The Online Payday Loan Premium

∗

Filipe Correia

†

Peter Han

‡

Jialan Wang

March, 2024

Abstract

Using data from a subprime credit bureau with nationwide coverage in the United States,

we test whether technology lowers the cost of credit in the expensive payday loan market.

Contrary to the idea that technology lowers consumer prices, we ﬁnd that online loans are

about 100% APR higher than storefront loans. This premium is not explained by observable

loan or customer characteristics including credit scores and traditional credit risk measures,

and we do not ﬁnd evidence of risk-based pricing on observables. Our evidence is consistent

with asymmetric information and diﬀerences in ﬁxed costs across the online and storefront

markets.

Keywords: Online Lending, Payday Loans, Fintech

JEL Codes: D18, G23, G52

∗

We are grateful to Heitor Almeida, Scott Baker, Laura Blattner, Tatyana Deryugina, Julia Fonseca,

Diego Garcia, Jitka Hilliard, Anson Ho, Rustom Irani, Timothy Johnson, Charles Kahn, Mathias Kronlund,

Tom Miller, Holger Mueller, Chao Liu, Nikhil Paradkar, George Pennacchi, Andrea Presbitero, Elizaveta

Sizova, Alexei Tchistyi, Sergio Vicente, Malcolm Wardlaw, Constantine Yannelis, and participants at the

Gies College of Business and Terry College of Business brownbag, the 28th Finance Forum, and the 7th

International Young Scholars Conference, the Edinburgh Economics of FinTech Conference, the HEC Mon-

treal Workshop on Household Finance, and the GA Tech - Atlanta Fed Household Finance Conference for

valuable suggestions. Renhao Jiang, Peichen Li, Yunrong Zhou, Hejia Liu, Kevin Tzeng, Kashish Mehta,

Raja Gajula, Matthew Boyd, Yiheng Yang, Kunal Bagali, and Avishi Biljani provided valuable research

assistance.

†

University of Georgia, Terry College of Business, B339 Amos Hall, 620 S Lumpkin St, Athens, GA 30602.

[email protected]

‡

University of Illinois at Urbana-Champaign, Gies College of Business, 1206 South 6th Street, Champaign,

IL 61820. [email protected]

University of Illinois at Urbana-Champaign & NBER. Gies College of Business, 1206 South 6th Street,

Champaign, IL 61820. [email protected]

1 Introduction

Payday loans gained popularity in the 1990s and have been a controversial credit product

ever since, partly due to high prices, generally ranging between 300% to 400% APR (PEW,

2012). While the industry is believed to have relatively low barriers to entry and modest

levels of concentration and proﬁt margins (Huckstep, 2007), it also faces high ﬁxed costs of

operation and high default rates (Ernst & Young, 2009). Upon the rise of online lending in

many consumer credit markets, ﬁnancial technology brought the potential to lower prices

and increase eﬃciency by reducing ﬁxed costs and improving default prediction. Alterna-

tively, the business ecosystem of online payday loans has costly lead generators, targeted

advertising, aﬃliates, and lead-scoring agents. Thus, it could easily result in higher prices.

These two economic forces, lower existing costs versus new ecosystem costs, lead to a natural

question: are online payday loan prices higher or lower?

This paper contributes to this debate by documenting and decomposing price diﬀerences

between online and storefront payday loans in the United States. We utilize a novel dataset

from Clarity, a ﬁrm specializing in data collection from the subprime lending market and

lead generation services, including the payday loan market. We observe loan applications

and loan records. We observe loan amounts, repayments, maturities for each loan, and

information on whether the loan originated at a store or online. We also have access to

an array of information about the borrower, such as their income, age, ZIP code and state

of residence, housing status, and homeownership status. Our dataset allows for a direct

comparison between online and storefront loans, with the same data collected about the two

business models.

We document the important empirical fact that online loans are $4 per $100 borrowed,

or 100 percentage points APR more expensive. This premium is robust to controls for ex-ante

predictors of risk, time and location ﬁxed eﬀects, and is consistent across diﬀerent samples.

Although realized default rates are higher online, controlling for ex-ante credit risk metrics

absorbs this diﬀerence. The premium is larger when consumer ﬁxed eﬀects are included,

with the same borrower having a larger propensity to default online.

We ﬁnd no evidence of risk-based pricing at the loan level. Predictors of default observed

by the lender do not strongly correlate with loan pricing. We do not detect any systematic

pattern in prices or quantities across the credit score spectrum. However, borrower pools

exhibiting larger default rates experience higher loan prices. This result suggests each lender

sets similar prices for various consumers, regardless of observable characteristics, such as

income. Lenders, however, price the credit risk of the pool of borrowers they face.

If payday loans were priced on an individual basis, we would argue that in a similar

fashion to Jagtiani and Lemieux (2018), online payday lenders could use alternative sources of

information and advanced algorithms to improve portfolio performance at a relatively lower

cost. This would lower loan prices relative to storefront loans. Lead generators provide

lead-scoring services and use other sources of information to enrich the lead. However, these

services are, costly and feed a whole information supply chain, adding to the burden of the

borrower endpoint.

We rationalize our ﬁndings by adapting the Stiglitz and Weiss (1981) framework of credit

rationing under information asymmetry. Information asymmetry is likely to generate credit

rationing in the storefront market. As a result of credit rationing, a riskier pool of borrowers

will endogenously sort into the online market. Our empirical tests verify the presence of

information asymmetry in the payday loan market. Information asymmetry is larger in the

online market, partly justifying diﬀerences in realized default rates, and loan pricing across

the two markets.

We contrast states that implemented statewide payday loan databases with states that

did not. Statewide databases allow all lenders to share and access the same information

about borrowers. This decreases information asymmetry in the payday loan market and

equalizes the information set between online and storefront lenders. We ﬁnd that, in states

with a database, the premium is about 50% smaller than that of states without a database.

States with a payday loan database provide a switch from lower to higher loan amounts

online, consistent with a credit rationing attenuation. We show that the online premium

is larger in markets with a higher storefront market share. In those markets, the online

borrower pool contains the riskiest borrowers. With a higher online market share, online

lenders capture safer borrowers and the risk of the online pool decreases.

Lastly, we decompose the online payday loan premium into three components. One is

the diﬀerences in risk of the borrower pool faced by each business model. Secondly, business

models have diﬀerent cost structures. Third, business models can diﬀer in losses induced

by credit events or recovery rates. We ﬁnd evidence that cost structures are likely the most

important determinant of the online payday loan premium.

Our paper directly contributes to the broader literature aiming at understanding the

subprime credit market (e.g., Adams, Einav and Levin (2009)), namely the payday loan mar-

ket (e.g., Nunez, Shaberg, Hendra, Servon, Addo and Mapillero-Colomina (2016)). Monteze-

molo (2013) reviews the legal framework of payday loans in the United States, and documents

abusive practices from the lenders’ angle. Wang and Burke (2022) ﬁnd a signiﬁcant drop

in payday lending volume following mandatory state disclosure. Within the payday lending

literature, our paper speaks to the interplay between pricing and business model innovation.

We show that despite enjoying some leeway for regulatory arbitrage, online lending does not

lead to lower loan prices.

Existing works on the payday lending market focus on welfare implications

. We speak

to this literature by documenting the existence of a premium on online payday loans, which

have become very prevalent in the last decade. Only recently, the debate on payday loan

welfare started devoting attention to payday loan pricing. Saldain (2023) ﬁnds that borrow-

ing limits and interest rate caps reduce household welfare. Relatedly, Sridhar (2023) shows

See Agarwal, Skiba and Tobacman (2009), Melzer (2011), Bertrand and Morse (2011), Morse (2011),

Carrell and Zinman (2014),Bhutta (2014), Bhutta, Skiba and Tobacman (2015), Carter and Skimmyhorn

(2017), Dobridge (2018), Skiba and Tobacman (2019), Gathergood, Guttman-Kenney and Hunt (2019), Li

and Sun (2021).

that existing rate caps are ineﬃciently regulated and that separating initial fees from rollover

fees can be welfare-improving. Allcott, Kim, Taubinsky and Zinman (2022) estimate that

while a pure payday loan ban likely generates welfare losses, limiting rollovers can increase

welfare. Our paper is the ﬁrst to allow for diﬀerent business models within the payday loan

industry and examine diﬀerences in loan pricing.

Our paper is related to the literature on costly search and price dispersion in the con-

sumer credit market. Stango and Zinman (2016) ﬁnd that consumer shopping behaviors

generate economically signiﬁcant dispersion in equilibrium price in credit card rates. Alexan-

drov and Koulayev (2018) document price dispersion in mortgage markets and a lack of rate

shopping by mortgage borrowers. Argyle, Nadauld and Palmer (2023) show that search

frictions in credit markets contribute to price dispersion and aﬀect loan sizes using evidence

from the auto loan market. To our knowledge, our paper is amongst the ﬁrst to document

price dispersion in the payday lending market and attempt to explain it.

We also contribute to the growing literature on how ﬁnancial technology shapes con-

sumer credit markets. So far, much attention has been paid to peer-to-peer lending (Morse,

2015; Li, Li, Yao and Wen, 2019), online mortgages (Fuster, Plosser, Schnabl and Vick-

ery, 2019; Di Maggio and Yao, 2021; Chava, Ganduri, Paradkar and Zhang, 2021). Bartlett,

Morse, Stanton and Wallace (2022) show that ﬁntech fails to mitigate consumer racial/ethnic

discrimination in lending. Buchak, Matvos, Piskorski and Seru (2018) attribute the rise of

the originate-to-distribute logic in credit to technological advancement and regulatory fric-

tions. Li, Liao, Wang and Xiang (2018) study consumption responses to online cash loans

and ﬁnd consumption increases in the digital ecosystem. Payday loans and their online

dimension have not yet been addressed in ﬁntech research. We show how fundamental dif-

ferences between the in-store and online business models lead to a price diﬀerential, where

online payday loans are more expensive.

Lastly, we contribute to the recent advances in the eﬀects of information asymmetry in

consumer credit markets with ﬁntech participants. DeFusco, Tang and Yannelis (2022) back

out welfare costs of information asymmetry, exogenously perturbing interest rates of a large

ﬁntech lender, in a randomized experimental setting. We add to this literature by showing

how the coexistence between online and storefront lending markets with diﬀerent degrees of

information asymmetry and homogeneous goods generates a pooling equilibrium diﬀerent

from that expected from a perfectly competitive market.

The rest of the paper is organized as follows: Section 2 provides institutional background

on payday lending and online payday lending; Section 3 describes the dataset, explains the

data processing procedure, and puts the data in context; Section 4 documents the lack of risk-

based pricing and provides evidence of the online payday loan premium; Section 5 documents

asymmetric information and explains the pooling equilibrium; Section 6 decomposes the

premium in diﬀerences in business model, and diﬀerences in credit risk; Section 7 concludes.

2 Background on the Payday Loan Market

Payday loans are a common source of short-term credit among low- to middle-income Amer-

icans. Between 2015 and 2019, about 2 percent of households reported using at least one

payday loan per year, with higher shares among lower-income groups and higher shares that

had ever used a payday loan (Kutzbach, Lloro, Weinstein and Chu, 2020). They are typ-

ically between $300 and $500 in principal and are structured as a single balloon payment

of the amount borrowed and fees, timed to coincide with the borrower’s next payday. Fees

generally average between $10 and $20 per hundred dollars borrowed and typically do not

vary with loan duration. A ﬂat $15 per hundred fee annualizes to nearly 400% APR for a

14-day loan corresponding to biweekly pay dates (Consumer Financial Protection Bureau,

2013).

The storefront payday industry expanded through the 1990s and early 2000s, driven in

part by the loosening of state usury laws and partnership structures between payday lenders

and banks to “import” regulations across state lines, a practice ended by the FDIC in the

mid-2000s.

The online payday industry grew from a small share of loans to a signiﬁcant

market share over the 2010s, reaching a steady state of between 35% to 45% of the overall

payday market between 2013 and 2019, with overall loan volumes including storefront and

online declining from $46 billion to $25 billion annually during this period (Hecht, 2014,

2018; Graham and Golden, 2019).

The payday industry has attracted controversy and regulatory scrutiny due to high

annualized costs and the high frequency of repeat borrowing. In recent years, state regula-

tors have imposed restrictions including loan size caps, fee caps, limits on roll-over activity,

cooling-oﬀ periods, and outright bans, among other measures (Kaufman, 2013). The Con-

sumer Financial Protection Bureau became the industry’s ﬁrst federal regulator in 2011 .

The CFPB issued rules governing the payday industry in 2017 which were largely rescinded

in 2020, so regulation still largely falls on the states

(Kirsch, Mayer and Silber, 2014).

