Upload
others
View
3
Download
0
Embed Size (px)
Citation preview
The Taxation of Recreational Marijuana:Evidence from a Reform in Washington State
Benjamin Hansen∗
University of Oregon, NBER, IZA
Keaton Miller∗
University of Oregon
Caroline Weber∗
University of Oregon
April, 2017PRELIMINARY
Abstract
A central question surrounding the legalization of marijuana is the rate at which itshould be taxed. We speak to this question by taking advantage of a unique naturalexperiment in Washington state. On July 1, 2015, Washington switched from taxing 25percent of gross receipts at each step of production (cultivation, processing, and retail)to a sole 37 percent excise tax at retail. We use administrative records from Washingtonthat record each step in the supply chain to assess how the market responded tothe change. We find the previous tax regime provided strong incentives for verticalintegration. Because the shift in tax regimes was approximately revenue-neutral – itonly meant to change who paid the taxes – we should have seen a fall in processorprices and no change in the tax-inclusive retail price. However, we find the processorprice only falls by about 25 percent of what classic tax-invariance theory would predict.The remaining increase in consumer taxes is split about 50-50 between consumers andretailers. We estimate the elasticity of demand at the market level to be -0.81. Weconclude a retail tax rate on marijuana of 37 percent remains on the correct side of theLaffer curve.
JEL Codes: H2, H3, H7, I1, K4.Keywords: Marijuana, Excise Taxes, Pass Through, Tax Incidence, Vertical Inte-
gration
∗University of Oregon, Eugene OR, 97403-1285. Hansen: [email protected]; Miller:[email protected]; Weber: [email protected]
1 Introduction
The United States reached a tipping point on the public attitude towards the legalization
of marijuana in 2015 (Motel, 2015). Prior to this time, the majority of adults opposed legal
marijuana. This could be a watershed moment for federal policy, as prior tipping points
on other social issues have typically been followed by broad social reforms. Examples come
from legislation regarding gay marriage, civil rights, the repeal of prohibition, or the recent
taxation of smoking. With median voters now favoring legalization, many have suggested it
is a question of ‘when’ rather than ‘if’ marijuana will be legalized at the federal level. Four
states, Alaska, Colorado, Oregon and Washington currently have legally operating recre-
ational marijuana markets. California, Maine, Massachusetts, and Nevada voted to legalize
recreational marijuana during the 2016 elections.
Taxation and revenue generation has been one of primary political and economic ar-
guments for legalizing recreational marijuana (Miron and Zwiebel, 1995). However, despite
the existence of sizable black and quasi-legal medical markets for marijuana, little is known
about what legal recreational markets for marijuana will look like or how they will evolve,
and what the optimal taxation policies may be. Previous research in alcohol and cigarette
markets have found evidence that tax pass-through rates can exceed 1 (Kenkel, 2005; Bar-
nett, Keeler, and Hu, 1996). In addition, consumer responsiveness to taxation can depend
on geographic proximity to other markets (Harding, Leibtag, and Lovenheim 2012), which,
in the absence of federal policy changes, may be particularly relevant as individual states
slowly change their policies over time.
Other concerns also arise with marijuana legalization and taxation proposals. Continuing
1
black markets and medical marijuana offer readily available substitutes if taxation drives
prices too high.1 Previous estimates and approaches in the literature have largely estimated
the elasticity of demand for marijuana in the illegal market to range from -.1 to -.5 (Jacobi
and Sovinsky, 2016; Miron and Zwiebel, 1995; Pacula et al., 2000; Clements and Zhao, 2009;
Donohue, Ewing, and Peloquin, 2011; Williams, van Ours, and Grossman, 2011). These
estimates have typically relied on survey responses and prices collected from drug seizures,
or crowd-sourced data.
We offer new evidence concerning these key policy and economic questions by examining
Washington state, where recreational marijuana was legalized on November 12, 2012. The
detailed tracking system implemented by the state government to regulate the legal market
affords us the unique ability to observe the evolution of product prices, quality, and variety
as producers, processors and retailers enter into this newly available legal marketplace. These
unique “seed-to-sale” administrative records track the cultivation, processing, and retail sale
of recreational marijuana, thereby offering us the opportunity to explore the evolution of
the legalized marijuana industry. In contrast to previous work, we estimate the demand for
marijuana in the legal market using a data set that represents the universe of recreational
marijuana transactions in one of the first legally operating recreational markets.
To examine the effect of market regulation and taxation on outcomes in the industry,
we take advantage of a major overhaul to Washington’s marijuana tax policy via House
Bill 2136. Prior to July 1, 2015, a 25 percent excise tax was assessed at each transfer of
marijuana (i.e. at cultivation, wholesale, and retail). This is a type of gross receipts tax
1Raising taxes too high could also perpetuate the black markets legalized marijuana aims to supplant –particularly if legal producers are able to divert a portion of their output.
2
because gross sales are taxed at each stage of production (and costs are not deducted). After
July 1, 2015, the only tax collected was a 37 percent tax at the retail level. Crucially, this
change was unexpected by market participants – it was passed during a special session of
the Washington Legislature on June 27, 2015 and signed by the Governor on June 30.2 This
allows us to estimate the effect of these large policy changes on key behaviors and decisions
made throughout the production process.
From the estimated tax-inclusive retail price increase and corresponding changes in quan-
tities purchased by consumers, we estimate the market-level elasticity of demand to be -0.81.
This is somewhat larger than most of the illegal or medical marijuana estimates. In the
current legal recreational markets more substitutes are available. For instance, recreational
marijuana consumers could switch to medical marijuana, black market marijuana, or other
drugs such as alcohol if the price were raised sufficiently via taxation. Furthermore, the
legal status of marijuana may encourage market participation by individuals who had not
previously consumed marijuana who are likely to have higher individual price elasticities.
Given these differences in demand behavior between legal and illegal markets, our estimates
are meaningful for policy makers in those states concerned with the appropriate taxation of
marijuana and where they sit on the Laffer curve.
Since the tax shift from producers and processors to retailers was designed to be approxi-
mately revenue-neutral, the classic tax-invariance result would predict that processors would
respond to the tax change by lowering their price by the full amount of the change. The tax-
inclusive retail price would, under this hypothesis, stay approximately constant. However,
2Contemporaneous media reports suggest that, although the change was supported by the industry, itwas not expected to pass, and market participants did not have confidence in their forecasts of future pricesat the time of the change (La Corte, 2015).
3
though processors paid an average of $1.03 in taxes for each gram sold prior to the change,
we find that processors only lower their price by an average of 24 cents per gram, leaving the
remaining burden to be split approximately 50-50 by retailers and consumers. This stands in
contrast to the literatures on cigarette and gasoline tax incidence, among others, which gen-
erally finds consumers bear the substantial majority of the tax burden (Kopczuk et al., 2016;
Harding, Liebtag, and Lovenheim, 2012). One plausible explanation for the difference is that
our estimated elasticity of demand for marijuana is higher than the consensus estimates for
cigarettes or gasoline. Alternatively, tight ownership and size restrictions in Washington’s
marijuana market may lead to differential market power effects or frictions relative to these
other markets.
We also find the original tax regime strongly encouraged vertical integration. Via Wash-
ington law, vertical integration was only possible for producers and processors (banning final
retail outlets). Before the change, roughly 90 percent of marijuana was processed through
vertically integrated firms. This drops following the tax change. Notably, this is primarily
due to a 300 percent increase in the processing of marijuana from non-vertical producers,
rather than a decrease in the processing of vertical marijuana production.
