Sage Journals: Discover world-class research

Abstract

This paper studies the effects of bank digitalization on bank transparency. We propose a novel, time-varying measure of bank digitalization based on variations in broadband access across bank depositors. Our baseline digital measure captures how digital a bank’s customers are using a deposit-weighted access to broadband metric. Analyzing a sample of over 5,800 banks, this paper finds that more digital banks are more transparent. This finding highlights the complementarity between bank digitalization and transparency and has implications for technology-driven financial fragility. We test alternative digitalization measures by taking advantage of variations in the timing of mobile app roll-out for medium and large banks. Using our novel digitalization measure, this paper also shows that larger banks are more digital and that banks are more likely to adopt mobile apps when they become more digital. Furthermore, we document that the positive effect of digitalization on transparency decreases with the size and financial health of the bank, which has important implications for policy aimed at increasing bank transparency. Overall, this paper fills a gap in the literature by documenting the positive impact of digitalization on bank transparency.

Plain language summary

As the economy digitalizes, an important question is how digitalization impacts the stability of the financial system. This is a broad question, and this paper focuses on the impact of digitalization on the banking sector by proposing a novel digital score measure. We test the research hypothesis that digitalization leads to a higher level of transparency in bank disclosures to the public. Bank transparency is important because transparency can help regulators and depositors gain information about a bank’s loan portfolio. To that end, this paper finds that bank digitalization increases transparency in the economy. Empirically, we propose a novel measure of bank digitalization by measuring the likely broadband access of a bank’s customers. We compute this digital score measure for a sample of 5,800 banks over time. In addition to documenting the positive impact digitalization has on bank transparency, we also study how digitalization interacts with other underlying characteristics of importance such as bank size and bank loan performance.

Keywords

digitalization digital economy financial institutions bank opacity bank transparency

Introduction

History tells us technological innovations spur changes in the banking sector. One example is the introduction of railroads. In the pre-Civil War Free Banking Era from 1834 to 1863, banknotes could only be redeemed at the specific bank that issued the note. This transaction friction was not lost on depositors. Banknotes circulating in the city of the issuer traded at par. However, banknotes from distant cities traded at a discount, reflecting the inconvenience of redemption. As railroads reduced travel times across cities, banknotes became more convenient to use as money (Gorton, et al., 2022). The historical evidence of technology impacting the banking sector motivates our focus on the impacts of bank digitalization.

In this paper, we ask the question: Are more digital banks more transparent? Transparency speaks to two streams of banking theories. On the one hand, transparency stabilizes the banking system by allowing sensitive depositors to vote with their feet. Thus, greater transparency allows for better monitoring by regulators and depositors and makes the disciplining mechanism more effective (Calomiris & Kahn, 1991). The alternative stream of banking theory emphasizes that greater transparency could introduce fragility in the banking system by making uninsured deposits more informationally sensitive (Dang et al., 2017). The importance of transparency in the banking theory literature motivates our focus on the link between digitalization and transparency. Do more digital banks lead to greater transparency? Or, are the two substitutes? The challenge to answering this research question is multi-fold. First, we must define a measure of transparency. Second, we must define a measure of digitalization. Third, we must account for omitted variables that could simultaneously affect a bank’s decision to be digital or to be transparent.

To address the first challenge, this paper adopts the transparency measure as introduced by Q. Chen et al. (2022). Chen et al. introduce a novel measure of transparency which reflects the ability of banks’ past disclosures in predicting future loan write-offs. The transparency measure (also referred to as the R² measure) captures the informativeness of bank disclosed earnings. All else equal, banks with higher R² measures are more transparent about their performance. Moreover, Q. Chen et al. (2022) link transparency to deposit flows and find that the sensitivity of uninsured deposit flows to bank performance are higher for more transparent banks. This central finding indicates that uninsured depositors respond to bank fundamentals.

To address the second challenge and to directly answer the research question of how transparency responds to digitalization, this paper proposes a novel measure of digital score. To do so, we make use of the Federal Communications Commission (FCC) dataset on broadband availability at the county level from 2008 to 2021. Using the location of bank branches from the Summary of Deposits (SOD) data, we compute a deposit-weighted measure of broadband access for a bank’s depositor base. Using this methodology, we compute time-varying digital scores for 5,843 unique banks with 40,158 bank-quarter observations. This is our primary measure of bank digitalization. For a subset of banks, we also collect information on whether they released mobile banking applications and the date of application release.

To address the third challenge, this paper addresses the concern that omitted variables could drive both broadband access in the regions that the bank is located in and the bank’s level of transparency by using a two-way fixed effects (TWFE) approach. We control for both time and bank fixed effects, in addition to a host of controls, to capture both time-variant and time-invariant bank characteristics that could be driving both digitalization and transparency. In particular, we establish a new descriptive fact that larger banks are more digital and include bank size as a control variable to rule out the size effect in driving our results.

The central finding of the paper is that an increase in likely broadband access of bank depositors has a positive effect on bank transparency. We find that for a 20% increase in broadband access amongst a bank’s depositors, the bank’s transparency score increases by 0.32, meaning that the R-squared score from regressing past bank Call Report information on future loan write-offs increases by .32. This finding is significant in illustrating the complementarity of bank digitalization and transparency. A more digital bank is also more transparent. Furthermore, this paper documents interesting heterogeneity in the impact of digitalization on bank transparency depending on the underlying bank characteristics. Using a TWFE regression design with interaction terms, we find that the positive impact of digitalization on bank transparency is decreasing in bank size and decreasing in the financial health of the bank, measured as whether the bank reports a loss in the current quarter.

Given our findings, the contributions of this paper are mainly divided into the following aspects. Frist, we propose a novel measure of bank digitalization based on deposit-weighted access to broadband in counties where bank branches are located. This measure captures a dynamic representation of digitalization and can be adopted by other researchers in studying the impact of digitalization both in the time-series and in the cross-section. Second, we use econometric methods to test the hypothesis that an increase in digitalization leads to an increase in bank transparency. Importantly, we show that this effect of digitalization is not uniform across all banks and is decreasing in bank size and bank financial health. Moreover, using these empirical results, we are able to comment on predictions from theory literature on bank transparency. In particular, we find evidence in support of the findings of Acharya and Mora (2015) that banks may choose to reduce transparency during periods of financial distress. The findings of this paper have important implications for policy targeted toward digitalization and bank regulation. In particular, our results suggest that policymakers could promote digitalization as an effective means of increasing transparency within the banking system.

In addition to literature on bank transparency (Acharya & Ryan, 2016; Beatty & Liao, 2014; Bushman, 2014), our paper relates to a burgeoning literature on the impact of digitalization on financial institutions. Recent literature has mostly focused on the impact of the entry of FinTech firms to traditional banks (Berg et al., 2022; Buchak et al., 2018; Gopal & Schnabl, 2020). This paper contributes to the discussion on how digitalization of traditional banks themselves impact the financial system. To our knowledge, this paper is the first to examine the impact of bank digitalization on transparency, in contrast to existing studies that primarily focus on its effects on deposits. Furthermore, this paper presents new facts surrounding bank digitalization and transparency.

