Pricing and Rarity Design of Blind Boxes with Random Items

Ming Hu
Rotman School of Management, University of Toronto, [email protected]

Taojie Qin
Rotman School of Management, University of Toronto, [email protected]

Shreyas Sekar
Rotman School of Management, University of Toronto, [email protected]

Blind box selling is a novel selling mechanism in which buyers purchase sealed packages containing unknown

items, with the chance of uncovering rare or special items. To unravel the economic motivation behind

blind boxes, we propose a stylized model encapsulating the core trade-offs a seller faces when designing a

blind box package, including the price for each box and drawing probabilities of different items. The model

distinguishes between two types of items: regular and special, with the latter valued higher by consumers

due to its rarity. The consumers’ expected valuation from opening the blind box decreases with each unique

item received, whether regular or special, but remains unchanged if they receive a duplicate item. Naturally,

consumers cease purchasing when the expected marginal utility becomes negative. We compare blind box

selling with traditional separate selling under no supply rationing, in terms of the optimal prices, sales

volume, and profit. Our findings indicate that item heterogeneity and moderate rarity are necessary for the

success of blind box selling. Surprisingly, we show that blind boxes are more profitable in extreme scenarios

where the manufacturing costs are either low or high. We also find that blind boxes perform better when

consumers are highly status-seeking. However, our study also illustrates that blind boxes are not a panacea

and that the seller has to deploy them carefully, especially when manufacturing costs are neither high nor

low. To examine the robustness of our results, we consider cases such as separate selling with rationing,

consumer heterogeneity, and lump-sum bonus upon collecting all items.

1. Introduction
Blind boxes, also known as mystery boxes, have achieved remarkable success1 as an innovative
selling mechanism, particularly for toys and collectibles. A blind box is a sealed package containing
a single item whose identity is unknown to buyers before purchase. Sellers typically populate these
blind boxes by randomly choosing from a pre-determined assortment, relying on buyers’ affinity
1
The global market for blind boxes is estimated to be US $24.3 billion and is expected to grow at a compound annual
growth rate of 5.5%. See https://www.verifiedmarketreports.com/product/blind-boxes-market/.

1
2

SECRET EDITION

Harry Potter-Expecto Patronum

Harry Potter- Hermione
Sirius Black George Weasley Sybill Trelawaney Peter Pettigrew
Lumos Maxima Granger

The probability of drawing

the secret figurine in this
series is 1/144

Harry Potter-The Monster Ron Weasley Draco Malfoy Fred Weasley

Remus Lupin Severus Snape
Book of Monsters

(a) Regular Items (b) Special Item

Figure 1 A Pop Mart blind box and the items contained in it

for surprise to drive sales. As an example, consider the Harry Potter blind box depicted in Figure 1
that is sold by Pop Mart2 —a leading vendor of designer toys. Each blind box retails for US $14.99.
Upon purchasing the blind box, a buyer has an equal chance of receiving any one of twelve regular
items (i.e., a figurine of some character from the series) or may draw a rare special item with a
probability of 1/144. As a result, buyers base their purchase decisions not just on the price of the
blind box but also on the probability of drawing each item, which is publicly known.
More generally, blind boxes can be viewed as a form of probabilistic (Fay and Xie 2008) or opaque
selling (Jerath et al. 2010, Elmachtoub et al. 2019) with some heterogeneity in the underlying
items. However, while opaque selling may be employed to “sell off less desirable items,” blind boxes
utilize randomness to incentivize repeat purchases. For example, buyers may purchase a blind box
multiple times until they get a desired item or collect the entire assortment. On the other hand, loot
boxes have gained popularity as a mechanism for randomly allocating items in video games (Chen
et al. 2021). While the underlying intent behind loot boxes is similar to blind boxes, the latter
is typically associated with physical and non-substitutable goods, which can lead to materially
different prices and purchasing behavior. Given the intricacies underlying the design of blind boxes,
it is natural to question which of these factors have contributed to its success and whether blind
boxes are suitable for every setting. These questions are particularly timely as similar mechanisms
are now being considered in other domains, including food and beverage, fashion, and books.
One defining characteristic of blind boxes is that they provide buyers with the opportunity to
obtain rare or special items alongside regular items, albeit with a lower probability. As illustrated
in Figure 1, these special items are comparable to the regular items in terms of their attributes,
and thus, their higher value can be ascribed to their perceived rarity or exclusivity. For example,

2
See https://www.popmart.com/us/products/379/Harry-Potter-and-the-Prisoner-of-Azkaban-Series-Figures.
 3

data from Qingdao, a marketplace for trading collectibles, suggests that consumers’ willingness
to pay for special toys is 11.4 times that of regular toys (see Online Appendix B). This strategy
of leveraging rarity to boost profits is not unique to blind boxes, and a number of works have
explored how scarcity drives status and consumption (Amaldoss and Jain 2005, Tereyağoğlu and
Veeraraghavan 2012, Rao and Schaefer 2013). Unfortunately, in many of these markets, artificially
limiting the supply of items in order to boost prices can lead to adverse consequences such as
consumer dissatisfaction and regulatory scrutiny (Anderson et al. 2006, Biraglia et al. 2021). In
contrast, the blind box model allows firms to regulate the supply of rare items in a more organic
fashion and utilize randomness to allocate goods fairly.
Clearly, consumer expectations and purchase decisions of probabilistic goods depend not just
on price but also on the probability of obtaining different items. Therefore, the success of a blind
box hinges critically on optimally calibrating its parameters, particularly the rarity of the special
item(s). For example, consumers may not pay a high price for the blind box if the special item is
too easy to collect; conversely, when the special item is too rare, consumers may forego purchasing
the blind box given its low expected value. In such environments, it may be in the seller’s best
interest to disregard the blind box strategy and instead revert to the more traditional mechanism
of selling items separately in a deterministic fashion. However, under this more common traditional
selling model, the seller may not always have the freedom to limit supply artificially. Motivated
by these subtleties, we study the problem faced by a profit-maximizing seller in designing a blind
box, i.e., determining the price as well as the rarity of each item. In particular, our work focuses on
when employing blind boxes can be more advantageous than the status-quo mechanism of separate
selling.

