F INANCE R ESEARCH S EMINAR
S UPPORTED BY U NIGESTION
Half Banked: The Real Effects of Financial
Exclusion on Firms
Elizabeth BERGER,
Cornell University
Friday, December 13, 2019, 10:30-12:00
Room 126, Extranef building at the University of Lausanne
Half Banked: The Real Effects of Financial Exclusion on
Firms
Elizabeth Berger∗
Cornell University
Nathan Seegert†
University of Utah
Friday 22nd November, 2019
Preliminary and Incomplete: Comments Welcome.
Abstract
This paper measures the economic effects of the warehouse function of bank
access on firms. Although the economics of access to the credit side of banking is well
understood, we know considerably less about how banking creates firm value by
storing wealth and providing payment services. This is due to the fact that almost all
firms have access to the warehouse function of banks. To address this empirical
challenge, we focus on firms in the marijuana industry that face hurdles to obtaining
financial services. Our results show that warehouse banking improves the business
strategies of firms. Because banking services allow firms to make different business
decisions, these firms are more profitable.
JEL: D22; D24; G21; L25; L26; O16; O43
Keywords: Banking; Financial Inclusion; Firm Outcomes; Small Business; Marijuana
Industry
∗
†
elizabeth.a.berger@cornell.edu
nathan.seegert@business.utah.edu
1
Introduction
Banks have served many roles over time. Their initial function was to store assets, i.e.
collateral and cash, for the wealthy. This warehousing function of banks led to the lending
function of banks (Donaldson, Piacentino, and Thakor (2018)). A broad literature shows
that the lending function of banks creates economic value by relieving financial constraints
(Adelino, Schoar, and Severino (2015), Beck, Demirguc-Kunt, and Martinez Peria (2006)).
In addition, banks play the role of resolving information asymmetry and monitoring
borrowers.1 Although we understand well the economic value of the lending and monitoring
functions of banks, we know little of the economic value of the warehousing function.
This paper measures the economic value of cash management (the warehousing function
of banks). Cash management should be economically important. It takes custody of assets,
it manages the payment obligations of firms, and it is an overall key component of a firm’s
financial stability. Moreover cash management supports the lending function of banks
(Donaldson et al. (2018)). This question is important due to the fact that technology is
quickly altering the banking sector. Many functions of banks, especially those tied to brick
and mortar, are becoming obsolete. Understanding the economic value of the cash
management function may alter the perceived costs of losing bank branches.
Although the question is straightforward, measurement is not. We know little of the
economic value of access to cash management, primarily because all firms (or almost all
firms) have access to cash management. Those that do not are outliers. This leads to a
setting in which there is no counterfactual. The key to answering this question is to
separate firms’ access to the lending function from access to the warehousing function.
We tackle this challenge using the marijuana industry as an experimental setting.
Firms in this industry face regulatory hurdles to obtaining financial services, which we
1
Related results in the literature include Berger, Miller, Petersen, Rajan, and Stein (2005); Brickley,
Linck, and Smith Jr (2003); Butler and Cornaggia (2011); Cetorelli and Strahan (2006); Gilje, Loutskina,
and Strahan (2016); Gilje (2017); Paravisini (2008); Petersen and Rajan (1994); Petersen and Rajan (2002);
Rice and Strahan (2010).
1
exploit as our identification strategy. We measure how the warehousing function of banks
affects firms. We measure interactions with both retail customers and supply chain
processors. Our findings show that firms with access to warehouse banking earn higher
profits, sell higher quantity, charge higher prices, and sell products with higher potency.
Marijuana legalization throughout the US provides variation in firms’ access to finance.
Marijuana is a federally prohibited substance and FDIC insured banks are subject to
federal laws. The federal government considers it money laundering to handle any money
from the marijuana industry. Efforts to clarify the laws pertaining to the interaction
between banks and the marijuana industry have only made existing laws more ambiguous.
Hence, FDIC insured banks have primarily chosen to avoid the marijuana industry.
However, the marijuana industry is economically large and growing rapidly. Despite its
size, business has been conducted almost entirely in cash. The lack of banking services to
help with the most basic tasks, such as cash management, is a growing problem.
In Washington state, four independent credit unions (which we refer to as 502 credit
unions) elected to provide cash management services to marijuana firms. To reduce
information asymmetry, these banks require site visits to ensure that cash from marijuana
firms is generated through legal activity (according to state law). Quarterly site visits help
banks verify soft information but so does knowledge about the firm that comes with
physical proximity. Distance has traditionally played a role in forming bank relationships
by enabling banks to gather soft information about firms.2 Hence, distance is a crucial
element of firm access to 502 credit unions. The combination of soft information needs,
required on-site visits, and the need for physical proximity mean that not all firms have
access to these financial institutions.
We use variation in firm distance from 502 credit unions as a proxy for financial access.
In our sample, we find that distance matters for financial access. The probability of using
2
Related results in the literature include Berger et al. (2005); Brickley et al. (2003); Butler and Cornaggia
(2011); Cetorelli and Strahan (2006); Gilje et al. (2016); Gilje (2017); Paravisini (2008); Petersen and Rajan
(1994); Petersen and Rajan (2002); Rice and Strahan (2010).
2
financial services is 23 and 21 percentage points higher if the dispensary is within 2 or 3
miles of a 502 credit union, respectively.
We build our dataset by combining two novel data sources (survey and transaction-level
data). Survey data of retail cannabis firms come from the Banking, Entrepreneurship,
Regulation, and Taxes (BERT) Study. These data include a module on access to finance.
Data from this survey reports that only 49% of dispensaries have access to a business bank
account, 30% indicate they do not use any financial services, and 18% use personal
accounts.
Transaction level data come from the state of Washington and are generated by the
state from filings based on reporting requirements for firms. Specifically, firms must report
details of all marijuana transactions at each establishment and provide addresses of each
establishment.
The results of the instrumental variables (IV) analysis show that firms with 502 credit
union access run businesses differently than firms without. Financial services increase firm
prices by 2% and reduce markups by 6 percentage points. Wholesale and retail prices are
higher. After controlling for fixed effects based on the strain of marijuana, the combination
of these results suggest that financial services increase prices. Further analysis of strains
potency shows that variation in the quality of product explain these differences. Firms
with access to banks sell higher quality products.
One potential concern with the IV design, however, is that distance to a 502 credit
union could be conflating other effects of proximity to geographic areas where financial
institutions locate. The conflating factors could bias the estimates and make them less
precise. If 502 credit unions are located in areas that are different from other areas in ways
that impact business decisions and outcomes, then the estimates from the IV specification
could be imprecise or biased.
To address these concerns, we use additional analysis on firms close to other credit
unions to control for these conflating factors in a panel data fixed effects model. This panel
3
data fixed effects model extends the IV specification by using additional variation in type
of credit union to control for other potential unobserved differences. The PDFE results
support the IV results. Firms within 2 miles of a 502 credit union have higher profits,
lower markups, and higher sales.
We explore the drivers of these results. The higher sales, lower markups, and higher
wholesale prices are consistent with two mechanisms. First, the cost savings from
dispensaries using banking services may partially pass-through upstream to suppliers,
leading to higher wholesale prices. Second, banking services may improve transactions with
suppliers, thereby improving their relationship, and may allow dispensaries that use
banking services to acquire higher quality products from their suppliers.
To disentangle these mechanisms, we analyze transaction level data with strain fixed
effects. These data include over 62 million observations of transactions between dispensaries
and retail consumers. Rather than aggregating transactions to the establishment level, this
analysis allows us to compare transactions for the same strain of marijuana. The evidence
suggests that some of the cost savings from banking services are passed upstream to
processors. Sales are on average 16 percent higher for firms that use banking services per
transaction. Overall, banking services allow firms to make different business decisions that
lead to an increase in profits of 23 to 25 percent higher per transaction. This evidence
suggests that firms that use banking services have different types of consumers—those that
consume more. These consumers may be less likely to be tourists or casual consumers.
The results in this paper add to the literature on that studies the economic impacts of
financial inclusion (Brown, Cookson, and Heimer (2017)). Our results shed light on the
impact of financial inclusion for firms. This area is understudied due to a lack of data. We
use a novel setting in which to identify how being ”unbanked” affects the way that firms
are managed.
In addition, our paper estimates the economic value of warehouse banking for firms
(Donaldson et al. (2018)). Using the fact that marijuana firms have access to only the cash
4
management function of banking, and not lending services, we are able to isolate the
economic value of the warehouse function. Our results reveal an economically meaningful
value of these services for firms.
Finally, our results contribute to the literature that explores the value of financial
integration for economic growth (Jayaratne and Strahan (1996)). Our results reveal that
adding financial services to an economy permits firms to improve the efficiency of
operations and the quality of supply chain relationships.
2
Background: Empirical Setting
2.1
Legal Status of Marijuana Businesses
Marijuana production and consumption is federally illegal in the United States.
However, in 2012 individual states started legalizing recreational marijuana for adults 21
years of age or older. As of 2019, 11 states have legalized recreational marijuana—each
with their own idiosyncratic rules. Following state-level marijuana legalization, the rules
about whether banks could provide services to marijuana firms were ambiguous. Unlike
marijuana, which is subject to state law, FDIC-insured financial institutions are subject to
federal law FIN-2014-G001 (2014).
Despite the fact that marijuana is legal at the state level, marijuana production and
consumption is classified as a Schedule I substance, making it federally illegal in the United
States. The Comprehensive Drug Abuse Prevention and Control Act of 1970 includes a
schedule of five tiers of controlled substances based on characteristics such as acceptable
medical use, potential for abuse, and general safety. Schedule I substances are defined as
drugs with no currently accepted medical use and a high potential for abuse; examples
include heroin, LSD and marijuana (Drug Enforcement Administration).3
Despite the clear federal regulations, states have legalized marijuana either for medical
3
https://www.dea.gov/druginfo/ds.shtml
5
or recreational purposes. Between 1996 and 2018, 33 states elected to permit production
and use of marijuana for medical purposes.4 Moreover, starting in 2012 with Colorado and
Washington, states began to legalize recreational marijuana for adults 21 years of age or
older.
Reconciling state and federal marijuana laws has been a challenge for regulators. The
Ogden Memorandum, announced on October 19, 2009 by Deputy United States Attorney
David W. Ogden, states that medical marijuana firms in compliance with state medical
marijuana laws would not be an enforcement priority for the Department of Justice. While
most medical marijuana businesses have not been raided or prosecuted following this
memorandum there have been several instances in Montana, California, Nevada, Colorado,
and Michigan where business property has been seized.
Financial institutions required specific direction from federal regulators FIN-2014-G001
(2014). On August 29, 2013, the FDIC issued guidance about how financial institutions
should interact with marijuana firms (Cole, 2013). Unfortunately, this guidance included
the provision that all rules could be revoked at any time. This guidance resulted in
confusion and many unanswered questions (Cole, 2013).