As of 2015, 15 states had banned traditional storefront lending. State payday laws are

complex, and jurisdiction over online lending remains contested in the court system, despite

many state and federal regulators enforcing laws that restrict online loans in states that also

regulate storefront lending (King and Standaert, 2013; Consumer Federation of America,

2010). In addition to regulatory considerations, other features that diﬀer between the online

and storefront payday loan markets include payment and collection mechanisms that involve

Automated Clearing House (ACH) transactions and bank account access instead of dated

checks and in-person payment, and the online advertising market (PEW, 2014).

Online payday loans have structural diﬀerences from the storefront business model in

that the former requires two additional layers of ﬁnancial intermediaries, which add to the

lending costs. In the ﬁrst step, online payday lending starts with online lead generation. A

lead is an opportunity to do business, proxied by an expression of interest by a potential

customer. Leads can be tracked and traded. A merchant that sells leads to potential lenders

See Mann and Hawkins (2007) for more information on the “rent-a-bank” model.

See https://www.consumerﬁnance.gov/payday-rule/

is a lead generator. As is common in digital marketing, lead generation is heavily targeted and

follows consumers in every action they perform online. According to a Wall Street Journal

piece

, 75% of online payday loan volume is sourced from lead generators, at the same time

that lead generators have a large space to conduct fraudulent and abusive behavior, such as

selling personal ﬁnancial information without consent.

In the second step, lead aggregators do auctions to sell their lead portfolios. These

auctions can be conducted via ping trees. Ping trees can be used when leads can be sold

more than once to diﬀerent buyers and when leads are exclusive and only sold once. They

are electronic queues lenders pay to deﬁne their priority in receiving a lead. As a lead

aggregator explains it on their website, ping tree placement can range from $2 to upwards

of $120. Online lenders set their price points, and as leads come in, they are shown to the

lenders willing to pay the highest price per lead. Lead information helps lenders decide

whether to purchase the lead. Lower ping tree bids result in a lower tree position and a

higher chance of lower lead quality. Higher returns from lending are promised in exchange

for a higher bid for a privileged pin tree position.

3 Data

Our storefront and online payday loan data are sourced from Clarity, an alternative credit

bureau and a subsidiary of Experian, one of the three major credit reporting agencies. Pre-

vious research using Clarity data includes Di Maggio, Ma and Williams (2020), Miller and

Soo (2020), Miller and Soo (2021), and Fonseca (2023). Blattner and Nelson (2021) use sim-

ilar data from FactorTrust, another alternative credit bureau. Clarity compiles application,

origination, and repayment information for subprime loans to help lenders make underwrit-

ing decisions. Its database includes about 63 million borrowers and over 70% of non-prime

consumers in the United States. Like other credit reporting agencies, Clarity relies on vol-

See ’Middlemen for Payday Lenders Under Fire’, The Wall Street Journal, April 7, 2014.

Link: https://www.wsj.com/articles/SB10001424052702304819004579487983000120324

untary reporting of inquiries, originations, and performance by its network of lenders and

data furnishers, which may not reﬂect the full universe of subprime loans or the universe

of information from all participating lenders. Nonetheless, it is among the best sources of

nationwide subprime credit activity.

The Clarity database contains information on various subprime credit products includ-

ing payday, rent-to-own, installment, auto, and auto title loans. Our focus is on storefront

and online payday loans in this study, which represent about 32% of inquiries and 47%

of tradelines in the full database. For each inquiry, Clarity reports information about the

type of loan applied for and some self-reported demographics including ZIP code and state

of residence, monthly income, age, housing status, months at the same address, and pay-

check frequency. While some lenders may employ income and identity veriﬁcation and fraud

detection mechanisms, the information reported in inquiries is self-reported by borrowers

and might not be veriﬁed before submission to Clarity. For originated tradelines, we ob-

serve loan type, highest credit, scheduled and actual payment amounts, payment dates, and

delinquency status.

We use two samples provided by Clarity in our analysis. The ﬁrst one, known hereafter

as the “standalone” or “random Clarity” sample, consists of 1 million consumers randomly

drawn from Clarity’s database from 2013 to 2017. According to the data provider, Clarity’s

full database consisted of about 63 million consumers as of 2020, so our sample contains

about 1.5% of the database. The sample of borrowers includes those who apply for payday

loans and other products, and only a subset of applications result in originated loans, which

we use in our main analysis. In 1 million unique borrowers who submitted an inquiry for

any subprime credit tradeline, 366,327 applied for an online or a storefront payday loan. Of

those, 65,733 originated a payday loan, comprising our ﬁnal sample.

The second sample, known hereafter as the ‘credit visible’ sample, consists of payday

borrowers matched to a random 1% sample of all consumers in the traditional Experian

credit report database as of 2018. All payday loans originated by 35,550 unique borrowers

between 2013 and 2019 are included in this sample. The random Clarity and credit visible

samples are drawn independently.

Because payday loan fees typically do not vary with duration, they are generally mar-

keted to customers in terms of cost per $100 borrowed. However, lenders are also required

to disclose prices in APR terms. Laws on interest rate caps are also commonly written using

interest rates or APR terms. Therefore, we examine both measures of loan prices. We do not

observe prices directly in the Clarity data, and infer them based on observed loan maturity,

highest credit amount, and repayment amount:

AP R =

365

LoanMaturity

Repayment − LoanAmount

LoanAmount

Cost per 100 = 100 ×

Repayment − LoanAmount

LoanAmount

(1)

Because payday loans have fairly simple and standardized structures, these basic formu-

las accurately capture realized prices for most loans. However, one caveat is that scheduled

payment amounts are missing in much of the data, so we need to use realized payments in-

stead. This means that prices will not be accurately captured for loans not repaid in full (e.g.

prices would be inferred to be zero for loans that are fully defaulted on). While defaults

represent a small fraction of loans, if defaulted loans are systematically priced diﬀerently

from repaid loans, our method would lead to measurement error that could be correlated

with our variables of interest.

However, as shown in the next section, risk-based pricing is very limited in the on-

line and storefront payday markets. Therefore, we do not think this potential source of

measurement error drives our results. To impute prices for defaulted loans, we employ a

waterfall methodology to match defaulted loans to the median price of similar non-defaulted

loans within cells by origination month, loan type, state, ZIP code, and terciles of loan and

See DeYoung and Phillips (2006a,b)

borrower characteristics. We try to match defaulted loans to non-defaulted ones in cells of

decreasing granularity until all loans are matched (e.g. ZIP code is matched ﬁrst, and if no

available priced loans are matched by ZIP code, then state-level matches are used). In the

analysis below, we show the results for both the full sample and the ‘non-imputed’ sample

of loans where we measure prices from equation (1) instead of via matching. We winsorize

APR and cost per $100 at the 99th percentile in all analyses to reduce the eﬀect of outliers.

3.1 Summary Statistics and External Validity

In this section, we report summary statistics for our sample. Table 1 presents summary

statistics for the random Clarity sample in Panel A and the credit visible sample of loans

matched to Experian consumer credit records in Panel B. Despite diﬀerences in the sample

periods and the existence of traditional credit reports between the two samples, the descrip-

tive statistics are extremely similar across our two samples. Figure 1 shows the geographical

distribution of loans by state in both of our samples, comparing the online and storefront

markets. Online loans are signiﬁcantly present in all ﬁfty states. Storefront loans are absent

in some sparsely populated states and those where state laws are likely to eﬀectively prohibit

traditional payday lending during our sample period (e.g. Montana, New Mexico, and much

of New England).

The random Clarity sample in Panel A of Table 1 consists of 336,690 loans from more

than 65,733 borrowers, 65% of which are online. The credit visible sample in Panel B includes

188,913 loans and 35,550 borrowers with a 70% online share. By scaling our random Clarity

sample by the size of the full Clarity universe and comparing to industry payday market size

estimates, we calculate that Clarity represents 8% of the storefront market and 23% of the

online payday market as of 2017, with coverage of the total payday market growing from 4%

to 15% of originated loan volume between 2015 and 2017 (Hecht, 2014, 2018; Graham and

Golden, 2019). The larger market share of online versus storefront loans likely reﬂects the

historical evolution of Clarity’s client base and online lenders’ greater use of reporting and

veriﬁcation systems to mitigate fraud risk.

The characteristics of loans and consumers in our samples are consistent with those

from previous literature and policy reports, with average loan amounts of $365 to $370 across

our two samples and average maturities of 19 to 20 days corresponding to a combination

of consumers with weekly, biweekly, and monthly pay dates (Skiba and Tobacman, 2008,

Consumer Financial Protection Bureau, 2013, PEW, 2014, Wang and Burke, 2022). While

loan and customer characteristics are fairly comparable to previous studies using storefront

payday data (e.g. Skiba and Tobacman 2008, Wang and Burke 2022), the average borrower

income of $2822 to $2849 is signiﬁcantly lower in the Clarity online payday data compared

with $4334 among online payday borrowers in an account aggregator sample studied by

Baugh (2016), which could reﬂect diﬀerences in income measurement or the likely higher

income of consumers included in account aggregator data.

We consider a lower bound for the default rate loans reported to Clarity as not paid

in full, by lenders. The average default rate of 7% in the full Clarity samples and 4% in

the storefront samples are comparable to those from previous studies using administrative

data from storefront payday lenders. Skiba and Tobacman (2008) report a 4% charge-oﬀ

rate and Wang and Burke (2022) report a 3% default rate. Both of these previous papers

report substantially higher delinquency rates than default rates, suggesting that the default

rate we measure in Clarity likely corresponds to ultimate charge-oﬀs and not to temporary

delinquency. We do not attempt to distinguish between delinquency and default, track the

ultimate recovery rate of defaulted loans, or account for reporting errors or reporting lags

(e.g. lenders failing to report defaults to Clarity).

Average default rates for online loans are 8% to 9% in our samples, about double that

of storefront loans. This contrasts with general demographics associated with lower credit

risk for online loans and borrowers and with a slightly lower propensity for late payments on

online loans (31%) versus storefront (35%). This discrepancy is consistent with technological

advances in the online sphere. ACH processing via routing and account numbers can be done

to avoid delays and cashing checks can generate payments posterior to due dates.

Online loans are signiﬁcantly smaller, and online borrowers report signiﬁcantly higher

income and home ownership and slightly longer months at address compared with storefront

borrowers in both Panels A and B. The main exception to this pattern is that online borrowers

in the credit visible sample are more than twice as likely to be unscoreable compared with

storefront borrowers (21% vs. 10%), and have lower Vantage scores conditional on being

scoreable (500 vs. 535). Even though we classify all borrowers with a traditional Experian

credit report as part of our ‘credit visible’ sample, some nonetheless lack valid Vantage

scores, which likely reﬂects borrowers with thin or potentially incomplete or incorrect credit

ﬁles (Blattner and Nelson, 2021). The higher income, lower age, and higher default risk

associated with online payday loans are consistent with previous survey evidence (PEW,

2014).

Based on the pricing formulas and imputation algorithm described in Section 3, we ﬁnd

average APRs of 373% in the random Clarity sample (Panel A) and 382% in the credit

visible sample (Panel B). The average cost per $100 is $16.6 in the random Clarity sample

and $16.9 in the credit visible sample. Despite the assumptions needed to calculate prices

in the Clarity data, these estimates are consistent with those from previous studies that use

prices directly observed in administrative data from storefront payday lenders. Skiba and

Tobacman (2008) report a cost per $100 of $17.9 using a sample from Texas, one of the most

expensive lending markets. Wang and Burke (2022) report an average cost per $100 of $12

in a multi-state sample and $20 in Texas, corresponding to APRs of 281% and 508%. By

comparison, average prices for storefront loans range from 295% to 300% APR and $12.4 to

$13.3 per $100 in our Clarity samples.

Price statistics for online payday loans are rarer, and we are unaware of previous aca-

demic studies on this topic. In a survey of lender websites, Consumer Federation of America

(2011) reports an average APR of 652% and cost per $100 of $25, which are substantially

higher than the average APRs of 416% - 417% and cost per $100 of $18.5 - $18.9 in the Clarity

samples. Despite Clarity’s substantial market share of online payday loans, less-compliant or

more predatory lenders who may charge higher prices and engage in other unfriendly prac-

tices toward consumers may be less likely to report to Clarity, causing a disparity relative to

the sample of lender websites. Nonetheless, the signiﬁcant price disparity between storefront

and online loans has been widely described in industry and policy reports, so we believe

the diﬀerence of well over 100% APR between these two loan types reﬂects true underlying

heterogeneity, even if its magnitude in the full universe of payday loans is unknown. To the

best our knowledge, this paper is the attempt to quantify the price diﬀerence between online

and storefront payday loans.

While we use the pricing formulas in equation (1) to measure prices for most loans, these

formulas rely on realized payments, which would not accurately measure prices for defaulted

loans. The fourth column of Table 1 shows statistics for the non-imputed loan sample. By

construction, the default rate in this sample is zero, and the proportion of loans with late

payments is lower. The average loan amount is also slightly lower and repayment amount

is signiﬁcantly higher, but other characteristics, including prices, are similar between the

imputed and non-imputed samples.