1.1. Model Overview and Summary of Contributions.

To encapsulate the core trade-offs faced by sellers when designing a blind box package, we consider a
stylized setting with n unique regular items and a single special item. A monopolistic seller can pro-
duce any desired quantity of these two types of items at the same unit procurement/manufacturing
cost. Throughout this work, we compare the seller’s profit under two distinct mechanisms:
Blind Box: Under this strategy, the seller selects the price of the blind box along with the
probability of drawing each of the n regular items and the special item. These probabilities
are publicly revealed to consumers.
Separate Selling: Under this strategy, the seller simply selects the per-unit price for each of
the regular items and the special item, respectively.
4

On the consumer side, we assume that the value for regular items is exogenously given (and
independent of any of the seller’s choices), whereas the valuation for the special item increases
linearly with its perceived rarity 3 relative to the regular item. When consumers are more status-
seeking, the valuation of the special item increases more rapidly with its rarity. Given the model
primitives, the consumers’ behavior can be described as follows. If the seller opts for separate
selling, then consumers make independent purchase decisions on each of the n + 1 items, i.e., buy
an item as long as their valuation is no less than its price. On the other hand, under the blind
box selling mode, consumers would purchase one box as long as the expected valuation of doing
so (from the items not yet received) is at least its price.

Base Model. We begin with the setup where all consumers are homogeneous in their valuations
and where there is no rationing under separate selling, i.e., the seller cannot deny the special item
to any buyer who is willing to pay. In practice, rationing a highly sought-after item in retail, such
as a special edition Nike sneaker, often results in the product falling into the hands of scalpers. The
no rationing assumption is motivated by the fact that in many industries, including collectibles,
sellers cannot arbitrarily restrict the supply of items without facing adverse consequences from
consumers. We relax this assumption in a later section. Note that the no rationing assumption
does not impact blind boxes as they control rarity in a more organic fashion through randomness,
i.e., every buyer has a fair chance of receiving the special item.
As a warm-up, we first investigate two special cases of our base model, designed to spotlight
specific aspects of our setting. First, we show that in the absence of a special item, separate selling
is always more profitable for the seller compared to blind boxes. Following up on this, we consider
the other extreme scenario (with a special item), where the number of regular items n approaches
infinity. Surprisingly, separate selling still outperforms blind boxes under this model. Specifically,
even though the special item’s rarity is significant in this environment, the perceived difficulty of
obtaining this item deters consumers from purchasing the blind box enough times to match the
profits from separate selling. Together, these results imply that while item heterogeneity and rarity
are crucial for the blind box strategy, one must carefully balance the value of the special item with
a buyer’s likelihood of actually receiving the item.
Building on these findings, we present our main result that characterizes the regimes in which
blind boxes or separate selling is the optimal selling scheme, providing insights into how the seller
can optimally choose the price and rarity under the blind box model. In particular, we show that for
3
We formally prove that consumers’ perceived rarity of the special item prior to purchase matches its actual rarity
observed in equilibrium.
 5

any given instance, blind boxes outperform separate selling if and only if the manufacturing costs
are either low or high (below or above certain thresholds). In contrast, in the intermediate cost
regime, separate selling dominates blind box selling. This non-monotonic behavior with respect to
cost can be explained in terms of two complementary drivers of blind box performance:
1. Low Cost - Special Item Drives Sales: When costs are low, sellers can benefit by setting a
relatively low price for the blind box and profit from the ensuing (large) expected sales volume,
i.e., consumers buy the blind box again and again hoping to receive the special item but may
end up purchasing the same regular items multiple times.
2. High Cost - Special Item Boosts Prices: As costs increase, the seller is forced to raise the
price of the blind box, which hurts sales. However, when the costs become sufficiently high,
the seller is able to leverage a different mechanism wherein the special item creates positive
externalities for the regular item. Specifically, the rarity of the special item ensures that the
consumer would purchase the blind box even if its price is higher than their valuation for
regular items. As a consequence, the seller is able to extract greater margins compared to
separate selling.
Beyond the cost dimension, we also explore the role of consumers’ status-seeking nature. Unsur-
prisingly, as the premium that consumers place on rarity increases, the performance gap between
blind boxes and separate selling widens.

Model Extensions. We also test the robustness of our findings by extending the core model
in several directions. First and foremost, we relax the no rationing assumption and examine the
effect of rationing under separate selling. In this scenario, blind box selling continues to outperform
separate selling in the low-cost regime. However, separate selling may dominate when costs are
high, as it is now possible to inflate the scarcity of the special item and set a high price. It is natural
to expect that when the seller can arbitrarily calibrate the supply of the items, blind boxes would
confer no advantage in markets with status-seeking consumers. Surprisingly, we argue this is not
the case by showing that for any fixed manufacturing cost level, blind box selling provides higher
profits as long as the premium consumers place on rarity is sufficiently high. In other words, all
things being equal, blind boxes are able to leverage the status-seeking nature of consumers better
and extract greater profits.
Heterogeneous Consumers. Second, we extend our model to settings where consumers are hetero-
geneous in their valuations and argue that consumer heterogeneity can further enhance the value
of blind boxes. Formally, we consider an extension of our base model with two consumer types
who differ in their valuations for the regular and special items. We show that the ratio of the
6