Meanwhile, legal marijuana was growing into a large industry. Firms grew into large
businesses in terms of size and sales volumes. In 2018, the marijuana industry generated
$367.4 million in annual revenues to the state of Washington. Despite its size, firms in this
industry conducted business entirely in cash.
Firm managers grew concerned about the safety and efficiency of an all cash business
(Weed, 2018). These concerns extended beyond individuals in the industry. Local
community leaders were increasingly uncomfortable with the amount of cash circulating
through the community. Overall, cash management became a greater concern as the
marijuana industry grew (Weed, 2018). In the state of Washington, four independent
credit unions, which we refer to as 502 credit unions, elected to provide cash management
4
Washing D.C. and Puerto Rico also permit medical use of marijuana.
6
services to marijuana firms.5 These credit unions were primarily responding to perceived
danger in communities as businesses accumulated more and more cash. Their interest was
to provide bank accounts as a safe storage facility for this cash.
2.2
Banking Literature and Practical Implications
Theoretically, financial access should be important for small businesses in general and
marijuana firms in particular. If holding cash on hand creates frictions, then financial
institutions provide an important function as storage facilities or “warehouse banks”
(Donaldson et al. (2018)). Because marijuana firms cannot borrow from financial
institutions, the value of financial institutions comes primarily from cash management
services. In addition, the household finance literature shows that mere exposure to
financial institutions increases financial inclusion, and thereby improves financial health
(Brown et al. (2017)).
Distance and bank size and scope have traditionally played an important role in
determining banking relationships. Figure 1, Panel (A) graphs the relationship between
firms with financial access and distance to a credit union. The majority of firms are located
within 5 miles of a credit union. Banks gather information about potential borrowers. The
information can be either hard information or soft information. To reduce information
asymmetry in banking relationships, banks gather soft information and form relationships
with borrowers that are physically near the bank (Petersen and Rajan (1994) and Petersen
and Rajan (2002)). Small, local banks rely more on soft information and relationship
capital about the borrower to assess a borrower’s creditworthiness (Williamson, 1967).
Banks in close physical proximity to borrowers have a comparative advantage in gathering
soft information.6
5
The accounts were called 502 bank accounts because retail marijuana was legalized in the Washington
Initiative 502 legislation.
6
Related results in the literature include Berger et al. (2005); Brickley et al. (2003); Butler and Cornaggia
(2011); Cetorelli and Strahan (2006); Gilje et al. (2016); Gilje (2017); Paravisini (2008); Petersen and Rajan
(1994); Petersen and Rajan (2002); Rice and Strahan (2010).
7
Information frictions potentially link local 502 credit unions and marijuana firms. Even
for simple bank accounts, financial institutions that work with marijuana firms in the
marijuana industry must provide verification that the deposits from marijuana firms
originate from legal activity. Hence, marijuana firms are more soft information intensive.
For example, 502 credit unions require quarterly site visits to verify that firms are
operating legally. Hence, physical distance between financial institutions and marijuana
firms should be important for establishing financial relationships.
Because soft information about the legality of business activities is important in the
marijuana industry, firms should rely more on local financial institutions and physical
distance should be a primary determinant of banking relationships. Hence local 502 credit
unions may better serve marijuana firms, which have higher information asymmetry
(Adelino et al. (2015)).
In this paper, we use these characteristics of the banking sector to design our empirical
tests that measure how financial access influences firms.
3
Data
We pair two novel data sets to provide evidence on how banking influences firms in the
cannabis industry. First, we use administrative records from the Washington State Liquor
and Cannabis Board (WSLCB). The administrative data contains the universe of
businesses and includes 62 million retail transactions from 2014 through 2017. Second,
during the period July 10 through October 31, 2016, we conducted a survey in three states,
Colorado, Oregon, and Washington. The survey provides detailed information on 462
individual marijuana businesses.
8
3.1
Washington State Liquor and Cannabis Board
Most states that have legalized recreational marijuana production and consumption
legislate extensive data reporting requirements.7 Firms are required to provide data on all
plants from planting, harvesting, production, and final retail sale. In terms of reporting,
BioTrackTHC is one company that provides technology that tracks cannabis from
“Seed-to-Sale,” with contracts in Delaware, New Mexico, Illinois, New York, Hawaii, and
for our purposes notably Washington. In terms of enforcement, the state of Washington
uses random audits with penalties and inventory destruction to ensure compliance with the
tracking system.
Processors record characteristics of each plant, including weight and test results for the
primary psychoactive ingredients tetrahydrocannabinol (THC), tetrahydrocannabinolic
acid (THC-A), and cannabidiol (CBD). They then seal the product into packages of
specific weights (e.g., 1 gram) and sell a retail lot to dispensaries, where each lot is
considered identical and given its own retail lot identification number.
At the retail level, firms record the strain, quantity (weight), retail price, and wholesale
price for each transaction. Because data are tracked at the transaction level, these data
allow us to construct transaction level variables. Specifically, we construct a profit by
transaction variable: Profit = (retail price − wholesale price) × quantity, and a markup by
transaction variable: Markup = (retail price − wholesale price)/retail price. Because most
firms operate in cash due to lack of financial services, most retail firms choose prices such
that the tax-inclusive price is a round number (Hansen et al., 2017). These pricing
decisions lower the transaction costs of using cash.
The WSLCB administrative data provides a unique look at business operations because
it records retail and wholesale prices, as well as sales volume. The average retail price is
$15 per gram and the median transaction is 2 grams. The average markup across
7
States require a large amount of disclosure due, in part, to a response to the Cole Memo (Cole 2013).
For more information on disclosure see Hansen, Miller, and Weber (2017).
9
dispensaries is 134%. Note, these markups do not include labor costs, rent, and other fixed
costs. The 25th and 75th percentile markups are 92.9% and 146.5%, respectively. Panel c
of Figure 1 shows the full distribution of markups. Table 3 summarizes the data for the
interactions between dispensaries and retail consumers.
The data also include information about the strains of marijuana—which can be
thought of as wine varieties such as a Giuseppe Rinaldi 2010 Brunate. The median
dispensary sold 422 different strains, and the 25th and 75th percentile dispensaries sold 257
and 673 strains, respectively.
The data include the latitude and longitude of every establishment. We pair this
geographic information with credit union data from the National Credit Union
Administration (NCUA) to calculate the distance to the nearest credit union and 502
credit union. We also calculate the number of other marijuana establishments
(competitors) within 1, 2, and 5 miles of each marijuana establishment.
3.1.1
Where are Dispensaries located?
The WSLCB administrative data include records on 434 unique medical and
recreational retail locations. These retail establishments are in 152 cities or towns within
Washington. Seattle, Spokane, and Tacoma are the three largest cities in Washington and
they have the most retail dispensaries. The percent of dispensaries in these cities is roughly
similar to their percentage of the population of Washington. For example, Seattle accounts
for 9% of Washington’s population and has 13% of the dispensaries in the state.
Dispensaries are typically in commercial areas and as a result close to financial
institutions. The average distance between a dispensary and a bank is 1.1 miles and the
maximum distance is 11.9 miles. The average distance between a dispensary and a credit
union is slightly further, 2.4 miles. The maximum distance between a dispensary and a
credit union is 46 miles, though, 95% are within 12.5 miles. Figure 1 shows that the
distribution of distance is highly skewed, which is why we use log distance in our empirical
10
analysis.
3.2
Banking, Entrepreneurship, Regulation, and Taxes (BERT)
Study
The second dataset used in this study is a survey that was part of the Banking,
Entrepreneurship, Regulation, and Taxes (BERT) Study. As part of the BERT study, we
included a survey module specifically about access to financial services. Building on
existing surveys of individuals and small businesses, we developed an initial set of questions
to include in the survey (Fairlie, 2014; US Bureau of Labor Statistics, Retrieved 2018;
Kauffman Firm Baseline Public Use Survey Data (KFS)). The questions cover a range of
topics from personal details about the business owner, such as education and occupation of
parents, and technical details of the business, such as whether the business uses banking
services and what type of banking institution it works with. We incorporated feedback
from academics and marketing experts into the survey over several iterations. We then
piloted the survey through in-person interviews with small businesses in different industries
to ensure participants and researchers had similar interpretations of the questions and to
minimize biases induced by the survey. Participants took an average of 40 minutes to
complete the survey. Participation entailed visiting a website and inputting answers to
survey questions. Participants were compensated $50 and entered to win a $500 reward,
which was given away randomly to two participants. To determine if participant fatigue
affected the quality of answers, we implemented two versions of the survey with different
ordering of the nondemographic sections. We find no evidence that the quality of response
rate differs depending on the ordering of the questions.
We contacted all retail cannabis dispensaries, medical and recreational, in Colorado,
Washington, and Oregon through several waves of contact.8 First, we mailed letters that
contained information and instructions on how to take the survey with an enclosed $2 bill
8
The BERT study information can be found at eccles.link/bert.
11
as a free gift to increase participation. Second, we called all businesses. If their given phone
number did not work we used internet searches to find updated numbers. Third, we
partnered with several industry groups that emailed their members information about the
survey. Fourth, we sent another wave of letters. Fifth, we had a research assistant
physically visit businesses for two months. Finally, we did an additional round of phone
calls.
After all waves of the survey, the response rate was 21% and our sample contains 325
firms. This response rate is similar to recent surveys of businesses. For example, Graham
and Harvey (2001) obtained 16% and Trahan and Gitman (1995) 12%—though these
surveys focused on CFOs and business executives of large firms.
Our goal was to have owners or managers fill out the survey. One of the first questions
asks the participant whether they are an owner, manager, both, or neither. In our sample
we have 65% of respondents were either an owner or owner and manager and 31% of
respondents were managers and the remaining eight responses were filled out by someone
else such as an accountant or office manager.
To supplement the survey evidence, we followed up with businesses in Washington that
had not replied to answer a short six question survey about their access to banking
services. These questions included several demographic questions and the banking module
from the full survey. In Washington, we identified 325 dispensaries. Of these, 295 answered
at least the banking module, a 90% response rate. We use these data in the first-stage
regression of our instrumental variable analysis.
3.2.1
Banking Services in the Marijuana Industry
Guidance for financial institutions seeking to provide services to marijuana-related
businesses came from the Financial Crimes Enforcement Network (“FinCEN”) and the
Cole Memo (2013) to clarify Bank Secrecy Act (“BSA”) expectations. The guidelines
suggested banks should know their clients to ensure they were not engaged in illegal
12
activities outlined in the Cole memo (Cole, 2013). The Drug Enforcement Administration
has formally warned banks that marijuana remains illegal at the federal level and Colorado
Bankers’ association formally urged its members to avoid serving marijuana businesses. As
a result of these clarifications and warnings, very few financial institutions are willing to
take the risk of potential money laundering from marijuana firms. Many financial
institutions will close personal and business bank accounts if they learn that the accounts
are linked to marijuana in any way.