In addition to contributing to the understanding of tax policy in the marijuana mar-
ket, our vertical integration findings also contribute to several broader long-standing public
finance questions. It is theoretically obvious that vertical integration would be a natural con-
sequence of a gross receipts tax and tax economists frequently come out vehemently against
gross receipts taxes for this and other reasons (e.g., McClure, 2005; Pogue, 2007; Testa and
Mattoon, 2007); however, this is the first paper we are aware of that provides compelling
empirical evidence of this behavior. Gross receipts taxes have begun to proliferate across
4
states in recent years, so this paper provides an important source of empirical evidence that
such taxes do, in fact, lead to inefficient levels of vertical integration. Our findings also con-
tribute to the discussion on tax-collection invariance on two dimensions (Kopczuk et al. 2016;
Chetty, Looney and Kroft, 2009). First, we highlight that the possibility of vertical integra-
tion under a gross receipts tax is another reason tax-collection invariance will no longer be
expected to hold. Second, we document that tax collection at the retail level has different
incidence implications than does tax collection at the processor level. In contrast to previous
work, there is no evidence in this case that the result is driven by different tax evasion pos-
sibilities at different points in the supply chain (Kopczuk et al. 2016), nor is it likely driven
by a relative lack of awareness of the tax by one side of the market (Chetty, Looney, and
Kroft, 2009). Instead, media reports and conversations with industry participants suggest
processors took advantage of a unique opportunity to increase margins.
Our findings additionally contribute to broader research on supply side interventions in
drug markets, which has included examinations of policies ranging from alcohol’s prohibi-
tion (see Miron and Zwiebel, 1991), to limitations on precursors to methamphetamines (see
Dobkin and Nicosia, 2009) and the legalization of marijuana sale and cultivation for medical
purposes (Anderson, Hansen and Rees, 2013). Our research is some of the first to study
legalized marijuana markets for recreational use, and the first to analyze the substantial nat-
ural experiment the tax change offers. And while prior papers have focused on the elasticity
of demand for illegal markets, we are the first, to our knowledge to study the market-level
elasticity of demand in a legal marijuana market.
The remainder of the paper proceeds as follows. In Section 2, we discuss the history
of marijuana’s legalization and the associated tax system in Washington. In Section 3 we
5
discuss the administrative data we utilize and the methods we use to estimate responses to
the policy change. We present our results in Section 4. We conclude by discussing the policy
and economic implications of our findings, and potential avenues for future work.
2 Background
Prior to 1938, marijuana was a legal substance in the United States. Indeed, it was listed in
the United States Pharmacopeia as a prescription for labor pains, nausea, and other condi-
tions. Since the passage of the Marijuana Taxation Act of 1938, marijuana has been consid-
ered illegal. The advent of Scheduled substances via the Controlled Substances Act of 1970
significantly strengthened the prohibition against marijuana, as it was quickly classified as a
Schedule I substance. This places it in the same category as heroin and methamphetamine,
substances with the highest potential for abuse and little known medical benefit.
In 1996, California became the first state to legalize marijuana for medical use. Soon
after, Washington also enacted its own medical marijuana law in 1998 under Washington
Initiative 692. Currently, 27 states and regions (including Washington D.C and Puerto Rico)
permit its cultivation and use for medical reasons. In response to the growing acceptance of
the medical use marijuana, in October 2009 the United States Department of Justice issued
a memorandum to United States Attorneys (Ogden, 2009) discussing the appropriate way to
allocate resources in states with legal medical marijuana markets. In particular, the memo
stated “federal resources in your States [should not be focused] on individuals whose actions
are in clear and unambiguous compliance with existing state laws providing for the medical
use of marijuana.” This was broadly interpreted as an effort to defer to states’ choices in
6
the absence of federal consensus (Stout and Moore, 2009), though the Department of Justice
emphasized a need to investigate and prosecute “drug traffickers who hide behind claims of
compliance with state law.”
In the November election of 2012, Washington voters faced ballot initiative 502. The ini-
tiative legalized the recreational possession and consumption of small amounts of marijuana
for adults over 21. Three types of licenses were created: producers, who are permitted to
grow and harvest cannabis plants, processors, who transform the harvested plant into usable
marijuana and other products for wholesale, and retailers, who may sell final products obtain
at wholesale to consumers. It also laid out that taxes would be collected with the revenue set
aside for education, healthcare, and substance abuse prevention. The initial tax structure as-
sessed a 25 percent tax on each marijuana transaction. This includes when grown marijuana
is sold to processors who convert the harvested plant material into usable marijuana, when
processors sell the usable marijuana to retailers and when retailers sell it to end consumers.
Producer licenses come with capacity constraints – each producer is allocated one of three
sizes of plant canopy 3 Firms which grow marijuana may also have a license to process it, and
vice versa. These vertically integrated firms did not owe any taxes when they transferred the
grown marijuana to their processing operation. However firms involved in the production or
processing of marijuana are forbidden from owning and operating a retail location. Holders
of a retail license may operate up to three locations. Broader ownership restrictions exist as
well – individuals are not allowed to have direct or familial financial interests in more than
one marijuana license.
3According to Washington law (WAC 314-55-010), “ ‘Plant canopy’ means the square footage dedicatedto live plant production, such as maintaining mother plants, propagating plants from seed to plant tissue,clones, vegetative or flowering area. Plant canopy does not include areas such as space used for the storageof fertilizers, pesticides, or other products, quarantine, office space, etc.”
7
The initiative gave regulatory authority to the newly-renamed Washington State Liquor
and Cannabis Board. One of the early actions the new board took was the implementation
of a state-operated traceability system which would track the cultivation, testing, process-
ing, and retail sales of marijuana throughout the state. We provide more details on these
administrative data in Section 3.
The federal Department of Justice responded to the changing environment in Washing-
ton and Colorado4 with an August, 2013 memo – commonly known within the industry as
the Cole Memo (Cole, 2013). As with its response to changing views on medical use, the
Department emphasized the prohibition on production and consumption of marijuana un-
der federal law, but provided guidance as to specific enforcement priorities. These included
concerns about diversion of products from the legal market to illegal markets or jurisdictions
without legal markets, as well as public health concerns associated with marijuana consump-
tion. Importantly, the Department set a clear expectation that “states and local governments
... will implement strong and effective regulatory and enforcement mechanisms.” Within the
industry, the traceability system is seen as the implementation of the expectations and pri-
orities laid out within the Cole Memo.
The tax reform analyzed in this paper was part of House Bill 2136 introduced during
the 2015 Regular Session of the Washington Legislature. As our identification rests on the
assumption that the policy change was effectively a set of exogenous cost shocks throughout
the supply chain, the details of the bill’s history are critical. Table 2 provides a detailed
4Colorado voters also approved a legalization effort in November, 2012. We focus on Washington in thispaper for two primary reasons. First, in contrast to Washington’s vertically separated market, Colorado’ssystem enforces mandatory vertical integration – each retailer sells product they have grown individually.Additionally, while both states collect similar traceability data, Washington’s legal framework requires publicdisclosure of the data and Colorado’s forbids it.
8
timeline of the bill’s progress through the Washington Legislature. The bill originated in the
House midway through the 2015 Regular Session, and accumulated a number of amendments
and substitutions before being passed by the full House. However, while a Senate committee
recommended passage on the last day of the session, the full Senate declined to consider the
bill during the session. A similar pattern occurred in the First Special Session – the House
quickly passed the bill, and the Senate chose to take action. It wasn’t until the very end of
the Second Special Session, June 27, that the bill received a vote by the full Senate. The
Governor signed it on June 30, and the law went into effect the next day.