seller’s optimal revenue under a blind box to that under separate sales is increasing in the level of
heterogeneity for many parameter regimes of interest, e.g., when consumers put a sufficiently high
premium on the special item (i.e., they exhibit more significant status-seeking behavior) and the
manufacturing cost is low enough. In those cases, blind boxes allow for more granular outcomes
under heterogeneity by mimicking targeted price discrimination, i.e., consumers with a higher val-
uation would purchase the blind box more times in expectation. On the other hand, under separate
sales, it is much harder for the seller to extract surplus from different segments using a uniform
price.
Lump-Sum Bonus. Third, we explore an additional force that drives motivated consumers to
purchase blind boxes in practice: the satisfaction associated with collecting all the items in the
assortment. We model this in terms of a lump-sum reward that a buyer receives upon obtaining all
the items. We find that a lump-sum bonus favors separate selling as opposed to blind boxes. This
is because, under separate selling without rationing, consumers can obtain all regular and special
items for sure and receive the lump-sum reward, which the seller can factor into their pricing and
extract such surplus. However, due to the nature of random purchases under blind boxes, consumers
may give up purchasing before collecting all items, with less surplus than the full lump-sum bonus
left for the seller to extract.
In summary, the multitude of results presented in this work provide a theoretical framework
aimed at better understanding the prevalence of blind boxes in practice. Specifically, our findings
emphasize how the seller can leverage rarity differently in low-cost versus high-cost environments,
i.e., boost transaction volume and increase price. Finally, we also shed light on how features such as
supply rationing, consumer heterogeneity, and lump-sum bonuses influence the selection between
optimal blind box vs. separate selling strategy.

1.2. Literature Review

Our work is closely tied to a number of distinct streams in the revenue management and marketing
literature, and we describe the commonalities below.

Mystery Boxes. Despite the growing popularity of the blind box model, the literature on this
topic—to the best of our knowledge—is still nascent. In this regard, there are two notable papers
that model blind boxes via an analytical framework (Feng et al. 2022, Chen and Fang 2023).
Specifically, on the one hand, Feng et al. (2022) study the impact of secondary markets on blind
boxes, focusing on the consumer behavior of buying and selling items in the secondary market.
On the other hand, Chen and Fang (2023) model the interaction between sellers and consumers in
 7

Gacha games4 as a Stackelberg game and characterize the seller’s optimal strategy and the impact
of subsidies on player behavior. Our work complements these papers by characterizing when it is
optimal for sellers to adopt a blind box strategy in the first place, as well as identifying driving
forces other than secondary markets that contribute to this success.
As mentioned previously, physical blind boxes are similar in spirit to loot boxes, which have
been used to allocate virtual items in video games. Given the loot box’s immense popularity in
practice, a number of papers in recent years have centered around the design and analysis of
loot boxes (Chen et al. 2021, Han et al. 2023, Huang and Yin 2021, Miao and Jain 2024). The
mechanism has also attracted scrutiny from regulators5 due to its adverse effects on consumers.
In terms of the findings, Chen et al. (2021) compare loot boxes to traditional separate sales as
well as unique loot boxes, which guarantee a new item with each purchase. Their results suggest
that unique loot boxes are more effective at generating revenue than traditional loot boxes as the
number of virtual items grows larger. However, while allocating unique items is feasible in virtual
settings, it is impossible to guarantee unique draws in the case of physical goods such as toys.
Therefore, it is not physically feasible to design blind boxes that avoid repeated draws, i.e., without
replacement. Moreover, Han et al. (2023) present techniques for optimizing the loot box price and
item probabilities under a model where the items are vertically differentiated. Huang and Yin
(2021) consider opaque selling involving two quality-differentiated products through a loot box.
In contrast, we consider horizontally differentiated regular items. Finally, Miao and Jain (2024)
consider two symmetric products and find that firms may assign asymmetric probabilities to ex-
ante symmetric products. Broadly speaking, our work differs from this literature on two fronts: (i)
we consider the sale of a finite set of physical goods with manufacturing costs, and (ii) we assume
that consumers’ willingness to pay (for the special item) is dependent on rarity, whereas all of the
above loot-box papers assume exogenous valuations. Both these choices lead to materially different
findings, e.g., new insights on how sellers can exploit the rarity of the special item in low-cost vs.
high-cost regimes.

Randomized and Opaque Selling. Naturally, blind boxes can be viewed as a special class of
probabilistic and opaque selling. Generally, in the case of probabilistic mechanisms, sellers offer a
composite product (i.e., a lottery among several goods) whose purchase results in the buyer obtain-
ing any one of the underlying items according to some prescribed probability. These mechanisms
4
Analogous to blind boxes, a Gacha game is an arcade game where a consumer pays to receive a mystery package
containing a toy or a novelty item.
5
See FTC Video Game Loot Box Workshop, https://www.ftc.gov/system/files/documents/reports/
staff-perspective-paper-loot-box-workshop/loot_box_workshop_staff_perspective.pdf.
8

