiaries of tech companies can seamlessly continue operating after a holding company files for default and how business partners can seamlessly switch their operations to another institution. Without doubt, as fintechs become more mature and interconnected, concerns about market risk and systematic risk rise. However, it should be noted that the collapse of Wirecard did not result in a financial turmoil comparable to the collapse of Lehman Brothers in 2008.
Moreover, having access to alternative financial products such as marketplace or mobile loans, which, to a lesser degree, are correlated with other loans and institutions, can reduce systemic risk in the financial industry (BIS, 2017). A greater share of fintech credit through marketplace loans or mobile loans could thus mitigate problems of too-big-to-fail or too-systemic-to-fail institutions.
Marketplace lending platforms operated by fintechs have minimal direct financial exposure to each other, a systemic benefit that might disappear if fintechs become more interconnected over time (BIS, 2017). Furthermore, the use of biometric information and other enhanced data security measures that fintechs implemented early on are considered to have improved data security, potentially lowering the risk of cyber-attacks. Finally, systemic risk could also be reduced through enhanced market transparency, which could result from the more extensive use of cloud computing and decentralization (European Banking Authority, 2017). We therefore hypothesize the following: H2b. The exposure of traditional financial institutions to systemic risk is negatively related to fintech start-up formations.

Dependent variables
To investigate whether fintech start-up formations affect incumbent institutions' performance and default risk, we consider eight dependent variables. For most of these variables, we need daily stock returns as a basis. For U.S. financial institutions, we obtained daily stock returns from the  Table A1 in the Appendix). For each listed financial institution, we collect adjusted prices or adjustment factors, the number of shares outstanding, the location of the headquarters, and calculated annual returns. 4 With adjusted prices and number of shares outstanding, we can compute market valuation.
We use returns and market valuation of financial institutions to compute value-weighted market return indices of the financial sector of each country. Because we need yearly financial institutionlevel variables to assess the impact of fintech formation on financial institutions' performance, we collapsed all daily firm data to yearly data. To test hypothesis 1, and in line with Phan et al. (2020), we calculate the net interest margin, return on assets (ROA), return on equity (ROE), and Tobin's Q as measures of financial institutions' performance. Tobin's Q traditionally measures the sum of the market value of equity and the book value of liabilities divided by the book value of total assets.
We compute financial institutions' performance also with a market measure. We chose to analyze annual stock returns because stock prices better reflect current information about and expectations of firms' future profitability and growth (Anilowski et al., 2007).
To test hypothesis 2a, we use accounting and market measures of risk in our analysis. The first measure of financial institution default risk is the Z-score of each financial institution, which equals the ROA plus the capital-asset ratio divided by the standard deviation of the ROA. The Z-score thus measures the number of standard deviations below the mean by which profits would have to fall to deplete the financial institution's equity capital completely (Boyd et al., 2006). The measure has a long tradition in the finance literature (Roy, 1952) and is still used in empirical research to capture a financial institution's distance from default (Laeven and Levine, 2009;Pathan, 2009;Houston et al., 2010, Jin et al., 2013Bhagat et al., 2015). A higher Z-score value indicates a lower default risk and greater stability of the respective financial institution. Because the Z-score is often highly skewed, we follow Laeven and Levine (2009) and use the natural logarithm of the Z-score in our estimations. Our second measure of financial institution default risk is the volatility of stock returns, which has been widely used in prior research (Pathan, 2009;Sun and Liu, 2014;Brown et al., 2015). It captures the market's perception of the risk inherent in banks' assets, liabilities, and off-balance-sheet positions (Pathan, 2009).
To test hypothesis 2b, we consider the marginal expected shortfall, which captures a financial institutions' exposure to systemic risk. It measures the average of individual stock returns on a subset of sample days that correspond with the 5% worst days of the equally weighted market index.