Using imputed prices allows us to investigate the important role of default rates in

loan pricing which is impossible in the non-imputed sample. To support the validity of this

analysis, we show that other results are remarkably similar across the imputed and non-

imputed loan samples due to the lack of risk-based pricing at the individual loan level. Few

loans realize default even in higher-default groups, allowing us to use non-defaulted loans

to impute prices for defaulted loans. Overall, the summary statistics in Table 1 establish a

basic level of external validity for our analysis and show that the pricing diﬀerences across

online and payday loans are not purely driven by sample selection or measurement error.

Our payday loan sample statistics are well-aligned with Liu, Lu and Xiong (2022),

who describe online loans as smaller, higher interest, earlier repayment, and more repeated

borrowing. Despite not studying payday loans, they suggest that online lenders specialize in

serving short-term liquidity needs, rather than providing long-term ﬁnancing solutions.

4 Online Payday Loan Premium

Online payday loans are contracted at a higher price than storefront payday loans. We

validate the existence this spread in graphical analysis along consumer credit risk predictors,

and in a regression analysis, where we control for a rich set of confounding eﬀects.

4.1 Risk-based Pricing in Payday Loan Markets

To understand the diﬀerences between online and storefront prices, we start by visually

exploring the distribution of prices across the two business models. Figure 2 plots the kernel

densities for APR in graphs (a) and (c) and cost per $100 borrowed in graphs (b) and (d)

for the two Clarity samples. As with other descriptive statistics, the distributions are almost

identical between the random Clarity and credit visible samples. This suggests that payday

loan pricing is not substantially diﬀerent for borrowers with and without existing credit

reports. This is consistent with the segmentation between traditional and alternative credit

markets.

Consistent with pricing schedules based on integer values of cost per $100, the distribu-

tions in graphs (b) and (d) exhibit several local modes for both online and storefront loans.

Even though both distributions have common support, higher price points are more common

for online loans. The pricing function is more continuous for online loans, where cost per

$100 is an exact integer in 17% of observations compared with 32% for storefront. The most

common integer values of cost per $100 are $15, $17, $20, and $25 for online and $8, $10,

$15, and $20 for storefront loans. The interaction of discrete price points for cost per $100

and common pay frequencies leads to a multi-modal distribution of APRs, especially for

storefront loans.

Potential reasons for higher prices online could be higher default incidence than at the

storefront or at least diﬀerences in default predictors such as income or credit score, across

the two business models. To shed initial light on these mechanisms, Figures 3 through 5

present unconditional binscatter plots of how prices and default rates change depending on

loan duration, customer income, and Vantage score.

Figure 3 shows binscatter plots of prices and default risk by loan duration, typically

driven by borrowers’ pay frequency. Conﬁrming our discussion above and typical practices

in the industry that present uniform prices across diﬀerent pay frequencies, Panel B shows

that cost per $100 is ﬂat and does not vary monotonically with loan duration. Despite

that, costs are uniformly higher for online loans at all levels of loan duration. Uniform

pricing by cost per $100 mechanically causes APRs to be strongly negatively correlated with

loan duration, as shown in Panel A. APRs are higher for online loans conditional on loan

duration. This disparity is greater in absolute terms for lower loan duration. Default risk is

slightly negatively correlated with loan duration only for online loans, but there is generally

a limited relationship between these two dimensions, outside accurately controlled setting as

in Hertzberg, Liberman and Paravisini (2018).

Next, Figure 4 shows the relationships between prices and default risk by self-reported

income. As with loan duration, Panel B shows that cost per $100 is ﬂat for both online and

storefront loans, across levels of borrower income. As shown in Panel A, APR is also ﬂat

across borrower income for online loans but is signiﬁcantly positively related to income for

storefront loans, which is driven by a strong negative correlation between income and loan

duration. The lack of price diﬀerentiation by income contrasts with a signiﬁcant negative

relationship between income and default risk, shown in Panel C, which is stronger for online

loans.

The greater use of credit reporting agencies such as Clarity, greater price dispersion,more

continuous pricing functions, and the high overall levels of credit risk could make under-

writing technology particularly valuable in this market. However, we do not observe this

sophistication resulting in risk-based pricing at the loan level or driven by default predictors.

Another potential demographic that drive credit risk is housing stability, as measured by the

number of months a consumer has lived at their current address, shown in Appendix Figure

A1, but we ﬁnd limited evidence of a relationship with either default or prices, possibly due

to noise in this self-reported measure.

While monthly income predicts default, credit scores encompass other dimensions of

consumer credit risk. Beer, Ionescu and Li (2018) show evidence of a limited correlation

between self-reported income and Vantage scores. Figure 5 presents binscatter plots of

prices and default risk by Vantage score, one of the most widely-used consumer credit scores.

Vantage advertises a particular ability to predict default risk for subprime and near-prime

consumers who are not scoreable by other widely-used models such as FICO. In all three

subﬁgures, consumers that are in the credit visible sample but without a valid Vantage score

in the year the loan was originated are pooled and shown in the leftmost data point on the x-

axis (marked as a Vantage score of 300) for comparison with scoreable consumers. The ﬁgure

shows that the Vantage score predicts default risk for storefront and online payday loans, even

conditional on taking out subprime credit. However, as with income, online payday loans

have signiﬁcantly greater credit risk at every level of Vantage. Despite its strong correlation

with credit risk, APR and cost per $100 are only weakly correlated with Vantage score in

the online and storefront markets, consistent with a general lack of risk-based pricing.

Lastly, we bring together loan prices and loan amounts, and correlate them with credit

score, in Figure 6. We do not ﬁnd any systematic pattern in pricing across the credit score

and loan amount spectrum. As in Dobbie and Skiba (2013), loan amounts can be seen either

as a proxy for income-based eligibility or adverse selection. Highest prices take place for

large loan amounts, but we cannot assert a monotonic trend between quantities and prices.

4.2 Measuring the Online Payday Loan Premium

The previously documented relationships between loan pricing, default, and credit risk pre-

dictors support the existence of an online payday loan premium, across the income, maturity,

and credit score spectrum, not explained by diﬀerences in default. The results are consistent

with each lender setting their optimal cost per $100 borrowed, not distinguishing among

borrowers. Establishing empirical evidence of this practice is very important to support our

choice of the theoretical framework used to rationalize in section 5 to rationalize our ﬁndings.

In this section, we use regression analysis, to control for variables observable to the

lender and to the econometrician. Only conditioning on those observables, we statistically

test for the existence and pin down the magnitude of the online payday loan premium. We

test for the existence of conditional pricing diﬀerences between the online and the storefront

business models using the following speciﬁcation:

ist

= α

+ α

+ βOnline + X

ist

+ 

ist

(2)

where Y

ist

is a price or default outcome for a given loan from customer i living in state

or ZIP code s originated at time t. All regressions include borrower-location ﬁxed eﬀects

at borrower-ZIP code or borrower-state level, depending on the speciﬁcation. α

are

time ﬁxed eﬀects for day of week, day of month, month of year, and calendar year. X

ist

is a vector of controls that includes deciles of loan duration, loan size, age, and income,

categorical variables for housing status and pay frequency, and number of inquiries per week

as a measure of time-varying credit demand. For variables that include missing values, we

include a separate category for missing values to maximize sample size. The regressions

include a dummy variable for online loans with the coeﬃcient of interest β. It estimates

diﬀerences in means of the outcome, conditional on all the controls and ﬁxed eﬀects employed.

Standard errors are clustered at the state level, as loans originated in the same state are

subject to similar legal and market circumstances.

Table 2 presents our results. The table includes three columns for each outcome: APR,

cost per $100, and a dummy variable indicating whether the loan defaults. The three diﬀerent

speciﬁcations per outcome variable include either state ﬁxed eﬀects, ZIP code ﬁxed eﬀects,

or customer ﬁxed eﬀects. Panel A shows results for the random Clarity sample, which

covers loans originated between 2013 and 2017. Panel B shows results for the credit visible

sample, which covers 2013 through 2019, and Panel C includes deciles of vantage score as an

additional control in the credit visible sample.

The online payday loan premium is very similar across the two samples, and robust

to the inclusion of Vantage score as a control. As shown in column (1), when state ﬁxed

eﬀects are included, the APR premium is between 92 to 98 p.p. across samples and models.

Surprisingly, the premium increases when ZIP code or customer ﬁxed eﬀects are included

instead of state ﬁxed eﬀects. The coeﬃcient on the online loan dummy ranges from 103

to 110 p.p. APR when including ZIP code ﬁxed eﬀects, and from 127 to 141 p.p. when

including customer ﬁxed eﬀects. As shown in columns (4) through (6), the online payday

loan premium is between $3.4 and $6.5 when expressed in terms of cost per $100. For

comparison, the descriptive statistics in Table 1 showed an unconditional online payday

loan premium of 117-121 p.p. APR and $5.2 to $6.5 per $100 borrowed, which are within

the range of the regression estimates. Overall, these results show that the online payday

loan premium is not explained by diﬀerences in observable loan or customer characteristics

between the two loan types.

As shown in columns (7) through (9), default risk is between 2.6 p.p. and 8.4 p.p. higher

for online loans, although the online coeﬃcient is imprecisely estimated in some models and

samples using the linear probability speciﬁcation. These estimates are within the range of

the unconditional diﬀerence of 4 to 5 p.p. in default probability from Table 1. The large

increase in both prices and default risk when including customer ﬁxed eﬀects reﬂects the fact

that only 2-3% of consumers have both online and storefront loans, and these customers on

average face both higher prices and higher default risk. Nonetheless, the results show that

even customers with both types of loans are more likely to default on an online payday than

on a storefront loan. Thus, diﬀerences in default risk are not explained by time-invariant

customer characteristics.

Consistent with the lack of risk-based pricing, the inclusion of controls for Vantage

score does not produce detectable diﬀerences in the estimated price premia, as shown in

Panel C. Interestingly, despite Vantage score being tightly correlated with default risk, its

inclusiondoes not explain the gap in default risk between online and storefront loans. Thus,

the ﬁndings in Panel C further conﬁrm that consumer characteristics explain neither the

price premium nor the default gap for online payday loans.

To maximize precision and sample size, we use all available loans in both Clarity samples

in the regression analysis. However, there were very few storefront payday loans in the Clarity

data in 2013, so we replicate the analysis dropping 2013 in Appendix Table A1, which shows

similar results to our main sample. We also replicate the analysis on the subsample of non-

defaulted loans, where we calculate prices directly using equation (1) instead of imputing

them from matching non-defaulted loans. These results are shown in Appendix Table A2.

While this sample by deﬁnition has a default rate of zero, the estimated price premia are

similar to those using the full sample that includes imputed and non-imputed prices.

Overall, the results in this section show that the online payday loan premium is not

driven by diﬀerences in consumer or loan characteristics or diﬀerences in pricing models

between online and storefront loans. Other factors such as diﬀerences in ﬁxed costs, the lead

generation system, and advertising and customer acquisition costs likely drive some of these

diﬀerences. So far we have not tested whether higher default rates of online loans – which

are even present within the same customer – are likely to be part of the explanation. To

do so, we run the speciﬁcation in equation (2), but with two major modiﬁcations: (i) we

collapse our data at the market level, where a market is deﬁned alternatively as a state or a

ZIP code in a given week, separately for online and storefront loans and (ii) we control for

both the lagged average propensity to default and for the lagged average propensity for a

loan to have a late payment, in that market.

In Table 3, we report the results. As shown in column (1) across the three samples,

the average storefront APR at the ZIP code level is around 300%, with estimates for the

premium being close to 100 p.p., even when controlling for past default and incidence of

late payments on loans in that ZIP code. Surprisingly, we observe in column (2) that the

average storefront APR at the state level is much higher, around 360%. Given the ZIP code

average, this much higher average is caused by some outlier loans with abnormally large

APRs, happening all across the nation. This leaves a smaller room for a spread between

storefront and online to be statistically detected, yielding a lack of statistical signiﬁcance

to the estimate from the random Clarity sample in Panel A, and a statistically signiﬁcant

estimate at 10% in Panel B. However, in our most conservative speciﬁcation that accounts

for the average Vantage score in a given market for a given week, the spread is sizeable (64.7

p.p.) and statistically signiﬁcant. Furthermore, outlier APRs are normally caused by short-

maturity loans, with the same ﬂat cost per $100, as discussed before. Hence, in columns (3)

and (4) we use the cost per $100, that is not contaminated by outlier maturities. The online

payday premium is detectable and very stable across all samples, and all market deﬁnitions,

ranging from around $3-4 per $100 borrowed at the state market denomination to $6-7 per

$100 borrowed.

Together with the graphical analysis in subsection 4.1, our results suggest an online

payday loan premium of about $4 per $100 borrowed, or 100 p.p. APR, not explained by

ex-ante predictors of default, nor by market-level realized default.