Most dispensaries have struggled to obtain financial services at some point due to these
ambiguities. Survey data reported in Table 1 reveal that only 49% of dispensaries have
access to a business bank account, 30% indicate they do not use any financial services, and
18% use personal accounts. Our pilot interviews bolster these statistics. Some participants
told us that they used personal bank accounts but that banks frequently closed these
accounts if they discovered that the account was related to a marijuana business. National
banks are especially likely to close accounts. One pilot participant discussed a strategy to
conceal the fact that money was coming from the marijuana industry. The business owner
sprayed Febreze (a scented laundry spray) on the money before depositing it at the bank to
conceal the scent of marijuana.
Most financial institutions that provide services to the marijuana industry do so
discretely, mostly through word of mouth, and with heightened oversight. In response to
federal guidance, financial institutions working with the marijuana industry require
businesses to disclose financial records, personal information of the owners including tax
records, and in some cases regular, on-site inspections of the businesses.
This data collection is costly for financial institutions and limits financial access for the
marijuana industry. At one credit union in Colorado, five of the 13 employees solely focus
on compliance in the marijuana industry (Mandelbaum, 2018). As a result, financial
institutions that work with dispensaries limit the number of dispensaries that they work
with and often have long waiting lists of marijuana dispensaries that need bank accounts.
13
Because of the nature of the oversight required between financial institution and
marijuana firm, the type of financial institution (i.e., local bank, national bank, and credit
union) matters. Table 1 reports that 69% of those dispensaries with financial services
obtained them through a community or regional credit union. In comparison, only 11% of
dispensaries report working with a national bank.
3.2.2
The Effects of Lack of Financial Services
One effect of limited financial services is that the marijuana industry is mostly a cash
business. Table 2 shows that in our survey, 49% of dispensaries report that they only
accept cash, 40% also accept checks, and only 3% accept credit cards. Even dispensaries
that have bank accounts with a financial institution may not be able to process credit card
transactions because of increased regulations.
The frictions associated with being a cash industry creates additional costs from
security, theft, and fees from financial institutions. Table 2 reports that in our survey, 30%
of dispensaries use a security company to transport their cash or product at a cost of $348
a month. Cash is also more vulnerable to theft. Dispensaries reported an average loss from
theft of $618 per month. Even normal business activities, such as payroll, are more costly
when transacted in cash. For example, many dispensaries have to pay employees in cash
rather than writing checks which puts a burden of risk on employees. Due to these costs,
dispensaries are willing to pay large fees to financial institutions to hold their cash. In our
survey, dispensaries report paying fees between $437 and $1,059 for access to financial
services.
4
Instrumental Variable Evidence
Identifying the effects of banking services on business outcomes and decisions is
challenging. In most industries there is very little variation across firms in the decision and
14
ability to use banking services.9 Moreover, the few businesses that do not use banking
services are likely very different and not comparable to businesses that do resulting in
selection issues in the decision to use banking services. Finally, a host of potential omitted
variables could bias the estimates.
In the marijuana industry, the choice not to use banking services is in large part driven
by search and transaction costs. Specifically, firms do not know which financial institutions
will provide these services, there may not be any financial institutions that provide these
services nearby, and there may be large costs in working with a financial institution far
away due to the information frictions associated with the mostly cash based nature of the
industry. These institutional details suggest that the decision to use banking services is
plausibly exogenous to other observable factors (e.g., firm business decisions).
In light of the industry-specific difficulties in accessing bank finance, the marijuana
industry provides a unique opportunity to study the impact of banking services on firm
outcomes and decisions. We specifically are interested in how banking services impact a
firm’s (1) markups, (2) retail price, (3) wholesale price, (4) profits, and (5) sales
volume—our dependent variables Yi,t . To investigate these relationships, we consider the
specification
Yi,t = β0 + β1 Banking Servicesi,t + βXi,t + αt + αj + εi,t
(1)
where we use month and location fixed effects and other controls to isolate the effect of
banking services on firm decisions. We control for the number of competing marijuana
firms within 5 miles in the regression specification. Standard errors are clustered at the
dispensary level. Under the null hypothesis that using banking services does not impact a
firm’s choices and outcomes, β1 should be small in magnitude and not statistically different
from zero.
If the decision to use banking services is related to unobserved differences between
firms, then β1 may be biased. Therefore, we instrument for using banking services with the
9
In fact, almost all businesses use banking services, though some through personal accounts ().
15
nearest distance between a dispensary and a financial institution that provides banking
services to the marijuana industry. Due to the search and transaction costs of financial
intermediation, this distance strongly affects a firm’s decision on whether to use banking
services (section 3.2.1 discusses this in more detail). In addition, due to several
institutional details, the distance between a firm and a financial institution that provides
these services is plausibly exogenous (section 4.2 discusses this in more detail).
In Washington, four independent credit unions decided to provide banking services to
the marijuana industry. We refer to these as 502 credit unions because they provide
banking under the code 502 that legalized marijuana in Washington. Other financial
institutions did not provide these services either due to risk or legal constraints, as the case
with federally chartered banks. We use the distance to the closest branch of one of these
four 502 credit unions as our instrument in the specification below,
Yi,t = β0 + β1 Distance to 502 Credit Unioni,t + βXi,t + αt + αj + εi,t .
(2)
Our baseline specifications limit the sample of firms to those within 10 miles of a 502 credit
union. This specification, therefore, uses variation in distance between 0 and 10 miles.
Using this cutoff provides a cleaner comparison group. The estimates are not sensitive to
this cutoff. The coefficient β1 captures the causal effect of banking services on business
decisions and outcomes, Yi,t , if the conditions for a valid instrumental variable are met,
which we discuss in more detail in the next two sections. We use month and location fixed
effects and other controls to isolate the effect of banking services on firm decisions. We
control for the number of competing marijuana firms within 5 miles in the regression
specification. Standard errors are clustered at the dispensary level.
16
4.1
Distance and Banking Services
For the instrument (distance to 502 credit union) to be valid, it must strongly affect a
firm’s decision to use banking services. This instrument builds on a large literature on
banking services that suggest that physical distance is an important predictor of access to
finance (Petersen and Rajan, 1994). The literature suggests that relationships in banking
are important and that firms close to a financial institution may have the opportunity to
build a better relationship (Berger et al., 2005).
Evidence from our survey suggests that distance, in the form of search and transaction
costs, is an important determinant of financial access. Our survey confirms that the
marijuana industry is a largely cash based industry (Table 2). Larger distances between
the firm and the financial institution may increase transaction costs because firms have to
physically transport cash. Moreover, because very few financial institutions provide
financial services to the marijuana industry, firms may not know which institutions would
provide service for them and may rely on nearby banks due to visibility.
In addition, the relationship between firm and financial institution may be more
important in the marijuana industry due to legal ambiguities. Using the survey, we run a
first-stage regression and find that distance to a 502 credit union predicts a firm’s decision
to use banking services. Specifically, 502 credit unions require quarterly site visits to all
client locations to determine the nature of the proceeds from the business.
We use the instrumental variable of distance to a 502 credit union to proxy for access to
a bank account. The first-stage results, presented in Table 4, demonstrate the effect of a
firm’s distance to a 502 credit union on its decision to use banking services. The dependent
variable is equal to one if a firm indicates it uses banking services (not necessarily with a
credit union) and zero otherwise. Column 1 of Table 4 shows that the probability a
dispensary uses financial services decreases by 9 percentage points for every additional mile
to the nearest 502 credit union. Columns 2 shows that the probability of using financial
17
services is 29 percentage points higher if a dispensary is within 1 mile of a 502 credit union.
Columns 3 and 4 show that the effect decreases with distance, as we would expect.
Specifically, the probability of using a financial service is 23 and 21 percentage points
higher if a dispensary is within 2 or 3 miles of a 502 credit union. The first-stage results
also suggest the instrument is strong. The F-statistic equals 22 and exceeds the standard
threshold of F=10 (Bound, Jaeger, and Baker, 1995; Staiger and Stock, 1997).
Overall, the first-stage results, evidence from the literature, and our survey evidence
indicate that distance to a financial institution has a strong effect on a firm’s decision to
use banking services.
4.2
The Exclusion Restriction Condition
A valid instrument should affect business outcomes only through the decision to use
banking services, i.e., the exclusion restriction, and not through any other channel.
Formally, the exclusion restriction states that the instrument is uncorrelated with the
residuals in equation 2.
The distance instrument abstracts from these selection issues, is unlikely to directly
impact profits, and increases the likelihood a firm uses financial services. The proximity to
502 credit union variable is bounded by 0 and 1 and equal to (10 - distance)/10. This
variable equals 1 for firms within 1 mile and 0 for firms 10 miles away. Firms more than 10
miles away from a 502 credit union are excluded from the sample.
The weakness of the simple IV specification is the additional noise in the estimates
because firm profits depend on many factors. Through a series of additional tests we add
precision to this estimate by controlling for these other factors.
We provide several pieces of collaborating evidence to alleviate concerns about the
exclusion restriction:
1. Timing of location choice: One concern is that firms with more financially savvy
managers choose to locate near financial institutions that provide banking services to
18
the marijuana industry. Financially savvy managers may make other decisions that
affect business outcomes we are interested in, such as profits, markups, prices, and
sales. To investigate whether this concern is problematic, we consider the timing of
when firms chose their location and when financial institutions began offering
banking services to the marijuana industry. Recreational marijuana was legalized in
Washington when voters passed Initiative 502 in November of 2012. At that time a
rush of firms applied for licenses. Washington regulations initially required that firms
have at most three locations and no one person could hold a stake in more than one
firm. As the need for banking services became apparent, two credit unions offered
banking services to the marijuana industry. Other banks decided not to provide these
services or, as the case with federally charted banks, were not permitted to do so.
These institutional details suggest that firms could not have chosen their location
based on the location of these financial institutions because that information was not
available when they made their location decisions.
2. Balance of observable characteristics: Another concern is that a firm that locates
near a financial institution that provides banking services to the marijuana industry
differs from a firm that does not. To investigate this concern, we test for the balance
in observable characteristics between these sets of firms. Table 3 reports that there
are no significant differences between these sets of firms across observable
characteristics such as, product potency, sales, and retail price.
4.3
Instrumental Variable Empirical Evidence
We begin the analysis of the impact of banking services on firms by studying
dispensary-level data. We aggregate transaction-level data to the dispensary level such that
the analysis examines how marijuana firms operate, on average.
Table 5 reports evidence at the dispensary level on the impact of banking services.
Column (1) considers how banking services impact the markup percentage firms set. The
19
administrative data from the state of Washington provides exact retail and wholesale prices
for every transaction. This novel data allows us to calculate (as opposed to estimate)
markups as the ratio of the difference in retail and wholesale price divided by the retail
price; markup = (pretail − pwholesale )/pretail for each transaction.