While the tax change is the most relevant part of the bill for our purposes, the bill, along
with companion legislation, contained a number of other measures, which arguably may have
played a larger role in the internal political process leading to its eventual passage. Within
the recreational market, increased funds from tax collections were made available to local
jurisdictions, on the condition that they participate in the market. Local jurisdictions also
obtained greater zoning flexibility for marijuana businesses. Finally, Washington’s medical
marijuana market, previously legal though essentially unregulated, was brought into the
regulatory framework created by the original initiative legalizing recreational use.
Today, seven states have legalized marijuana for recreational use. Table 1 delineates the
tax structure within each state. Notably, Washington applies the highest tax rate of 37% –
the next highest is neighboring Oregon, with a 17% tax. Not all states apply taxes at retail –
Nevada applies its tax of 15% at wholesale. California applies a cultivation tax designed as a
fee for each ounce of dried plant flowers and leaves, while Colorado applies a tax at wholesale
based on the average price per gram state-wide. All of these taxes lead to a lower effective
rate than Washington’s – implying that if Washington is on the left side of the Laffer curve,
9
these other states almost certainly are as well.
3 Data and Methods
Our data consist of administrative records from the Washington State Liquor and Cannabis
Board (WSLCB). As a part of the legislative effort to legalize and regulate recreational
marijuana production and consumption in Washington, the state implemented a traceability
system (also known as a seed-to-sale system) produced by BioTrackTHC. Its purpose is
to track each step in the marijuana supply chain, enabling regulators to collect taxes and
prevent diversion to the black market. The state provides an API to market participants and
requires timely reporting of many details of the production process. Firms generally use one
of several commercially available software packages to report data to the traceability system
and comply with state data-reporting requirements. The end result is data that tracks a
marijuana plant’s planting, harvest, production into usable goods, and final retail sale. Along
the way, certain products are also selected and tested for tetrahydrocannabinol (THC),
tetrahydrocannabinolic acid (THC-A), and cannabidiol (CBD), the primary psychoactive
ingredients, as well as foreign contaminants and moisture content to comply with product
quality requirements.
We obtain an extract of the state’s database in SQL format that removes most personal
information and other information about the supply chain that are subject to security and
privacy concerns.5 Firms, locations, and production rooms are given unique identifiers. Each
5For example, our extract does not include data on individual employees. Additionally, while firms arerequired to report itineraries and planned routes for marijuana transfer operations (e.g. when a wholesalermakes a delivery to a retailer), our extract does not include them.
10
plant is registered at the time of planting. Firms record the provenance of the plant material
(e.g. a clone or a seed) as well as the strain6 and a log tracks its movement through the
growth process. Once harvested, flowers and other plant material are generally collected into
lots and converted into a new inventory unit which is assigned a unique identifier. These
intermediate products may progress through several processing steps including conversions,
combinations, and divisions before becoming a final retail lot for wholesale. Along the way,
the database records these transformations, generating new inventory ids as necessary.
When wholesale lots are sold to retail locations, the tracking system records the date,
quantity and price of the transaction. Transfer manifest identifiers allow us to observe whole-
sale transactions that comprise multiple product lots. Similarly, the system tracks individual
dispensing events, linking the prices and quantity of different items, as well as the transaction
time,7 in a dispensing transaction to the relevant inventory lots, allowing us to trace sales
back through to a set of original plants.8
We consider both tax-exclusive and tax-inclusive prices. The tax-inclusive prices include
both the marijuana-specific excise tax and Washington’s general and locality-specific sales
taxes.9 Given the retail firms’ inability to access traditional financial services markets, almost
6Strains are defined by the producer and consequently are the “dirtiest” field in our data. Additionally,the system allows producers to input the seed-grown offspring of two cloned parents as a member of thatclone strain, even though the offspring are not identical.
7Washington’s regulatory framework does not require retailers to have a constant connection to thetracking system. Many connect to the system at the end of their business day and upload their transactionswithin one session. While the order of transactions is maintained and other information in the system(specifically the inventory log file) may allow us to track the specific time of each transaction, we choose thedaily level as our most granular view of activity in the industry.
8Given the details of the production process discussed in this section, it is not possible to preciselyidentify which plant or plants a particular retail package of marijuana came from – we can only identifythe set of plants which contributed to the creation of the wholesale lot from which the retail package wasdivided.
9The tax reform also changed the way firms reported prices in the traceability system. Prior to thereform, firms reported fully tax-inclusive prices. Afterwards, firms reported fully tax-exclusive prices.
11
all choose to set tax-exclusive prices that lead to round numbers when taxes are included
(this lowers the cash handling costs for firms). As a consequence, the posted prices, and the
prices faced by consumers, include all taxes and therefore, in contrast to many other settings,
the sales taxes are salient to consumer-level decision making.
We analyze the effects of the tax change on a number of observable behaviors of market
participants through a series of regression exercises. We restrict our analysis at the processor
and retail levels to the “usable marijuana” product category – by far the most popular
category.10 For each component of the supply chain—producers, processors, and retailers—
we collapse the data by firm-day after performing minor cleaning steps. We drop a small
number of duplicate sales in the processor data and we drop all retail transactions that were
deleted or refunded. We also drop all processor and associated retail transactions in which
the processor price or total transaction amount was less than or equal to one cent. It is
our understanding that these were generally samples given to the retailer by the processor;
they were given a price of one cent so that these movements of marijuana could be tracked
within the traceability system. Lastly, we drop some remaining extreme outliers in the data.
In particular, we drop all observations at the producer level if the number of plantings,
harvestings, or days from plant-to-harvest are outside the 0.5th or 99.5th percentiles of their
respective distributions. We also drop all transactions at the processor and retail level in
10‘Usable marijuana’ is defined by Washington state law as “dried marijuana flowers, [excluding]marijuana-infused products [and] marijuana concentrates.” In practice, usable marijuana is consumed ei-ther through the use of a fixed apparatus or by rolling the flower into a “joint” with paper produced forthe purpose. Though the traceability system has a unique code for products that fall into this category, itcontains two types of products: both raw dried flowers and pre-rolled joints, which include some value-add.Pre-rolled joint products are also listed under a different inventory type code. As we cannot cleanly dis-tinguish between raw flower and pre-rolled joints in the “usable marijuana” category, our analysis includesthe entire “usable marijuana” category as well as any pre-rolled joint products we can identify across othercategories.
12
which the price or weight are outside the 0.5th or 99.5th percentiles of their respective
distributions. Lastly, we censor the THC content data at its 0.5th or 99.5th percentiles.
Our primary target in our regression analyses is the response to the tax change. As
the change took place within the broader context of the market’s non-linear evolution, our
estimating equations include a polynomial in time. Furthermore, as the change was unex-
pected by market participants, we include several dummy variables in the days immediately
surrounding the tax change, to account for short-term adjustment effects. The detail of the
data collected by the traceability system allows us to analyze behaviors at the firm-day level,
though cyclical patterns throughout the course of a week and a month cause us to include
additional fixed effects.
Our analyses, therefore, use the following template:
log(yit) =α0 + α1taxit +6∑
j=1
α3jtax dayj +6∑
k=1
α3kdowk +31∑l=1
α4ldoml
+5∑
m=1
α5mdatemit + uit.
(1)
where yit is our outcome variable for firm (producer, processor, or retailer) i at date t, taxit
is an indicator variable that is one after the tax change took place on July 1, 2015 and zero
before, tax dayj are indicator variables for June 30 - July 5 to absorb local responses to the
tax change and the 4th of July, dowj are day of the week fixed effects, domk are day of the
month fixed effects, and datelit is a polynomial in the date of sale.