Explanatory variables
The data source for our explanatory variable of interest is the CrunchBase database, which contains detailed information on fintech start-up formations and their financing. The database is assembled by more than 200,000 company contributors, 2,000 venture partners, and millions of web data points 5 and has recently been used in scholarly articles (Cumming et al., 2016;Bernstein et al., 2017;Haddad and Hornuf, 2019). We retrieved the data for our analysis on July 9, 2019. Because CrunchBase might collect some of the information with a time lag, the observation period in our sample ends on December 31, 2018. Overall, we identified 12,549 fintech start-ups from 87 countries for our relevant sample period.
To account for financial institution and cross-country heterogeneity, we consider several variables frequently used as controls in the bank performance literature (Agoraki et al., 2011;Tabak et al., 2012;Phan et al., 2020). Following Pathan and Faff (2013), Shaban and James (2018), Dietrich and Wanzenried (2014) and Berger et al. (2017), we control for total assets as a measure of average firm size, the capital ratio, the cost income ratio, the interest income margin, and the book-tomarket ratio. All variables came from the CRSP and Compustat databases. To address country-time-specific heterogeneity, we consider several macroeconomic indicators. We control for gross domestic product (GDP), because it might influence bank performance through the business cycle.
When the economy faces a recession or an economic crisis, the quality of borrowers deteriorates, which in turn worsens banks' loan portfolio and affects their performance. On the loan demand side, borrowers are less willing to invest in long-term projects in times of crisis and often cut spending. Not surprisingly, the empirical literature shows that economic growth also stimulates the financial system (Athanasoglou et al., 2008;Albertazzi and Gambacorta, 2009). We also account for inflation as a measure of financial institutions' performance, because research shows a positive relationship between inflation and profits (Kasman et al., 2010;Trujillo-Ponce, 2013). However, if inflation is anticipated and financial institutions fail to adjust their interest rate, costs can increase faster than profits, which negatively affects bank performance. Therefore, the effect of inflation on bank performance is ambiguous.
To control for the extent to which countries' political decisions affect bank performance, we include the variable size of government, which combines five components: government consumption, transfers and subsidies, government enterprises and investment, top marginal tax rate, and state ownership of assets. The variable ranges from 0 to 10, with higher values indicating that countries rely more on personal choice and markets rather than government budgets and political decision-making. To control for differences in the efficiency of legal protection and enforcement of laws across economies, we consider the variable legal protection curated by the Fraser Institute database. It entails several legal system components, including rule of law, security of property rights, an independent and unbiased judiciary, and impartial and effective enforcement of the law. These components are indicators of how effectively the protective functions of the legal system are performed. The variable ranges from 0 to 10, with higher values indicating better government efficiency in terms of legal protection.
Finally, we control for the impact of the concentration of banks on bank performance. Empirical research still shows ambiguous results for this variable. In the European context, Delis and Tsionas (2009) find that firms with market power tend to operate inefficiently, because managers enjoy monopoly profits. Maudos and De Guevara (2007) find no evidence of a significant relationship between firm concentration and performance. The measure came from the World Bank database and reflects the sum of market share in terms of total assets of the three largest banks. Table A2 in the Appendix provides definitions of all variables and their sources.

Model specifications
Our empirical approach is motivated by recent research estimating determinants of bank performance (Köster and Pelster, 2017;Shaban and James, 2018;Phan et al., 2020). To test our hypotheses, we use a two-step generalized method of moments (GMM) system dynamic panel estimator (Arellano and Bover, 1995;Blundell and Bond, 1998 where RISK represents one of three different dependent variables: Z-score, stock return volatility, and marginal expected shortfall. We use year dummies in all models to account for business cycle effects. In addition, we use heteroskedasticity-robust standard errors clustered at the financial institution level. Table 1 reports the baseline regression. 6 Columns represent the five dependent variables measuring performance: net interest margin, ROA, ROE, Tobin's Q, and annual stock return. We find that sector-specific and macro-level variables have an economically meaningful and statistically significant impact on financial institutions' performance. The control variables that are significant performance predictors in three models are lagged cost income ratio and market-to-book ratio.

Benchmark model
Inflation and bank concentration are statistically significant in two of the five models, while the interest income margin is significant in one model.