Column (1) reports that firms closer to 502 credit unions have markups that are 6
percentage points lower. This evidence is consistent with several conjectures. First,
banking services allow firms to be more competitive and set lower prices to increase sales.
Second, firms that use banking services may have better relationships with their suppliers
and customers that allow them to sell higher quality products that have lower markups.
Third, lower costs from using banking services could be partially passed through to their
suppliers causing wholesale prices to be higher and as a result markups to be lower. Fourth,
there could be confounding factors with distance such that firms in areas where 502 credit
unions all charge lower markups. We examine each of these conjectures. Specifications in
Tables 10 and 11 alleviate concerns about the fourth conjecture. To separate the first three
conjectures, we consider retail price, wholesale price, and sales volume separately.
Columns (2) and (3) of Table 5 reports that firms close to 502 credit unions have retail
prices that are on average 1.8 percent higher and wholesale prices that are 4.2 percent
higher. Higher retail prices suggest that firms are not passing on cost savings through
banking services onto customers. Higher wholesale prices are consistent with banking
services causing firms to sell higher quality products—higher wholesale prices in
combination with higher retail prices. Banking services may ease transactions with
suppliers, improve that relationship, and lead dispensaries to buy higher quality products.
Table 6 shows that dispensaries with financial access do indeed sell products with
different chemical content. Dispensaries with financial access sell products with higher
average potency. Within the two main chemical compounds in marijuana, i.e., THC and
CBD, these dispensaries sell products with higher THCA levels than dispensaries without
financial access.
20
This evidence is also consistent with some of the cost savings from using banking
services being passed through to suppliers. We will investigate these mechanisms in section
6 using transaction level data and marijuana strain fixed effects. Differences between the
dispensary and strain level results will indicate whether firms are paying and charging
higher prices generally or if they are selling higher quality products that have a higher
wholesale and retail price.
Taken together, the evidence using distance as an instrumental variable suggests that
banking services lead to lower markups and higher prices. One potential concern with the
instrumental variable design, however, is that distance to a 502 credit union could be
conflating other effects of proximity to geographic areas where financial institutions locate.
The conflating factors could bias the estimates and make them less precise. Specifically,
these estimates may be an underestimate due to attenuation bias from measurement error.
To address these concerns, we use additional analysis on firms close to other credit unions
to control for these conflating factors in a panel data fixed effects model. Tables 7 and 8
will provide additional specifications to control for confounding factors that could be
causing this estimate to be attenuated and imprecisely estimated.
5
Panel Data Fixed Effects Evidence
If 502 credit unions are located in areas that are different from other areas in ways that
impact business decisions and outcomes, then the estimates from the IV specification could
be imprecise or biased. To control for these differences, we implement a panel data fixed
effects (PDFE) model. If credit unions locate in areas that provide some advantage (or
disadvantage) to marijuana dispensaries, then we might be concerned that distance alone is
not capturing the effect estimated in the IV specification.
The model uses within group variation (Cameron and Trivedi, 2005; Bloom, Sadun, and
Van Reenen, 2012), with groups close and far and type of credit union (ones that provide
21
or do not provide services to the marijuana business). The PDFE specification, however,
controls for that by including a distance measure to any credit union,
Yi,t =β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t
+ β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
(3)
The close 502 CU variable is one if a firm is within 2 miles of a 502 credit union and
zero otherwise. The close non-502 CU variable is one if a firm is within 2 miles of a credit
union that does not provide services to the marijuana industry and zero otherwise. The far
non-502 CU variable is one if a firm is between 2 and 10 miles of a credit union that does
not provide services to the marijuana industry and zero otherwise. The omitted group is
firms that are 2 to 10 miles from a credit union that does provide services to the marijuana
industry. Firms that are not within 10 miles of a credit union are excluded from the sample
to provide a clean comparison group. The estimates are not sensitive to this restriction.
The coefficient of interest β1 captures the causal effect of banking services on business
decisions and outcomes if the differences in error terms is zero, discussed further below.
5.1
Panel Data Fixed Effects Identification
For the panel data fixed effects model to causally identify the effect of banking services
on business decisions and outcomes, the effect cannot vary differently across groups. This
approach removes omitted variable bias from differences between groups due to strategic,
non-random, or confounded selection into our groups (Bloom et al., 2012). To formalize
j
this point, let our outcome variables be denoted yi,k
and depend on observable groups
j = {0, 1} and k = {0, 1}, following the notation of Cameron and Trivedi (2005). Define
group variables Dj , Dk , and Dkj such that Dj = 1 if j = 1, Dk = 1 if k = 1, Dkj = 1 if both
22
k and j equal 1 and Dj = 0, Dk = 0, and Dkj = 0 otherwise. This leads to the specification
j
j
yi,k
= α + αj Dj + αk Dk + βDi,k
+ εji,k ,
(4)
j
where εji,k is a zero-mean constant-variance error term. The coefficient on Di,k
is the policy
impact parameter of interest. To make concrete how β identifies the policy impact,
consider the four group outcomes,
0
yi,1
= α + αk + ε0i,1
(5)
1
yi,1
= α + αk + αj + β + ε1i,1
(6)
0
yi,0
= α + ε0i,0
(7)
1
yi,0
= α + αj + ε0i,0 .
(8)
The coefficients αk and αj change the outcome for groups k = 1 and j = 1 and β
additionally changes the outcome for firms in group k = 1 and j = 1. All reasons why an
outcome may differ across groups are captured by the coefficients αk and αj . The effect of
being in both groups is captured by β. In our context, β captures the effect of banking
services, which we instrument as being close to a credit union and that credit union being
one that provides banking services to the marijuana industry.
The differences in outcomes across groups is given by
1
0
yi,1
− yi,1
= αj + β + ε1i,1 − ε0i,1
(9)
and
1
0
yi,0
− yi,0
= αj + ε1i,0 − ε0i,0 .
(10)
The level effects αk and αj are eliminated by taking the differences in and between
23
equations (9) and (10). In our context, this demonstrates that the differences we explicitly
modeled between being close to a credit union and that credit union providing banking
services are eliminated in the fixed effects specification. We recover the effect of banking
services β by taking the difference between equations (9) and (10),
1
0
1
0
(yi,1
− yi,1
) − (yi,0
− yi,0
) = β + (ε1i,1 − ε0i,1 ) − (ε1i,0 − ε0i,0 )
(11)
and assuming that E[(ε1i,1 − ε0i,1 ) − (ε1i,0 − ε0i,0 )] = 0 for identification. Said differently, the
model allows for there to be differences between firms that are close or far to financial
institutions and differences between firms that are close (far) to 502 credit unions and firms
that are close (far) to non 502 credit unions. The model is unable to identify β if the
difference between being close to a 502 credit union relative to being far from a 502 credit
union is different than the difference between being close to a non 502 credit union relative
to being far from a non 502 credit union. Note that β is consistently estimated even if the
estimates of the level differences, e.g., αk , are not (Cameron and Trivedi, 2005).
This panel data fixed effects model extends the instrumental variable specification by
using additional variation in type of credit union to control for other potential unobserved
differences.
5.2
Panel Data Fixed Effects Empirical Evidence
Tables 7 and 8 report evidence using our panel data fixed effects specification. These
specifications ensure that conflating factors with distance in our instrumental variable do
not drive our IV results. These specifications can also add precision to the estimates. Table
7 compares firms within 2 miles of a 502 credit union to firms between 2 and 10 miles of
one of these 502 credit unions and firms within 2 miles of and 2 to 10 miles away from a
credit union that does not provide these services.
In Table 11 we explicitly model the incremental effect of distance by extending this
24
comparison to firms within 2 miles, 2–4 miles, 4–6 miles, 6–8 miles, and 8–10 miles. Table
11 reports our preferred estimates because they model the most comprehensive distance
controls and as a result, the cleanest estimates.
Column (1) of Tables 7 and 8 report that being within 2 miles of a 502 credit union
increases a firm’s profits by 18.1 percent and 18.8 percent, respectively. The consistency in
these estimates across specifications adds to the validity of these specifications and
demonstrates that the value of the distance controls is mostly to add precision as opposed
to correct for bias. This evidence reinforces the IV estimate that banking services can
substantially increase profits.
Tables 7 and 8 show that profits are higher despite lower markups because sales volumes
are higher for firms that use banking services. In particular, sales are 16 to 17 percent
higher and these estimates are precisely estimated. These estimates, again, follow that
pattern that they are strongest for firms within 2 miles from a credit union that provides
these services and decrease the further away from a firm is from a 502 credit union. The
transaction level estimates in section 6, can also provide additional insights into the sources
of these higher sales volumes. In particular, do firms with banking services engage in more
individual transactions or do they sell a higher quantity of product per transaction?
We estimate that markups are 10 to 11 percentage points lower for firms within 2 miles
of 502 credit unions (Table 8). Table 8 shows that the markups increase with distance,
which is consistent with our first stage estimates that firms are more likely to use banking
services if they are closer to credit unions that provide these services. In particular, we find
markups increase from negative 11 percentage points 2 to 4 miles away, to negative 7
percentage points from 4 to 6 miles away. These estimates control for other factors that
might be correlated with distance from credit unions generally. In fact, we find that
markups decrease with distance from credit unions that do not provide services to the
marijuana industry.
Markups are lower for firms that use banking services because their wholesale prices are
25
7 percent higher, reported in column (3) of Tables 7 and 8. This evidence is consistent with
two mechanisms. First, the cost savings from dispensaries using banking services may
partially pass-through upstream to suppliers, leading to higher wholesale prices. Second,
banking services may improve transactions with suppliers, thereby improving their
relationship, and may allow dispensaries that use banking services to acquire higher quality
products from their suppliers. To investigate these two mechanisms, the analysis in section
6 uses transaction level estimates with strain fixed effects.
6
Transaction Level Evidence
We use a dataset of transaction level data to explore the channels through which these
effects occur. The data provide a unique look at business operations because it records
retail and wholesale prices, as well as sales volume. The data contain the universe of
businesses and includes 62 million retail transactions from 2014 through 2017. The data
include the latitude and longitude of every establishment.
Tables 9–11 repeat the empirical specifications in Tables 5–8 using transaction-level
data and strain fixed effects. These estimates provide complementary evidence to the
dispensary-level evidence by detailing the mechanisms underlying the results at the
dispensary level. For example, the dispensary-level evidence suggests that markups are
lower and profits are higher for firms that use banking services. The transaction-level
evidence can inform us as to whether banking services cause firms to charge a lower
markup on all products or to sell a different set of products that have a lower markup.
Column (2) and (3) of Tables 9–11 report that firms that use banking services charge 6
percent higher retail prices (Tables 10 and 11) and pay 3 to 6 percent higher wholesale
prices than other firms for the same strain. This evidence suggests some of the cost savings
from using banking services are passed upstream to processors.