provide sellers with the opportunity to segment the market without any explicit price discrimina-
tion, which can be particularly valuable when there is uncertainty about demand (Fay and Xie
2008, Huang and Yu 2014, Fay and Xie 2015).
Additionally, the term opaque selling is employed in relation to specific industries such as travel
and tourism and is associated with the practice of probabilistically selling items with differentiated
features in order to distribute demand more evenly across the assortment. We briefly summa-
rize the related papers below while noting that this literature is too vast to cover exhaustively.
One stream of this literature has centered around comparing the revenue (or profit) of opaque
pricing to other commonly studied mechanisms such as advance selling (Fay and Xie 2010), last-
minute selling (Jerath et al. 2010), and discriminatory pricing (Elmachtoub and Hamilton 2021).
A related stream has sought to present consumer behavior models that lead to the outperformance
of opaque selling, which include consumer flexibility (Elmachtoub et al. 2019), forward-looking
behavior (Chen et al. 2014), and pessimism (Elmachtoub and Hamilton 2021). Finally, some papers
have also proposed more sophisticated versions of probabilistic selling, including dynamic poli-
cies (Huang and Yin 2021) and deferred product assignment strategies (Fay and Xie 2015). It is
worth noting here that most works on opaque selling involve consumers who are heterogeneous in
their intrinsic preferences across products, whereas, in our base model, consumers’ valuations are
homogeneous. At a more fundamental level, there are some key differences in the rationale behind
why sellers adopt opaque pricing or blind boxes. In simple terms, while opaque selling is useful
to boost the sales of less desirable items, blind boxes are employed to induce scarcity in a more
organic fashion and incentivize buyers to engage in repeat purchases. As a consequence, unlike
opaque selling, blind boxes are more valuable in scenarios where the products being sold are not
perfect substitutes.
On a different note, there have also been papers studying randomized mechanisms other than
lotteries, three of which are highlighted below. In Yang and Yu (2023), the authors model surprise
clearance in supermarkets as a form of opaque selling where consumers face uncertainty about the
number of units contained in the package. In other settings, the purchase behavior is driven by the
randomness in the underlying price dynamics. For example, Yang et al. (2023) investigate a form of
social commerce where the price of an item is reduced randomly in response to consumers’ referrals.
However, while the possibility of a windfall drives consumer behavior in both these papers, they do
not capture the notion that consumer valuations for certain items are endogenous to their rarity;
that is, the seller can influence the expected valuation for specific items through rarity optimization,
and hence, the expected value of the package. Sheng et al. (2024) examine the strategies for selling
 9

bonus actions in video games, considering the uncertainties of tool necessity and value at the time
of use. They show how the distribution of skill among players and the inherent randomness of the
game influence selling strategies, including advance and spot sales or a combination of both.
In terms of deterministic mechanisms, some sellers may rely on assortment rotation (Bernstein
and Martı́nez-de Albéniz 2017, Lei et al. 2023) to induce freshness or spread demand across multiple
products. Although we model strategic consumers similar to (Bernstein and Martı́nez-de Albéniz
2017), in our setting, product rotation is achieved via randomization. Moreover, the emphasis on
rarity in our work leads to fundamentally different insights.

Conspicuous Consumption. Our assumption of status-seeking consumers is aligned with a

rich literature on conspicuous consumption for status. In many settings, buyers often purchase
goods that are “hard to find” to conspicuously display their exclusivity and social status (Amaldoss
and Jain 2005, Tereyağoğlu and Veeraraghavan 2012, Rao and Schaefer 2013). For instance, Koford
and Tschoegl (1998) point out consumers demonstrate a strong preference for items simply due
to their rarity, often willing to pay a premium for such goods. As with our setting, consumers’
preference for exclusivity in these papers is typically captured in the form of a rarity bonus, which
in turn depends on their perceived beliefs about inventory levels. Additionally, an item’s scarcity
can also serve as a signal of its quality (Stock and Balachander 2005), which highlights a secondary
mechanism through which sellers can benefit from rarity. Despite the similarities in the underlying
consumer models, we note that our work fundamentally differs from this literature, given our
specific focus on randomized selling mechanisms, namely the blind box strategy.

2. Model
In this section, we present our base model that captures the trade-offs faced by a seller when
designing a blind box. More precisely, we consider a monopolistic seller who controls an assortment
of n regular items labeled [n] = {1, 2, . . . , n} as well as a single premium or special item denoted by
s. The cost of manufacturing a single unit of either the regular or the special item is c, indicating
that the special item is similar to the regular item in terms of its attributes and derives any
additional value from its perceived rarity.
On the other side of the market, there is a unit mass of infinitesimal consumers. Under the
base model presented in this section, the consumers are homogeneous, and their valuation for each
regular item is v, which is exogenously specified and independent of the selling strategy chosen by
the seller. If a consumer obtains more than one copy of a regular time, we assume the consumer
draws no utility out of any repeated items beyond the first one that is valued at v. Further, the
10

consumer’s valuation over multiple unique items is additive, i.e., if the consumer obtains k ∈ [n]
unique regular items, the overall value derived is kv. In contrast, the valuation for the special item
is endogenous and status-driven, i.e., it depends on the available supply of the special item relative
to individual regular items. Formally, the valuation for the special item, vs (r), is given by

vs (r) = (1 + (1 − r)α) · v, (1)

where r ∈ [0, 1] denotes the perceived rarity6 of the special item and the slope α ≥ 0 is a status
parameter and captures the (multiplicative) gain in a consumer’s value for the special item as it
becomes rarer. Broadly speaking, larger values of α imply that consumers are willing to pay a
greater premium for the status of owning a rare item. Next, we define the rarity parameter r as the
ratio of the number of special items to the average number of regular items in the market, where
the average is over each unique regular item. A smaller value of r ∈ [0, 1] implies that the special
item is rarer. The motivation for such a model stems from the social comparison theory posited
by Festinger (1954) and used widely in a number of fields (Roels and Su 2014, Momot et al. 2020,
Feylessoufi et al. 2023, Mai and Hu 2023). In our context, one can imagine that owners of special
items compare their endowment with others in the market and derive an additional value of α
when they find that the item is exclusively in their possession, which occurs with probability 1 − r.
Finally, in the case of randomized mechanisms such as the blind box, consumers may not know
the exact realization of the number of special and regular items in equilibrium. Therefore, we
assume that consumers’ purchase decisions are made in terms of the perceived rarity r (that can be
shaped by the announced drop rates of special items vs. regular items), which represents their ex-
ante beliefs. We provide a formal mathematical definition of the rarity parameter once we introduce
the selling mechanisms.