5 Information Asymmetry and Payday Loan Prices

The absence of face-to-face interactions and the heightened risk of fraud and identity theft in

online transactions can lead to increased prices and a greater likelihood of credit risk. This

scenario could arise due to several factors, including the potential for selecting customers

with higher unobservable risk or placing online loans lower in the repayment hierarchy for

a given customer. Moreover, the complexities underlying higher default rates are further

compounded by variations in collection mechanisms. It is widely believed that collection

methods tend to be more assertive for online loans, given that lenders have direct access to

consumers’ bank accounts through the ACH network (CFPB, 2016). Our ﬁnding that online

loans are less likely to result in a late payment corroborates this fact.

Previous literature has empirically examined how information asymmetry in payday

loan markets aﬀects the relationship between quantities of credit (loan amounts) and default

(Dobbie and Skiba, 2013). In this section, we bring information asymmetry to rationalize

prices in the payday loan market, how they relate to default, and to explain the online

payday loan premium. To our knowledge, this paper is the ﬁrst to conduct such an exercise.

5.1 Conceptual Framework and Testable Hypotheses

Thus far we have documented that: (i) online payday loans are more expensive than store-

front payday loans; (ii) ex-ante predictors of default are not priced at the loan-level; (iii)

default rates are higher online than at the storefront.

Selecting a conceptual framework to analyze payday loan prices and default presents a

non-trivial decision. A conventional approach that directly links credit risk to credit prices

would be at odds with our ﬁndings, which indicate that lenders do not engage in risk-based

pricing.

An alternative would be to use an industrial organization framework to consider com-

petition features within and between markets for homogeneous goods or services. However,

it would not be empirically testable in our setting, since we do not observe lender identi-

ﬁers, and are therefore unable to directly measure lender-level market shares, market power

concentration, or elements of their cost structures.

To incorporate information asymmetry in our setting, and still be able to speak to how

the industry organizes itself and to credit prices, we build on the framework from the seminal

work of Stiglitz and Weiss (1981).

In their setting, information is asymmetric, and thus lenders cannot price each individual

loan. Therefore, lenders choose a rate to charge every customer. The expected proﬁt of each

loan depends on the chosen rate, and on its riskiness. In turn, riskiness itself depends on

the chosen rate, as an excessively high rate makes the repayment too large and defaulting

becomes optimal or the consumer experiences hardship. Borrowers are heterogeneous in risk,

and each borrower is only willing to borrow up to a certain price. So, as the rate increases,

the pool of borrowers gets riskier. The assumptions that need to hold and how they apply

to the payday loan market are described in the Appendix.

The main result of this framework is credit rationing at storefront payday lenders, who

need to choose a one-size-ﬁts-all fee per $100 borrowed, and customers self-selecting, likely

adversely, to enter their stores and inquire for a loan. Instead of choosing the rate at which

supply equals demand the storefront lender chooses the rate for which expected proﬁt is

maximized, under asymmetric information. This rate leaves some loan demand unattended

in the storefront market. Customers not served by storefront lenders, then resort to online

payday lenders.

Arnold and Riley (2009) note that under mild assumptions within the Stiglitz-Weiss

setting, if there is credit rationing at a lower rate, then a second rate is also optimal. That

optimal rate is higher and clears the market, being paid by higher-risk borrowers. In the

same fashion, Cassar and Wydick (2012) incorporate present bias rather than undertaking

risky projects as the borrower source of risk under asymmetric information, as is likely the

case in the payday loan market. They assert that a new lending technology would facilitate

an equilibrium where high-risk borrowers borrow at higher rates, and low-risk borrowers

borrow at a low rate.

Altogether, and put in the context of the storefront-online dichotomy, it would have to be

the case that online lenders would simultaneously (i) face higher asymmetry of information,

(ii) maximize their proﬁt at a higher rate and (iii) lend to a riskier borrower pool. These

are our testable hypotheses. Intuitively, the loss of soft information that can be collected in

person at the store can be claimed as aggravated asymmetric information in online payday

loan markets, especially if we consider the increased risk of identity theft and fraud.

The existence of the online payday loan premium establishes that online lenders likely

maximize their proﬁts at higher rates. This can be due to their superior ability to collect

repayment through the ACH network, as shown by the lower occurrence of late payments,

mitigating some of the moral hazard stemming from practicing higher rates. Another ex-

planation is the ability to better diversify their portfolio, in quantities originated and geo-

graphically, due to the online nature of their lending activity.

Online prices are empirically veriﬁed to be higher, and online borrowers exhibit higher

default rates. The only hypothesis lacking a formal test so far is that of establishing that

the online market has a higher degree of information asymmetry. We do so in the next

subsection.

5.2 Validating Testable Hypotheses

We hypothesize that there is information asymmetry in payday loan markets. We explicitly

remain agnostic as to whether it manifests through adverse selection, moral hazard, or both.

As is common in cases where asymmetric information is claimed, in this section, we test for

correlations between the outcome of interest (loan prices) and default. A positive correlation

indicates information asymmetry, either because riskier borrowers adversely select loans with

higher rates, or because too high of a rate might make it optimal to default on a loan. Our

goal is to test whether asymmetric information happens to a greater extent in the online

market than in the storefront market.

To establish a comparable set of loan prices and defaults, we bin our loans with two

diﬀerent schemes. In the ﬁrst scheme, we bin loans according to age × income quintiles,

totaling 25 bins for online and 25 bins for storefront market. This is done in both the

standalone and credit visible samples. In the second scheme, we bin loans according to

credit score quintiles, only possible in the credit visible sample. This procedure ensures we

always compare similar loans between markets, and within the same risk class.

Our ﬁrst exercise is to ensure the online premium still holds within bins (ruling out

the alternative explanation that the online premium is driven by a diﬀerent composition of

borrowers in the two markets). In Appendix Table A3, we report the results of running the

speciﬁcation in eq. 2, but including bin ﬁxed eﬀects. Therefore the coeﬃcient on the online

dummy is identiﬁed by average diﬀerences between the storefront and online prices within

the same borrower bin. We observe an online premium of 91-124 p.p. APR, or $4-5 per

$100 borrowed, which are very consistent magnitudes with the previous tables. All in all,

our binning scheme does not eliminate any of the online payday loan premiums.

With this binning scheme, we test for asymmetric information, and to what extent it

is more prevalent online. A direct test of asymmetric information relates loan pricing with

default, and tests for a correlation. Furthermore, we test this hypothesis as suggested by

Chiappori and Salanie (2000) and implemented by Dobbie and Skiba (2013). If similar

borrowers obtain diﬀerent prices on their loans, and higher prices are associated with higher

default rates, there is evidence of asymmetric information in this market. Additionally, we

interact the price variables with the online dummy to test for diﬀerences between the two

business models. The speciﬁcation is the following:

Default

isbt

= α

+α

+β

P rice

ibst

+β

Online

ibst

+β

O,P

Online

ibst

×P rice

ibst

+

ibst

(3)

where the coeﬃcient β

tests for the correlation between loan prices and default in storefront

markets, and the coeﬃcient β

O,P

tests to what extent that correlation is higher online than

at the storefront.

Table 4 reports the results. In Columns (1)-(6), we ﬁnd a positive correlation between

loan prices and default. Again, we do not claim a direction: it could be that within the

same bin, borrowers more likely to default choose higher rates, or that a higher repayment

makes borrowers more likely to default. This association is more accentuated in online loans

as reported by the interaction terms, most of them statistically signiﬁcant at 10%, across

diﬀerent samples, outcomes, and binning schemes. Importantly, due to the lack of individual-

level risk-based-pricing, we can have high conﬁdence that the correlation shown in Table 4

is not a simple reﬂection of the risk premium imposed by the lenders on riskier borrowers.

We take the evidence in Table 4 as support for asymmetric information in payday loan

markets and that such asymmetry is higher online. In practice, payday lenders do not per-

form hard credit pulls from credit bureaus of prime credit products. This is particularly acute

in the online market, where lenders do not observe beyond noisy self-reported information

until far later in the process, making these results reasonably expected.

Graphically, we explore this relation separately for online and storefront loans using

binscatter plots of each loans’ APR relative to their bins’ average APR and default. What

we aim to compare is the slope of the relation. The steeper it is, the more robust is the

evidence of asymmetric information in credit markets. We report our ﬁndings in the credit

visible sample in Figure 7. On the two left panels, we show the binscatter plot for storefront

loans. On the two right panels we report the same graph for online loans. Panels (a)-

(b) report correlations between APR (relative to bin average) and default. Panels (c)-(d)

report correlations between cost per $100 borrowed (relative to bin average) and default.

The results are consistent with Table 4. There is an upward-sloping relationship between

APR and credit risk within the same bin, and it is stronger in the online market, suggesting

a higher degree of asymmetric information online. These results are robust to sampling,

as shown Appendix Figure A2, where we plot the same analysis using the random Clarity

sample.

To complement our results, we use an alternative test, as proposed by Chiappori and

Salanie (2000). We directly estimate the correlation between price and credit risk via bivari-

ate probit estimation. The advantage of this approach is the absence of a stance regarding

the dependent or independent variable. The disadvantage is that it requires two binary vari-

ables. We use an indicator variable of ex-post loan default as the endogenous variable for

credit risk. We use an indicator of whether the loan price lies above the average price of the

borrowers’ bin as the endogenous variable for loan price.

We report results for the estimate for correlation between loan price and default in Table

5. In Columns (1) and (3) we report correlations between prices and defaults in the storefront

market, and in Columns (2) and (4) we report correlations for the online market. Each of

panels A, B, and C reports estimations on the random Clarity sample, on the credit visible

sample, and the credit visible sample controlling for vantage score, respectively. Across all

samples and all outcomes, it is consistently true that there is a positive correlation between

credit prices and default. That correlation is higher in the online payday loan market.

Lastly, if our interpretation of the pooling market equilibrium is correct, we can make

two testable claims. First, a smaller online market share, leads to an online borrower pool

extremely risky, compared to the larger, much safer storefront borrower pool. Second, as a

result, the online premium should be larger when the online market share is lower. In Figure

8, we plot binscatter plots of default rates of each pool (online vs storefront) in each market

(state-year), in panels (a) and (c), and the online premium in each market, in panels (b)

and (d), sorted by online market share. We observe exactly what our predictions indicate:

a larger online market share, captures safer borrowers to the online pool, and makes the

riskiness of the two pools similar, reducing the online payday loan premium.

5.3 Evidence from Statewide Payday Loan Databases

We have quantiﬁed the online payday loan premium and conducted empirical tests to detect

asymmetric information within the payday loan market. Furthermore, we have demonstrated

that the heightened degree of asymmetric information inherent in online payday loans be-

comes apparent when subjected to classic tests for information asymmetry.

We argue that information asymmetry serves as a rationale for the observed higher

realized default rates and loan prices online, particularly in a market where default predictors

are not factored into loan pricing.

The crucial question that arises is whether attenuating information asymmetry would

impact the magnitude of the online payday loan premium. Most importantly, if both online

and brick-and-mortar lenders had access to identical information, would such a signiﬁcant

disparity in their prices still exist?

To test this, we use cross-sectional variation across state laws that establish a statewide

payday loan database, where lenders are mandated to report the loans they provide, and

which can be accessed through the payment of a small fee. There are 13 states which have

this database in place: Alabama, Delaware, Florida, Illinois, Indiana, Kentucky, Michigan,

North Dakota, Oklahoma, South Carolina, Virginia, Washington, and Wisconsin, where

19% of the loans of our sample are originated. These databases have two eﬀects: ﬁrst, they

diminish information asymmetry for the market as a whole, as all the lenders in a state

can now observe the same information. Second, they decrease the diﬀerences in information

asymmetry between online and storefront lenders.

A limitation arising from our samples ranging 2013-2017 and 2013-2019 is that we

cannot directly exploit the eﬀects of introducing a database in a given state, as the 13 states

implemented databases during 2002-2012. Notwithstanding, the cross-sectional variation

still allows us to test the asymmetric information explanation. To do so, we estimate the

following speciﬁcation:

ist

= α

+ α

+ β

Online

ist

+ β

DID

Online

ist

× Database

+ X

ist

+ 

ist

(4)

where Y

ist

is the price of a loan i, both measured as APR (in p.p.) or cost per $100

borrowed (in dollars), for a customer living in state s originating at time t. We run the

speciﬁcation with and without time ﬁxed eﬀects α

, and loan and customer level controls

ist

. We cannot include state ﬁxed eﬀects, as the cross-sectional variation of the variable

Database

is state-level. The three most important coeﬃcients are β

, which measures the

online premium in states without a database, where asymmetric information is higher and

diﬀerent between storefront and online lenders; and β

DID

the diﬀerence in online premia

between states with and without databases. The size of β

DID

, relative to β

allows us to

speak to the economic signiﬁcance of information asymmetry in explaining the two-price,

dual-credit risk equilibrium.

Table 6 reports the results for the estimation of equation 4 on both the random Clarity

sample (Panel A) and the credit visible sample (Panel B). The online payday loan premium

ranges between 102-122 APR percentage points or $4.1-$5.8 per $100 dollars borrowed in

states without a database. Even though accessing the database incurs a cost for the lender,

legislation provisions such as Wis.Stat. § 138.14(10)(b)1. state “A licensee may not assess a

customer any fee or charge for database access or usage.”