This paper is organized as follows. Section 2 discusses relevant literature and research hypotheses. Section 3 describes the data, regression specifications, and computation of the digitalization measure. Section 4 discusses key results and contextualizes our findings. Section 5 concludes. Section 6 discusses limitations. Section 7 proposes directions for future research.

Literature Review

Digitalization of the Banking Sector

There is a wealth of literature on the digitalization of the financial system which closely track infrastructure developments over time. Taking the United States as an example, the banking sector has evolved with the introduction of new technologies such as railroads (1850s), credit cards (1950s), ATMs (1960s), and online banking (2000s). Recent studies on the digitalization of the banking sector have increasingly focused on the rise of FinTech as a new form of financial intermediation (Tepe et al., 2021; Tiberius et al., 2022). Broadly, the literature has converged on two main theoretical arguments for the existence of new financial institutions. First, FinTech companies that conduct their businesses online rather than via physical branches benefit from decreasing costs through their online operational set-up (Arslanian & Fischer, 2019). Second, from a policy perspective, these digital institutions are often cited as a promising solution for promoting financial inclusion, especially in areas without physical bank branches presence (Glushchenko et al., 2019; Ramdani et al., 2020).

The literature also examines the interaction of traditional banks with their digital competitors. Given that digital and traditional firms often compete in the same industry or market segments, researchers contend that there exists a substitution effect. For example, Fuster et al. (2019) document an increased market share of FinTech lenders relative to traditional mortgage lenders in the United States. Similar substitution arguments are made in Buchak et al. (2018) and Erel and Liebersohn (2022). Conversely, researchers also document complementarity between the two forms of financial intermediation. For example, newer forms of financial institutions such as neo banks often work closely with traditional banks in order to offload their balance sheet risk, and traditional banks also benefit from balance sheet diversification (Hornuf et al., 2021).

However, less explored in the literature are topics on how digitalization within traditional banks themselves drive changes in banking businesses. Most closely related to our work are research on the impact of digitalization on deposit flows. For example, Benmelech et al. (2024) show that deposit outflows following the collapse of SVB was the most salient for banks with low branch density. In an effort to explain this phenomenon, the authors attribute this to the observation that banks with low branch density likely have high online presence, which allows for a faster movement of capital for their depositors. Related to the study of deposit outflows for banks with less physical presence, Erel et al. (2024) document that deposit flows are more sensitive to fed fund rate announcements for banks with more online presence. As opposed to studying the flow of deposits around banking sector shocks such as the SVB collapse, the work by Erel et al. (2024) show that deposits systematically over-react in banks with a more digital operational framework.

Our work contributes to existing literature along two dimensions. First, as opposed to juxtaposing new entities with traditional banks, we show that traditional depository institutions themselves are experiencing digitalization. Hence, we contribute to the literature by documenting a novel measure of digitalization for traditional banks. Second, and most importantly, compared with the existing papers examining the impact of bank digitalization on deposit flows, this paper fills an opening in the literature by connecting bank digitalization with bank transparency. Much of the existing work focuses on the influence of bank digitalization on the flight of deposits in response to shocks or monetary policy. However, to our knowledge, this is the first paper which studies the impact of bank digitalization on bank transparency, which is a topic of great policy interest particularly following the 2023 banking crisis.

Theory of Bank Transparency and Opacity

Our paper is related to several streams of literature. First, this paper is related to banking and accounting literature on the role of transparency in the banking system. To fix ideas, transparency can be understood as the availability of relevant information to outside stakeholders including investors, regulators, and policymakers. Bank transparency, therefore, represents the availability of performance and risks-related information to outside stakeholders, including depositors (Bushman, 2014). Bank opacity, on the other hand, refers to the propensity of banks in withholding information and can be understood as the opposite of bank transparency. For example, when bank portfolio choice is not publicly available, banks take advantage of opacity to take on more risk (Matutes & Vives, 2000). Bank transparency are understood as highly relevant to policymakers and regulators, and there has been a renewed interest in the topic following the most recent Silicon Valley Bank (SVB) failure.

A large literature exists in exploring the impact of bank transparency versus opacity on financial stability. Earlier banking literature suggests that transparency promotes bank stability by enhancing the market discipline of banks’ risk-taking decisions (Blum, 2002; Cordella & Yeyati, 2003; Rochet, 1992). Under this stream of thought, ex-ante to a bank’s risk-taking behavior, managers recognize transparency allows market discipline and hence reduces moral hazard induced behaviors. Ex-post, market could discipline the bank via impact on security prices or depositors voting with their feet (Flannery, 2011; Stephanou, 2010). Hence, the broader financial literature understand that transparency can act as a regulatory tool in order to mitigate risks and moral hazard in the banking system. Given this theoretical background, it is important to understand the drivers of transparency and how bank transparency evolves over time. This is where the current paper seeks to make a contribution by understanding the impact of digitalization on bank transparency.

An alternative theory of bank transparency and opacity emphasizes the potential drawbacks of transparency and highlights the benefits for banks to be opaque. This second stream of theoretical hypotheses posit that there exist social benefits of limited disclosure in two ways. First, Y. Chen and Hasan (2006) and Morris and Shin (2002) argue that bank transparency could result in coordination failures on part of bank depositors. That is, when banks are overly transparent in disclosing negative earnings information such as unprofitable investments or losses, depositors may withdraw their money in response to such disclosure. In equilibrium, depositors may pool in the bank run equilibrium, leading to an increase in the likelihood of a banking crisis. Clearly, this argument provides an alternative perspective to the benefit of transparency to overall banking sector stability. Similarly, others have argued that bank transparency could lead to a decrease in general consumer confidence toward the banking sector (Morrison & White, 2013) and could therefore undermine a bank’s ability to issue private money (Gorton, 2013).

Despite these alternative theories, it is understood and agreed upon that bank transparency and opacity are of central importance to regulators because of their implications for financial stability and risk-sharing in the banking sector (Alvarez & Barlevy, 2021; Dang et al., 2017). Hence, this paper makes a contribution both in the cross-section by describing how transparency differs between more and less digital banks. This paper also makes a contribution in the time-series by examining how bank digitalization evolves over time, which has implications for bank transparency over time.

Measurement of Bank Transparency and Opacity

Given the importance of bank transparency and opacity within the theory literature and their implications for policymakers and regulators, the literature has also made progress on measuring these theory concepts empirically. In contrast to the theoretical literature highlighted in the first section, the empirical development in the field lags by many years, and only recently have researchers been able to convincingly capture transparency using bank call reports information (Beatty & Liao, 2011; Q. Chen et al., 2022; Yue et al., 2022). Two dominant ways of measuring transparency have emerged.

In the first attempts of capturing transparency empirically, Beatty and Liao (2011) introduced a notion of transparency as timeliness in information release. That is, a financial institution is thought of as being more transparent if there are not significant delays in reporting of key financial information such as expected loss recognition. In their seminal paper, Beatty and Liao (2011) argue that delay in loan write-off reporting can serve as a reliable indicator of bank transparency by capturing the speed at which bank managers reveal their private information about loan quality. Relatedly, Bushman and Williams (2015) also adopt similar measures of timeliness of accounting reporting as an empirical basis for capturing bank transparency. However, one drawback in using a measure of timeliness as an empirical metric for bank transparency lies in the omission of measurement of information quality. That is, banks could release timely disclosures without adequately disclosing the magnitude or severity of losses in events of loss-reporting.