Pricing Strategies
We consider a profit-maximizing seller who can choose between one of two mechanisms7 for selling
items: separate sales and blind boxes. In addition, we assume that the seller can produce any
desired quantity of each item to meet the realized demand—e.g., this can occur if there is low lead
time or if the seller can store excess supply at low cost. However, we note that the seller always
incurs a fixed manufacturing/procurement cost of c ≤ v for each item produced/procured. We now
define the two pricing mechanisms studied in this work.
6
In Lemma 2, we show that this perceived rarity equals the actual rarity of the item in equilibrium in the case of
blind boxes.
7
In reality, the seller may have an infinite space of mechanisms to choose from. However, given that the goal of this
work is to compare blind boxes to a more traditional selling method, we restrict our focus to these two strategies.
 11

Separate Selling Without Rationing. Under separate sales, the seller sets the prices for both
regular and special items, and consumers make independent purchase decisions on each item. Math-
ematically, suppose that pr and ps denote the prices of the regular and special items, respectively.
Then, each consumer purchases each regular item as long as v ≥ pr and the special item as long as
vs (r) ≥ ps . Given that the regular items are all symmetric (in terms of the seller’s valuation), it is
optimal for the seller to set the same price pr for each of the n regular items, and we take this for
granted in this rest of this paper.
Additionally, we make the assumption that the seller cannot ration the items or deny selling
the item to any buyer whose willingness to pay is no less than the price. As mentioned in the
Introduction, this assumption reflects the negative externalities associated with artificial scarcity
in many industries (Anderson et al. 2006). Since consumers are homogeneous in their valuation
of the special item, the no-rationing assumption implies that the only feasible solutions are those
where the seller either sells the special items to all buyers (i.e., vs (r) ≥ ps ) or none (vs (r) < ps ).
We relax this assumption in later sections via two natural modeling extensions: a) allowing the
seller to ration the item among consumers in Section 5.1, and b) incorporating heterogeneity in
consumers’ valuation for the regular and special items in Section 5.2, so that not all consumers are
willing to purchase the item at a higher price or even if they do, they may not necessarily be able
to obtain the item.
Blind Box. The seller’s task is to design an optimal blind box package, characterized by the
parameters (pb , q), which are revealed to all consumers (e.g., see Figure 1). Here, pb denotes the
price of a single blind box, and q represents the probability of drawing a special item. Once again,
we assume symmetry on the regular items; conditional on the blind box not containing a special
item (i.e., with probability 1 − q), all n regular items are equally likely to be drawn.
Given a blind box (pb , q), the consumer’s valuation for purchasing the blind box for the first time
can be expressed as (1 − q) · v + q · vs , where vs is the consumer’s valuation for the special item.
Naturally, the consumer may purchase this blind box multiple times with diminishing returns. After
j purchases, let nr ∈ [0, n] and ns ∈ {0, 1} denote the number of unique regular and special items
that the consumer has in her possession. Then, the consumer’s expected valuation for purchasing
the j + 1-st blind box can be written as

n − nr
vB (nr , ns ) = (1 − q) v + q(1 − ns )vs . (2)
n

Based on the blind box parameters, the probability of obtaining a special item is q, whereas that
of obtaining any regular item is (1 − q)/n. The ratio of these quantities, q/((1 − q)/n), represents
12

the consumer’s perceived rarity of the special item, r. As we prove in Lemma 2, this matches the
ratio of the equilibrium purchased quantity of the special item to that of any regular item, in
expectation over all consumers. Finally, using Equation (1) and substituting r with q/((1 − q)/n),
vs can be derived as follows:  
q
vs = 1 + (1 − )α · v. (3)
(1 − q)/n
Remark 1. The no-rationing argument trivially holds under the blind box policy, as buyers are
always able to purchase a blind box whenever their willingness to pay is no lower than its price.

2.1. Seller Profits

We conclude this section by briefly characterizing the optimal policies of the consumer and the
seller under each pricing rule.
Separate Selling. The seller’s goal is to select prices for the regular and special item—i.e., pr and
ps —in order to maximize its profit. Denote by xr and xs the quantities sold for each regular and
special item, respectively. Given these prices, the seller’s profit ΠS under separate selling can be
expressed as
ΠS (pr , ps ) = n · xr (pr − c) + xs (ps − c).

In order to characterize the optimal prices, first note that a consumer will purchase all n regular
items as long as v ≥ pr and the special item if vs (r) ≥ ps . We assume that each consumer only
buys each product at most once, as there is no extra utility to be derived from subsequent direct
purchases. Therefore, the seller’s profit is maximized when both the regular and special items are
priced at pr = ps = v. Recall that under the no rationing argument, ps > v is not possible as it
would require the seller to ration the special item, i.e., vs (r) = v(1 + α(1 − r)) > v implies that
r < 1. Further, under this set of prices, each buyer purchases all n regular items and one special
item. Thus, the optimal profit under separate selling, Π∗S , is given by

Π∗S = (n + 1)(v − c).

Blind Box. The seller’s expected profit for the blind box (pb , q) is the profit margin, pb − c, multi-
plied by the expected number of blind boxes purchased by a consumer, which we characterize below.
After purchasing the blind box for a given number of times, suppose that the consumer has received
nr ∈ {0, 1, 2, . . . , n} regular items and ns ∈ {0, 1} special items. Then, applying Equation (2), the
consumer will continue purchasing the blind box as long as

n − nr
(1 − q) v + q(1 − ns )vs ≥ pb ,
n
 13

where vs is as defined in Equation (3). Clearly, the consumer would cease purchasing once the
expected marginal valuation falls below the price. Moreover, even if the consumer obtains the
special item, she may continue to buy the blind boxes to collect more unique regular items. The
following lemma provides precise conditions for the optimal stopping time.