Sales volumes are 14 to 16 percent higher for firms that use banking services per
26
transaction, reported in column (5) of Tables 9–11. This evidence suggests that firms that
use banking services have different types of consumers—those that consume more. These
consumers may be less likely to be tourists or casual consumers.
We find strong evidence that at the transaction level, profits are higher, retail and
wholesale prices are higher, and sales are higher. In contrast, markups are not as strongly
negative at the transaction level as they are at the firm level. Taken together, this evidence
suggests that some of the lower markups at the firm level are due to these firms selling a
different set of products that have lower markups.
Banking services allow firms to make different business decisions that lead to an
increase in profits of 23 to 24 percent higher per transaction. The estimates from Table 10
and 11 provide the strongest evidence combining the panel data fixed effects method with
extensive distance controls.
7
Conclusion
In this paper we measure the economic effects of financial inclusion on firms. To
address the empirical challenges of answering this question, we use the marijuana industry
as a unique setting. Most firms in our sample struggled to obtain access to financial
institutions at some point during the sample period. Some firms were successful and others
unsuccessful at achieving financial inclusion.
In an IV setting, the results show that firms with 502 credit union access have lower
markups and higher retail and wholesale prices. When we control for the strains of
marijuana that a firm sells, the results suggest that access to 502 credit unions increases
profits, which are due to higher sales of higher quality products. Results from a panel data
fixed effects model confirm the IV results and document higher profits and sales volumes
among firms. In all, firms within 2 miles of a 502 credit union have higher profits, lower
markups, and higher sales.
27
The cost savings from dispensaries using banking services may partially pass-through
upstream to suppliers, leading to higher wholesale prices. Alternatively, banking services
may improve transactions with suppliers, thereby improving their relationship, and may
allow dispensaries that use banking services to acquire higher quality products from their
suppliers.
We analyze transaction level data with strain fixed effects to disentangle these effects.
The evidence suggests that some of the cost savings from banking services are passed
upstream to processors. In addition, access to 502 credit unions lead to an increase in
profits of 23 to 24 percent higher per transaction. The evidence suggests that firms with
banking services have different types of consumers—those that consume more. These
consumers may be less likely to be tourists or casual consumers.
In developed economies, unbanked firms are rare. In contrast to the abundance of
literature surrounding credit access and firm outcomes, there is no evidence about access to
the warehouse banking function. Using the marijuana industry, we are able to provide a
first glimpse of how financial inclusion affects firms. In addition, as the number of bank
branches declines and banking moves online, the results shed light on the potential impact
of reduced local bank access on firms.
28
References
Adelino, M., A. Schoar, and F. Severino (2015). House prices, collateral, and
self-employment. Journal of Financial Economics 117 (2), 288–306.
Beck, T., A. Demirguc-Kunt, and M. S. Martinez Peria (2006). Banking services for
everyone? Barriers to bank access and use around the world. The World Bank.
Berger, A. N., N. H. Miller, M. A. Petersen, R. G. Rajan, and J. C. Stein (2005). Does
function follow organizational form? evidence from the lending practices of large and
small banks. Journal of Financial economics 76 (2), 237–269.
Bloom, N., R. Sadun, and J. Van Reenen (2012). The organization of firms across
countries. The Quarterly Journal of Economics 127 (4), 1663–1705.
Bound, J., D. A. Jaeger, and R. M. Baker (1995). Problems with instrumental variables
estimation when the correlation between the instruments and the endogenous
explanatory variable is weak. Journal of the American Statistical Association 90 (430),
443–450.
Brickley, J. A., J. S. Linck, and C. W. Smith Jr (2003). Boundaries of the firm: evidence
from the banking industry. Journal of Financial Economics 70 (3), 351–383.
Brown, J. R., J. A. Cookson, and R. Z. Heimer (2017). Courting economic development.
The World Bank Economic Review 30 (Supplement 1), S176–S187.
Butler, A. W. and J. Cornaggia (2011). Does access to external finance improve
productivity? evidence from a natural experiment. Journal of Financial
Economics 99 (1), 184–203.
Cameron, A. C. and P. K. Trivedi (2005). Microeconometrics: methods and applications.
Cambridge university press.
Cetorelli, N. and P. E. Strahan (2006). Finance as a barrier to entry: Bank competition
and industry structure in local us markets. The Journal of Finance 61 (1), 437–461.
Donaldson, J. R., G. Piacentino, and A. Thakor (2018). Warehouse banking. Journal of
Financial Economics 129 (2), 250–267.
FIN-2014-G001 (February 14, 2014). Bsa expectations regarding marijuana-related
businesses. Department of the Treasury; Financial Crimes Enforcement Network .
Gilje, E. P. (2017). Does local access to finance matter? evidence from us oil and natural
gas shale booms. Management Science 65 (1), 1–18.
Gilje, E. P., E. Loutskina, and P. E. Strahan (2016). Exporting liquidity: Branch banking
and financial integration. The Journal of Finance 71 (3), 1159–1184.
29
Graham, J. R. and C. R. Harvey (2001). The theory and practice of corporate finance:
Evidence from the field. Journal of Financial Economics 60 (2-3), 187–243.
Hansen, B., K. Miller, and C. Weber (2017). The taxation of recreational marijuana:
Evidence from washington state. National Bureau of Economic Research working paper .
Jayaratne, J. and P. E. Strahan (1996). The finance-growth nexus: Evidence from bank
branch deregulation. The Quarterly Journal of Economics 111 (3), 639–670.
Mandelbaum, R. (January 4, 2018). Where pot entrepreneurs go when the banks just say
no. The New York Times.
Paravisini, D. (2008). Local bank financial constraints and firm access to external finance.
The Journal of Finance 63 (5), 2161–2193.
Petersen, M. A. and R. G. Rajan (1994). The benefits of lending relationships: Evidence
from small business data. The journal of finance 49 (1), 3–37.
Petersen, M. A. and R. G. Rajan (2002). Does distance still matter? the information
revolution in small business lending. The journal of Finance 57 (6), 2533–2570.
Rice, T. and P. E. Strahan (2010). Does credit competition affect small-firm finance? The
Journal of Finance 65 (3), 861–889.
Staiger, D. and J. Stock (1997). Instrumental variables regression with weak instruments.
Econometrica 65 (3), 557–586.
Trahan, E. A. and L. J. Gitman (1995). Bridging the theory-practice gap in corporate
finance: a survey of chief financial officers. The Quarterly Review of Economics and
Finance 35 (1), 73–87.
30
Figure 1: Distance to Nearest Credit Union
(a) Distribution of Distance
0
0
.1
.1
Density
.2
Density
.2
.3
.3
.4
.5
.4
(b) Distribution of Log Distance
0
10
20
30
40
Distance to nearest credit union (Miles)
50
kernel = epanechnikov, bandwidth = 0.3028
−4
−2
0
2
Log distance to the nearest credit union
4
kernel = epanechnikov, bandwidth = 0.3309
0
.002
Density
.004
.006
.008
(c) Distribution of Markups
0
100
200
300
400
Retail Markup Over Whole Sale Price (Percent)
kernel = epanechnikov, bandwidth = 14.9626
31
Table 1: Banking in the Marijuana Industry
This table shows survey responses from marijuana firms in Washington state in 2016 about whether they use banking services
and from what type of financial institution. The survey questions are shown in Appendix A.
Credit Union
Bank
All Types
(1)
Community
(2)
Regional
(3)
Community
(4)
Regional
(5)
National
(6)
Other
Business
49%
30%
36%
5%
14%
12%
2%
Personal
18%
27%
26%
2%
9%
8%
5%
Other
3%
No Acct.
30%
33%
36%
3%
14%
11%
3%
Total
32
Table 2: Transactions and Financial Services in the Marijuana Industry
This table shows survey responses from marijuana firms in Washington state in 2016 about
transactions and financial services. The survey questions are shown in Appendix A.
What forms of payment do you accept? (Please mark all that apply).
Cash Only
49%
Checks
40%
Debit Cards
14%
Credit Cards
3%
Do you pay for a security company that transports cash and/or product?
Yes
30%
Dollars per month
$348
How much do you lose due to theft?
Dollars per month
$618
How much does your business pay per month, in dollars, to use all financial services at
Bank or credit union
Non-bank financial institution
Other
$748
$1,059
$437
33
Table 3: Summary Statistics: Marijuana Dispensaries
The table reports summary statistics for the sample of marijuana dispensaries as of February
2017. The sample consists of marijuana dispensaries with both retail and medical marijuana
licenses. Column 1 reports means of variables for the set of dispensaries within 3 miles
of a special financial institution and Column 2 reports means for the set of dispensaries
more than 3 miles from special financial institutions. Column 3 reports the difference in
means between Columns 1 and 2 and the p-value of the t-statistic of these differences, with
standard errors clustered by dispensary. Number of financial institutions is a variable that
measures how many financial institutions are within 1 mile (5 miles) of the dispensary.
The potency variables measure the chemical contents of each strain of marijuana that a
dispensary sells. Markup percentage measures the percentage by which the price charged
exceeds the cost of the marijuana. Sales measures the number of transactions at a dispensary
in each month, Sales Volume measures the total product weight of these transactions, and
Number of strains sold counts the total number of unique strains that the dispensary sells
in these transactions. The average retail price measures the price paid by the consumer to
the dispensary and the average wholesale price measures the price paid by the dispensary to
the processor (distributor).