We take the logs of all of our outcomes, unless otherwise specified, because the outcomes
we examine are essentially log-normally distributed (with the added benefit of allowing us
to interpret the estimated coefficients on the binary regressors as semi-elasticities). Standard
13
errors are clustered by firm location. Given the details of our estimating equations, we
interpret coefficients on the tax change indicator variable to represent the average medium-
term response to the change.
4 Results
We first report our findings on the producer market, then discuss the processor market, and
finish with the retail market. Across markets, we present Tables consisting of point estimates
for each of the outcomes we examine, and Figures detailing the data behind outcomes of
particular significance. In the Figures, the solid line is a fitted model based on the equation
(1), while the hollow circles represent average outcomes at the producer level for the days
leading up to and after the tax change.
4.1 Producers
We first focus on producers – those firms that plant and harvest raw marijuana flowers.
Note that vertical integration is rampant, as over 95 percent of growers also have a license
to process marijuana. Table 3 provides baseline summary statistics for growers. The July 1
policy reform decreased their transaction tax from 25 percent to zero. However, we note that
around the time of the tax change, over close to 95 percent of the producers are vertically
integrated with processors. With the incentives this provides firms to charge prices close to
zero to avoid taxes, we focus on the counts of the number of new plantings, and the total
quantities of marijuana harvested.11
11Due to this vertical integration, the administrative data does not contain meaningful prices for manyof the transfers of plant material between producers and processors. in the administrative data many of
14
Table 4 provides point estimates for three outcomes of interest: the number of plantings,
the number of harvests, and the average number of growth days for plants harvested. The
left panel of Figure 1 illustrates the the counts of the plants harvested around the July 1
tax change. The harvest rate does not change in any significant way after the tax change.
However, the number of plants harvested are rare enough with outliers that detecting shifts
in growing behavior via the number of plantings might be challenging from a statistical
power perspective. For instance, when we examine the the number of days from planting
until harvesting, we do find evidence that number of days until harvest falls, by roughly
0.2 percent. So although some firms might be trying to harvest marijuana more quickly,
practically the response is minor. The empirical results are not sensitive to bandwidth or
allowing for an adjustment period from June 27 through July 5.
These estimates suggest that in the short run, firms did not drastically alter their planting
or harvesting in response to the tax shift. In some ways, given the vast amount of vertical
integration and that firms are producing at capacity due to strong demand, this is not entirely
surprising – particularly given the firms did not have a long period to adjust their production
process since the law was only passed days before the tax change went into effect.
4.2 Processors
The tax reform also eliminated the transaction taxes faced by processors – those firms which
take raw marijuana flowers and plant material as inputs and transform them into usable
marijuana and other products. Importantly, before the change, the processors paid the taxes
the processor prices are missing, are filled in with prices arbitrarily close to zero. Due to the difficulty ofinterpreting missing values, we focus on quantity outcomes that auditors routinely verify among growers.
15
after the transaction, implying the retailers paid the tax-inclusive price, while processors
received the after-tax price. Table 6 provides baseline summary statistics for producers.
As shown in Figure ??, the after-tax price for processors increases dramatically following
the tax change. This suggest that 25 percent tax change was in large part a huge boon to
processors, while it simultaneously reduced the variable costs of retail firms. At the same,
the quantity of marijuana processed and sold to retail firms did not alter significantly around
the tax change. This suggests two things. First, it provides strong evidence the firms did not
anticipate the taxes. The substantial shift in taxes should give firms incentives to hold back
production and sales to retail firms until after the processor taxes are eliminated. Only the
day before the tax change do we find the quantity of marijuana substantially decrease (indeed
June 30 exhibits a true outlier, consistent with tax avoidance on the part of processors, albeit
extremely short-sighted avoidance).
In Table 7, we provide point estimates of the effect of the tax change on the following
potential mechanisms through which processors might have altered their behavior: prices
(those charged to retailers), after tax prices (those paid by the processor), weight (total
weight of marijuana sold in grams), sales (total number of sales), after-tax revenue, and
THC levels. The only significant findings are on prices – we find that average tax-exclusive
prices (the price that retailers paid to processors) fell by 6 percent. Simultaneously, the
tax-inclusive price (the price received by processors) increased by 22 percent.12 We also find
the the quantity of sales transactions, total weight of marijuana sold, and THC levels were
essentially unchanged.13 This suggests that processors received the majority of the benefits
12The log change in the tax rate is 28% (i.e. log((1-.75)/1)=0.287).13Within the “usable marijuana” product segment, THC levels are effectively fixed by the production
process. However, processors may choose to differentially allocate plant material of varying quality to differentproduction processes (including those creating products falling outside the “usable marijuana” category and
16
from the tax realignment in Washington, although retailers also received about 25 percent
of the benefits of the reduction in the processors’ taxes.
Table 8 details several variations of our preferred specification for the significant after-tax
price response observed in columns (1) and the insignificant change in weight observed in
column (3) of Table 7.
Figure 5 illustrates the fraction of transactions that are vertically integrated, and the log
weight of transactions which are not vertically integrated. Point estimates are provided in
Table 9, with robustness checks for the outcomes illustrated in the Figure provided in Table
10. Vertical integration appears to fall, which is driven by a nearly 300 percent increase in
the number of transactions that occur between non-vertically integrated firms. This pro-
vides evidence that the deadweight loss of the processor tax was realized in part through
discouraging otherwise efficient trades between producers and processors from occurring.
In summary, our findings suggest processors (not surprisingly) widely benefited from the
elimination of their transaction taxes with retailers. The tax benefits are shared between both
the retailers, whose price paid for marijuana fell by 6 percent on average, and the processors,
whose price received after paying taxes rose by 22 percent. We find in the medium run
(when using monthly data), that the number of non vertically integrated transactions rose
significantly, although the vertically integrated transactions continue to dominate the market
in absolute size. We also find no significant increase the quantity of marijuana supplied from
processors to retailers, other than a 1 day anticipation effect and a few days adjustment period
(in which the sales withheld the day before the tax are sold to retail firms). This suggests
that in the short run, the tax changes did not fundamentally alter the supply of marijuana.
thus outside of our analysis). Our null result here suggests firms did not substitute in this way.
17
This altogether suggests that in the short-run window surrounding the tax change, supply
is relatively inelastic – not surprising given the requirements of the production process.
4.3 Retail Market
We now use similar estimating equations to examine the consumer retail market. Given that
we can also observe a portion of any given retailer’s input prices and those of their nearest
competitors, we also include as controls prices paid to processors (their variable costs) and
the prices of local competitors. Summary statistics are provided in Table 11.
Point estimates across a variety of outcomes are detailed in Table 12. We include esti-
mates for tax-exclusive and -inclusive prices, weight, the number of sales, tax-exclusive and
-inclusive revenue, and THC content. We find evidence that tax-exclusive prices, the prices
retailers received, fell by 6.4 percent. In our preferred specification, we find tax-inclusive
prices, the prices paid by consumers, rise by 2.7 percent. Given that we know the tax rate is
increasing from 25 to 37 percent, this amounts to a 9.6 percent increase in the tax rate (1+τ).
Combining our estimates on the change in tax-inclusive price with the change in weight, the
implied, average market-level elasticity of demand is -.81. This suggests Washington is on the
part of the Laffer curve where higher taxes on the margin would increase revenue – supported
by column (6), which reports a slightly positive effect on tax-inclusive revenue. However, the
decrease in tax-exclusive price leads to a significant decrease in revenue received by retail-
ers. Additionally, the average level of THC in the products sold decreased significantly as
well, which, when combined with the null results for THC in the processor analysis, suggests
consumers engaged in a degree of product-level substitution in response to the increase in
18
tax-inclusive prices – suggesting the product-level price-elasticities faced by firms throughout
the supply chain may be higher. Figure 6 illustrates the shifts in tax-inclusive price and total
weight surrounding the tax reform.