Given this drawback, recent literature has focused on measuring bank transparency by capturing the information quality of bank disclosures. On this front, Q. Chen et al. (2022) provide a significant contribution to literature by providing a time-varying measure of transparency reflecting the quality of bank disclosure. This is, to our knowledge, the first paper to measure the quality of information release using bank call reports data. In particular, the paper measures transparency as the adjusted R-squared from a regression of bank loan write-offs on the past components of disclosed earnings. Banks with higher R-squared measures when regressing future realizations on previous disclosure are more transparent than banks with lower R-squared measures. Furthermore, Q. Chen et al. (2022) connects their transparency measure with deposit flows and shows that uninsured deposit flow is more sensitive in more transparent banks.

In our paper, we adopt the transparency measure as constructed in Q. Chen et al. (2022) for the following reasons. First, this is the first measure which captures the extent to which information in released call reports explain the uncertainty around future loan defaults, which offers a deeper understanding of transparency compared to simple measures like timeliness. Second, this measure more closely aligns with theoretical understandings of transparency, which centers on reducing information asymmetry and providing stakeholders with clear insights into a bank’s potential risks (Blum, 2002; Cordella & Yeyati, 2003). Third, given we are interested in the impact of digitalization on bank transparency, adopting this measure allows us to make statements about how our novel digitalization measure impacts the quality of disclosure, which is of interest to policymakers and regulators. Given the existing theoretical research on bank transparency and opacity, we propose the following testable hypothesis:

H1. Banks with higher levels of digitalization, measured by deposit-weighted broadband access, exhibit higher levels of transparency.

H2. The positive relationship between digitalization and transparency weakens as bank size increases.

H3. The positive effect of digitalization on transparency is reduced for banks that report losses.

Data and Methodology

In this section, we begin with a description of the data sources used and then expand on the empirical strategies in order to test the hypotheses mentioned above.

Sample and Data

Bank-Level Quarterly Data

We conduct analysis at the bank-quarter level. For our baseline analysis, we collect data on commercial banks from December 2007 to December 2021, totaling 40,158 bank-quarter observations. Note that data at collected at the commercial bank level, not the bank holding company level, following Q. Chen et al. (2022). Following Berger and Bouwman (2009), we classify banks with assets above 3 billion as large banks and banks with assets between 500 million and 3 billion as medium banks. Banks with assets less than 500 million are classified as small banks. To rule out the impact of mergers and acquisitions on bank fundamentals which may impact its transparency score, we drop bank-quarter observations during quarters where the bank experiences an asset growth of greater than 10%. Following Acharya and Mora (2015), we also winsorize all continuous variables at the 1% and 99% level. For all 4,927 banks in our sample, we collect data from the following sources.

Branch-Level Deposits Data

FDIC Summary of Deposits (SOD) is an annual survey of branch office deposits for all FDIC-insured institutions, including insured U.S. branches. We collect all SOD reports for all U.S. banks from 2007 to 2019. For a given bank, we collect data related to the location of its branch and the amount of deposits at the branch. We match branch-level FDIC data to bank-level call reports using the FDIC bank identifier.

Census Data

The U.S. Census asks questions about the computers and devices that people use, whether people access the internet, and how people access the internet. Computer and internet use questions were added to the ACS in 2013 and modified in 2016. The particular metric we are interested in is the percent of households with a broadband internet subscription. While newer questions have been added regarding mobile phone usage, we want to take advantage of the longer time-series of the internet usage variable. There exists significant variation in access to broadband across counties. For instance, within the state of Pennsylvania, 76.2% of households in Cameron County have access to internet subscription while 91.9% of households in Chester County have access to internet subscription. We collect internet usage of households at the county level.

FCC Broadband Data

The Federal Communications Commission (FCC) reports county-level broadband access data from December 2008 to December 2021. All facilities-based broadband providers are required to file data with the FCC twice a year (Form 477) on where they offer internet access service at speeds exceeding 200 kbps in at least one direction. For a given county, we collect data on the level of internet access, defined as in Table 1. We focus in particular on the availability of broadband connections with speed of at least 200 kbps for two reasons. First, the FCC changed its reporting of tiers of internet access over the time period of our bank-level data. From 2008 to 2013, the FCC defined internet tiers as Tier 1 (200 kbps), Tier 2 (768 kbps), Tier 3 (3 mbps), and Tier 4 (10 mbps). From 2013 to 2021, the FCC defined internet tiers as Tier 1 (200 kbps), Tier 2(10 mbps), Tier 3 (25 mbps), and Tier 4 (100 mbps). To maintain consistency across our time period, we select access to broadband with a downstream speed of 200 kbps as our variable of interest. Second, 200 kbps is known to be fast enough to support a single individual’s online banking activities.

Table 1.

Residential Fixed Broadband Connections with a Downstream Speed of at Least 200 kbps.

County score	Percentage of households with at least 200 kbps broadband
0	0
1	0 < X < 20
2	20 < X < 40
3	40 < X < 60
4	60 < X < 80
5	80 < X < 100

App Store Data

To further address the shift to mobile-banking for financial institutions, we turn to the app store for data on mobile app adoption by banks. We collect data on the initial release date of a bank’s mobile app, the number of reviews on the app, and the distribution of reviews for the app. For example, we document that Southside Bank, a bank primarily servicing Eastern Texas, first adopted a mobile app on December 12 of 2011. The Southside Bank mobile banking app has 2,995 reviews on the app store, with an average rating of 4.66.

In order to connect the adoption of mobile applications with our proposed digital score, we implement a staggered difference-in-differences (DiD) strategy to examine how improvements in deposit-weighted broadband access across counties influence the timing of mobile app adoption by U.S. banks. This staggered DiD approach allows us to account for the fact that different banks adopted mobile apps at different points in time, based on their digital environment. The banks that had not yet adopted mobile apps serve as the control group, while those that adopt mobile apps during the observation period constitute the treatment group. We reserve the details of this specification in Section 4, when we discuss new stylized facts around bank digitalization.

Digital Measure

We compute digital scores at the bank level by using a deposit-weighted measure of access to broadband in the counties where bank branches are located according to Equation 1. The intuitive way to understand this score is that it represents the percent of bank deposit base with likely broadband access.

$D i g i t a l S c o r e_{i, t} = \sum_{c} \frac{D e p_{i, c, t}}{D e p S u m_{i, t}} \times B r o a d b a n d S c o r e_{c, t}$ (1)

$De p_{{i, c, t}}$ is the amount of deposits for bank i located in county c at time t. $Dep Su m_{{i, t}}$ is the sum of all deposits managed by bank i at time t. $Broadband Scor e_{c, t}$ represents the 0 to 5 score as reported in Table 1 for county c at time t. Hence, $Digital Scor e_{i, t}$ represents a deposit-weighted measure of broadband penetration for counties in which a given bank’s customers reside. As an example, consider Mercer County State Bank, a community bank servicing Philadelphia with 550 million dollars in assets. Figure 1 describes the evolution of the digital score for Mercer County State Bank. We select the earliest data point in our sample, 2008, and the last data point in our sample, 2019. We see that in 2008, Mercer County State bank had a digital score of 2.91, reflecting that close to 40% of its depositor base has access to broadband internet with a speed of at least 200 kbps. In 2019, the digital score for Mercer County State Bank is 3.71, reflecting that close to 60% of its depositor base has access to broadband internet with a speed of at least 200 kbps.