Dispensaries With:
Number of Fin Institutions within 5 miles
Number of Fin Institutions within 1 mile
Potency (Total)
Potency (CBD)
Potency (THC)
Potency (THCA)
Markup Percentage
Ln(Sales)
Ln(Sales Volume)
Ln(Number of Strains Sold)
Average Unit Price (Retail)
Ln(Avg Unit Price, Retail)
Ln(Avg Wholesale Price)
Number of Observations
Special Finance = 0
Special Finance = 1
Difference
32.516
(35.521)
3.366
(5.698)
18.616
(3.353)
0.316
(0.436)
2.358
(3.117)
16.204
(4.583)
165.221
(59.679)
7.888
(0.797)
8.861
(0.928)
4.324
(0.465)
41.438
(8.599)
3.694
(0.209)
2.793
(0.289)
81.019
(50.110)
9.944
(9.457)
18.731
(3.029)
0.275
(0.245)
1.778
(2.554)
16.792
(3.946)
151.827
(56.838)
8.135
(0.962)
9.005
(0.910)
4.378
(0.470)
41.007
(7.693)
3.689
(0.180)
2.851
(0.260)
48.502
(0.006)
6.579
(0.007)
0.115
(0.211)
-0.041
(0.225)
-0.580
(0.122)
0.588
(0.126)
-13.394
(0.197)
0.248
(0.086)
0.143
(0.182)
0.054
(0.340)
-0.430
(0.291)
-0.005
(0.560)
0.057
(0.139)
186
108
34
Table 4: IV: First Stage—Banking Services and Distance
This table shows how the probability of using banking services changes with proximity to a
credit union. Observations are at the marijuana firm level. The sample combines survey data
and administrative data from the state of Washington (see section 3.2.2). The dependent
variable, ✶(Use Banking Services), is an indicator variable equal to one if in the survey the
firm indicated that it used banking services and zero otherwise. Proximity to credit union
is the log of the distance between a marijuana firm and the closest credit union. Within 1
mile of credit union is an indicator variable equal to one if the distance to the nearest credit
union is less than 1 mile and zero otherwise. Similarly, Within 2 miles of credit union and
Within 3 miles of credit union are indicator variables equal to one if the distance to a credit
union is less than 2 or 3 miles, and zero otherwise. Standard errors appear in parentheses
and *, **, and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: ✶(Use Banking Services)
Proximity to credit union
(1)
A
-0.094***
(0.020)
Within 1 miles of credit union
(2)
B
(3)
C
0.293***
(0.056)
Within 2 miles of credit union
0.229***
(0.065)
Within 3 miles of credit union
F-statistic
Adj. R-Square
Observations
(4)
D
0.210***
(0.079)
22.302
0.068
295
27.355
0.082
295
12.303
0.037
295
7.011
0.020
295
35
Table 5: Average Monthly Dispensary Markups, Prices, and Profits - Instrumental Variable
This table reports the effects of banking services on marijuana dispensaries using an instrumental variable specification at the
dispensary-month level. The instrumental variable specification uses distance to a 502 credit union as an instrument for whether
a dispensary uses banking services. The specification includes controls for whether the firm is within 2 miles of a 502 credit
union, the number of dispensaries (competitors) within 5 miles, as well as month and county fixed effects, αt and αj .
Yi,t = β0 + β1 Distance to 502 Credit Unioni,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) markup percentage, which is defined as the difference between
retail and wholesale price divided by retail price; Column (2) the log of the average retail price per unit, where the retail price
is observed; Column (3) the log of the average wholesale price per unit, where the wholesale price is observed; Column (4) log
profit, where profit is defined as the difference between retail and wholesale price multiplied by the sales volume; Column (5)
the log of usable weight of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the
dispensary level. ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Proximity to 502 CU
Firms within 2mi of 502 CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
Markup
Price
(1)
Markup Percentage
(2)
(3)
Ln(Avg Unit Price, Retail) Ln(Avg Wholesale Price)
-6.313∗ ∗ ∗
(2.261)
1.144
(3.124)
-0.048
(0.061)
0.501
5,703
280
Yes
Yes
Yes
Dispensary
0.018∗
(0.009)
0.029
(0.021)
-0.001
(0.000)
0.649
5,703
280
Yes
Yes
Yes
Dispensary
Profits
0.042∗ ∗ ∗
(0.014)
0.024
(0.021)
-0.000
(0.000)
0.496
5,703
280
Yes
Yes
Yes
Dispensary
(4)
Profit
(5)
Ln(Ussble Weight)
0.013
(0.030)
0.159∗∗
(0.070)
-0.002
(0.001)
0.017
(0.021)
0.113∗∗
(0.053)
-0.001
(0.001)
0.371
5,703
280
Yes
Yes
Yes
Dispensary
0.102
5,703
280
Yes
Yes
Yes
Dispensary
36
Table 6: Average Monthly Dispensary Markups, Prices, and Profits - Instrumental Variable
This table reports the effects of banking services on marijuana dispensaries using an instrumental variable specification at the
dispensary-month level. The instrumental variable specification uses distance to a 502 credit union as an instrument for whether
a dispensary uses banking services. The specification includes controls for whether the firm is within 2 miles of a 502 credit
union, the number of dispensaries (competitors) within 5 miles, as well as month and county fixed effects, αt and αj .
Yi,t = β0 + β1 Distance to 502 Credit Unioni,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) markup percentage, which is defined as the difference between
retail and wholesale price divided by retail price; Column (2) the log of the average retail price per unit, where the retail price
is observed; Column (3) the log of the average wholesale price per unit, where the wholesale price is observed; Column (4) log
profit, where profit is defined as the difference between retail and wholesale price multiplied by the sales volume; Column (5)
the log of usable weight of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the
dispensary level. ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Proximity to 502 CU
Firms within 2mi of 502 CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
Total
CDB
THC
THCA
(1)
Potency (Total)
(2)
Potency (CBD)
(3)
Potency (THC)
(4)
Potency (THCA)
0.266
(0.167)
0.592
(0.379)
0.002
(0.007)
-0.328
(0.373)
0.339
(0.467)
0.008
(0.007)
0.137
5,703
280
Yes
Yes
Yes
Dispensary
0.380
5,703
280
Yes
Yes
Yes
Dispensary
0.261∗∗
(0.117)
0.151
(0.174)
-0.002
(0.003)
0.370
5,703
280
Yes
Yes
Yes
Dispensary
0.499∗ ∗ ∗
(0.140)
0.177
(0.259)
-0.003
(0.004)
0.326
5,703
280
Yes
Yes
Yes
Dispensary
37
Table 7: Average Monthly Dispensary Markup, Prices, and Profits - Panel Data Fixed Effect
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The
panel data fixed effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries
and 502 credit unions across counties within Washington. The data are monthly averages of a dispensary’s total transactions
within a given month. The control variables include: number of firms within 2 miles of the 502 credit union to measure financial
access, number of other dispensaries within 5 miles of the dispensary to measure product market competition, and the number of
firms within 2 and 10 miles of any credit union to measure potential for financial access. In addition the regression specification
includes county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) markup percentage, which is defined as the difference between
retail and wholesale price divided by retail price; Column (2) the log of the average retail price per unit, where the retail price
is observed; Column (3) the log of the average wholesale price per unit, where the wholesale price is observed; Column (4) log
profit, where profit is defined as the difference between retail and wholesale price multiplied by the sales volume; Column (5)
the log of usable weight of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the
dispensary level. ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Markup
(1)
Markup Percentage
Firms within 2 of 502 CU
Firms 2 - 10 mi from any CU
Firms within 2 of any CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
-10.316∗
(5.299)
-8.833∗
(5.245)
10.238∗∗
(4.490)
-0.086
(0.062)
0.499
5,703
280
Yes
Yes
Yes
Dispensary
Price
Profits
(2)
Ln(Avg Unit Price, Retail)
(3)
Ln(Avg Wholesale Price)
(4)
Profit
0.035
(0.045)
-0.001
(0.040)
0.010
(0.023)
-0.000
(0.000)
0.074∗
(0.042)
0.033
(0.039)
-0.031
(0.028)
-0.000
(0.000)
0.181∗
(0.097)
0.025
(0.074)
0.072
(0.064)
-0.002
(0.001)
0.648
5,703
280
Yes
Yes
Yes
Dispensary
0.490
5,703
280
Yes
Yes
Yes
Dispensary
0.372
5,703
280
Yes
Yes
Yes
Dispensary
(5)
Ln(Ussble Weight)
0.164∗∗
(0.068)
0.050
(0.046)
0.014
(0.043)
-0.001
(0.001)
0.102
5,703
280
Yes
Yes
Yes
Dispensary
38
Table 8: Average Monthly Dispensary Transactions: Markups, Prices, and Profits - Panel Data Fixed Effects All distances
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The
panel data fixed effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries
and 502 credit unions across counties within Washington. The data are monthly averages of a dispensary’s total transactions
within a given month. The control variables include: number of firms within 2 miles of the 502 credit union to measure financial
access, number of other dispensaries within 5 miles of the dispensary to measure product market competition, and the number of
firms within 2-4, 4-6, 6-8, and 8-10 miles of any credit union to measure potential for financial access. In addition the regression
specification includes county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) markup percentage, which is defined as the difference between
retail and wholesale price divided by retail price; Column (2) the log of the average retail price per unit, where the retail price
is observed; Column (3) the log of the average wholesale price per unit, where the wholesale price is observed; Column (4) log
profit, where profit is defined as the difference between retail and wholesale price multiplied by the sales volume; Column (5)
the log of usable weight of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the
dispensary level. ∗ , ∗∗ , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Markup
(1)
Markup Percentage
Firms within 2mi of 502 CU
Firms within 2-4mi of 502 CU
Firms within 4-6mi of 502 CU
Firms within 6-8mi of 502 CU
Firms within 8-10mi of 502 CU
Firms within 2mi of any CU
Firms within 2-4mi of any CU
Firms within 4-6mi of any CU
Firms within 6-8mi of any CU
Marijuana firms w/in 5 miles
39
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
-11.593∗∗
(5.176)
-11.754∗∗
(5.566)
-7.456
(6.585)
-3.268
(8.007)
-5.594
(7.045)
-32.087∗ ∗ ∗
(10.493)
-40.495∗ ∗ ∗
(11.706)
-51.395∗ ∗ ∗
(12.167)
-33.267∗∗
(15.593)
-0.076
(0.064)
0.504
5,703
280
Yes
Yes
Yes
Dispensary
Price
Profit
(2)
Ln(Avg Unit Price, Retail)
(3)
Ln(Avg Wholesale Price)
0.030
(0.046)
-0.002
(0.042)
-0.056
(0.051)
-0.021
(0.059)
0.024
(0.044)
-0.099
(0.069)
-0.105
(0.075)
-0.114
(0.077)
-0.111
(0.077)
-0.000
(0.000)
0.075∗
(0.044)
0.045
(0.043)
-0.025
(0.049)
-0.011
(0.051)
0.043
(0.041)
0.026
(0.019)
0.053
(0.034)
0.092
(0.057)
0.004
(0.063)
-0.000
(0.000)
0.649
5,703
280
Yes
Yes
Yes
Dispensary
0.493
5,703
280
Yes
Yes
Yes
Dispensary
(4)
Profit
0.188∗
(0.099)
0.032
(0.080)
-0.016
(0.114)
0.002
(0.120)
0.005
(0.086)
-0.025
(0.069)
-0.045
(0.112)
-0.133
(0.098)
-0.340∗∗
(0.167)
-0.002
(0.001)
0.374
5,703
280
Yes
Yes
Yes
Dispensary
(5)
Ln(Ussble Weight)
0.178∗∗
(0.070)
0.065
(0.048)
0.062
(0.072)
0.026
(0.055)
0.001
(0.056)
0.154∗ ∗ ∗
(0.053)
0.176∗∗
(0.076)
0.149∗∗
(0.072)
-0.117
(0.128)
-0.001
(0.001)
0.107
5,703
280
Yes
Yes
Yes
Dispensary
Table 9: Monthly Transactions: Markups, Prices, and Profits - Instrumental Variables
This table reports the effects of banking services on marijuana dispensaries using an instrumental variable specification at
the transaction-month level. The instrumental variable specification uses distance to a 502 credit union as an instrument for
whether a dispensary uses banking services. The specification includes controls for whether the firm is within 2 miles of a 502
credit union, the number of dispensaries (competitors) within 5 miles, as well as month and county fixed effects, αt and αj .