With estimates of the changes in prices for both the processor and retailer markets
in hand, we turn to the incidence implications of these results. Figure 8 summarizes the
components of the tax-inclusive price charged by the average firm for a gram of marijuana
before and after the tax change in the left and right columns. We consider how much goes to
processor and retail taxes as well as how much is spent to purchase a gram, on average, from
the firm. Before the tax change, all prices and taxes (in dollars) are based on the average
prices the month prior to the tax change. After the tax change, these numbers are the pre-
tax change prices adjusted by our estimated changes caused by the tax changes. This holds
constant the composition of the market and eliminates any secular trends in prices.
The tax changes that were implemented on July 1, 2015 were designed to be approx-
imately revenue neutral, and, in practice, they were. We calculate that the average taxes
collected on a gram of marijuana before the tax change was $3.52 ($1.03 paid by the pro-
cessor and $2.49 paid by the consumer) and the average taxes paid on a gram of marijuana
after the tax change was $3.45 (all paid by the consumer).14 If the classic result of tax-
collection invariance holds in this setting, we would expect processor prices to fall by the
amount of the tax formerly paid by processors and retail prices to remain approximately
constant after the tax change – this outcome is illustrated as the middle column of Figure
8. Our findings are somewhat different. First, we found in Section 4.2 that processors cut
14We ignore the taxes applied to producers in this discussion because over 90 percent of the market wasvertically integrated, and thus did not have to pay the tax and our price data for the remaining producersis not of high-enough quality to consider this part of the tax change directly.
19
their prices to retailers by only 5.9 percent; hence, they keep about 75 percent of the tax
cut for themselves. Put another way, of the 96 cents increase in taxes on consumers due to
this reform, processors pay only 24 cents, on average, or 25 percent of this tax burden. The
other 72 cents are split between consumers and retailers. We find in this section, on average,
retailers increase their prices to consumers by 2.7 percent, which translates into a 34 cent
increase of the average price of a gram of marijuana; the retailer covers the other 38 cents
of the tax increase. Of the portion of the increased tax on consumers not covered by proces-
sors, the retailers and consumers split the burden roughly 50-50. This stands in contrast to
the literatures on cigarette and gasoline tax incidence, among others, which generally finds
consumers bear the substantial majority of the tax burden (Kopczuk et al., 2016; Harding,
Liebtag, and Lovenheim, 2012).
One plausible explanation for the difference is that our estimated elasticity of demand
for marijuana is higher than the consensus estimates for cigarettes or gasoline, and, in the
short-run, the supply from processors is inelastic, given the agricultural cycle. Contempo-
raneous media reports and interviews with market participants suggest processors saw the
reform as a unique opportunity to ‘claw back’ margins from retailers (La Corte, 2015). Al-
ternatively, though the state contains hundreds of producers, processors, and retailers, each
individual retail firm is connected to a relatively small number of suppliers, and consumers,
particularly in cities such as Seattle, may have access to far more retailers. This unbalanced
level of competitive may force retailers, particularly in the short term, to accept higher prices
from processors (relative to the after-tax price received from those processors before the tax
change) without being able to pass those prices on fully to consumers.
20
4.3.1 Firm-Level Retail Analysis
Given the richness of our data, we can estimate the tax-inclusive price and weight responses
separately for each retail firm in our data set. Figure ?? highlights the large degree of
heterogeneity in pass-thru rates of the increased excise tax, both positive and negative, to
consumers. In Figure ?? and Table ??, we explore the underlying firm characteristics that
determine the wide variation in pass-thru rates that we have observed at the firm level. Both
the table and figure highlight that processor pass through is a key predictor. In other words,
the more of the tax benefits the processor shared with a retail firm, the more of the tax
increase the firm was willing to absorb rather than passing it through to consumers. Table
?? estimates that a 10 percent decrease in processor pass-thru decreases retail pass-thru by
2.99 percent, and with city fixed effects included in the model, this estimate approximately
doubles. Likewise, firms that on average charged higher prices before the tax change passed
through less of the tax change to their consumers. Furthermore, retailers further away from
the Oregon border and with greater populations passed through more of the tax to their
customers.
5 Conclusion
Following California’s approval of marijuana legalization during the 2016 elections, mari-
juana is now legally available in states affecting 66,224,419 of the residents in the United
States, or 21 percent of the population. Given the broad societal shift in support for the
legalization of marijuana driven in part by the desire to increase state revenues, it is crucial
to accurately gauge how revenue will respond to the changes in marijuana taxation, on the
21
margin. While previous literature has studied the elasticity of demand for marijuana within
the black market, ours is the first to focus exclusively on the legal market for recreational
marijuana. And while some recent studies, notably Jacobi and Sovinksy (2016), have esti-
mated the elasticity of marijuana to be -.2, our findings suggest the elasticity of demand is
-.81.
While all of our evidence suggests that we are to the left of the peak of the Laffer curve,
our findings also suggest the recreational market market demand may be more elastic than
previous studies of the illegal market. Given the number of readily available substitutes
(including medical marijuana, black market marijuana, and other drugs) perhaps a higher
elasticity of demand is not entirely surprising. We also find firm level demand is more elastic
than the market-level demand, a finding consistent with the idea that these firms seek to
profit-maximize, despite the fact that most retailers are run by individuals with relatively
little experience in retailing or as an entrepreneur. We also find evidence of product-level
substitution in response to the tax change which suggests that specific products are more
elastic than marijuana overall. The substitution across products reduces the revenue gener-
ated by the taxes. This type of behavior has been noted before in markets for tobacco, while
we are the first to document this in legal marijuana markets. Further research, however, is
required to estimate price elasticities at the product-level.
Lastly, we find the retailers bear most of the incidence of the tax change in the short run.
If the elasticity of demand is inelastic, we might wonder why they do not pass more of the
tax to consumers. The simple answer to this that all the demand is inelastic, supply is even
more inelastic, at least in the short run. Although producers begin to plant more marijuana
and harvest it quicker, in the short run processors were unable to yield any more usable
22
marijuana (although are estimates suggests they held back shipments the day before the tax
change). Because supply does not shift in the short-run, retailers end up bearing most of the
tax incidence, despite the demand curve being inelastic.
With recreational marijuana’s continued expansion in the United States, our estimates
offer the first evidence on tax incidence and the elasticity of demand in recreational markets.
Given that the tax rates that were passed in California (15 percent), Massachusetts (12
percent), and Nevada (15 percent) are considerably lower than Washington’s tax, our findings
suggests considerable state revenue may be left on the table given the market level elasticity
of demand. It remains to be seen how these tax rates will change (e.g. will we observe state’s
competing) when or if federal policy shifts to allow marijuana’s transportation across state
boundaries.
23
References
[1] Anderson, D. Mark, Benjamin Hansen, and Daniel I. Rees. (2013). “Medical Marijuanalaws, traffic fatalities, and alcohol consumption.”Journal of Law and Economics. 56.2 :333-369.
[2] Anderson, D.M., B. Hansen and D.I. Rees. (2015). “Medical Marijuana Laws andTeenage Marijuana Use.” American Law and Economics Review. 17 (2): 495-528.
[3] Barnett, P. G., Keeler, T. E., & Hu, T. W. (1995). “Oligopoly Structure and the Inci-dence of Cigarette Excise Taxes.”Journal of Public Economics, 57(3), 457-470.