Figure 1.

Comparison of digital score for Mercer County State Bank in 2008 and 2019. (a) displays broadband access of Mercer County State Bank branch locations in 2008. (b) displays braodband access of Mercer County State Bank branch locations in 2019.

The takeaway from Figure 1 is that first, there is considerable cross-sectional variation in broadband access within a given geography. Second, there is considerable time-series variation in broadband access for one given bank. The combination of different levels of broadband access in geographies and different locations of bank branches give rise to the key heterogeneity that we exploit in this paper.

Baseline Regressions

In this section, we present our main analysis of the effect of digitalization on bank transparency. We test our hypothesis H1 that bank digitalization leads to an increase in bank transparency using the following panel regression specification:

$\begin{matrix} Transparenc y_{i, t} = α_{i} + γ_{t} + β_{1} Digital Score \\ + β_{2} Siz e_{i, t} + β_{3} LL P_{i, t - 1} + β_{4} CA P_{i, t} + β_{5} LOS S_{i, t} \\ + Interactions + ϵ_{i, t} \end{matrix}$ (2)

The dependent variable is the transparency score, measured as regressing past loan loss provisions, earnings before loan loss provisions, and non-performing loans on future loan write-offs. We follow the rolling specification employed in Q. Chen et al. (2022). The variable Digital Score is the baseline digital score capturing the likely access to broadband of a bank’s depositor base. We include the following control variables: Size is the natural logarithm of total assets; $LL P_{t - 1}$ is the loan loss provisions in the previous quarter; CAP is the ratio of shareholder equity to total assets; LOSS is an indicator variable if a bank reports negative net income in the current quarter. Because we are potentially interested in the interactions of digitalization with other characteristics of the bank, we include these controls both on their own, and interacted with Digital Score in our regression. Lastly, we include bank fixed affects and time fixed effects to help account for both the time-invariant and time-varying characteristics of a bank which may be related to its underlying transparency. The standard errors are clustered at the bank level.

This research design answers the hypotheses of the paper in the following ways. The coefficient $β_{1}$ captures the relationship between our proposed digitalization measure and bank transparency. A statistically significant and positive $β_{1}$ coefficient would support H1 that more digital banks are also more transparent banks. A statistically insignificant coefficient would indicate that there is no systematic relationship between bank digitalization and transparency. The coefficient on the interaction terms would test our H2 and H3 on the heterogeneous impact of bank digitalization on bank transparency. In particular, if the coefficient term on the interaction between Digital Score and Size is statistically significant and positive, this would refute our hypothesis H2 that the relationship between digitalization and transparency weakens as bank size increases. If the coefficient term on the interaction between Digital Score and LOSS is statistically significant and negative, this would support our hypothesis H3 that the impact of digitalization on transparency is weaker for banks reporting losses in the current quarter.

Additional Econometric Specifications

The second specification performs a robustness test for our hypotheses H1-H3 by including an additional set of controls. Importantly, we keep the bank and time fixed-effects as in Equation 2 to control for time-varying and time-invariant bank characteristics.

$\begin{matrix} Transparenc y_{i, t} = α_{i} + γ_{t} + β_{1} Digital Scor e_{i, t} \\ + β_{2} Siz e_{i, t} + β_{3} LL P_{i, t - 1} + β_{4} CA P_{i, t} + β_{5} LOS S_{i, t} \\ + β_{6} W F_{i, t} + β_{7} U C_{i, t} + β_{8} R E_{i, t} + Interactions + η_{i, t} \end{matrix}$ (3)

We include the following additional set of controls: WF is the amount of wholesale funding scaled by total assets, UC is the ratio of total unused commitments to the sum of total loans and unused commitments, and RE is the real estate loan share calculated as the amount of loans secured by real estate divided by total loans. Similarly, because we are interested in how being more or less digital interacts with these additional sets of bank characteristics, we include interaction terms of these controls with digital score in our regression.

This additional set of controls are motivated by the specification outlined in Acharya and Mora (2015). The inclusion of the unused commitments variable reflects the fact that a bank with a higher amount of outstanding commitments is more vulnerable to these commitments being drawn down, especially in times of financial stress. Thus, this captures an additional relationship between a bank’s potential exposure to liquidity demands with its overall transparency. The inclusion of the wholesale funding variable is yet another liquidity and solvency measure and reflects the net wholesale borrowing by a bank, including repos less reverse repos. The inclusion of the real estate variable captures the real-estate exposure of a bank’s loan portfolio. This variable controls for changes in transparency which may be explained via real-estate loan holdings for a given bank. For example, one might hypothesize that banks with high nonperforming loans, high real estate loans, and low capital may choose to be more opaque about their loan portfolio, resulting in a lower transparency score.

Given that the above specification maintains the same definition for Digital Score, we want to conduct another econometric test to ensure that our results are robust to different definitions of bank digital-ness using the digital score metric we defined in this paper. In order to accomplish this, we also present results for a third econometric specification where instead of using a time-varying digital score, we assign a digital ranking to a bank corresponding to how the bank ranks against the cohort of banks at the first time when it appears in our sample. For example, if a bank first appears in our sample in 2012, we assign it a Digital Category based on how it compares against all other banks in our sample in 2012. Using this newly defined Digital Category variable, we estimate the following regression specification:

$\begin{matrix} Transparenc y_{i, t} = β_{1} Digital Categor y_{i} \\ + β_{2} Siz e_{i, t} + β_{3} LL P_{i, t - 1} + β_{4} CA P_{i, t} + β_{5} LOS S_{i, t} \\ + Interactions + ϵ_{i, t} # \end{matrix}$ (4)

By using this static Digital Category variable, we assess whether a bank’s initial digital ranking continues to influence its transparency throughout the sample period. This specification provides a robustness check on our primary results while helping us isolate the effect of starting conditions in the digitalization process. A statistically significant and positive $β_{1}$ coefficient would support H1 that more digital banks are also more transparent banks.

Results and Discussion

In this section, I proceed in the same order as the Methodology section. First, I present the results from the panel regression with bank and time fixed effects and discuss the significance of the results. Then, I discuss the policy and academic implications of our findings.

New Facts Around Bank Digitalization

Given the digitalization measures we proposed, we are able to document new stylized facts on bank digitalization and transparency. These facts are important to highlight because 1. they provide a cross-sectional view of the bank digitalization landscape and 2. they motivate our reduced-form specifications. We present three new facts surrounding bank digitalization and transparency. First, larger banks are more digital. Second, subject to a large jump in digital score, banks are more likely to adopt mobile apps. Third, more digital banks are strictly more transparent over less digital banks from 2007 to 2019. This third fact establishes a positive correlation between bank digitalization and bank transparency, which motivates our reduced-form regressions to test our hypotheses H1 to H3 formally.