Lag effect of fintech start-up formations on bank performance and risk-taking
In Table 2 --- Table 2 About Here ---Next, we test whether the effect of fintech start-up formations on financial institutions' performance differs for large and small institutions. Recent research suggests that financial characteristics of institutions are important predictors of their performance (Dietrich and Wanzenried, 2011;Köster and Pelster, 2017;Talavera et al., 2018). We treat the market value of a financial institutions as a proxy to differentiate large, universal financial institutions from small, specialized financial institutions. On the one hand, we expect large financial institution to adapt their business models at a slower rate than small financial institution, which presumably have already specialized in business models such as BaaP and BaaS. On the other hand, large financial institutions often have deeper pockets and can more forcefully pursue change through acquisitions and in-house experimentation. Our results show a positive and significant association between the formation of fintech start-ups and large financial institutions' performance. The results in Table 3 show that for financial institutions with above-median market value, fintech start-up formations have a positive and robust effect on three of the five measures for financial institutions' performance-ROA, ROE, and annual stock return. For financial institutions with below-median market value, we do not observe any significant association between fintech formations and financial institutions' performance. Large financial institutions might benefit from alliances with fintechs, for example, through product-related corporations or partial acquisitions of fintechs, which help them gain specialized knowledge and improve their performance .
This result does not necessary imply that small financial institutions are reluctant to change. Indeed, these institutions might already possess a more modern IT infrastructure and thus benefit only at the margin from fintech start-ups.
--- Table 3  recent event study shows that at least in the short run, the market perceives announcements of bank-fintech alliances negatively . In a next step, we therefore divide our sample into two subsamples and test whether the development of fintech start-ups has a differential impact on financial institutions' performance for the periods 2005-2011 and 2012-2018.
The results in Table 4  performance. Thus, the pressure resulting from fintech start-ups following the financial crisis appears to have vanished over time potentially as a result of more cooperative business models, though the positive association between performance and fintech formations has not entirely disappeared in recent years.
--- Table 4 About Here ---In Table 5, we test whether fintech start-up formations predict the default risk of financial institutions. The columns report estimates for our dependent variables of interest-Z-score, stock return volatility, and marginal expected shortfall.
Fintech start-up formations have a negative and statistically weak significant effect on the accounting measure Z-score. If this result stems from a lack of statistical power, it would indicate that fintech start-ups are associated with a higher probability of default of financial institutions.
However, the results we obtain for the Z-score are weakly significant and thus should be interpreted with caution. First, the Z-score computation is based on accounting data, which are only as good as the underlying accounting and auditing framework. The case of Wirecard constitutes a recent example that accounting measures might not reflect the actual situation of a financial institution.
Second, if financial institutions are able to smooth out the reported data, the Z-score provides an overly positive assessment of the financial institution's stability.
Using a market measure for our dependent variable, we find that the development of the fintech sector has decreased financial institutions' stock return volatility. This is in line with Hypothesis 2a that fintech start-up formations decrease financial institutions' default risk.
Finally, the stock return volatility assesses each financial institution separately, neglecting that a default of one financial institution may cause losses to other financial institutions in the system.
During the 2007-2008 financial crises, it became evident that many financial institutions were interconnected and market contagion occurred as a domino effect. Using the marginal expected shortfall as our dependent variable, we capture the effect of fintech formations on financial institutions' exposure to systemic risk. We find that the development of fintech start-ups decreases incumbents' exposure to systemic risk, which is in line with Hypothesis 2b. Not only does the spread of fintechs result in more competition and better performance of traditional financial institutions, but it also increasingly diversifies the use and execution of financial services over different market players. In this sense, the rise of fintechs might, to some degree, counteract the too-systemic-to-fail problem.
--- Table 5 About Here ---To test the robustness of our results, we calculate the number of fintechs founded per year and country and divide them by the total number of start-ups founded during that year in the respective economy as an alternative measure of fintech start-up formations. The data for start-up formations came from the Crunchbase database. As Table 6 reports, the results are similar to previous findings that fintech formations positively predict bank performance. With regard to default risk, we also find that fintech start-up formations negatively affect financial institutions' default risk, as indicated by the decrease in stock return volatility.

Discussion and conclusion
The article investigates whether fintech start-up formations affect financial institution' performance and default risk. We evidence that fintech start-up formations improve financial institutions' performance in terms of accounting and market measures. These findings are in line with previous research (Vives, 2019) that posits that banks rethink and reshape their business model when confronted with competitive pressure. One potential way for financial institutions to improve performance when confronted with fintechs is by cooperating with and integrating the new players in their organization . Moreover, we use the marginal expected shortfall as a measure of systemic risk and find that financial institutions' exposure to systemic risk decreases when more fintech start-ups enter the market. This finding sheds light on how financial institutions can benefit from technology spillovers when confronted with novel technological solutions developed by fintechs (Blalock and Gertler, 2008;Newman et al., 2015).
, −1 + 2 , −2 + 3 , + 4 , + 5 , + 6 , + 7 , + 8 , + 9 , + 10 , + 11 , + 12 , + , . In this regression, performance respectively represents one of the five different dependent variables: net interest margin (NIM), ROA, ROE, Tobin's Q, and RETURNS. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. , = + 1 , −1 + 2 , −1 + 3 , −2 + 4 , + 5 , + 6 , + 7 , + 8 , + 9 , + 10 , + 11 , + 12 , + 13 , + , . In this regression, performance respectively represents one of the five different dependent variables: net interest margin (NIM), ROA, ROE, Tobin's Q, and RETURNS. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. High market value contains the top-half financial institutions with the highest market value, while low market value includes the bottom-half financial institutions with the lowest market value. These categorizations are based on the median market values. The regression model takes the following form: , = + 1 , −1 + 2 , −1 + 3 , −2 + 4 , + 5 , + 6 , + 7 , + 8 , + 9 , + 10 , + 11 , + 12 , + 13 , + , . In this regression, performance respectively represents one of the five different dependent variables: net interest margin (NIM), ROA, ROE, Tobin's Q, and RETURNS. The descriptions of the control variables are provided in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. , = + 1 , −1 + 2 , −1 + 3 , −2 + 4 , + 5 , + 6 , + 7 , + 8 , + 9 , + 10 , + 11 , + 12 , + 13 , + , . In this regression, performance respectively represents one of the five different dependent variables: net interest margin (NIM), ROA, ROE, Tobin's Q, and RETURNS. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. , + , . In this regression, RISK respectively represents one of the three different dependent variables: Ln Z-score, volatility, and marginal expected shortfall. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively. Table 6 Robustness check. Alternative measure of fintech formation consists on the ratio of the number of fintech founded in a year and country divided by the total number of start-ups founded in a year and country.