Lemma 1. For a given blind box (pb , q), there exist thresholds n ≤ n̄ such that any consumer
stops purchasing if one of the following conditions is met:
(i) if a consumer has not drawn the special item and has collected exactly n̄ unique regular items
(i.e., ns = 0, nr = n̄);
(ii) if a consumer has drawn the special item and has collected exactly n unique regular items (i.e.,
ns = 1, nr = n);
(iii) if the consumer’s latest purchase results in a special item, they have collected at least n and
at most n̄ − 1 unique regular items, and they have not received the special item in any prior
purchase (i.e., ns = 1, nr ∈ [n, n̄ − 1]).

As mentioned above, the inherent randomness in blind boxes can lead consumers to obtain
repeated items, though they intend not to, which can be avoided in separate selling but is unavoid-
able under blind box selling. Moreover, Lemma 1 shows an additional consequence of this random-
ness: for a given price and drawing probability, the stopping condition dynamically depends on the
purchase process. Specifically, if a consumer has yet to draw a special item, she tends to purchase
a greater number of regular items before stopping, in contrast to the scenario where a special
item is drawn early on. This indicates that obtaining the special item significantly influences the
consumer’s decision on when to stop purchasing.
In order to formally define the expected number of purchases before the stopping point is reached,
consider any sample path of purchases for a given buyer and define the (random) variable Yj = 1 if
the consumer satisfies one of the three conditions mentioned above for the first time after buying
the blind box j times (i.e., Yi = 0 for all i < j). Denote the profit under the blind box by ΠB . Then,
the problem under the blind box can be characterized as
∞
X
max ΠB = (pb − c) · j · Pr(Yj = 1). (4)
pb ,q
j=0

Let p∗b and q ∗ denote the optimal price and drawing probability, and Π∗B represents the optimal
profit to problem (9). Given the multiplicity of stopping conditions, it is challenging to obtain
a closed-form expression for the expected number of purchases under the optimal pricing policy.
In later sections, we overcome this tractability issue by showing that the optimal price follows a
14

natural structure and then deriving lower and upper bounds on the expected number of purchases
that closely match the actual value.
We end this section by formalizing the notion that consumers’ perceived beliefs about rarity under
the blind box strategy are consistent with the actual realized purchased quantities in equilibrium.

Lemma 2 (Belief is Consistent with Outcome). For any given blind box (pb , q), the ratio
of the expected number of special items collected by consumers in equilibrium to the expected quantity
of items purchased equals the drawing probability q.

On the surface, it would appear that the different stopping conditions exhibited by consumers
(e.g., see Lemma 1) would distort the realized purchase quantities in equilibrium. However, as we
argue in the proof of the lemma, along every sample path and with each purchase, a fraction q
of the consumers who have not yet reached their stopping condition receive the special item. As
a result, the corresponding rarity, r = q/((1 − q)/n), is also maintained in equilibrium. A detailed
proof of this property is provided in Online Appendix D.

3. Preliminary Analysis and Warm-up Results

Before we dive into characterizing the performance of blind box selling and separate selling for our
base model—in Section 4—we first consider some special cases to illustrate better how different
aspects of blind boxes influence their performance. In particular, in this section, we investigate two
well-motivated cases of our base model: (i) when there is no special item and (ii) when there is an
infinitely larger number of regular items. In both these cases, we prove that it is optimal for the
seller to use separate selling to maximize profits. The mechanisms underlying these results provide
new insights into what drives the performance of the blind box strategy, which we leverage in later
sections. We begin by analyzing the seller’s pricing strategy for blind boxes and tying this to the
consumer’s purchase behavior, which was characterized in Lemma 1.

3.1. Optimal Blind Box Pricing and Stopping Conditions

In Lemma 1, we identified three conditions under which a consumer would no longer continue to
purchase a blind box, essentially corresponding to cases where the buyer stops right after drawing
the special item or collects a sufficient number of unique regular items with or without the special
item. Specifically, recall from Lemma 1 that n and n̄ denote the maximum number of unique regular
items—with and without the special item, respectively—that the consumer can obtain beyond
which their marginal value for the next purchase is less than the given price pb . We first establish
a simple relationship between n and n̄, which we then use to derive a simple structural form for
the optimal blind box price p∗b .
 15

As per Lemma 1, under the blind box price pb , a consumer will stop purchasing once they have
acquired (i) one special item and n unique regular items8 or (ii) n̄ (≥ n) unique regular items
without a special item or (iii) a consecutive stream of unique regular items whose count is between
n and n̄ with the last purchase resulting in a special item.9 The first stopping condition based on
n implies that
n−n n−n+1
(1 − q) · v < pb ≤ (1 − q) · v. (5)
n n
The right-hand side of the above inequality denotes the consumer’s marginal value for the blind
box when they have received exactly n − 1 unique regular items and a special item from prior
purchases. If pb were greater than (1 − q)(n − n + 1)v/n, then the consumer would stop purchasing
at n − 1 regular items, which would contradict our definition of n̄. Similarly, the second stopping
condition based on n̄ implies that

n − n̄ n − n̄ + 1
(1 − q) + qvs < pb ≤ (1 − q) · v + qvs , (6)
n n

where vs = (1 + (1 − qn/(1 − q))α) · v. Once again, the right-hand side above denotes the consumer’s
marginal value from the blind box when prior purchases have yielded n̄ − 1 regular items and no
special item. Denote by n+ ∈ [n, n̄ − 1] the number of unique regular items possessed by a consumer
when the special item is drawn for the first time. The third stopping condition implies that

n − n+ n − n+
(1 − q) · v < pb ≤ (1 − q) · v + qvs .
n n

The left-hand side above indicates that the consumer continues to purchase after receiving the n+ -th
unique regular item, while the right-hand side implies that the consumer stops upon obtaining the
special item. Figure 2 provides an illustration of the relationship between the marginal valuations
and the stopping conditions.
We now establish a relationship between n̄ and n for a given blind box (pb , q). Suppose that
κ = nqvs /(1 − q)v denotes the relative value of a special item in the blind box compared to that of
a regular item. Then, it follows from Equations (5) and (6) that n̄ = n + min{n, ⌊κ⌋}, where ⌊κ⌋
denotes the largest integer that is less than or equal to κ.
The introduction of κ allows us to characterize the optimal price in terms of the stopping
conditions, which we do in the following lemma.