Larger Banks Are More Digital

Figure 2 shows a cross-sectional view of the distribution of digital scores for banks in our sample in the year 2021. The multimodal nature of the distributions is a function of the data we have from the FCC, which is categorical in nature, as described in Table 1. More interesting is the mass of digital scores concentrated around 5 for the blue distribution, corresponding to the digital scores for the 377 large banks in our sample. This suggests that the majority of large banks have a deposit base with 80% to 100% access to broadband internet. Furthermore, the distribution of digital scores for large banks has zero mass around digital scores 2 and 3, suggesting that the lower tail of internet access for depositors at large banks is still bounded by 60%.

Figure 2.

Distribution of digital scores for large, medium, and small banks using 2021 vintage of FCC broadband data.

In contrast, we see that the mass of the green distribution, corresponding to the 3,204 small banks in our sample are concentrated around 4. this suggests that the majority of small banks have a deposit base with 60% to 80% access to broadband internet in 2021. The biggest difference between the small and medium banks (N = 1,346) banks in our sample is the concentration of digital score around the score of 3. While the distribution of digital scores for medium banks have near zero mass concentrated around 3, the distribution of digital scores for small banks has a non-negligible mass concentrated around 3, meaning that a non-negligible fraction of small banks have a deposit base with 40% to 60% access to broadband internet, while that is not the case for medium or large banks.

In sum, this cross-sectional view of digital scores for banks by size illustrates that large banks service the most digital customer base, followed by medium banks, and finally small banks, which service the least digital customer base. We also include an illustration of how these digital score distributions have evolved since 2008 in Figure 3, and the same pattern of large banks servicing more digital customers has been the case since the beginning of the FCC data-set.

Figure 3.

Evolution of digital scores for large, medium, and small banks from 2008 to 2021. Scores are computed based on FCC reported access to broadband per 1,000 households at the county level.

Banks are More Likely to Adopt Mobile Apps when Their Customer Base is More Digital

This descriptive fact connects the baseline digital score used in this paper with a new dataset we collect on mobile app releases of commercial banks. To do so, we conduct a Staggered DiD event study using the specification denoted in Equation 5.

$H a s A P P_{i, t} = \sum_{τ \neq 1} β I_{i, τ} \times D_{i, t} + S i z e_{i, t} + ϵ_{i, t}$ (5)

$Has AP P_{i, t}$ denotes whether a commercial bank in our sample has a consumer facing banking application available on either the Apple Store or Google Play at time t. $D_{i, t}$ denotes whether the bank i experienced its largest increase in digital score at time t. Treated is the absorbing state, that is, $D_{i, t} = 1 \Rightarrow D_{i, s} = 1 \forall s > t . Siz e_{i, t}$ denotes the assets of bank i at time t.

We show the results from the Callaway and Sant’Anna (2021; CS) Estimator in Figure 4. The CS solution is applicable in our context as we use “not yet treated” banks as controls. For example, if Bank A released a mobile banking application in August of 2012, then prior to the release of the banking application, Bank A is a member of the control group. In August of 2012, Bank A is treated as a member of the treatment group and stays in that group until the end of our sample. In particular, the CS estimator re-aggregates treatment effects with appropriate weighting when compared to a straightforward two-way fixed effect (TWFE) specification. In our estimator, larger and more precise estimators are upweighted, while those based on a smaller number of observations are down-weighted.

Figure 4.

Staggered DiD event study using the Callaway and Sant’Anna Estimator.

One central assumption for a valid DiD design is no-existence of pre-trends. From Figure 4, we see there is no evidence to reject the parallel trend assumption for banks in the control group and banks in the treatment group, as the confidence interval of the estimated coefficients pre-treatment intersects the ATT = 0 axis. Furthermore, once a bank experiences its largest jump in digital score, the bank is more likely to adopt a consumer facing application, seeing as the coefficient for the lead indicator is statistically significant and non-zero. For subsequent periods, we also observe positive lead coefficients, although the standard errors around these coefficients does intersect the ATT = 0 axis.

Finally, one may be curious when the largest jump in digital score occurred for banks in our sample. In Figure 5, we document the distribution of timing for when banks experienced their largest increase in digital score. For example, if a given bank’s digital score for the years 2008, 2009, 2010, and 2011 are respectively 3, 3, 4, 4, then the year recorded for the bank is 2011. Note that if a bank experiences equally large increases in digital score in two or more years, we record the first year of such a jump as the corresponding entry for that bank. In Figure 5, we notice something interesting. Most banks experience the largest jump in digital score from 2009 to 2010. This provides support for our digital score as it likely reflects a lagged reaction to the Broadband Act of 2008 to ensure that broadband capabilities are being deployed to all Americans on a reasonable and timely basis. The banks experience another jump in digital score, albeit a smaller number of them, in the year 2016. Post 2016, the number of banks who experience a largest number in digital score decreases, likely reflecting the maturity of the technology in areas that banks service.

Figure 5.

Timing of treatment for the staggered DiD event study.

The conclusion from this descriptive fact should offer confidence in the digitalization measure we put forth in this paper as it illustrates the connection between our measure and the more observable action of banks releasing mobile applications. One possible underlying mechanism is that as broadband rolls out to areas where bank depositors are located, these depositors in turn has greater demand for mobile and digital services provided by their financial institutions. Therefore, the positive and statistically significant coefficient on the lead indicator in Figure 4 reflects banks responding to the demand from a more digital customer base. We highlight the connection between our digital score and app adoption to further provide a robustness check that our digital score captures fundamental characteristics of how digital a bank’s customers are.

More Digital Banks are More Transparent

In Figure 6, we plot the transparency scores over time for the most digital and the least digital banks. The most digital banks, banks with digital scores in the top quantile at time t, are plotted in red and the least digital banks, banks with digital scores in the bottom quantile at time t, are plotted in blue. Something striking in the figure is the strict dominance of the red line over the blue line. That is, over the entire period of 2008 to 2019, the more digital banks have consistently been more transparent than the less digital banks. While this provides descriptive evidence in support of our hypothesis H1 that more digital banks are more transparent, we formally test this relationship using our baseline regressions with time and bank fixed effects as well as a host of controls.

Figure 6.

Time-series of transparency scores for the most digital banks and the least digital banks.

Regression Results

Table 2 reports the OLS results without instrumenting for digital score. Column (2) includes the subset of controls, LogAssets, $LL P_{t - 1}$ , CAP, LOSS. Column (1) includes all of the above controls and additionally: wholesale funding scaled by total assets (Wholesale Ratio), the ratio of total unused commitments to the sum of total loans and unused commitments (Unused Commitments), and real estate loan share calculated as the amount of loans secured by real estate divided by total loans (Real Estate Loan). Standard error estimates, reported in parentheses, are clustered at the bank level.

Table 2.

Regression Results from the Baseline and Robustness Specifications.

