EC 560: Theories of IO

Professor Keaton Miller Masters Project

1 Introduction
In this assignment you will be replicating the primary analysis of Hansen, Miller, Seo, and
Weber (2020). In that paper, we analyze the potential outcome of a potency-based tax
on cannabis products. We introduce a simple model of the cannabis market and use it
to motivate an empirical examination of sales in Washington state. We use the model to
perform a counterfactual experiment: what would happen if we introduced a potency tax
and removed the current tax?

2 Theory
Consider a monopolist retailer selling a single good. The firm sets the tax-inclusive price of
the good p and also a single product ‘potency’ characteristic (i.e. THC or alcohol content)
x. The firm faces a per-unit potency tax τx (in units of dollars per unit of potency per unit
of the good). Consumers pay an ad valorem tax τp (expressed as a percentage of the tax-
exclusive price) and the firm sets the tax-inclusive price p (so that the firm earns p/(1 + τp )
per unit sold). Consumer demand q(p, x) is a function of the price and the potency and is
assumed to feature constant demand elasticities (εp , εx ) with respect to both characteristics.
Marginal costs vary with the potency via
mc(x) = c + (γ + τx )x. (1)
In this equation, c represents marginal costs which are independent of potency and γ is the
marginal cost of an additional unit of potency. The marginal costs of additional potency are
either costs associated with growing more potent marijuana (e.g. allowing plants to grow
for a longer period of time) or with manipulation during the testing process (e.g. sending in
samples that are more likely to give a particular THC reading).
Putting these pieces together, the firm’s profit function is
 
p
π(p, x) = − c − (γ + τx )x · q(p, x) (2)
1 + τp
where q(p, x) = Apεp xεx .
1. Write the first order conditions for profit maximization. You may write these with
respect to partial derivatives of q, instead of plugging in the demand function.
2. Use the fact that the price and potency elasticities of demand are constant to find
a closed form solution for x and p. Hint: If you used partial derivatives for q in
the previous problem, you should be able to multiply and divide to turn those into
elasticities.
3. What is the relationship between taxes on price and the potency of the product? Define
the potency-adjusted price as p/x. What is the relationship between taxes on price
and this object?
4. Suppose you had estimates of demand elasticities, and knew the prices, potency, and
current tax regime from the data. How could you use the model to estimate c and γ?

1 of 4
 EC 560: Theories of IO
Professor Keaton Miller Masters Project

3 Empirics
To examine the role of potency in recreational cannabis markets, we analyze public records
from Washington’s internal regulatory “traceability” system designed to track each mar-
ijuana product from “seed to sale.” Broadly speaking, Washington’s regulatory system
creates a three-step supply chain. Producers grow Cannabis plants and harvest the raw
plant material. Processors, which may be vertically integrated with producers, convert that
material into “usable marijuana” (i.e. dried flower) and other cannabis products (i.e. edibles
and concentrates), package those products and combine them into homogenous “inventory
lots.” Usable marijuana must be packaged into units of pre-set sizes (e.g. 1 gram, 2 grams,
or 3.5 grams). Retailers, which must be financially independent from producers and proces-
sors, purchase wholesale inventory lots from processors and sell individual products to final
consumers. Each inventory lot identifier therefore represents a unique combination of strain,
manufacturing batch, package size, processor, and retailer. This means that inventory lot
identifiers, while playing the same role within retailers’ point-of-sale data systems as a UPC
in other retail markets, are far more specific. For example, a 3.5 gram package of usable
marijuana from the “Jack Herer” strain produced by “Keaton’s Kush Garden” and sold by
“Caroline’s Cannabis Collective” will have a different inventory lot identifier if the flower
comes from plants grown and processed in different batches.
We then aggregate our data to the inventory-lot-week level to avoid cyclical day-to-
day variation in sales and idiosyncratic differences in reporting behaviors – the level of our
observations is therefore similar to demand analyses using scanner data aggregated to the
UPC-store-week level. We restrict our data set to the most recent year of data we have
available – fiscal year 2016 (July 2016 - June 2017) – to obtain estimates that are closer to
what will happen in the long-run equilibrium of the market.
Note: For the following problems, submit your answers as well as any code you wrote to
generate the answers. You may use Stata, Python, R, or Julia. If you wish to use a different
package, please ask me.