Yi,t = β0 + β1 Distance to 502 Credit Unioni,t + βXi,t + αt + αj + εi,t .
Dependent variables are at the transaction level for each dispensary month. Each column reports the effects of banking on the
following dependent variables: Column (1) markup percentage, which is defined as the difference between retail and wholesale
price divided by retail price; Column (2) the log of the average retail price per unit, where the retail price is observed; Column
(3) the log of the average wholesale price per unit, where the wholesale price is observed; Column (4) log profit, where profit is
defined as the difference between retail and wholesale price multiplied by the sales volume; Column (5) the log of usable weight
of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the dispensary level. ∗ , ∗∗ , and
∗∗∗
denote significance at the 10%, 5%, and 1% levels.
Markup
(1)
Markup Percentage
Proximity to 502 CU
Firms within 2mi of 502 CU
Competitors within 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Strain FE
Cluster Variable
-4.433∗
(2.526)
2.441
(2.959)
-0.097∗
(0.058)
0.303
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
Prices
(2)
Ln(Avg Unit Price, Retail)
Profits
(3)
Ln(Avg Wholesale Price)
0.020
(0.012)
0.057∗∗
(0.024)
-0.001∗
(0.000)
0.039∗∗
(0.018)
0.046∗
(0.024)
-0.000
(0.000)
0.090
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
0.055
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
(4)
Profit
0.027
(0.029)
0.194∗ ∗ ∗
(0.064)
-0.002∗∗
(0.001)
0.053
1,706,318
280
Yes
Yes
Yes
Yes
Dispensary
(5)
Ln(UsableWeight)
0.026
(0.022)
0.110∗ ∗ ∗
(0.040)
-0.001
(0.001)
0.034
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
40
Table 10: Monthly Dispensary Transactions: Markups, Prices, and Profits - Panel Data Fixed Effects
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The
panel data fixed effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries
and 502 credit unions across counties within Washington. The data are each transaction of a dispensary within a given month.
The control variables include: number of firms within 2 miles of the 502 credit union to measure financial access, number of
other dispensaries within 5 miles of the dispensary to measure product market competition, and the number of firms within
2 and 10 miles of any credit union to measure potential for financial access. In addition the regression specification includes
county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are measured at the transaction level. Each column reports the effects of banking on the following dependent
variables: Column (1) markup percentage, which is defined as the difference between retail and wholesale price divided by retail
price; Column (2) the log of the average retail price per unit, where the retail price is observed; Column (3) the log of the
average wholesale price per unit, where the wholesale price is observed; Column (4) log profit, where profit is defined as the
difference between retail and wholesale price multiplied by the sales volume; Column (5) the log of usable weight of marijuana
by volume, where the usable weight is observed. Standard errors are clustered at the dispensary level. ∗ , ∗∗ , and ∗∗∗ denote
significance at the 10%, 5%, and 1% levels.
Firms within 2 of 502 CU
Firms 2 - 10 mi from any CU
Firms within 2 of any CU
Competitors within 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Strain FE
Cluster Variable
Markup
Prices
(1)
Markup Percentage
(2)
(3)
Ln(Avg Unit Price, Retail) Ln(Avg Wholesale Price)
-0.106
(6.046)
-0.445
(5.862)
7.434∗
(4.454)
-0.119∗∗
(0.056)
0.302
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
0.065∗
(0.036)
0.000
(0.032)
-0.007
(0.025)
-0.001
(0.000)
0.089
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
Profits
0.063∗∗
(0.028)
0.000
(0.024)
-0.036
(0.028)
-0.000
(0.000)
0.054
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
(4)
Profit
0.230∗ ∗ ∗
(0.088)
0.030
(0.068)
0.046
(0.057)
-0.002∗
(0.001)
0.053
1,706,318
280
Yes
Yes
Yes
Yes
Dispensary
(5)
Ln(UsableWeight)
0.141∗ ∗ ∗
(0.054)
0.024
(0.039)
0.019
(0.038)
-0.001
(0.001)
0.034
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
41
Table 11: Monthly Dispensary Transactions: Markups, Prices, and Profits - Panel Data Fixed Effects - All
Distance
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The panel data fixed
effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries and 502 credit unions across counties
within Washington. The data are each transaction of a dispensary within a given month. The control variables include: number of firms within 2
miles of the 502 credit union to measure financial access, number of other dispensaries within 5 miles of the dispensary to measure product market
competition, and the number of firms within 2-4, 4-6, 6-8, and 8-10 miles of any credit union to measure potential for financial access. In addition
the regression specification includes county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are measured at the transaction level. Each column reports the effects of banking on the following dependent variables: Column
(1) markup percentage, which is defined as the difference between retail and wholesale price divided by retail price; Column (2) the log of the average
retail price per unit, where the retail price is observed; Column (3) the log of the average wholesale price per unit, where the wholesale price is
observed; Column (4) log profit, where profit is defined as the difference between retail and wholesale price multiplied by the sales volume; Column
(5) the log of usable weight of marijuana by volume, where the usable weight is observed. Standard errors are clustered at the dispensary level. ∗ , ∗∗ ,
and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Markup
(1)
Markup Percentage
Firms within 2mi of 502 CU
Firms within 2-4mi of 502 CU
Firms within 4-6mi of 502 CU
Firms within 6-8mi of 502 CU
Firms within 8-10mi of 502 CU
Firms within 2mi of any CU
Firms within 2-4mi of any CU
Firms within 4-6mi of any CU
Firms within 6-8mi of any CU
Competitors within 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Strain FE
Cluster Variable
-0.986
(5.980)
-3.110
(6.231)
2.212
(7.053)
-0.957
(6.991)
2.243
(7.087)
-21.993∗ ∗ ∗
(0.297)
-26.072∗ ∗ ∗
(5.066)
-41.976∗ ∗ ∗
(6.284)
-16.074∗
(8.384)
-0.112∗
(0.060)
0.305
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
Prices
(2)
Ln(Avg Unit Price, Retail)
0.066∗
(0.037)
0.009
(0.035)
-0.029
(0.044)
-0.026
(0.041)
0.003
(0.046)
-0.058∗ ∗ ∗
(0.004)
-0.046
(0.031)
-0.050
(0.037)
-0.081∗ ∗ ∗
(0.030)
-0.001
(0.000)
0.090
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
Profits
(3)
Ln(Avg Wholesale Price)
0.068∗∗
(0.030)
0.020
(0.029)
-0.037
(0.034)
-0.025
(0.031)
-0.008
(0.034)
0.021∗ ∗ ∗
(0.004)
0.045
(0.028)
0.111∗
(0.059)
-0.037
(0.047)
-0.000
(0.000)
0.055
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
(4)
Profit
0.245∗ ∗ ∗
(0.093)
0.053
(0.078)
0.038
(0.101)
-0.024
(0.089)
-0.009
(0.086)
-0.062∗ ∗ ∗
(0.008)
-0.078
(0.076)
-0.126∗
(0.071)
-0.355∗ ∗ ∗
(0.110)
-0.002∗
(0.001)
0.053
1,706,318
280
Yes
Yes
Yes
Yes
Dispensary
(5)
Ln(UsableWeight)
0.156∗ ∗ ∗
(0.056)
0.044
(0.041)
0.051
(0.052)
-0.003
(0.040)
-0.015
(0.043)
0.048∗ ∗ ∗
(0.005)
0.034
(0.044)
0.066
(0.058)
-0.210∗∗
(0.084)
-0.001
(0.001)
0.035
1,707,936
280
Yes
Yes
Yes
Yes
Dispensary
42
Table A.1: IV: First Stage—Banking Services and Distance With Controls
This table shows how the probability of using banking services changes with proximity to a
credit union controlling for distance to a bank. Observations are at the marijuana firm level.
The sample combines survey data and administrative data from the state of Washington (see
section 3.2.2). The dependent variable, ✶(Use Banking Services), is an indicator variable
equal to one if in the survey the firm indicated that it used banking services and zero
otherwise. Proximity to credit union is the log of the distance between a marijuana firm and
the closest credit union. Proximity to bank is the log of the distance between a marijuana
firm and the closest bank. Within 1 mile of credit union is an indicator variable equal to one
if the distance to the nearest credit union is less than 1 mile and zero otherwise. Similarly,
Within 2 miles of credit union and Within 3 miles of credit union are indicator variables
equal to one if the distance to a credit union is less than 2 or 3 miles, and zero otherwise.
Within 1 mile of bank, Within 2 mile of bank, and Within 3 mile of bank are similarly defined
as indicator variables equal to one if the distance to the nearest bank is within 1, 2, or 3
miles, and zero otherwise. Standard errors appear in parentheses and *, **, and *** indicate
significance at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: ✶(Use Banking Services)
Proximity to credit union
Proximity to bank
(1)
-0.161***
(0.022)
(2)
(3)
0.147***
(0.024)
Within 1 miles of credit union
0.439***
(0.063)
Within 1 miles of bank
-0.317***
(0.069)
Within 2 miles of credit union
0.384***
(0.076)
Within 2 miles of bank
-0.344***
(0.093)
Within 3 miles of credit union
0.287***
(0.092)
Within 3 miles of bank
F-statistic
Adj. R-Square
Observations
(4)
-0.183*
(0.110)
30.781
0.168
295
25.110
0.141
295
13.251
0.077
295
4.903
0.026
295
43
Table A.2: IV: First Stage—Placebo Test
This table shows a placebo test that shows how the probability of using banking services
changes with proximity to a bank, despite banks typically not working with marijuana
businesses. Observations are at the marijuana firm level. The sample combines survey data
and administrative data from the state of Washington (see section 3.2.2). The dependent
variable, ✶(Use Banking Services), is an indicator variable equal to one if in the survey the
firm indicated that it used banking services and zero otherwise. Proximity to bank is the
log of the distance between a marijuana firm and the closest bank. Within 1 mile of bank,
Within 2 mile of bank, and Within 3 mile of bank are similarly defined as indicator variables
equal to one if the distance to the nearest bank is within 1, 2, or 3 miles, and zero otherwise.
Standard errors appear in parentheses and *, **, and *** indicate significance at the 10%,
5%, and 1% levels, respectively.