Dependent variable	Transparency
	(1)	(2)
Digital score	0.322*** (0.076)	0.283*** (0.072)
LOSS	0.115*** (0.036)	0.122*** (0.036)
LLP	0.049** (0.025)	0.044* (0.025)
LogAssets	0.093*** (0.028)	0.105*** (0.027)
CAP	1.28* (0.714)	1.27* (0.708)
Unused commitments	−1.106 (0.320)
Wholesale ratio	−0.029 (0.186)
Real estate loan	0.404*** (0.108)
Bank characteristics × digital score	Yes	Yes
Time FE	Yes	Yes
Bank FE	Yes	Yes
No. of observations	40,158	40,158

Note.***, **, * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

The coefficient for Digital Score with the regression specification including all controls is 0.322, suggesting that for a 20% increase in digital score, a bank would experience a 0.322 increase in its transparency score. Put differently, for a 20% increase in digital score, the information released in a bank’s call reports would explain an additional 32.2% of the variation in the bank’s future loan write-offs. An increase of similar magnitude is observed when we consider our baseline regression containing a subset of the controls. The coefficient of 0.283 reported in column (2), significant at the 1% level, indicates that for a 20% increase in digital score, a bank would experience a 0.283 increase in its transparency score. Put differently, for a 20% increase in digital score, the information released in a bank’s call reports would explain an additional 28.3% of the variation in the bank’s future loan write-offs. Thus, the positive and statistically significant coefficient on Digital Score allows us to reject the statement that bank digitalization has no impact on bank transparency in favor of our hypothesis H1 that an increase in bank digitalization, holding all else constant, leads to an increase in bank transparency. Furthermore, this result is highly significant and the size of the coefficient is meaningful.

From Table 2, a strong piece of evidence in support of our hypothesis that more digital banks are more transparent is the statistical significance of the digital score coefficient with the inclusion of LogAssets as a control. As established by our first descriptive fact, large banks service the most digital customer base, followed by medium banks and small banks. As of 2021, 80% to 100% of large banks’ deposit base have access to broadband, compared with 60% to 80% for small banks. Hence, if the digital score of a bank is still able to explain transparency after the inclusion of bank size as a control, this gives us confidence that it is not simply bank size driving the transparency of the bank. In other words, the digital score metric, reflecting the likely access to broadband of a bank’s customers, captures additional explanatory power for bank transparency in addition to what is captured by the size of the bank. Comparing the coefficient for LogAssets of 0.093 with the coefficient for digital score of 0.322 in the first specification (and 0.105 vs. 0.283 in the second specification), we see that for a given bank, holding all other characteristics constant, an increase in one unit of digital score, corresponding to an increase in broadband access for 20% of a bank’s deposit base, results in a larger increase in bank transparency compared to an increase in one unit of LogAssets, corresponding to an increase in bank assets by 10 times.

The coefficient for the LOSS term in Table 2 is equally interesting. Recall that LOSS represents whether the bank reports a loss in the current quarter. The coefficient of 0.115 suggests that if a bank reports loss in the current quarter, its transparency score increases by 0.115, holding all other variables constant. In other words, if a bank reports loss in the current quarter, the information it releases in its call reports are able to explain an additional 11.5% of variation in its future loan write-offs. One possible explanation for the negative coefficient is that banks tend to be more transparent in periods when they are experiencing financial difficulties, perhaps to manage depositor expectations.

Table 3 shows the relevant interaction terms from both specifications in Equations 2 and 3. The first column of the table corresponds to the regression specification including all controls, and the second column of the table corresponds to the inclusion a subset of controls as used in Yue et al. (2022). We focus our analysis on regression coefficients significant at the 5% level. First, we see under both specifications a negative coefficient term for the interaction of digital score with the LOSS variable. The negative coefficient suggests that the positive relationship between digital score and bank transparency is weakened when the bank reports a loss. In other words, for banks that report a loss, the increase in transparency associated with a higher digital score is less by 0.029 units than for banks that do not report a loss. This finding supports our hypothesis H3 that while we observe a positive relationship between bank digitalization and bank transparency, this relationship is weakened when a bank reports a loss in the current quarter, holding all else equal. When a bank reports a loss, this might not directly impact the broadband access of its customers. However, the bank’s financial difficulties could lead to changes in how it leverages this digital connectivity. For example, the bank might reduce investments in digital platforms or online services, which could affect the positive influence of digital connectivity on transparency.

Table 3.

Interaction Terms of Digital Score with Bank Characteristics.

Dependent variable	Transparency
	(1)	(2)
Digital score × LOSS	−0.028* (0.103)	−0.030* (0.010)
Digital score × LLP	−0.006 (0.007)	−0.005 (0.007)
Digital score × LogAssets	−0.019*** (0.006)	−0.020*** (0.005)
Digital score × CAP	−0.203 (0.187)	−0.218 (0.186)
Digital score × unused commitments	0.049 (0.086)
Digital score × wholesale ratio	0.049 (0.050)
Digital score × real estate loan	−0.101*** (0.027)
Time FE	Yes	Yes
Bank FE	Yes	Yes
No. of observations	40,158	40,158

Note.***, **, * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

The corresponding coefficient for the interaction term between digital score and log assets further shed light on how the digital-ness of a bank interacts with its underlying characteristics. As seen in Table 3, the coefficient term for the interaction term between digital score and log assets is −0.020 under the baseline specification and significant at the 1% level, corresponding to the interpretation that the relationship between digital score and transparency changes with the size of the bank, as proxied via log assets. Furthermore, the negative coefficient supports our hypothesis H2 that the positive effect of bank digitalization on bank transparency diminishes as the size of the bank increases. This result is not at odds with what we observe in the positive coefficient term for log assets in Table 2 for the following possible reasons. First, this may be because larger banks, despite having customers with good broadband access, might not be able to leverage this advantage as effectively as smaller banks due to their size, complexity, or a different customer engagement strategy. Hence, although larger banks tend to be more transparent and while they generally have higher digital scores, corresponding to depositor bases with more broadband access, these banks may be less efficient in using these digital channels to improve transparency, compared with smaller, more nimble banks (which generally focus on targeted communities and geographies). However, we note that the magnitude of the interaction term is small, −0.019 for the full specification, which suggests that the impact of digital score interacted with log assets on transparency is, while statistically significant, less economically significant when compared to the contributions of size and digitalization to transparency alone.

Finally, we provide some comments for the interaction term between digital score and real estate loan, which is also significant at the 1% level. Note that this interaction term is only relevant for the full specification of the model as in Equation 3. The negative coefficient for the interaction term suggests that the positive impact of the digital score on bank transparency is less in banks with a higher proportion of real estate loans. This could be due to the nature of real estate lending involving more physical collateral and being subject to more stringent regulatory requirements. Hence, bank digitalization via increased broadband access by bank’s customers, while valuable in enhancing transparency in many banking areas, might not significantly contribute to these specific transparency requirements in real estate lending. However, this does not distract from the overall finding of bank digitalization leading to higher bank transparency, even when considering a host of controls and interaction terms.

Table 4 presents the results from our second robustness check where we substitute the time-varying Digital Score with a bank’s Digital Category, which is a static variable based on the bank’s initial digital score when it first appeared in the sample. The purpose of this specification is to test the robustness of our main results. In Table 4, I present the coefficients most relevant to testing hypotheses H1-H3.

Table 4.

Regression Results from the Second Robustness Specification.