1. Download the potency data for 560 file from Canvas. It is available in either Stata
or CSV format (which can be imported from any statistical package). Summarize the
data: make a table reporting the mean, 10th percentile, median, and 90th percentile
of the weight sold, the retail price, the wholesale price, and the THC and CBD con-
centrations, all in levels (the data come in natural logs so you will have to convert it).
See the end of this assignment for a data dictionary.

2. The model above leads to a natural empirical model of demand. For some cannabis
inventory lot j and week t, we model the quantity of grams sold qjt as function of
potency and fixed effects via

log(yjt ) = β0 + β1 log(pricejt ) + β1 · log(T HCj ) + β2 · log(CBDj ) + F Xjt + jt . (3)

Estimate this model with the following combinations of fixed effects. For all models,
instrument for log price with log wholesale prices. Report coefficients and standard
errors for the log price and the log THC content. For Stata users, I recommend the
ivreghdfe package. For R users, the lfe package may be useful.

2 of 4
 EC 560: Theories of IO
Professor Keaton Miller Masters Project

• Week-by-retailer and strain-by-retailer

• Week-by-retailer, strain-by-retailer, and producer
• Week-by-retailer, strain-by-retailer, and processor
• Week-by-retailer, strain-by-retailer, producer, and processor
An example table may be seen below.
Table 1: The association between alcohol content, prices, and quantities for
liquor

Log liters sold

(1) (2)

Log price -1.1493*** -1.3343***

(0.0496) (0.0636)

Log alcohol content 1.0619*** 1.0670***

(0.0326) (0.0818)

Fixed effects
Unit size Yes Yes
Week * Retailer Yes Yes
Brand * Retailer Yes Yes
Feature, Display No Yes

Obs. 3,148,163 491,989

First stage F-stat. 629.61 267.76

Notes: An observation is a retail sale of a product per store-week. The dependent variable is the log of liters sold. Log price is
the log of the average tax-inclusive retail price per liter in each store-week, and it is instrumented by non-contiguous Hausman
instruments. Alcohol content is reported as percentage ethanol by volume on a scale from 0 to 100. We consider brand equivalent
to producer-processor combination in cannabis. Standard errors, reported in parentheses, are robust to heteroskedasticity and
calculated adjusting for clustering at the store level. Observation counts differ across specifications due to singleton observations,
lack of instruments for some observations, and Feature and Display variables which exist since 2013.

3. The data in this paper come from period in which the state and local sales tax rate
was 8.4% and the ad valorem tax rate was 37%, so the total tax rate was 45.4%. Using
the data, calculate the average tax-revenue per inventory-lot-week.
4. Suppose the government believed that changing from an ad-valorem tax to a potency
tax would not affect demand or supply at all. What would be the potency tax rate
(in terms of dollars per percentage point of THC per gram of cannabis) that generates
the equivalent tax revenue to the current policy? Note: We are only talking about
replacing the ad valorem tax, not the state and local sales tax.
5. Suppose the government implemented the rate you found in the previous question.
What would be the new tax-inclusive price? The new quantity? The new potency?
The new tax revenue? Again, we are only talking about replacing the ad valorem tax,
not the state and local sales tax. Hint: You will need to use your cost estimation
technique from the Theory section to answer this question.

3 of 4
 EC 560: Theories of IO
Professor Keaton Miller Masters Project

Table 2: Data Dictionary for potency data for 560

Variable Type Description

salesloc Integer Identifies the retail location at which the inventory lot
was sold.
inventoryid String The ID of the inventory lot (similar to a UPC code).
monthly Integer Months since January 1960 (in Stata, use display format
%tM).
weekly Integer Weeks since the first week of January 1960 (in Stata, use
display format %tW).
procorg Integer Identifies the firm which processed the raw flower into
usable marijuana.
lab license Byte Identifies the lab which performed the potency test
strain code Integer Identifies the strain of cannabis that this inventory lot
consists of. Labels are available in Stata.
prodorg code Integer Encodes the firm which grew the raw flower. −1 means
unknown, 925 means multiple firms. Labels are available
in Stata.
package size Float The size of the product package in grams. Common ex-
amples include 1g, 3.5g, and 7g.
log thc Float The natural logarithm of the percentage of THC by dry
weight.
log cbd Float The natural logarithm of the percentage of CBD by dry
weight.
log grams Float The natural logarithm of the number of grams sold from
that inventory lot in that week.
log ppg Float The natural logarithm of the average price per gram for
purchases made in that week.
log inst wholesale Float The natural logarithm of the wholesale price per gram
for that inventory lot. Use as an instrument for price.

4 of 4