In states with a database, our point estimates indicate that the online premium has

about half of the magnitude of the premium in states without a database. The β

DID

co-

eﬃcient is statistically signiﬁcant at the 10% level in all speciﬁcations for cost per $100

borrowed, but is not, when using APR as the dependent variable.

As an extended test, in columns (5)-(6), we use loan amounts as the dependent variable.

In states without a database, loan amounts are smaller online. However, in states with a

database, loans are larger online. This is consistent with databases reducing information

asymmetries online, leading to a lesser extent of credit rationing.

In our best attempt to use all the time series variation we can, we study pricing dynamics

following the implementation of a database, and whether the records added to it through

time have any impact on (i) loan prices on the market as a whole, and (ii) in the online

payday loan premium. To do so, we document the year in which each state implemented

its database, and for each loan in our sample, we compute the number of years since the

implementation of the database, yielding variations along the time series dimension. We then

re-estimate equation 4, but we decompose the Database

variable into dummy variables that

indicate whether state s has implemented a database less than 6 years before, 7-8, 8-10, 10-

12, or more than 13 years ago. This binning scheme is the most granular we can have where

3 or more states are used to identify the coeﬃcient for each bin.

In Figure 9, we report the results for our two price outcomes, estimated in our two

diﬀerent samples. The Baseline estimates are estimates of online (green) and storefront (or-

ange) prices, and the premium (black) in states without a database. We then plot estimated

prices for each market and the premium (obtained from the estimates in Appendix Table

A4) for each time bin relative to database implementation in the state.

Three expected patterns arise: (i) The equilibrium interest rates in the states without

payday loan database are higher than those in the states without the database. This pattern

is consistent with the estimations in Table 6. (ii) As time passes, after the introduction of

payday loan database, the average equilibrium interest rates in both the online and storefront

markets decrease. This pattern is consistent with the increase in market eﬃciency in payday

loan markets via the reduction in information asymmetry when all the lenders share the

same information set; and (iii) the gap between online prices and storefront prices remains

statistically indistinguishable from zero as time goes by. The standard deviation of the

interest rate gap is large in the ﬁrst few years and drops to a much lower value as time

passes. This pattern is consistent with the stabilization of the payday lending market after

the changes from introducing a shared database. All three patterns are robust across samples

and diﬀerent deﬁnitions of price.

6 Decomposing the Online Payday Loan Premium

Asymmetric information explains the pooling equilibrium evident in the payday loan market,

encompassing both online and brick-and-mortar business models. The presence of informa-

tion asymmetry alone can give rise to an equilibrium where two distinct loan prices coexist.

However, what remains uncertain is whether the observed premium in information asymme-

try justiﬁably accounts the magnitude of the online premium. In this section, we delve into

the signiﬁcance of other factors in elucidating the price diﬀerential across various business

models.

We start by formalizing the relationship between two quantities that we observe: the

price and default rate in a given market, for each business model. We model cost per $100

borrowed, as it is independent of loan maturity. Business model i prices their loans following:

(λ

) = π

+ θ

+ ε

(5)

where c

is the cost per $100 of business model i in market m, and λ

is the default

rate experienced by business model i in market m. Note that we allow each business model

to have diﬀerent default rates in a given market, as that matches our empirical evidence. π

is the gross proﬁt per $100 that business model i requires, even in the absence of any default.

Economically, it has to cover all the costs the business model incurs to originate a loan. θ

is the compensation business model i demands per unit of risk. This term will be a function

of recovery rates, among others: a larger loss in default, leads to a higher compensation

demanded for that risk. A business model i is then deﬁned by the pair {π

, θ

}, i = O, S.

Using a variation of the Oaxaca-Blinder decomposition, we can write the online payday

loan premium as the diﬀerence between the two pricing equations:

− c

= (π

− π

) + λ

(θ

− θ

) + θ

(λ

− λ

) + (ε

− ε

) (6)

Equation 6 decomposes the premium into diﬀerences in business model fundamentals,

diﬀerences in credit risk, and diﬀerences in unexplained factors. Diﬀerences in business

model fundamentals can be broken into diﬀerent cost structures and diﬀerent compensations

for default events. The decomposition assumes E(ε

− ε

) = 0.

To estimate the contribution of each component, we deﬁne a market with three alter-

native binnings: state-year, income deciles by Vantage deciles by state, Vantage deciles by

year by month. The latter two binnings are only feasible for the credit visible sample.

Table 7 reports the estimates for the decomposition of unconditional diﬀerences in cost

per $100 between online and storefront payday loan business models. The diﬀerence in loan

prices ranges between $3.1 and $5.2, depending on binning schemes. Using columns solely

(3)-(4) to assess the qualitative consistency of our estimates, we rely on columns (1)-(2) as

robust deﬁnitions of a market, and comparable across samples.

While diﬀerences in default rates carry statistical and economic signiﬁcance, they ac-

count for only about 25% of the online payday loan premium, with less than 25% of this

premium attributed to discrepancies in default losses. Surprisingly, the primary component

of the online payday loan premium stems from variation in cost structures between business

models, explaining approximately 50% of the online premium.

This discovery aligns with the notion that online payday lenders cater to borrowers

who are visibly similar to those served by brick-and-mortar lenders but face higher default

rates. It also resonates with the idea that online lenders encounter greater susceptibility

to fraud and recoup smaller portions of defaulted amounts. We interpret the signiﬁcance

of diﬀerences in cost structures as a consequence of the intricate network of aﬃliates, ping

trees, lead generators, and aggregators. This aspect stands as a fundamental distinction

between online and brick-and-mortar lending models, directly inﬂuencing cost structures,

and in turn, loan prices.

7 Conclusion

This paper presents novel evidence of an online payday loan premium. Using data from a

national subprime credit bureau, we show that despite the potential for online technology

to lower ﬁxed costs and increase lending eﬃciency, online payday loans incur a higher cost

by $4 per $100 or 100 p.p. APR, even conditioning on loan and customer characteristics.

Interestingly, neither brick-and-mortar nor online payday loans appear to utilize risk-based

pricing, and available metrics of default risk fail to elucidate the price premium.

Default rates are about double for online payday loans compared with storefront loans,

and customers with both types of loans are much more likely to default on online loans.

However, predictors of default and realized default do not absorb the premium. From an

economic standpoint, the patterns we observe are reconciled in an information asymmetry

framework that generates credit rationing at the storefront and clears the market with riskier

borrowers facing higher rates in the online payday loan market.

Our ﬁndings underscore the signiﬁcance of disparities in cost structures between both

business models as the predominant component of the online payday loan premium. This

paper speaks to the relevance of evaluating the online and brick-and-mortar credit mar-

kets concurrently, rather than in isolation, as similar dynamics can arise in the presence of

ﬁnancial frictions.

This is the ﬁrst paper asserting the existence of a premium in online payday loans,

measuring, and decomposing it. The study of online lending business models is pivotal for

understanding their pricing mechanisms and the impact of FinTech players on consumer

credit, particularly in markets catering to the most vulnerable borrowers.

References

Adams, William, Liran Einav, and Jonathan Levin, “Liquidity constraints and imper-

fect information in subprime lending,” American Economic Review, 2009, 99 (1), 49–84.

Agarwal, Sumit, Paige Marta Skiba, and Jeremy Tobacman, “Payday loans and

credit cards: New liquidity and credit scoring puzzles?,” American Economic Review,

2009, 99 (2), 412–17.

Alexandrov, Alexei and Sergei Koulayev, “No shopping in the us mortgage market:

Direct and strategic eﬀects of providing information,” Consumer Financial Protection

Bureau Oﬃce of Research Working Paper, 2018, (2017-01).

Allcott, Hunt, Joshua Kim, Dmitry Taubinsky, and Jonathan Zinman, “Are high-

interest loans predatory? theory and evidence from payday lending,” The Review of Eco-

nomic Studies, 2022, 89 (3), 1041–1084.

Argyle, Bronson, Taylor Nadauld, and Christopher Palmer, “Real eﬀects of search

frictions in consumer credit markets,” The Review of Financial Studies, 2023, 36 (7),

2685–2720.

Arnold, Lutz G and John G Riley, “On the possibility of credit rationing in the stiglitz-

weiss model,” American Economic Review, 2009, 99 (5), 2012–21.

Bartlett, Robert, Adair Morse, Richard Stanton, and Nancy Wallace, “Consumer-

lending discrimination in the FinTech era,” Journal of Financial Economics, 2022, 143

(1), 30–56.

Baugh, Brian, “Payday borrowing and household outcomes: Evidence from a natural

experiment,” 2016.

Beer, Rachael, Felicia Ionescu, and Geng Li, “Are income and credit scores highly

correlated?,” FEDS Notes, 2018.

Bertrand, Marianne and Adair Morse, “Information disclosure, cognitive biases, and

payday borrowing,” The Journal of Finance, 2011, 66 (6), 1865–1893.

Bhutta, Neil, “Payday loans and consumer ﬁnancial health,” Journal of Banking & Fi-

nance, 2014, 47, 230–242.

, Paige Marta Skiba, and Jeremy Tobacman, “Payday loan choices and conse-

quences,” Journal of Money, Credit and Banking, 2015, 47 (2-3), 223–260.

Blattner, Laura and Scott Nelson, “How costly is noise? Data and disparities in con-

sumer credit,” Unpublished Working Paper, arXiv preprint arXiv:2105.07554, 2021.

Blinder, Alan S, “Wage discrimination: reduced form and structural estimates,” Journal

of Human resources, 1973, pp. 436–455.

Buchak, Greg, Gregor Matvos, Tomasz Piskorski, and Amit Seru, “Fintech, reg-

ulatory arbitrage, and the rise of shadow banks,” Journal of Financial Economics, 2018,

130 (3), 453–483.

Carrell, Scott and Jonathan Zinman, “In harm’s way? Payday loan access and military

personnel performance,” The Review of Financial Studies, 2014, 27 (9), 2805–2840.

Carter, Susan Payne and William Skimmyhorn, “Much ado about nothing? New

evidence on the eﬀects of payday lending on military members,” Review of Economics and

Statistics, 2017, 99 (4), 606–621.

Cassar, Alessandra and Bruce Wydick, “Credit rationing with behavioral foundations:

Revisiting Stiglitz and Weiss,” Unpublished Working Paper, 2012.

Chava, Sudheer, Rohan Ganduri, Nikhil Paradkar, and Yafei Zhang, “Impact of

marketplace lending on consumers’ future borrowing capacities and borrowing outcomes,”

Journal of Financial Economics, 2021, 142 (3), 1186–1208.

Chiappori, Pierre-Andr´e and Bernard Salanie, “Testing for asymmetric information

in insurance markets,” Journal of Political Economy, 2000, 108 (1), 56–78.

Consumer Federation of America, “States Have Jurisdiction over Online Payday

Lenders,” Technical Report 2010.

, “CFA Survey of Online Payday Loan Websites,” Technical Report 2011.

Consumer Financial Protection Bureau, “Payday Loans and Deposit Advance Prod-

ucts: A White Paper of Initial Data Findings,” Technical Report 2013.

DeFusco, Anthony A, Huan Tang, and Constantine Yannelis, “Measuring the wel-

fare cost of asymmetric information in consumer credit markets,” Journal of Financial

Economics, 2022, 146 (3), 821–840.

DeYoung, Robert and Ronnie J Phillips, “Payday loan pricing,” Networks Financial

Institute, 2006, pp. 09–07.

and Ronnie Phillips, “Strategic pricing of payday loans: Evidence from Colorado,

2000-2005,” Networks Financial Institute Working Paper, 2006.

Di Maggio, Marco and Vincent Yao, “Fintech borrowers: lax screening or cream-

skimming?,” The Review of Financial Studies, 2021, 34 (10), 4565–4618.

Dobbie, Will and Paige Marta Skiba, “Information asymmetries in consumer credit

markets: Evidence from payday lending,” American Economic Journal: Applied Eco-

nomics, 2013, 5 (4), 256–82.

Dobridge, Christine L, “High-cost credit and consumption smoothing,” Journal of Money,

Credit and Banking, 2018, 50 (2-3), 407–433.

Ernst & Young, “The Cost of Providing Payday Loans in a US Multiline Operator Envi-

ronment,” Technical Report 2009.

Fonseca, Julia, “Less mainstream credit, more payday borrowing? Evidence from debt

collection restrictions,” The Journal of Finance, 2023, 78 (1), 63–103.

Fuster, Andreas, Matthew Plosser, Philipp Schnabl, and James Vickery, “The

role of technology in mortgage lending,” The Review of Financial Studies, 2019, 32 (5),

1854–1899.

Gathergood, John, Benedict Guttman-Kenney, and Stefan Hunt, “How do payday

loans aﬀect borrowers? Evidence from the UK market,” The Review of Financial Studies,

2019, 32 (2), 496–523.