8
When we say a certain number of unique regular items, these unique items may include repeat purchases.
9
We suppress the dependence of n̄ and n on pb for conciseness.
16

i i 1 1 · · · · · · · · · · · · 𝑛𝑛� 𝑛𝑛�
𝑣𝑣 𝑣𝑣
𝜅𝜅 ⋅𝜅𝜅 1⋅ −1 𝑞𝑞− 𝑞𝑞
𝑛𝑛 𝑛𝑛

1 12 23 34 4 · · · · · · 𝑛𝑛 𝑛𝑛 𝑠𝑠 𝑠𝑠 ii ii 1 1 · · 𝑠𝑠 𝑠𝑠 · · 𝑛𝑛 𝑛𝑛
𝑣𝑣 𝑣𝑣
1 −1 𝑞𝑞− 𝑞𝑞 𝑞𝑞𝑣𝑣𝑠𝑠𝑞𝑞𝑣𝑣𝑠𝑠
𝑛𝑛 𝑛𝑛
iiiiii 1 1 · · · · · · · · 𝑠𝑠 𝑠𝑠

𝑛𝑛𝑟𝑟 𝑛𝑛∈𝑟𝑟 {𝑛𝑛,

∈ {𝑛𝑛,
𝑛𝑛 +𝑛𝑛 1,
+… 1,,…
𝑛𝑛� ,−𝑛𝑛� 1}
− 1}

(a) Valuations of Items in a Blind Box (b) Stop Conditions

Figure 2 A depiction of the consumer’s purchase behavior and marginal utility under the blind box model.

Notes. Label ‘s’ (solid red) refers to the special item, whereas the numbers {1, 2, . . . , n} (solid green) refer to the regular

items. Part 2(a) illustrates the consumer’s marginal valuation for each item in the blind box, with the special item

valued at κ times that of a regular item. Part 2(b) depicts the three stopping conditions from Lemma 1 respectively.

Lemma 3. For any given blind box where the probability of drawing the special item is given
by q < 1/(n + 1), the optimal price pb has the following form: pb (q, k) = (1 − q)(k + 1)v/n, where
k ∈ {0, 1, . . . , n − 1} ∪ {κ − 1, κ, . . . , κ + n − 1}.

Note that it is not necessary for k to be an integer when κ is not an integer. In such a case,
under the prices pb (q, κ − 1) and pb (q, ⌊κ⌋ − 1), the consumer purchases the blind box for the same
number of times in expectation. However, since pb (q, κ − 1) > pb (q, ⌊κ⌋ − 1), the former price is more
profitable. Moreover, given a blind box with price pb (q, k), it is clear that the choice of k directly
determines the consumer’s stopping condition. Further, if k is sufficiently small, the consumer may
not stop purchasing until they obtain the special item, e.g., when n̄ > n in Lemma 1. We now
present a simple example to illustrate better the relationship between the price of the blind box
and the condition under which the consumer stops purchasing.
Example 2. Consider an instance with n = 4 regular items and a single special item. Further,
suppose that the probability of drawing the special item under the blind box q is fixed such that
κ = qvs /((1 − q)v/n) = 2; one can always adjust the free parameters α and v to achieve such an
outcome. For simplicity, let ν = (1 − q)v/4. The initial valuation of the blind box is 6ν. By Lemma 3,
the optimal price of the blind box must be in the form of pb = (1 − q)(k + 1)v/n = (k + 1)ν, where
k ∈ {0, 1, 2, 3, 4, 5}. If the consumer obtains a unique regular item or a special item for the first
time, the expected marginal valuation of continuing to buy a blind box decreases by ν or 2ν,
respectively. We analyze three price scenarios: (1) low price, where k ∈ {0, 1}; (2) moderate price,
 17

where k ∈ {2, 3}; (3) high price, where k ∈ {4, 5}, and discuss how consumer behavior differs in
each case. The optimal price will be chosen from one of those cases.
(1) Low Price: In this case, the consumer will not stop if she does not obtain the special item
because pb ≤ qvs = 2ν. First, let pb = ν, and so k = 0. In view of pb < qvs , the consumer is
driven not to stop purchasing until she draws the special item. Further, even after drawing
the special item, the consumer will continue purchasing until she receives n = 4 unique regular
items. Therefore, for this choice of price pb = ν, the consumer will stop purchasing only if she
possesses 4 regular unique items and the special item. Second, if pb = 2ν, the consumer will
stop until she obtains at least n = 3 unique regular items and the special item.
(2) Moderate Price: First, let pb = 3ν implying that k = 2. In this case, pb > qvs , and thus, the
consumer has different stopping conditions based on whether or not she has already obtained
the special item. On the one hand, if the consumer has already drawn the special item, then
she will continue purchasing the blind box until the expected marginal valuation falls below
the price pb —this implies that n = 2. On the other hand, if the consumer has not drawn the
special item, then she will purchase the blind box until she receives n̄ = 4 unique regular items.
Second, if pb = 4ν, we have n = 1 and n̄ = 3.
(3) High Price: In this case, the consumer will stop once she obtains the special item because
the expected marginal valuation of keeping going falls below the high price. First, let pb = 5ν,
which implies that k = 4. Once again, there are two cases: (i) if the consumer has obtained
2 unique regular items, the valuation of the blind box falls below the price pb even if the
consumer has not yet received the special item; (ii) If she has obtained no more than 2 unique
regular items first (i.e., 1 unique regular item), she will stop once she obtains the special item.
Therefore, for this price, n = 0 and n̄ = 2. Second, if pb = 6ν, we have n = 0 and n̄ = 1.
We provide a more generalized analysis of the stopping conditions in Online Appendix C.