[4] Chaloupka, F. J., Grossman, M., & Tauras, J. A. 1999. “The demand for cocaine andmarijuana by youth. In The economic analysis of substance use and abuse: An inte-gration of econometrics and behavioral economic research”(pp. 133-156). University ofChicago Press.
[5] Clements, K. W., & Zhao, X. (2009). Economics and Marijuana: Consumption, Pricingand Legalization. Cambridge University Press.
[6] Dobkin, C., & Nicosia, N. (2009). “The war on drugs: methamphetamine, public health,and crime.”The American Economic Review, 99(1), 324-349
[7] Donohue III, J. J. (2013). “Drug Prohibition and Its Alternatives. Lessons from the Eco-nomics of Crime: What Reduces Offending? ”Lessons from the Economics of Crime’inLessons from the Economics of Crime, edited by PJ Cook, SJ Machin, O. Marie, andG. Mastrobuoni, MIT Press.
[8] Farrelly, M. C., Bray, J. W., Zarkin, G. A., Wendling, B. W., & Pacula, R. L. (1999).The effects of prices and policies on the demand for marijuana: Evidence from theNational Household Surveys on Drug Abuse (No. w6940). National Bureau of EconomicResearch.
[9] Harding, M., Leibtag, E., & Lovenheim, M. F. (2012). “The Heterogeneous Geographicand Socioeconomic Incidence of Cigarette Taxes: Evidence from Nielsen Homescan Data.”American Economic Journal: Economic Policy, 4(4), 169-198.
[10] Hall, W., & Lynskey, M. (2016). “Evaluating the public health impacts of legalizingrecreational cannabis use in the United States. ”Addiction, 111: 1764961773.
Kenkel, D. S. (2005). “Are Alcohol Tax Hikes Fully Passed through to Prices? Evidencefrom Alaska. ”The American Economic Review, 95(2), 273-277.
[11] Laffer, A. (2004). “The Laffer Curve: Past, Present, and Future.”
[12] Jacobi, L. & M. Sovinsky. 2016. “Marijuana on Main Street? Estimating Demand inMarkets with Limited Access.”American Economic Review. 106(8): 2009-45.
24
[13] McLure, C., 2005. “Why Ohio Should Not Introduce a Gross Receipts Tax Testimonyon the Proposed Commercial Activity Tax.” State Tax Notes, Apr. 18, 2005, p. 213.
[14] Miron, J., & Zwiebel, J. (1991). “Alcohol Consumption During Prohibition.”AmericanEconomic Review, 81(2), 242-247.
[15] Miron, J. A., & Zwiebel, J. (1995). The economic case against drug prohibition. TheJournal of Economic Perspectives, 9(4), 175-192.
[16] Motel, S. 2015. Six Facts about Marjuana. accessed fromhttp://www.pewresearch.org/fact-tank/2015/04/14/6-facts-about-marijuana/ on10/22/2016.
[17] Pogue, T. (2007). “The Gross Receipts Tax: A New Approach to Business Taxa-tion,”National Tax Journal, 60(4), 799-819.
[18] Testa, W. & R. Mattoon (2007). “Is There a Role for Gross Receipts Taxation?”NationalTax Journal, 60(4), 821-840.
[19] Williams, J., Van Ours, J. C., & Grossman, M. (2011). Why do some people want tolegalize cannabis use? (No. w16795). National Bureau of Economic Research.
25
6 Figures and Tables
Figure 1: Producer Harvests and Days-to-Harvest
These figures are based on the estimates in Columns (2) and (3) of Table 4. The solid line plots the estimated response to thetax change plus the polynomial. The scatterplot is the corresponding daily average of the dependent variable with estimates ofthe processor fixed effects removed. The vertical dashed line marks the day of the tax change, July 1, 2015.
Figure 2: Producer Harvests and Days-to-Harvest Bandwidth Sensitivity
These figures consider the sensitivity of the estimates in Columns (2) and (3) of Table 4 to the number of weeks of data weinclude on either side of the tax change. The dots mark the estimates for each bandwidth choice and the lines mark the 95percent confidence intervals around these estimates.
26
Figure 3: Processor After-Tax Prices and Revenue
These figures are based on the estimates in Columns (2) and (3) of Table 7. The solid line plots the estimated response to thetax change plus the polynomial. The scatterplot is the corresponding daily average of the dependent variable with estimates ofthe processor fixed effects removed. The vertical dashed line marks the day of the tax change, July 1, 2015.
Figure 4: Processor After-Tax Prices and Revenue Bandwidth Sensitivity
These figures consider the sensitivity of the estimates in Columns (2) and (3) of Table 7 to the number of weeks of data weinclude on either side of the tax change. The dots mark the estimates for each bandwidth choice and the lines mark the 95percent confidence intervals around these estimates.
27
Figure 5: Vertical Integration
These figures are based on the estimates in Columns (1) and (2) of Table 9. The solid line plots the estimated response to thetax change plus the polynomial. The scatterplot is the corresponding daily average of the dependent variable with estimates ofthe processor fixed effects removed. The vertical dashed line marks the day of the tax change, July 1, 2015.
Figure 6: Retail Tax-Inclusive Prices and Weight
These figures are based on the estimates in Columns (2) and (3) of Table 12. The solid line plots the estimated response tothe tax change plus the date polynomial. The scatterplot is the corresponding daily average of the dependent variable withestimates of the day-of-week, day-of-month, day indicators for June 30 - July 5, and retail fixed effects removed. The verticaldashed line marks the day of the tax change, July 1, 2015.
28
Figure 7: Retail Tax-Inclusive Prices and Weight Bandwidth Sensitivity
These figures consider the sensitivity of the estimates in Columns (2) and (3) of Table 12 to the number of weeks of data weinclude on either side of the tax change. The dots mark the estimates for each bandwidth choice and the lines mark the 95percent confidence intervals around these estimates.
Figure 8: The Average Price of One Gram of Marijuana and Tax Incidence acrossMarkets
In this figure, we plot the average retail firm’s price of one gram of marijuana both before and after the tax change. We thenconsider how much goes to processor and retail taxes as well as how much is spent to purchase a gram, on average, from thefirm. Before the tax change, all prices and taxes (in dollars) are based on the average prices the month prior to the tax change.After the tax change, these numbers are the pre-tax change prices adjusted by our estimated changes caused by the tax changes.This holds constant the composition of the market and eliminates any secular trends in prices. Note that if the market wasperfectly competitive, the theory of tax-invariance would predict the middle column in which processor prices fall and the restof the market is left unchanged. This is contrasted with what actually happens after the tax changes in the right column.
29
Table 1: Marijuana taxes by state
State Tax rate NotesCalifornia 15% Cultivation tax of $9.25/oz on dried flowers
and $2.75/oz on dried leaves.Colorado 10% Additional 15% tax applied at wholesale
based on average market rate in the stateMaine 10%
Massachusetts 3.75% Localities may impose additional 2% taxNevada 15% Tax applied at wholesaleOregon 17% Localities may impose additional 3% tax
Washington 37%Note: All taxes applied to retail sales unless otherwise noted
Table 2: House Bill 2136 legislative history summary
Date ActivityRegular Session
February 17 Introduced in WA House, referred to committeeFebruary - March Passed by committees, substituted twice
April 10 Passed by House 67-28, referred to SenateApril 24 Accepted by Senate Committee, referred back to House,
Regular Session Ends
First Special SessionApril 29 House passes 70-25, referred to SenateMay 1 Referred to Senate committeeMay 28 Referred back to House, First Special Session Ends
Second Special SessionMay 29 Reintroduced in House, referred to committeeJune 26 Removed from committee, passed by House 59-38June 27 Passed by Senate 36-7June 30 Signed by Governor
Source: http://app.leg.wa.gov/billsummary?BillNumber=2136&Year=2015
Table 3: Producer Summary Statistics
Obs. Mean Std. Dev. Mean>0 Min. Max.