Graham, Karen and Elaine Golden, “Financially Underserved Market Size Study 2019,”

Technical Report 2019.

Hecht, John, “Alternative Financial Services: Innovating to Meet Customer Needs in an

Evolving Regulatory Framework,” Technical Report 2014.

, “Short Term Lending Update: Moving Forward with Positive Momentum,” Technical

Report 2018.

Hertzberg, Andrew, Andres Liberman, and Daniel Paravisini, “Screening on loan

terms: evidence from maturity choice in consumer credit,” The Review of Financial Stud-

ies, 2018, 31 (9), 3532–3567.

Huckstep, Aaron, “Payday lending: Do outrageous prices necessarily mean outrageous

proﬁts,” Fordham J. Corp. & Fin. L., 2007, 12, 203.

Jaﬀee, Dwight M and Thomas Russell, “Imperfect information, uncertainty, and credit

rationing,” The Quarterly Journal of Economics, 1976, 90 (4), 651–666.

Jagtiani, Julapa and Cathy Lemieux, “The Roles of Alternative Data and ML in Fintech

Lending: Evidence from LendingClub Consumer Platform,” Unpublished Working Paper,

2018.

Kaufman, Alex, “Payday Lending Regulation,” Finance and Economics Discussion Series,

01 2013, 2013, 1–38.

King, Uriah and Diane Standaert, “Eﬀective State and Federal Payday Lending En-

forcement: Paving the Way for Broader, Stronger Protections,” Technical Report, Center

for Responsible Lending 2013.

Kirsch, Larry, Robert N Mayer, and Norman I Silber, “The CFPB and payday

lending: New agency/old problem,” Journal of Consumer Aﬀairs, 2014, 48 (1), 1–16.

Kutzbach, M, A Lloro, J Weinstein, and K Chu, “How America Banks: Household

Use of Banking and Financial Services, 2019 FDIC Survey,” 2020.

Li, Emma, Li Liao, Zhengwei Wang, and Hongyu Xiang, “Individual Consumption

Response to Expanding Credit Access: Evidence from online cash loan platform,” PBCSF-

NIFR Research Paper, 2018.

Li, Tsung-Hsien and Jan Sun, “The Payday Loan Puzzle: A Credit Scoring Explana-

tion,” Available at SSRN 3954233, 2021.

Li, Zhiyong, Ke Li, Xiao Yao, and Qing Wen, “Predicting prepayment and default

risks of unsecured consumer loans in online lending,” Emerging Markets Finance and

Trade, 2019, 55 (1), 118–132.

Liu, Lei, Guangli Lu, and Wei Xiong, “The big tech lending model,” National Bureau

of Economic Research, 2022.

Maggio, Marco Di, Angela T Ma, and Emily Williams, “In the Red: Overdrafts,

Payday Lending and the Underbanked,” 2020.

Mann, Ronald J and Jim Hawkins, “Just Until Payday,” UCLA Law Review, 2007, 54,

855.

Melzer, Brian T, “The real costs of credit access: Evidence from the payday lending

market,” The Quarterly Journal of Economics, 2011, 126 (1), 517–555.

Miller, Sarah and Cindy K Soo, “Does increasing access to formal credit reduce payday

borrowing?,” 2020.

and , “Do neighborhoods aﬀect the credit market decisions of low-income borrowers?

evidence from the moving to opportunity experiment,” The Review of Financial Studies,

2021, 34 (2), 827–863.

Montezemolo, Susanna, “Payday lending abuses and predatory practices,” The State of

Lending in America & Its Impact on US Households, Center for Responsible Lending,

2013.

Morse, Adair, “Payday lenders: Heroes or villains?,” Journal of Financial Economics,

2011, 102 (1), 28–44.

, “Peer-to-peer crowdfunding: Information and the potential for disruption in consumer

lending,” Annual Review of Financial Economics, 2015, 7, 463–482.

Nunez, Stephen, K Shaberg, Richard Hendra, Lisa Servon, Mina Addo, and

A Mapillero-Colomina, “Online payday and installment loans: Who uses them and

why? A demand side analysis from linked administrative, survey, and qualitative interview

data,” Subprime Lending Data Exploration Project, 2016.

Oaxaca, Ronald, “Male-female wage diﬀerentials in urban labor markets,” International

economic review, 1973, pp. 693–709.

PEW Charitable Trusts, “Payday Lending in America: Who Borrows, Where They

Borrow, and Why,” Washington, DC, 2012.

, “Fraud and abuse online: Harmful practices in internet payday lending,” Washington,

DC, 2014.

Saldain, Joaquin, “High-Cost Consumer Credit: Desperation, Temptation and Default,”

Unpublished Working Paper, 2023.

Skiba, Paige and Jeremy Tobacman, “Payday Loans, Uncertainty and Discounting: Ex-

plaining Patterns of Borrowing, Repayment, and Default,” Vanderbilt Law and Economics

Research Paper No. 08-33 08 2008.

Skiba, Paige Marta and Jeremy Tobacman, “Do payday loans cause bankruptcy?,”

The Journal of Law and Economics, 2019, 62 (3), 485–519.

Sridhar, Sharada, “Liquidity Based Contracting: A Path to Greater Eﬃciency in Payday

Lending,” Unpublished Working Paper, 2023.

Stango, Victor and Jonathan Zinman, “Borrowing high versus borrowing higher: price

dispersion and shopping behavior in the US credit card market,” The Review of Financial

Studies, 2016, 29 (4), 979–1006.

Stiglitz, Joseph E and Andrew Weiss, “Credit rationing in markets with imperfect

information,” The American Economic Review, 1981, 71 (3), 393–410.

Wang, Jialan and Kathleen Burke, “The eﬀects of disclosure and enforcement on payday

lending in Texas,” Journal of Financial Economics, 2022, 145 (2), 489–507.

A Appendix

Four assumptions need to hold in the Stiglitz and Weiss (1981) framework. We address

them one by one in explaining how the payday loan market is empirically likely to verify

them. The ﬁrst assumption is that there is a rate at which the proﬁtability of a given loan

is maximized. Beyond that rate, the lender selects risky borrowers, high repayment makes

it optimal to default, or both.

∗

This is likely to hold in the payday loan market. As empirical evidence, the absence of

risk-based pricing, even at the storefront, supports the idea of a market where information

is asymmetric, therefore generating this pattern. The second assumption is that borrowers

are heterogeneous in risk, and given their risk, there is a rate at which each borrower exits

the borrower pool, leaving the remaining pool of borrowers with higher average riskiness.

# of borrowers

pool risk

Altogether, the two ﬁrst assumptions, in the simplest context of this model, generate

a hump-shaped expected proﬁt function of the interest rate charged, with a well deﬁned

optimal interest rate, for which the expected proﬁt of the loan portfolio is maximized.

The two last assumptions are conventional. The third assumption is that lenders com-

pete for loanable funds, and that under a competitive environment, ultimately they will oﬀer

their expected return as compensation to the investor, who is willing to supply more loanable

funds for a larger the return, leading to a zero net proﬁt situation. The supply of loanable

funds is therefore increasing in proﬁtability. Payday loans are not funded by deposits, like

bank loans. They have to compete for loanable funds, and being a high credit risk product,

the competition is necessarily through the return they are able to oﬀer. Lastly, the fourth

assumption is that demand for loans is decreasing in interest rate. Like any usual demand

function, also likely to hold in payday loan markets.

The equilibrium determination happens with tension between the usual Walrasian in-

terest rate, at which the supply of credit equals the demand of credit, and a credit rationing

scenario as in Jaﬀee and Russell (1976), where the lender ultimately chooses an interest rate

that maximizes their proﬁt, even if it implies supplying a quantity of credit inferior to the

demanded quantity.

Figure 1: Geographic Distribution of Loans per Capita

Panel A: Random Clarity sample

(a) Online (b) Storefront

Panel B: Credit visible sample

Note: In this ﬁgure we map the geographical distribution of payday loans by state. Graphs show total

numbers of online and storefront loans per 1000 people in the state population based on the 2010 Census.

The random Clarity sample presented in Panel A consists of a random sample of 1 million unique borrowers

that submitted loan inquiries in Clarity’s full database between 2013 and 17. Only originated payday loans

from this sample of consumers are included in the analysis sample. The credit visible sample shown in Panel

B consists of payday borrowers that are matched to a random 1% sample of all consumers in the Experian

credit bureau database in 2018. All loans originated by matched borrowers between 2013 and 2019 are

included in this sample.

Figure 2: Online and Storefront Price distributions

Panel A: Random Clarity sample

(a) APR (b) Cost per $100

Panel B: Credit visible sample

Note: This ﬁgure shows the distributions of prices for online and storefront payday loans. The random Clarity

sample presented in Panel A consists of a random sample of 1 million unique borrowers that submitted loan

inquiries in Clarity’s full database between 2013 and 17. Only originated payday loans from this sample

of consumers are included in the analysis sample. The credit visible sample shown in Panel B consists of

payday borrowers that are matched to a random 1% sample of all consumers in the Experian credit bureau

database in 2018. All loans originated by matched borrowers between 2013 and 2019 are included in this

sample.

Figure 3: Prices and Default Rates by Loan Duration

Panel A: APR

(a) Random Clarity sample (b) Credit visible sample

Panel B: Cost per $100

Panel C: Default

(e) Random Clarity sample (f) Credit visible sample

Note: The ﬁgure presents binscatter plots of APR, cost per $100 borrowed, and default rates in the credit

visible sample in graphs (b), (d), and (f), and in the standalone sample in graphs (a), (c), and (e). The

x-axis represents the loan maturity.

Figure 4: Prices and Default Rates by Borrower Income

Panel A: APR

(a) Random Clarity sample (b) Credit visible sample

Panel B: Cost per $100

Panel C: Default

(e) Random Clarity sample (f) Credit visible sample

Note: The ﬁgure presents binscatter plots of APR, cost per $100 borrowed, and default rates in the credit

visible sample in graphs (b), (d), and (f), and in the standalone sample in graphs (a), (c), and (e). The

x-axis represents the borrower’s monthly income.

Figure 5: Prices and Default Rates by Vantage Score

(a) APR (b) Cost per $100

Note: The ﬁgure presents binscatters of APR, cost per $100 borrowed, and default rates in the credit visible

sample. Missing Vantage scores are set to 300 and outcomes for these borrowers are shown as the leftmost

data point in each graph.

Figure 6: Absence of Risk-Based Pricing

(a) APR (%)

(b) Cost per $100

Note: We plot loan prices (in color scale) versus loan amounts and credit scores. In panel (a), loan price is

measured as APR, and in panel (b) loan price is measured as cost per $100.

Figure 7: Prices and Default

(a) APR vs Default, Storefront (b) APR vs Default, Online

Note: The ﬁgure presents binscatter plots of default by bins of price relative to the bin average price in

the credit visible sample. Prices are expressed as APR in panels (a) and (b) and cost per $100 borrower in

panels (c) and (d). Bins are age-income bins.

Figure 8: Online Market Shares, Pool Risk, and the Online Premium

(a) Pool risk, Random Clarity Sample (b) Online Premium, Random Clarity Sample

Note: The ﬁgure presents binscatter plots of default rates online and storefront, in panels (a) and (c), and

the online premium, in panels (b) and (d), along markets with diﬀerent online market shares. Panels (a)

and (b) are built from the random Clarity sample, and panels (c) and (d) are built from the credit visible

sample.

Figure 9: Statewide Payday Loan Database

(a) APR, Random Clarity Sample (b) APR, Credit Visible Sample

Note: The ﬁgure presents coeﬃcients for loan-level regressions of prices and 95% conﬁdence intervals, in APR

and cost per $100 borrowed, on a dummy variable that indicates whether the loan is an online loan, and

dummy variables for each bin of time since implementation of a statewide payday loan database, and their

interaction. Regressions contain loan and customer-level controls, and time ﬁxed eﬀects. Robust standard

errors are clustered at the state level.