Dependent Variable: ✶(Use Banking Services)
Proximity to bank
(1)
0.056**
(0.023)
Within 1 miles of bank
(2)
(3)
-0.072
(0.064)
Within 2 miles of bank
-0.088
(0.081)
Within 3 miles of bank
F-statistic
Adj. R-Square
Observations
(4)
-0.009
(0.097)
6.016
0.017
295
1.245
0.001
295
1.181
0.001
295
0.008
-0.003
295
44
Table A.3: IV: First Stage—Other Covariates
This table shows how the probability of using banking services changes with whether your parents ran a
company and your level of education. The sample combines survey data and administrative data from the
state of Washington (see section 3.2.2). The dependent variable, ✶(Use Banking Services), is an indicator
variable equal to one if in the survey the firm indicated that it used banking services and zero otherwise.
Standard errors appear in parentheses and *, **, and *** indicate significance at the 10%, 5%, and 1% levels,
respectively.
Dependent Variable: ✶(Use Banking Services)
(1)
(2)
Proximity to credit union
Parents ran company
0.308***
(0.061)
Highschool degree (or equivalent)
Assoc. or technical degree
College degree
Professional or doctorate
F-statistic
Adj. R-Square
Observations
25.453
0.090
247
-0.069
(0.141)
-0.090
(0.135)
0.326**
(0.137)
-0.100
(0.152)
9.397
0.120
247
(3)
(4)
-0.095***
(0.021)
0.283*** 0.266***
(0.058)
(0.056)
-0.086
-0.055
(0.135)
(0.130)
-0.126
-0.080
(0.130)
(0.125)
0.271**
0.272**
(0.132)
(0.127)
-0.139
-0.091
(0.145)
(0.140)
13.042
15.276
0.197
0.258
247
247
45
Table A.4: IV: First Stage—Other Covariates
This table shows how the proximity to a credit union changes with whether your parents ran a company
and your level of education. The sample combines survey data and administrative data from the state of
Washington (see section 3.2.2). The dependent variable, Proximity to credit union, is the log of the distance
between a marijuana firm and the closest credit union. Standard errors appear in parentheses and *, **,
and *** indicate significance at the 10%, 5%, and 1% levels, respectively.
Dependent Variable: Proximity to credit union
Parents ran company
(1)
-0.189
(0.172)
Highschool degree (or equivalent)
Assoc. or technical degree
College degree
Professional or doctorate
F-statistic
Adj. R-Square
Observations
1.205
0.001
247
(2)
0.317
(0.404)
0.466
(0.387)
-0.023
(0.392)
0.486
(0.433)
1.622
0.010
247
(3)
-0.175
(0.172)
0.328
(0.404)
0.488
(0.387)
0.011
(0.394)
0.510
(0.434)
1.503
0.010
247
46
Table A.5: Monthly Dispensary Transactions: Sales - Instrumental Variable
This table reports the effects of banking services on marijuana dispensaries using an instrumental variable specification at the
dispensary-month level. The instrumental variable specification uses distance to a 502 credit union as an instrument for whether
a dispensary uses banking services. The specification includes controls for whether the firm is within 2 miles of a 502 credit
union, the number of dispensaries (competitors) within 5 miles, as well as month and county fixed effects, αt and αj .
Yi,t = β0 + β1 Distance to 502 Credit Unioni,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) total transactions, which is defined as the total number of retail
transactions that occur at a dispensary in a given month; Column (2) the log of sales volume, which is the total number of
transactions weighted by the usable weight of each sale; and Column (3) the log of the number of strains sold, which is a count
of the total number of strains that the dispensary sold in a given month. Standard errors are clustered at the dispensary level.
∗ ∗∗
, , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Sales
Proximity to 502 CU
Firms within 2mi of 502 CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
(1)
Ln(Total Transactions)
(2)
Ln(Sales Volume)
0.061
(0.064)
0.076
(0.141)
-0.000
(0.003)
0.098
(0.065)
0.187
(0.142)
-0.001
(0.002)
0.326
5,703
280
Yes
Yes
Yes
Dispensary
0.330
5,703
280
Yes
Yes
Yes
Dispensary
(3)
Ln(Number of Strains Sold)
0.079∗∗
(0.039)
0.138
(0.089)
-0.002
(0.002)
0.324
5,703
280
Yes
Yes
Yes
Dispensary
47
Table A.6: Monthly Dispensary Transactions: Sales - Panel Data Fixed Effects
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The
panel data fixed effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries
and 502 credit unions across counties within Washington. The data are monthly averages of a dispensary’s total transactions
within a given month. The control variables include: number of firms within 2 miles of the 502 credit union to measure financial
access, number of other dispensaries within 5 miles of the dispensary to measure produce market competition, and the number of
firms within 2 and 10 miles of any credit union to measure potential for financial access. In addition the regression specification
includes county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) total transactions, which is defined as the total number of retail
transactions that occur at a dispensary in a given month; Column (2) the log of sales volume, which is the total number of
transactions weighted by the usable weight of each sale; and Column (3) the log of the number of strains sold, which is a count
of the total number of strains that the dispensary sold in a given month. Standard errors are clustered at the dispensary level.
∗ ∗∗
, , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Sales
Firms within 2 of 502 CU
Firms 2 - 10 mi from any CU
Firms within 2 of any CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
(1)
Ln(Total Transactions)
(2)
Ln(Sales Volume)
(3)
Ln(Number of Strains Sold)
-0.068
(0.237)
-0.172
(0.213)
0.234
(0.143)
-0.000
(0.003)
0.140
(0.235)
-0.082
(0.210)
0.226
(0.161)
-0.001
(0.002)
0.196
(0.173)
0.028
(0.160)
0.038
(0.097)
-0.001
(0.002)
0.329
5,703
280
Yes
Yes
Yes
Dispensary
0.330
5,703
280
Yes
Yes
Yes
Dispensary
0.321
5,703
280
Yes
Yes
Yes
Dispensary
48
Table A.7: Monthly Dispensary Transactions: Sales- Panel Data Fixed Effects - All distance
This table reports the results of the effects of 502 credit unions on the markup, prices, and profits of marijuana dispensaries. The
panel data fixed effects identification strategy relies on cross-sectional variation in the distance between marijuana dispensaries
and 502 credit unions across counties within Washington. The data are monthly averages of a dispensary’s total transactions
within a given month. The control variables include: number of firms within 2 miles of the 502 credit union to measure financial
access, number of other dispensaries within 5 miles of the dispensary to measure produce market competition, and the number of
firms within 2-4, 4-6, 6-8, and 8-10 miles of any credit union to measure potential for financial access. In addition the regression
specification includes county and year fixed effects.
Yi,t = β0 + β1 Close 502 CUi,t + β2 Close Non-502 CUi,t + β3 Far Non-502 CUi,t + βXi,t + αt + αj + εi,t .
Dependent variables are aggregated to the dispensary month level from transaction-level data. Each column reports the effects
of banking on the following dependent variables: Column (1) total transactions, which is defined as the total number of retail
transactions that occur at a dispensary in a given month; Column (2) the log of sales volume, which is the total number of
transactions weighted by the usable weight of each sale; and Column (3) the log of the number of strains sold, which is a count
of the total number of strains that the dispensary sold in a given month. Standard errors are clustered at the dispensary level.
∗ ∗∗
, , and ∗∗∗ denote significance at the 10%, 5%, and 1% levels.
Sales
(1)
Ln(Total Transactions)
Firms within 2mi of 502 CU
Firms within 2-4mi of 502 CU
Firms within 4-6mi of 502 CU
Firms within 6-8mi of 502 CU
Firms within 8-10mi of 502 CU
Firms within 2mi of any CU
Firms within 2-4mi of any CU
Firms within 4-6mi of any CU
Firms within 6-8mi of any CU
Marijuana firms w/in 5 miles
R-squared
Number of Observations
Number of Clusters
Year FE
County FE
Location Type FE
Cluster Variable
-0.082
(0.242)
-0.218
(0.233)
-0.253
(0.307)
-0.021
(0.304)
-0.064
(0.283)
-1.839∗ ∗ ∗
(0.023)
-2.167∗ ∗ ∗
(0.183)
-1.766∗ ∗ ∗
(0.180)
-2.743∗ ∗ ∗
(0.158)
0.000
(0.003)
0.337
5,703
280
Yes
Yes
Yes
Dispensary
(2)
Ln(Sales Volume)
0.146
(0.245)
-0.106
(0.234)
-0.170
(0.299)
0.016
(0.284)
-0.031
(0.272)
-1.784∗ ∗ ∗
(0.015)
-2.057∗ ∗ ∗
(0.191)
-1.677∗ ∗ ∗
(0.192)
-3.074∗ ∗ ∗
(0.174)
-0.001
(0.003)
0.342
5,703
280
Yes
Yes
Yes
Dispensary
(3)
Ln(Number of Strains Sold)
0.196
(0.178)
-0.013
(0.172)
0.106
(0.235)
0.292
(0.249)
0.002
(0.210)
-0.831∗ ∗ ∗
(0.048)
-0.915∗ ∗ ∗
(0.137)
-0.677∗ ∗ ∗
(0.127)
-1.383∗ ∗ ∗
(0.129)
-0.002
(0.002)
0.329
5,703
280
Yes
Yes
Yes
Dispensary
49
Appendix A
Survey
Survey Questions
What type of bank account does [your business] use? Please mark all the apply.
• Business bank account
• Owner’s personal bank account
• No bank account
• Other
If [your business] does bank, which best describes [your business]’s banking? (Please
mark all that apply).
• Community credit union
• Regional credit union
• Community bank
• Regional bank
• National bank
• Other
How much does [your business] pay per month, in dollars, to use all financial services?
These are defined as a checking account, savings account, money market account, and
access to loans from a Bank or credit union, a Non-bank financial institution (finance
companies, insurance companies, brokerage or mutual fund companies, leasing companies,
mortgage banks, private equity, etc.), or Other (friends, family, other companies, other
individuals, etc.).
• Bank or credit union
• Non-bank financial institution
• Other
What forms of payment do you accept? (Please mark all that apply).
• Cash
• Check
50
• Debit cards
• Credit cards
Do you pay for a security company that transports cash and/or product (if yes how
much per month, in dollars, do you pay to your security company?
What is the highest level of education you have?
• Less than 9th grade
• Some highschool but no diploma
• Highschool graduate (diploma or equivalent diploma GED)
• Technical, trade, or vocational degree
• Some college, but no degree
• Assoicate’s degree
• Bachelor’s degree
• Some graduate school but no degree
• Master’s degree
• Professional school or doctorate
Did either of your parents ever own or run a company?
• Yes
• No
Appendix B
Additional Results
Tables 4–6 report detailed evidence on sales decomposed into total transactions, sales
volume, and number of strains sold using our IV and PDFE specifications. These tables
suggest that banking services increase sales volume and the number of stains sold, the
evidence on transactions is mixed. In particular, we find that sales volume is 12 to 16
percent higher for firms using banking services across different specifications. Similarly we
find that the number of strains sold is 9 to 19 percent higher for firms using banking
services across different specifications. The estimates, however, are noisy and are
statistically significant at conventional levels only in the instrumental variable specification.
51