Dependent variable	Transparency
Digital category	0.255** (0.114)
Digital score × LOSS	−0.061*** (0.022)
Digital score × LogAssets	−0.017* (0.009)
Additional controls	Yes
No. of observations	40,158

Note.***, **, * indicate statistical significance at the 1, 5, and 10 percent levels, respectively.

From Table 4, the coefficient for Digital Category is 0.255 and significant at the 5% level, suggesting that a higher initial level of digitalization is associated with a positive and significant increase in a bank’s transparency score. This supports hypothesis H1 that banks with greater access to digital infrastructure are more transparent. The coefficient for the interaction term Digital Score × LOSS is −0.061 and significant at the 1% level. This result indicates that the positive relationship between a bank’s digitalization and transparency decreases when the bank reports a loss in the current period. Similar to the results in Table 3, this suggests that banks may engage in strategic withholding of information when facing financial difficulties. This supports the notion that digitalization’s positive impact on transparency is conditional upon the bank’s financial health as stated in hypothesis H2. Finally, the coefficient for the interaction term Digital Score × LogAssets is −0.017 and statistically significant, reinforcing our earlier finding that the relationship between digitalization and transparency diminishes as bank size increases.

Taken together, the results from Table 4 provide strong support for H1, H2, and H3. The positive and significant coefficient for Digital Category confirms more digital banks are more transparent. The interaction terms underscore that the relationship between digitalization and transparency is not uniform across all banks, but rather contingent on factors such as financial health and bank size. These findings enhance the robustness of our analysis by demonstrating that the impact of digitalization on transparency holds even under alternative definitions of digitalization.

Practical and Theoretical Implications

The findings of this paper have important practical implications. Our results suggest that bank digitalization plays a critical role in improving bank transparency. However, we find that digitalization disparately impacts large versus small banks, and its effect depends on the financial health of the bank in the given quarter. These empirical findings have two important policy implications. First, policymakers could promote digitalization as an effective means of increasing transparency within the banking system. Furthermore, these policies should be tailored to the specific characteristics of banks, such as bank size or bank financial health, or include provisions that account for these differences.

Turning to the theoretical implications of our empirical findings, while our results validate the beneficial role of digitalization in promoting transparency, they also challenge the assumption that transparency is universally beneficial across all contexts, particularly for banks in financial distress. Our finding that banks reporting losses tend to show weaker transparency improvements from digitalization complicates the theoretical claim that more transparency always leads to better outcomes (Morris & Shin, 2002). Our results better support the argument made in Acharya and Mora (2015) that banks may intentionally reduce transparency during periods of financial distress to avoid triggering negative stakeholder reactions, such as depositor withdrawals or runs. Thus, our paper provides evidence that banks may strategically opt for opacity even when their level of digitalization should imply a higher level of transparency.

Another area where our findings both validate and challenge existing theory is in the relationship between bank size and the effectiveness of digitalization in promoting transparency. Traditional theories of technological adoption often suggest that larger firms benefit more from new technologies (Berger & Bouwman, 2009). However, our findings indicate that smaller and medium-sized banks tend to realize more significant transparency gains from digitalization, while the effect of digitalization on transparency for larger banks is more limited, holding all other variables equal. This finding contributes to the literature by refining existing theories on technological adoption in banking. Rather than assuming that larger banks will automatically benefit more from digitalization, our results suggest that smaller and more agile institutions see greater benefits from digitalization when transparency is the outcome of interest.

Conclusions

This paper investigates how bank digitalization impacts bank transparency. Our central finding is that for a 20% increase in bank depositors who have access to broadband, bank transparency increases by 0.32, reflecting that the information a bank releases is able to explain 32% more variation in a bank’s future loan write-offs. Furthermore, we investigate how bank digitalization interacts with other underlying bank characteristics such as whether the bank reports a loss in the current quarter and the size of the bank. In addition to documenting that more digital banks are more transparent, we document interesting heterogeneities along bank characteristics. We find that positive effect of bank digitalization on bank transparency diminishes as the size of the bank increases and if the bank reports a loss. We offer some explanations for these interesting interactions between digitalization, transparency, and bank characteristics.

To arrive at the main finding that more digital banks are more transparent, this paper defines a novel digital score metric which could be applied to any FDIC-insured institutions found in the Summary of Deposits. This is also a main contribution of the paper as it captures the deposit-weighted likely broadband access of bank’s depositors. Using the novel digital measure proposed, this paper is also able to quantitatively document descriptive facts and show that they hold for the entire sample period rather than any snapshot in time. We document that large banks have consistently been more digital than medium and small banks across the sample period using our digital measure. We also show that more digital banks are consistently more transparent, without controlling for possible confounders, over the entire sample period. The paper also considers how mobile app releases of banking institutions interact with the digital score proposed. We document that following a large jump in the novel digital score, a bank is more likely to adopt a mobile app. This fact illustrates the connection between our digital score and realized digitalization efforts at the bank-level.

Overall, this paper fills an important gap in literature by documenting how transparency, a fundamental topic of interest to policy makers and academics, responds to an increasingly digital economy. Moreover, this paper speaks to the digitalization of banks, large financial actors with a long history in the U.S. economy. We acknowledge that the results of this paper focus on U.S. depository institutions given limitations in the data available, but we believe the findings can still help researchers understand how digitalization impacts measures contributing to financial stability and fragility.

Research Limitations

The results of this study are subject to some limitations. First, the findings of this paper are limited to understanding the impact of digitalization on the transparency of U.S. banks. Given the main data source for bank-level deposits information is the FDIC Summary of Deposits (SOD), we do not make comments about the impact of digitalization on banks which do not appear on the SOD. This means that this paper does not seek to comment on the impact of digitalization on transparency for other countries. However, we believe that the computation of the digital score would be robust to other geographies given the prevalence of broadband penetration and the availability of such data or related measurements for other countries. We do acknowledge, however, that more thought needs to be given to the measurement of transparency when removed from the U.S. context given that other countries likely experience different banking regulation and reporting requirements. Provided such a measure, the methodologies of this paper can be extended to additional geographies. Second, the findings of this paper comments on the impact of digitalization through the lens of broadband penetration. In order to measure the level of broadband expansion, we rely on the data reported from the FCC, which categorizes the level of broadband penetration of a given county. Hence, if there is a large amount of variation in broadband penetration within a given county, we are unable to comment on the granularity of broadband penetration specific to the exact location of bank branches. However, we believe that this is less of a concern given that a county likely has similar amounts of broadband access. Finally, given the limitations of the data, we make comments on the impact of digitalization on bank transparency from 2007 to 2021. Given this was also a period of high digital improvements and infrastructure improvements, we likely capture a very interesting and relevant time period of study. However, if there is data available on the period prior to 2007, this would also be able to serve as an interesting extension to the time period studied in this paper.

Suggestions for Future Study

Given the above limitations, a related and interesting avenue for future research is to investigate whether the relationship between digitalization and transparency persists across geographies and different regulatory environments. The challenge to making the extension is twofold. First, one must acquire data on bank-branch level information for foreign banks. At the same time, one must be able to observe both the reporting behavior of foreign banks as well as the realized loan performances of these banks. Second, one also needs to observe a time-series of the level of broadband penetration in areas where banks are located. Provided the availability of these two data series, it would be interesting to investigate whether the impact of digitalization on transparency is consistent across different regulatory environments. If digitalization impacts transparency with a similar sign and magnitude for foreign banks, this would provide more evidence that the relationship documented in this paper is reflecting a fundamental mechanism.