Table 1: Summary Statistics

Panel A: Random Clarity Sample (2013-2017)

Subsample: All Non-imputed Online Storefront

Mean Median SD Mean Mean Mean

Loan Characteristics

Loan Amount ($) 365 260 265 344 316 456

Repayment Amount ($) 369 300 285 397 303 490

Loan Maturity (days) 20 15 9 19 20 19

Default 7% 0% 26% 0% 9% 4%

Late Payment 32% 0% 47% 26% 31% 35%

APR 373% 322% 217% 363% 416% 295%

Cost per $100 ($) 16.6 17.3 6.3 16.1 18.9 12.4

Online Loan 65% 100% 48% 62% 100% 0%

Self-Reported Information

Owns Home 15% 0% 35% 13% 19% 5%

Age 42.5 41.0 14.0 43.1 39.9 47.3

Months at Address 29.1 24.0 23.9 28.6 29.5 25.6

Net Monthly Income 2545 2200 1490 2533 2849 1970

# of Loans 336,690 272,220 217,596 119,094

# of Unique Borrowers 65,733 46,010 49,877 17,484

Panel B: Credit Visible Sample (2013-2019)

Subsample: All Non-imputed Online Storefront

Mean Median SD Mean Mean Mean

Loan Characteristics

Loan Amount ($) 370 255 284 342 332 460

Repayment Amount ($) 372 300 299 396 320 494

Loan Maturity (days) 19 15 9 19 19 19

Default 7% 0% 26% 0% 8% 4%

Late Payment 32% 0% 47% 26% 32% 34%

APR 382% 336% 208% 372% 417% 300%

Cost per $100 ($) 16.9 17.5 5.8 16.4 18.5 13.3

Vantage score 510 538 113 512 500 535

Unscoreable 18% 0% 38% 18% 21% 10%

Online Loan 70% 100% 46% 68% 100% 0%

Self-Reported Information

Owns Home 16% 0% 36% 14% 20% 6%

Age 42.1 41.0 13.5 42.4 40.1 46.6

Months at Address 30.5 24.0 24.2 30.0 31.0 26.0

Net Monthly Income 2578 2244 1514 2576 2844 1956

# of Loans 188,913 149,458 132,520 56,393

# of Unique Borrowers 35,550 24,654 27,473 9,097

Note: Table contains summary statistics for two samples of online and storefront payday loans from Clarity.

The random Clarity sample presented in Panel A consists of a random sample of 1 million unique borrowers

that submitted loan inquiries in Clarity’s full database between 2013-17. Only inquiries resulting in originated

payday loans are included in the analysis sample. The credit visible sample shown in Panel B consists of

payday borrowers who are matched to a random 1% sample of all consumers in the Experian credit bureau

database in 2018. All inquiries and loans originated by matched borrowers between 2013 and 2019 are

included in this sample. The two samples are drawn independently. Each panel shows statistics for the

full set of loans, ‘non-imputed’ loans where prices are calculated directly from loan-level terms instead of

imputed based on loans with similar characteristics (see text for details), online payday loans, and storefront

payday loans.

Table 2: Online Payday Loan Premium

(1) (2) (3) (4) (5 (6) (7) (8) (9)

Panel A: Random Clarity Sample (2013-2017)

Outcome: APR Cost / 100 Default

Storefront mean: 295 12.4 0.041

Online dummy 98.4 110.3 141.2 4.35 4.65 6.52 0.030 0.026 0.084

(36.0) (41.7) (54.3) (2.20) (2.52) (3.07) (0.026) (0.032) (0.038)

[0.009] [0.011] [0.012] [0.053] [0.071] [0.039] [0.253] [0.422] [0.032]

0.534 0.615 0.775 0.582 0.663 0.812 0.139 0.202 0.511

N 336,690 332,937 305,775 336,690 332,937 305,775 336,690 332,937 305,775

Panel B: Credit Visible Sample (2013-2019)

Outcome: APR Cost / 100 Default

Storefront mean: 300 13.3 0.044

Online dummy 91.9 103.1 127.1 3.36 3.70 4.91 0.036 0.034 0.060

(25.8) (33.7) (54.2) (1.48) (1.78) (2.38) (0.020) (0.027) (0.028)

[0.001] [0.004] [0.023] [0.027] [0.042] [0.045] [0.080] [0.202] [0.041]

0.531 0.605 0.743 0.543 0.623 0.757 0.121 0.167 0.320

N 188,913 186,687 171,518 188,913 186,687 171,518 188,913 186,687 171,518

Panel C: Credit Visible Sample with Vantage

Online dummy 91.9 103.1 127.2 3.37 3.71 4.91 0.036 0.035 0.060

(25.8) (33.7) (54.2) (1.48) (1.78) (2.38) (0.020) (0.027) (0.029)

[0.001] [0.004] [0.023] [0.027] [0.042] [0.044] [0.077] [0.199] [0.041]

0.531 0.605 0.743 0.544 0.623 0.757 0.125 0.171 0.320

N 188,913 186,687 171,518 188,913 186,687 171,518 188,913 186,687 171,518

State FE Yes No No Yes No No Yes No No

Zip FE No Yes No No Yes No No Yes No

Consumer FE No No Yes No No Yes No No Yes

Note: The table presents coeﬃcient estimates of the online loan dummy from regressions of payday loan

prices, default probability, and late payment probability for the random Clarity sample in Panel A and

the credit visible sample in Panels B and C. All regressions include ﬁxed eﬀects for either state, ZIP code,

or customer; ﬁxed eﬀects for day of week, day of month, month of year, and calendar year; and controls

for deciles of loan duration, loan size, age, and income, categorical variables for housing status and pay

frequency, and number of inquiries per week. Panel C additionally includes controls for decile of Vantage

score. Robust standard errors clustered at the state level are in parentheses, and p-values are in brackets.

Table 3: Online Payday Loan Premium - Market Level

(1) (2) (3) (4)

Panel A: Random Clarity Sample (2013-2017)

Outcome: APR Cost / 100

Storefront mean: 304 360 12.7 16.3

Online dummy 120.2 31.5 6.6 2.7

(7.3) (29.4) (0.3) (1.2)

[0.000] [0.290] [0.000] [0.027]

Market Border ZIP Code State ZIP Code State

R2 0.621 0.543 0.720 0.512

N 220,140 10,574 220,140 10,574

Panel B: Credit Visible Sample (2013-2019)

Outcome: APR Cost / 100

Storefront mean: 306 357 13.7 16.6

Online dummy 126.7 62.2 6.1 3.5

(8.7) (30.9) (0.4) (1.3)

[0.000] [0.049] [0.000] [0.008]

Market Border ZIP Code State ZIP Code State

R2 0.619 0.546 0.723 0.551

N 120,152 11,623 120,152 11,623

Panel C: Credit Visible Sample with Vantage

Outcome: APR Cost / 100

Online dummy 126.7 62.2 6.1 3.5

(8.7) (30.9) (0.4) (1.3)

[0.000] [0.049] [0.000] [0.008]

Market Border ZIP Code State ZIP Code State

R2 0.619 0.546 0.723 0.551

N 120,152 11,623 120,152 11,623

Note: The table presents coeﬃcient estimates of the online loan dummy from regressions of payday loan

prices for the random Clarity sample in Panel A and the credit visible sample in Panels B and C. The unit

of observation is market-week, and markets are deﬁned as ZIP codes in columns (1) and (3) and states in

columns (2) and (4). All regressions include ﬁxed eﬀects for either state or ZIP code and week ﬁxed eﬀects

and controls for averages of loan duration, loan size, age, income, number of inquiries per week, and lagged

default rates and lagged percentage of late payments in the market. Panel C controls for average Vantage

score of applicants in that market in that week. Robust standard errors clustered at the market level are in

parentheses, and p-values are in brackets.

Table 4: Asymmetric Information - APR and Default

(1) (2) (3) (4) (5) (6)

Propensities (p.p.)

Outcome: Default

Online Dummy x APR 0.005 0.003 0.000

(0.001) (0.001) (0.000)

[0.000] [0.000] [0.001]

APR 0.003 0.003 0.000

(0.000) (0.001) (0.000)

[0.000] [0.000] [0.000]

Online Dummy x Cost per $100 0.007 0.061 0.079

(0.028) (0.035) (0.034)

[0.793] [0.079] [0.021]

Cost per $100 0.334 0.185 0.162

(0.019) (0.024) (0.024)

[0.000] [0.000] [0.000]

Online Dummy 5.950 5.360 0.054 6.730 5.060 4.760

(0.286) (0.380) (0.004) (0.566) (0.692) (0.684)

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

R2 0.094 0.075 0.078 0.095 0.075 0.077

N 304,404 157,400 157,400 304,404 157,400 157,400

Sample Stand-Alone Credit Visible Credit Visible Stand-Alone Credit Visible Credit Visible

Binning Quintiles Age-Income Age-Income Credit Score Age-Income Age-Income Credit Score

State FE Yes Yes Yes Yes Yes Yes

Year-Week FE Yes Yes Yes Yes Yes Yes

Bin FE Yes Yes Yes Yes Yes Yes

Note: The table presents coeﬃcient estimates of correlations between payday loan prices and default for

the random Clarity sample in Columns (1) and (4) and the credit visible sample in Columns (2)-(3) and

(5)-(6). The unit of observation is loan, and bins are deﬁned by quintiles of age and income in columns

(1)-(2) and (4)-(5) , and credit score in columns (3) and (6). All regressions include ﬁxed eﬀects for bin and

week ﬁxed eﬀects and controls for averages of loan duration, loan size, age, income, number of inquiries per

week. Robust standard errors clustered at the market level are in parentheses, and p-values are in brackets.

Table 5: Asymmetric Information - Bivariate Probit

(1) (2) (3) (4)

Panel A: Random Clarity Sample (2013-2017)

APR > Bin Avg. Cost per $100 > Bin

vs. Default Avg. vs. Default

Rho (ρ) 0.053 0.222 0.057 0.297

(0.010) (0.005) (0.010) (0.005)

[0.000] [0.000] [0.000] [0.000]

N 105,002 199,402 105,002 199,402

Sample Storefront Online Storefront Online

Panel B: Credit Visible Sample (2013-2019)

Rho (ρ) 0.095 0.191 0.029 0.271

(0.014) (0.007) (0.014) (0.007)

[0.000] [0.000] [0.039] [0.000]

N 46,937 110,463 46,937 110,463

Sample Storefront Online Storefront Online

Panel C: Credit Visible Sample with Vantage

Rho (ρ) 0.154 0.191 0.012 0.267

(0.014) (0.007) (0.014) (0.007)

[0.000] [0.000] [0.400] [0.000]

N 46,937 110,463 46,937 110,463

Sample Storefront Online Storefront Online

Note: The table presents bivariate Probit coeﬃcient estimates of correlations between payday loan prices

and default for the random Clarity sample in Panel A, in Panel B and C the credit visible sample, being

Panel C controlled for credit score. In columns (1) and (2) we present correlation estimates for the bivariate

distribution of default and a dummy if the loan has an APR larger than the average APR within its bin. In

Columns (3) and (4), instead of APR, we use cost per $100 borrowed. Standard errors are in parentheses,

and p-values are in brackets.

Table 6: Statewide Payday Loan Database - Diﬀerence in Diﬀerences

(1) (2) (3) (4) (5) (6)

Panel A: Random Clarity Sample (2013-2017)

Outcome: APR (p.p.) Cost per $100 ($) Loan Amounts ($)

Online x Database - 46.6 - 49.0 - 4.9 - 3.6 55.6 107.3

(50.6) (38.3) (1.6) (1.5) (29.6) (42.8)

[0.357] [0.202] [0.002] [0.016] [0.062] [0.013]

Online 122.4 109.1 5.8 5.2 - 17.2 - 50.4

(36.6) (26.5) (1.5) (1.4) (25.0) (25.5)

[0.001] [0.000] [0.000] [0.000] [0.493] [0.049]

Controls No Yes No Yes No Yes

FE No Yes No Yes No Yes

R2 0.199 0.520 0.545 0.573 0.166 0.261

N 336,690 336,690 336,690 336,690 336,690 336,690

Panel B: Credit Visible Sample (2013-2019)

Outcome: APR (p.p.) Cost per $100 ($) Loan Amounts ($)

Online x Database - 39.2 - 34.1 - 4.1 - 2.5 79.4 134.9

(36.7) (31.1) (1.3) (1.4) (28.1) (37.2)

[0.287] [0.274] [0.001] [0.072] [0.005] [0.000]

Online 120.7 101.5 4.8 4.1 - 27.3 - 61.3

(27.3) (20.3) (1.2) (1.1) (22.3) (20.4)

[0.000] [0.000] [0.000] [0.000] [0.222] [0.003]

Controls No Yes No Yes No Yes

FE No Yes No Yes No Yes

R2 0.207 0.518 0.486 0.530 0.153 0.225

N 188,913 188,913 188,913 188,913 188,913 188,913

Note: The table presents coeﬃcient estimates of loan-level regressions of prices, in APR and cost per $100

borrowed, and loan amounts, on a dummy variable that indicates whether the loan is an online loan, and a

dummy variable that indicates whether the state where the tradeline originated has a statewide payday loan

database, and their interaction. Columns (2), (4), and (6) contain loan and customer-level controls, and

time ﬁxed eﬀects. The samples used are in Panel A and the random Clarity sample, in Panel B the credit

visible sample. Robust standard errors clustered at the state-year level are in parentheses, and p-values are

in brackets.