Plantings 45,726 8.90 66.47 93.22 0 3,646Harvesting 30,921 7.71 66.27 63.31 0 4,153Plant to Harvest Days 3,766 116.29 40.75 116.29 2 292
30
Table 4: Producer Response to Tax Change
(1) (2) (3)Plantings Harvests Days-to-Harvest
Tax Change 0.239 0.042 −0.077∗∗
(0.163) (0.141) (0.036)
Observations 45,726 3,766 3,766R-squared 0.369 0.564 0.570Months Pre-Post 2 2 2Polynomial Order 3rd 3rd 3rdProducer Locations 383 255 255The following variables are included, but not reported in these regressions:day indicator variables for June 30 - July 5, day of the week and day of themonth indicator variables, and producer location fixed effects. All dependentvariables are in logs; the dependent variable is listed directly below the col-umn number. AT stands for after-tax. Standard errors clustered by producerlocation are in parentheses.
Table 5: Robustness Checks for Producer Response to Tax Change
(1) (2) (3) (4) (5) (6)
Harvests 0.042 0.059 0.087 0.049 −0.038 0.034(0.141) (0.147) (0.121) (0.140) (0.224) (0.138)
Days-to-Harvest −0.077∗∗ −0.092∗∗ −0.063∗∗ −0.055∗ −0.009 −0.080∗∗
(0.036) (0.037) (0.031) (0.032) (0.050) (0.036)
Observations 3,766 3,766 3,766 3,766 3,766 3,766Locations 255 255 255 255 255 255
Indicators for June 27-29 No Yes No No No NoIndicators for July 2-5 Yes Yes No Yes Yes YesDay of Month Indicators Yes Yes Yes No Yes YesPolynomial Order 3rd 3rd 3rd 3rd 5th 3rdCounty x Linear Trend No No No No No YesThe top row reports an estimate for the dependent variable log of the tax-inclusive price per gram. Thenext row reports an estimate for the dependent variable log weight. Dependent variables are logged (orthe log of 1 plus the variable if there are zeros) except in the last column in which we take the inversehyperbolic sine. The following variables are included, but not reported in these regressions: day indicatorvariables for June 30 - July 5, day of the week and day of the month indicator variables, and processorlocation fixed effects. All dependent variables are in logs. Standard errors clustered by processor locationare in parentheses.
Table 6: Processor Summary Statistics
Obs. Mean Std. Dev. Mean>0 Min. Max.
Price per Gram 6,485 3.67 1.06 3.67 0.30 10Weight (in grams) 27,591 339.64 1,089.78 1,445.03 0 25,058.50Number of Sales 27,591 3.28 10.35 13.95 0 266Firm Revenue 27,591 1,262.08 4,018.2 5,369.62 0 95,747.17Fraction of Vertically Integrated Firms 6,477 0.93 0.22 0.93 0 1
31
Table 7: Processor Response to Tax Change
(1) (2) (3) (4) (5) (6) (7)Price AT Price Weight Sales Rev. AT Rev. THC
Tax Change −0.059∗∗ 0.229∗∗∗ 0.003 −0.006 −0.068 0.219∗∗ −0.024(0.026) (0.026) (0.083) (0.052) (0.096) (0.096) (0.092)
Observations 6,485 6,485 6,485 6,485 6,485 6,485 6,363R-squared 0.380 0.424 0.278 0.351 0.254 0.270 0.834Months Pre-Post 2 2 2 2 2 2 2Polynomial Order 3rd 3rd 1st 1st 1st 1st 3rdProcessor Locations 227 227 227 227 227 227 226The following variables are included, but not reported in these regressions: day indicator variables for June 30 - July 5,day of the week and day of the month indicator variables, and processor location fixed effects. All dependent variablesare in logs; the dependent variable is listed directly below the column number. AT stands for after-tax. Standard errorsclustered by processor location are in parentheses.
Table 8: Robustness Checks for Processor Response to Tax Change
(1) (2) (3) (4) (5) (6)
After-Tax Price 0.229∗∗∗ 0.230∗∗∗ 0.225∗∗∗ 0.240∗∗∗ 0.227∗∗∗ 0.233∗∗∗
(0.026) (0.028) (0.022) (0.025) (0.037) (0.026)After-Tax Revenue 0.219∗∗ 0.215∗∗ 0.216∗∗ 0.187∗∗ 0.370∗∗∗ 0.222∗∗
(0.096) (0.097) (0.090) (0.077) (0.105) (0.097)
Observations 6,485 6,485 6,485 6,485 6,485 6,485Processor Locations 227 227 227 227 227 227
Indicators for June 27-29 No Yes No No No NoIndicators for July 2-5 Yes Yes No Yes Yes YesDay of Month Indicators Yes Yes Yes No Yes YesPolynomial Order 3rd/1st 3rd/1st 3rd/1st 3rd/1st 5th/3rd 3rd/1stCounty x Polynomial No No No No No YesThe top row reports an estimate for the dependent variable log of the tax-inclusive price per gram. The next row reportsan estimate for the dependent variable log after-tax revenue. The following variables are included, but not reported inthese regressions: day indicator variables for June 30 - July 5, day of the week and day of the month indicator variables,and processor location fixed effects. Standard errors clustered by processor location are in parentheses.
32
Table 9: Vertical Integration Response to Tax Change
(1) (2) (3) (4)Vertical NVWeight NVPrice Vertical
Tax Change −0.018 0.594∗∗ 0.087 −0.029(0.018) (0.240) (0.070) (0.133)
July 0.022∗ −0.286∗ −0.054 0.066(0.012) (0.147) (0.043) (0.099)
Observations 3,261 3,570 567 284R-squared 0.700 0.627 0.311 0.016First Month Only No No No YesMonths Pre-Post 6 6 6 6Polynomial Order 3rd 3rd 3rd 3rdProcessor Locations 447 447 123 284Processor location fixed effects (Columns (1) to (3) only) and a polynomial in theprocessor sale month are included, but not reported. All dependent variables are in logs;the dependent variable is listed directly below the column number. NVWeight standsfor non-vertical weight. NVPrice stands for the ratio of the average vertical wholesaleprice over the average price for each processor location. Standard errors clustered byprocessor location are in parentheses for Columns (1) to (3) and heteroskedastic robuststandard errors are in parentheses for Column (4).
Table 10: Robustness Checks for Vertical Integration Response to Tax Change
(1) (2) (3) (4) (5)
Fraction Vertically Integrated −0.018 0.004 −0.026 −0.022 −0.017(0.018) (0.012) (0.017) (0.026) (0.017)
Non-Vertical Weight 0.594∗∗ 0.322∗ 0.467∗∗ 0.355 0.441∗∗
(0.240) (0.171) (0.199) (0.327) (0.209)
Observations 3,261 3,261 3,261 3,261 6,2603,570 3,570 3,570 3,570 6,743
Indicator for July No Yes No No NoPolynomial-Order 3rd 3rd 1st 1st 3rdTax Change Polynomial Interaction No No No Yes NoMonths Pre-Post 6 6 6 6 12Processor Locations 447 447 447 447 575The top row reports an estimate for the dependent variable fraction of vertically integrated transac-tions by firm. The next row reports an estimate for the dependent variable log non-vertical weight.Processor location fixed effects and a polynomial in the month are included. All dependent variablesare in logs. Standard errors clustered by processor location are in parentheses.