I-Lag effect of Fintech firms on bank performance
(1) (2) (3) (4) , + , . In this regression, performance respectively represents one of the five different dependent variables: net interest margin (NIM), ROA, ROE, Tobin's Q, and RETURNS. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively.

II-Lag effect of Fintech firms on bank risk-taking
(1) (2) In this regression, RISK respectively represents one of the three different dependent variables: Ln Z-score, volatility, and marginal expected shortfall. The descriptions of the control variables are noted in Table A2 of the Appendix. The estimation method is the two-step GMM system dynamic panel estimator. Year dummies are included in the model to account for heterogeneity across time. p-values are computed by the heteroskedasticity-robust standard errors clustered for bank level. The Arellano-Bond test for serial correlation is based on the null hypothesis of the second-order autocorrelation in the first differenced residuals. The p-value associated with the Hansen test for determining the validity of the overidentifying restrictions is reported. Finally, † , **, and *** denote significance at the 10%, 5%, and 1% levels, respectively Table A2 List of variables

Dependent variables
Annual stock return Annual stock return derived from daily returns using the ascol STATA command. Source: CRSP/Compustat database and own calculation.

Net interest margin
The net interest margin is the ratio of the net interest income to total assets. Source: CRSP/Compustat databases and own calculation.

ROA
The ROA is the ratio of net income to total assets. Source: CRSP/Compustat database and own calculation.

ROE
The ROE is the ratio of net income to total equity. Source: CRSP/Compustat database and own calculation.
Tobin's Q The sum of the market value of equity plus the book value of liabilities divided by the book value of total assets. Source: CRSP/Compustat database and own calculation.
Z-score Computed as (ROA + CAR)/STD(ROA), where ROA is earnings before taxes and loan loss provisions divided by assets, CAR represents the capital asset ratio, and STD(ROA) is the standard deviation of ROA over the period studied. Source: CRSP/Compustat database and own calculation.

Volatility
Volatility is the standard deviation of daily stock returns over one year. Source: CRSP/Compustat database and own calculation.

Marginal expected shortfall (MES)
The marginal expected shortfall is the marginal contribution of firm j to the expected shortfall of the financial system. Formally, marginal expected shortfall for firm j is the expected value of the stock return ̃ conditional on the market portfolio return ̃ being at or below the sample qpercent quantile. Source: CRSP/Compustat database and own calculation.

Size
The number of fintech start-ups founded by year and country. Source: Crunchbase and own calculation The natural logarithm of total assets in millions of USD. Source: CRSP/Compustat database and own calculation.

Capital ratio
The capital ratio is calculated as the firm's equity over its total assets. Source: CRSP/Compustat database and own calculation.

Cost income ratio
The cost income ratio is total expenses over total generated revenues. Source: CRSP/Compustat database and own calculation.

Interest income margin
The interest income margin is the total interest income over total income. Source: CRSP/Compustat database and own calculation.
Market-to-book ratio The market-to-book ratio is the market capitalization over the book value. Source CRSP/Compustat database and own calculation.
GDP growth Country-level annual GDP growth rate. Source: World development indicators database.

Inflation
Country-level annual inflation rate. Source: World development indicators database.
Size of government Includes five political system measure components: government consumption, transfers and subsidies, government enterprises and investment, top marginal tax rate, and state ownership of assets. The variable ranges from 0 to 10, with higher ratings indicating that country relies more on personal choice and markets rather than government budgets and political decision-making. Source: The Fraser institute database.
Legal protection Includes nine legal system measure components: rule of law, security of property rights, an independent and unbiased judiciary, and impartial and effective enforcement of the law. The nine components in this area are indicators of how effectively the protective functions of government are performed. The variable ranges from 0 to 10, with higher ratings indicating better government efficiency in terms of legal protection. Source: The Fraser institute database.
Bank concentration Raw data are from Bankscope. (Sum(data2025) for three largest banks in Bankscope) / (Sum(data2025) for all banks in Bankscope). Only reported if number of banks in Bankscope is 3 or more. Calculated from underlying bankby-bank unconsolidated data from Bankscope. Source: World development indicators database.