Table 7: Oaxaca-Blinder Decomposition

(1) (2) (3) (4)

21.3 19.7 20.8 19.1

16.4 16.5 17.1 13.9

− c

4.8 3.1 3.7 5.2

(0.2) (0.2) (0.2) (0.2)

[0.000] [0.000] [0.000] [0.000]

(λ

− λ

) 1.4 0.8 1.0 1.0

(0.1) (0.1) (0.1) (0.1)

[0.000] [0.000] [0.000] [0.000]

λS(θ

− θ

) 1.1 0.6 0.8 0.5

(0.1) (0.1) (0.1) (0.1)

[0.000] [0.000] [0.000] [0.000]

(π

− π

) 2.3 1.7 1.9 3.8

(0.3) (0.2) (0.2) (0.2)

[0.000] [0.000] [0.000] [0.000]

Online # Bins 2,336 2,990 3,686 660

Storefront # Bins 1,017 1,336 1,263 565

Sample Random Clarity Credit Visible Credit Visible Credit Visible

Bins State-Month State-Month State-Income-Vantage Vantage-Month

Note: This table contains estimates for the Oaxaca-Blinder decomposition of the unconditional diﬀerences

between online and storefront payday loan prices. We decompose them in diﬀerences in default rates θ

(λ

−

), diﬀerences in default losses λS(θ

− θ

), and diﬀerences in cost structures (π

− π

). In columns (1)

and (2) we use state-month bins in the random Clarity, and the credit visible sample, respectively. Columns

(3) and (4) allow for binnings using deciles of Vantage score, using the credit visible sample.

Figure A1: Prices and Default Rates by Months at Address

Panel A: APR

(a) Random Clarity sample (b) Credit visible sample

Panel B: Cost per $100

Panel C: Default

(e) Random Clarity sample (f) Credit visible sample

Note: The ﬁgure presents binscatter plots of APR, cost per $100 borrowed, and default rates in the credit

visible sample in graphs (b), (d), and (f), and in the standalone sample in graphs (a), (c), and (e). The

x-axis contains the number of months living at the same address for the loan applicant, at the application

date.

Figure A2: Prices and Default - Random Clarity Sample

(a) APR vs Default, Storefront (b) APR vs Default, Online

Note: The ﬁgure presents binscatter plots of default by bins of price relative to the bin average price in the

stand-alone sample. Prices are expressed as APR in panels (a) and (b) and cost per $100 borrower in panels

Table A1: Online Payday Loan Premium: Excluding 2013

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Random Clarity Sample (2014-2017)

Outcome: APR Cost / 100 Default

Storefront mean: 294 12.4 0.041

Online dummy 92.2 106.4 133.0 4.13 4.48 6.17 0.026 0.020 0.062

(37.0) (42.7) (56.0) (2.14) (2.47) (3.10) (0.026) (0.031) (0.036)

[0.016] [0.016] [0.021] [0.060] [0.076] [0.052] [0.320] [0.528] [0.091]

R2 0.538 0.627 0.779 0.613 0.698 0.825 0.099 0.129 0.398

N 312,298 309,186 286,945 312,298 309,186 286,945 312,298 309,186 286,945

Panel B: Credit Visible Sample (2014-2019)

Outcome: APR Cost / 100 Default

Storefront mean: 300 13.3 0.044

Online dummy 85.3 96.2 116.4 3.18 3.49 4.56 0.034 0.029 0.042

(25.4) (32.3) (53.5) (1.42) (1.68) (2.30) (0.020) (0.025) (0.027)

[0.001] [0.005] [0.035] [0.029] [0.044] [0.053] [0.094] [0.265] [0.131]

R2 0.533 0.608 0.746 0.556 0.635 0.764 0.096 0.140 0.287

N 178,953 177,197 163,898 178,953 177,197 163,898 178,953 177,197 163,898

Panel C: Credit Visible Sample with Vantage

Online dummy 85.4 96.2 116.5 3.19 3.50 4.57 0.034 0.029 0.042

(25.4) (32.3) (53.5) (1.42) (1.68) (2.30) (0.020) (0.025) (0.027)

[0.001] [0.005] [0.034] [0.029] [0.044] [0.053] [0.089] [0.259] [0.132]

R2 0.533 0.608 0.746 0.557 0.635 0.764 0.100 0.143 0.287

N 178,953 177,197 163,898 178,953 177,197 163,898 178,953 177,197 163,898

State FE Yes No No Yes No No Yes No No

Zip FE No Yes No No Yes No No Yes No

Consumer FE No No Yes No No Yes No No Yes

Note: The table presents coeﬃcient estimates of the online loan dummy from regressions of payday loan

prices and default probability for the random Clarity sample in Panel A and the credit visible sample in

Panels B and C, excluding loans made in 2013. All regressions include ﬁxed eﬀects for either state, ZIP code,

or customer; ﬁxed eﬀects for day of week, day of month, month of year, and calendar year; and controls

for deciles of loan duration, loan size, age, and income, categorical variables for housing status and pay

frequency, and number of inquiries per week. Panel C additionally includes controls for decile of Vantage

score. Robust standard errors clustered at the state level are in parentheses, and p-values are in brackets.

Table A2: Online Payday Loan Premium: Non-Imputed Sample

(1) (2) (3) (4) (5) (6)

Panel A: Random Clarity Sample (2013-2017)

Outcome: APR Cost / 100

Storefront mean: 298 12.4

Online dummy 98.7 111.0 119.2 3.45 3.80 4.80

(33.7) (40.4) (47.9) (1.88) (2.27) (2.94)

[0.005] [0.008] [0.016] [0.073] [0.101] [0.109]

R2 0.595 0.686 0.823 0.641 0.734 0.870

N 272,220 269,436 252,430 272,220 269,436 252,430

Panel B: Credit Visible Sample (2013-2019)

Outcome: APR Cost / 100

Storefront mean: 300 13.0

Online dummy 87.3 99.8 114.1 2.70 3.13 3.85

(23.7) (30.9) (57.8) (1.30) (1.70) (2.34)

[0.001] [0.002] [0.054] [0.043] [0.072] [0.107]

R2 0.593 0.671 0.801 0.602 0.692 0.837

N 149,458 148,104 138,710 149,458 148,104 138,710

Panel C: Credit Visible Sample with Vantage

Online dummy 87.3 99.9 114.1 2.70 3.13 3.85

(23.7) (30.9) (57.8) (1.30) (1.70) (2.34)

[0.001] [0.002] [0.054] [0.043] [0.072] [0.107]

R2 0.593 0.672 0.801 0.603 0.692 0.837

N 149,458 148,104 138,710 149,458 148,104 138,710

State FE Yes No No Yes No No

Zip FE No Yes No No Yes No

Consumer FE No No Yes No No Yes

Note: The table presents coeﬃcient estimates of the online loan dummy from regressions of payday loan

prices and default probability for the random Clarity sample in Panel A and the credit visible sample in

Panels B and C, excluding defaulted loans and those with missing information in the pricing formula in

equation (1). All regressions include ﬁxed eﬀects for either state, ZIP code, or customer; ﬁxed eﬀects for day

of week, day of month, month of year, and calendar year; and controls for deciles of loan duration, loan size,

age, and income, categorical variables for housing status and pay frequency, and number of inquiries per

week. Panel C additionally includes controls for decile of Vantage score. Robust standard errors clustered

at the state level are in parentheses, and p-values are in brackets.

Table A3: Online Payday Loan Premium - Bin Level

(1) (2) (3) (4) (5) (6)

Outcome: APR Cost / 100

Online Dummy 91.3 99.4 123.6 4.8 4.0 4.0

(1.4) (1.9) (1.6) (0.0) (0.1) (0.0)

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

Binning Quintiles Age-Income Age-Income Credit Score Age-Income Age-Income Credit Score

Sample Stand-Alone Credit Visible Credit Visible Stand-Alone Credit Visible Credit Visible

0.291 0.297 0.273 0.571 0.54 0.54

N 304,404 157,400 157,400 304,404 157,400 157,400

State FE Yes Yes Yes Yes Yes Yes

Year-Week FE Yes Yes Yes Yes Yes Yes

Bin FE Yes Yes Yes Yes Yes Yes

Note: The table presents coeﬃcient estimates of the online loan dummy from regressions of payday loan

prices for the random Clarity sample in Columns (1) and (4) and the credit visible sample in Columns (2)-(3)

and (5)-(6). The unit of observation is loan, and bins are deﬁned by quintiles of age and income in columns

(1)-(2) and (4)-(5), and credit score in columns (3) and (6). All regressions include ﬁxed eﬀects for bin and

week ﬁxed eﬀects and controls for averages of loan duration, loan size, age, income, number of inquiries per

week, and lagged default rates and lagged percentage of late payments in the bin. Robust standard errors

clustered at the market level are in parentheses, and p-values are in brackets.

Table A4: Statewide Payday Loan Database - Time Series

(1) (2) (3) (4) (5) (6) (7) (8)

Panel A: Random Clarity Sample (2013-2017) Panel B: Credit Visible Sample (2013-2019)

Outcome: APR Cost / 100 APR Cost / 100

Online × Database 6 or less years ago - 66.4 - 115.3 - 7.7 - 8.6 - 127.8 - 167.3 - 5.9 - 6.7

(105.3) (112.0) (4.8) (5.1) (140.6) (143.8) (4.9) (5.3)

[0.529] [0.304] [0.106] [0.095] [0.364] [0.245] [0.225] [0.207]

Online × Database 7 - 8 years ago - 89.7 - 67.6 - 3.8 - 3.8 - 59.2 - 45.7 - 1.0 - 0.9

(48.4) (38.7) (1.9) (1.9) (52.9) (37.6) (2.2) (1.8)

[0.065] [0.082] [0.045] [0.053] [0.264] [0.225] [0.654] [0.614]

Online × Database 9 - 10 years ago - 161.2 - 179.1 - 6.9 - 7.7 - 123.3 - 134.5 - 6.3 - 5.9

(37.3) (28.3) (2.5) (2.3) (30.6) (27.7) (1.5) (1.3)

[0.000] [0.000] [0.006] [0.001] [0.000] [0.000] [0.000] [0.000]

Online × Database 11 - 12 years ago - 150.0 - 118.3 - 6.8 - 5.6 - 141.1 - 111.0 - 5.9 - 4.8

(71.0) (55.0) (1.4) (1.3) (72.1) (60.3) (1.6) (1.4)

[0.036] [0.033] [0.000] [0.000] [0.051] [0.066] [0.000] [0.001]

Online × Database 13 or more years ago - 148.3 - 136.9 - 8.7 - 6.7 - 112.0 - 95.1 - 8.2 - 6.0

(67.1) (59.7) (1.5) (1.5) (48.2) (46.5) (1.5) (1.5)

[0.028] [0.023] [0.000] [0.000] [0.021] [0.042] [0.000] [0.000]

Online 149.9 129.1 7.9 7.4 146.0 118.2 6.7 5.7

(21.2) (20.8) (1.1) (1.1) (21.8) (19.3) (1.3) (1.1)

[0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000] [0.000]

Database 6 or less years ago 242.5 228.1 12.5 11.4 289.4 267.8 11.0 9.7

(119.0) (127.6) (5.6) (5.8) (158.9) (159.5) (5.6) (6.0)

[0.043] [0.075] [0.027] [0.052] [0.069] [0.094] [0.048] [0.107]

Database 7 - 8 years ago 45.6 27.0 2.7 2.0 51.4 39.9 1.5 0.7

(20.9) (22.2) (1.0) (1.1) (21.4) (21.2) (1.3) (0.9)

[0.030] [0.226] [0.006] [0.074] [0.017] [0.061] [0.228] [0.453]

Database 9 - 10 years ago 85.5 65.5 4.1 3.8 53.0 25.7 3.3 1.9

(29.3) (29.5) (2.6) (2.4) (23.0) (21.7) (1.3) (1.2)

[0.004] [0.027] [0.112] [0.122] [0.022] [0.238] [0.014] [0.103]

Database 11 - 12 years ago 74.9 62.8 4.2 3.0 55.8 40.2 2.7 1.7

(61.8) (48.2) (1.0) (0.9) (64.8) (54.4) (1.4) (1.1)

[0.226] [0.193] [0.000] [0.002] [0.390] [0.460] [0.052] [0.109]

Database 13 or more years ago 32.5 59.9 4.4 3.1 - 2.5 7.7 3.7 1.7

(48.0) (39.3) (1.1) (0.7) (38.1) (36.4) (1.3) (1.0)

[0.499] [0.129] [0.000] [0.000] [0.948] [0.832] [0.005] [0.090]

Controls No Yes No Yes No Yes No Yes

FE No Yes No Yes No Yes No Yes

R2 0.111 0.465 0.330 0.401 0.111 0.454 0.256 0.346

N 336,690 336,690 336,690 336,690 188,913 188,913 188,913 188,913

Note: The table presents coeﬃcient estimates of loan-level regressions of prices, in APR and cost per $100 borrowed, on a dummy variable that

indicates whether the loan is an online loan, and dummy variables for each bin of time since implementation of a statewide payday loan database,

and their interaction. Columns (2), (4), (6), and (8) contain loan and customer level controls, and time ﬁxed eﬀects. The samples used are in Panel

A and the random Clarity sample, in Panel B the credit visible sample. Robust standard errors clustered at the state-year level are in parentheses,

and p-values are in brackets.