33
Table 11: Retail Summary Statistics
Obs. Mean Std. Dev. Median Min. Max.
Price per Gram 16,095 9.62 1.54 9.52 3.74 19.20Weight (in grams) 16,095 510.20 455.40 387.65 1 6,639.53Number of Sales 16,095 240.89 213.87 186 1 2008Firm Revenue 16,095 4571.76 4019.19 3,480.32 6.73 39,105.71Days from Wholesale 16,095 20.07 29.33 17.52 -369 28THC Potency 16,094 19.67 1.41 19.75 10.42 29.60CBD Potency 16,095 0.40 0.28 0.34 0 11.12Number of Competitors 16,095 5.74 4.60 5 1 22Distance to Nearest U.S. Border 16,095 113.44 65.86 124.03 3.82 243.37
Table 12: Retail Response to Tax Change
(1) (2) (3) (4) (5) (6) (7)Price TI Price Weight Sales Rev. TI Rev. THC
Tax Change −0.065∗∗∗ 0.027∗∗ −0.022 −0.003 −0.070∗∗∗ 0.022 −0.014∗∗
(0.012) (0.012) (0.016) (0.017) (0.015) (0.015) (0.006)
Observations 16,095 16,095 16,095 16,095 16,095 16,095 16,094Retailer Locations 138 138 138 138 138 138 138R-squared 0.835 0.816 0.862 0.883 0.871 0.873 0.495Months Pre-Post 2 2 2 2 2 2 2Polynomial Order 5th 5th 3rd 3rd 3rd 3rd 5thThe following variables are included, but not reported in these regressions: log processor price, whether any competitors,log competitors’ processor price, day indicator variables for June 30 - July 5, day of the week and day of the month indicatorvariables, and retail location fixed effects. All dependent variables are in logs; the dependent variable is listed directlybelow the column number. TI stands for tax-inclusive. Standard errors clustered by retail location are in parentheses.
34
Table 13: Robustness Checks for Retail Response to Tax Change
(1) (2) (3) (4) (5) (6) (7) (8)
Tax-Inclusive Price 0.027∗∗ 0.026∗∗ 0.022∗∗ 0.030∗∗∗ 0.011 0.028∗∗ 0.020 0.025∗∗
(0.012) (0.012) (0.010) (0.010) (0.011) (0.011) (0.013) (0.011)
Firm-Strain TI Price 0.027∗∗∗ 0.027∗∗∗ 0.025∗∗∗ 0.030∗∗∗ 0.027∗∗∗ 0.027∗∗∗ 0.029∗∗∗ 0.027∗∗∗
(0.006) (0.006) (0.005) (0.006) (0.006) (0.006) (0.009) (0.006)
Weight −0.022 −0.017 −0.014 −0.029∗ −0.018 −0.019 −0.021 −0.018(0.016) (0.017) (0.014) (0.016) (0.017) (0.017) (0.020) (0.016)
Implied Elasticity -0.80 -0.65 -0.57 -0.98 -0.66 -0.68 -0.77 -0.63
Observations 16,095 16,095 16,095 16,095 16,095 16,095 16,095 16,095Firm-Strain Observations 637,621 637,621 637,621 637,621 637,621 637,621 637,621 637,621Retailer Locations 138 138 138 138 138 138 138 138
Indicators for June 27-29 No Yes No No No No No NoIndicators for July 2-5 Yes Yes No Yes Yes Yes Yes YesDay of Month Indicators Yes Yes Yes No Yes Yes Yes YesProcessor Prices Included? Yes Yes Yes Yes No Yes Yes Yes# Competitors Included? No No No No No Yes No NoPolynomial Order 5th/3rd 5th/3rd 5th/3rd 5th/3rd 5th/3rd 5th/3rd 7th/5th 5th/3rdCounty x Polynomial No No No No No No No YesThe top row reports an estimate for the dependent variable log of the tax-inclusive price per gram. The next row reportsan estimate for the dependent variable log of the tax-inclusive price per gram where the data are aggregated by retaillocation-strain-producer-day instead of retail location-day. The third row reports an estimate for the dependent variablelog weight. The elasticity of demand implied by these estimates is calculated by dividing the weight estimate by the priceestimates in the second row. The following variables are included, but not reported in these regressions unless otherwisespecified: log processor price, log competitors’ processor price, whether any competitors (these first three covariates areincluded only for the first and third rows), day indicator variables for June 30 - July 5, day of the week and day of themonth indicator variables, and retail location fixed effects. All dependent variables are in logs. Standard errors clustered byretail location are in parentheses.
35
Table 14: Heterogeneous Retail Tax-Inclusive Price Response to Tax Change
(1) (2) (3) (4) (5)
Tax Change 0.019∗∗ 0.011 0.046∗∗∗ 0.036 0.059(0.008) (0.023) (0.015) (0.028) (0.052)
Log Distance x Tax Change 0.002 0.002 −0.002(0.005) (0.005) (0.010)
Log Number of Competitors x Tax Change −0.018∗∗ −0.018∗∗ −0.022∗∗
(0.008) (0.008) (0.009)Log Distance −0.014 −0.015 −0.014 −0.015 −0.050∗∗
(0.013) (0.014) (0.013) (0.014) (0.023)Log Number of Competitors −0.048∗∗∗ −0.048∗∗∗ −0.040∗∗∗ −0.040∗∗∗ −0.041∗∗∗
(0.011) (0.011) (0.010) (0.010) (0.012)
Observations 637,621 637,621 637,181 637,181 637,181Retailer Locations 138 138 137 137 137R-squared 0.219 0.219 0.222 0.222 0.227Polynomial-Order 5th 5th 5th 5th 5thMonths Pre-Post 2 2 2 2 2Distance Quartiles x Polynomial No No No No YesThe dependent variable is log price per gram. The data for this regression are aggregated by retail location-strain-producer-day instead of retail location-day. The following variables are included, but not reported in these regressions: day indicatorvariables for June 30 - July 5, day of the week and day of the month indicator variables, income shares by county, andthe log of county population. Standard errors clustered by retail location are in parentheses.
Table 15: Heterogeneous Retail Weight Response to Tax Change
(1) (2) (3) (4) (5)
Tax Change −0.041 −0.040 −0.075 −0.070 −0.125(0.026) (0.109) (0.050) (0.120) (0.216)
Log Distance x Tax Change −0.000 −0.001 0.013(0.025) (0.025) (0.046)
Log Number of Competitors x Tax Change 0.024 0.024 0.018(0.029) (0.029) (0.030)
Log Distance −0.217∗∗∗ −0.217∗∗∗ −0.217∗∗∗ −0.217∗∗∗ −0.225∗∗∗
(0.058) (0.063) (0.058) (0.062) (0.079)Log Number of Competitors 0.070 0.070 0.056 0.056 0.117
(0.091) (0.091) (0.096) (0.096) (0.095)
Observations 16,095 16,095 16,095 16,095 16,095Retailer Locations 138 138 138 138 138R-squared 0.344 0.344 0.344 0.344 0.350Polynomial-Order 3rd 3rd 3rd 3rd 3rdMonths Pre-Post 2 2 2 2 2Distance Quartiles x Polynomial No No No No YesThe dependent variable is log of weight (in grams). The following variables are included, but not reported in theseregressions: log processor price, log competitors’ processor price, whether any competitors, day indicator variablesfor June 30 - July 5, day of the week and day of the month indicator variables, income shares by county, and thelog of county population. Standard errors clustered by retail location are in parentheses.
36