Footnotes

Not applicable.

ORCID iD

Qi Cui

Ethical Considerations

Not applicable.

Funding

The author(s) disclosed receipt of the following financial support for the research,authorship,and/or publication of this article: The research is thanks to the support of Research Project of Academic Degrees and Graduate Education Teaching Reform at Jiangxi Science and Technology Normal University (KSDYJG-2023-20).

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research,authorship,and/or publication of this article.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analyzed during the current study.

References

Acharya

V. V.

Mora

(2015). A crisis of banks as liquidity providers. Journal of Finance, 70(1), 1–43.

Acharya

V. V.

Ryan

S. G.

(2016). Banks’ financial reporting and financial system stability. Journal of Accounting Research, 54, 277–340.

Alvarez

Barlevy

(2021). Mandatory disclosure and financial contagion (NBERJournal of Economic Theory, 194, 1–41

Arslanian

Fischer

(2019). The future of finance: The impact of FinTech, AI, and crypto on financial services. Springer.

Beatty

Liao

(2011). Regulatory capital ratios, loan loss provisioning and procyclicality. Journal of Accounting and Economics, 52(2–3), 1–20.

Beatty

Liao

(2014). Financial accounting in the banking industry: A review of the empirical literature. Journal of Accounting and Economics, 58(2-3), 339–383.

Benmelech

Yang

Zator

(2024). Bank branch density and bank runs. NBER Working Paper No. w31462https://doi.org/10.3386/w31462.

Berger

A. N.

Bouwman

C. H. S.

(2009). Bank liquidity creation. Review of Financial Studies, 22(9), 3779–3837.

Berg

Fuster

Puri

(2022). FinTech lending. Annual Review of Financial Economics, 14(1), 187–207.

10.

Blum

(2002). Subordinated debt, market discipline, and banks risk taking. Journal of Banking and Finance, 26(7), 1427–1441.

11.

Buchak

Matvos

Piskorski

Seru

(2018). Fintech, regulatory arbitrage, and the rise of shadow banks. Journal of Financial Economics, 130(3), 453–483.

12.

Bushman

R. M.

(2014). Thoughts on financial accounting and the banking industry. Journal of Accounting and Economics, 58(2), 384–395.

13.

Bushman

R. M.

Williams

(2015). Delayed expected loss recognition and the risk profile of banks. Journal of Accounting Research, 53(3), 511–553.

14.

Callaway

Sant’Anna

P. H. C.

(2021). Difference-in-differences with multiple time periods. Journal of Econometrics, 225(20), 200–230.

15.

Calomiris

C. W.

Kahn

C. M.

(1991). The role of demandable debt in structuring optimal banking arrangements. American Economic Review, 81(3), 497–513.

16.

Chen

Goldstein

Huang

Vashishtha

(2022). Bank transparency and deposit flows. Journal of Financial Economics, 146(2), 475–501.

17.

Chen

Hasan

(2006). The transparency of the banking system and the efficiency of information-based bank runs. Journal of Financial Intermediation, 15(3), 307–332.

18.

Cordella

Yeyati

E. L.

(2003). Bank bailouts: Moral hazard vs. value effect. Journal of Financial Intermediation, 12(4), 300–330.

19.

Dang

T. V.

Gorton

Holmström

Ordoñez

(2017). Banks as secret keepers. American Economic Review, 107(4), 1005–1029.

20.

Erel

Liebersohn

(2022). Can fintech reduce disparities in access to finance? Evidence from the paycheck protection program. Journal of Financial Economics, 146(1), 90–118.

21.

Erel

Liebersohn

Yannelis

Earnest

(2024). Monetary policy transmission through online banks. Fisher College of Business Working Paper.

22.

Flannery

M. J.

(2011). The faces of market disciple. Journal of Financial Services Research, 20, 107–119.

23.

Fuster

Plosser

Schnabl

Vickery

(2019). The role of technology in mortgage lending. Review of Financial Studies, 32(5), 1854–1899.

24.

Glushchenko

Hodasevich

Kaufman

(2019). Innovative financial technologies as a factor of competitiveness in banking. SHS Web of Conferences, 69, 00043.

25.

Gopal

Schnabl

(2020). The rise of finance companies and FinTech lenders in small business lending. NYU Stern School Bus.

26.

Gorton

(2013). The development of opacity in U.S. banking NBER Working Paper No. w19540. https://doi.org/10.3386/w19540.

27.

Gorton

Ross

(2022). Making money. NBER Working Paper No. w29710. https://doi.org/10.3386/w29710.

28.

Hornuf

Klus

M. F.

Lohwasser

T. S.

Schwienbacher

(2021). How do banks interact with fintech startups? Small Business Economics, 57(3), 1505–1526.

29.

Matutes

Vives

(2000). Imperfect competition, risk taking, and regulation in banking. European Economic Review, 44(1), 1–24.

30.

Morrison

A. D.

White

(2013). Reputational contagion and optimal regulatory forbearance. Journal of Financial Economics, 110(3), 642–658.

31.

Morris

Shin

H. S.

(2002). The social value of public information. American Economic Review, 92(5), 1521–1534.

32.

Ramdani

Rothwell

Boukrami

(2020). Open banking: The emergence of new digital business models. International Journal of Innovation and Technology Management, 17(5), 1–19.

33.

Rochet

J. C.

(1992). Capital requirements and the behaviour of commercial banks. European Economic Review, 36(5), 1137–1178.

34.

Stephanou

(2010). Rethinking market discipline in banking. Lessons from the financial crisis (Working Paper 5227). World Bank Policy Research.

35.

Tepe

Geyikci

U. B.

Sancak

F. M.

(2021). FinTech companies: A bibliometric analysis. International Journal of Financial Studies, 10(1), 2.

36.

Tiberius

Gojowy

Dabić

(2022). Forecasting the future of robo advisory: A three-stage delphi study on economic, technological, and societal implications. Technological Forecasting and Social Change, 182, 121824.

37.

Yue

Zhang

Zhong

(2022). The politics of bank opacity. Journal of Accounting and Economics, 73(2–3), 101452–101475.

The Impact of Digitalization via Broadband Expansion on Bank Transparency

Abstract

Plain language summary

Keywords

Introduction

Literature Review

Digitalization of the Banking Sector

Theory of Bank Transparency and Opacity

Measurement of Bank Transparency and Opacity

Data and Methodology

Sample and Data

Bank-Level Quarterly Data

Branch-Level Deposits Data

Census Data

FCC Broadband Data

App Store Data

Digital Measure

Baseline Regressions

Additional Econometric Specifications

Results and Discussion

New Facts Around Bank Digitalization

Larger Banks Are More Digital

Banks are More Likely to Adopt Mobile Apps when Their Customer Base is More Digital

More Digital Banks are More Transparent

Regression Results

Practical and Theoretical Implications

Conclusions

Research Limitations

Suggestions for Future Study

Footnotes

ORCID iD

Ethical Considerations

Funding

Declaration of Conflicting Interests

Data Availability Statement

References