3.2. Warm-up Case: Absence of the Special Item

As mentioned in the introduction, rarity plays a crucial role in accentuating the value of a blind
box, which in turn can drive further sales. To illustrate this point, we consider a special case of our
model characterized by the absence of rarity. Formally, we study a market with n regular items in
the absence of a status-driven special item.10 We show that in this scenario, separate selling always
outperforms the blind boxes in terms of profitability.
First, recall that under separate selling, the seller sets the price pr for each regular item to
maximize the profit ΠS . Since consumers prefer all regular items equally, the seller can set an
10
Alternatively, one can perceive this to be a market where consumers are not status-driven, and so α = 0.
18

optimal price, p∗r = v, to extract the entire consumer surplus. Each consumer then purchases all
n regular items, yielding a total profit Π∗S = n(v − c). Next, in the absence of a special item, the
seller’s decisions under the blind box model revert to setting only the price pb (k) = (k + 1)n/v for
some integer 0 ≤ k ≤ n − 1 as per Lemma 3. Moreover, in response, the consumer would continue
purchasing the blind box until they obtain n − k unique regular items. The seller’s profit under the
blind box for price pb (k) can be represented as
∞
X
ΠB (k) = (pb (k) − c) · jPr(Yj = 1). (7)
j=0

As defined in Section 2.1, Yj is a random variable that equals one if the consumer satisfies the
stop condition for the first time after buying the blind box j times (i.e., Yi = 0 for all i < j). Given
the setup above, we now compare the optimal profit under both schemes without the special item
in the following proposition.

Proposition 1 (No Special Item). The optimal profit under separate selling, Π∗S , strictly
dominates the optimal profit under blind box selling, Π∗B , in the absence of the special item.

To understand the intuition behind Proposition 1, it is important to note that lowering the price
of the blind box can lead to a non-linear increase in the expected number of purchases. For example,
consider the two prices pb (k) = (k + 1)n/v and pb (k − 1) = kn/v. Following a price decrease from
pb (k) to pb (k − 1), the consumer would only stop purchasing the blind box after they receive one
additional regular item—in this case, the number of unique regular items before stopping increases
from n − k to n − k + 1. However, the probability of drawing a new regular item can be modeled
in terms of a geometric distribution. For example, the probability of the consumer drawing the
(n − k +1)-st unique regular item is k/n, and therefore, the expected number of additional purchases
after the price drop is n/k. In Lemma C4 in Online Appendix D, we leverage this insight to derive
a closed form expression m(k) for the expected number of purchases under price pb (k), which we
state below:
n
X n
m(k) = .
i=k+1
i
Unfortunately, even though the seller benefits from lower prices via increased volume, this effect
is counterbalanced by the decrease in price. In other words, we show that under smaller prices,
the low margins (i.e., (pb (k) − c)) for the blind box strategy puts it at a relative disadvantage to
separate selling, where the seller is able to extract full surplus under a price of pr = v.
We note that Proposition 1 complements the findings in Chen et al. (2021) by identifying
conditions under which separate selling outperforms blind boxes. More specifically, Chen et al.
 19

(2021) consider an alternative setting without special items where the value of each regular item is
drawn from a distribution. In the absence of costs, they show that the exact relationship between
the revenue obtained under the blind box (loot box in their work) and separate selling depends
on the nature of the valuation distribution. In contrast, we focus on a more specific valuation
structure—deterministic and symmetric valuations—where separate selling is optimal even when
one incorporates costs.
The failure of the blind box model without the special item raises a fundamental question: is it
possible for the seller to benefit from the increased purchase volume at higher prices? In Section 4,
we will show that the presence of the special item allows the seller to exploit its rarity to enhance
the valuation of the special item, thereby increasing the value of the blind box. In other words, the
seller can set a relatively higher price for the same stop condition than in scenarios without the
special item.

3.3. Warm-up Case: Infinite Regular Items

Our result in Section 3.2 underscores the importance of item heterogeneity and rarity in boosting
the value of a blind box relative to selling items separately. In order to develop a more nuanced
understanding of this effect, we now study another special case of the base model with an infinite
number of regular items (i.e., n → ∞). This provides us with the opportunity to analyze another
extreme case in terms of rarity, as increasing the number of regular items allows the seller to make
the special item increasingly rarer under the blind box model.
Formally, we consider the same model developed in Section 4 and present results for the case
where the number of regular items becomes infinitely large while still retaining a single special
item. In particular, we compare the ratio of the profit under separate selling to that of the blind
box model as n → ∞. Recall that the optimal profit under separate selling for a given value of n
can be written as Π∗S = (n + 1) · (v − c). Similarly, for a given n, the profit under blind boxes can
be written as ΠB (k) = (pb (k) − c) · m(k), where we use m(k) as shorthand for the expected number
of purchases by consumers under a price of pb (k) = (1 − q)(n − k)v/n. We now present our main
result for this special case.

Proposition 2 (Infinite Regular Items). As the number of regular items approaches

infinity, lim Π∗B /Π∗S ≤ 1, where Π∗B and Π∗S represent the optimal profits under blind box selling
n→∞
and separate selling, respectively.

The intuition for this result is straightforward. The probability of drawing the special item
q < 1/(n + 1) in order to maintain its relative rarity. Therefore, as n → ∞, q → 0. However, in the
20

process, the expected value of the special item under the blind box qvs → 0 as the maximum value
vs is bounded from above by v(1 + α). In other words, consumers lose interest in the special item
because they perceive it as too difficult to obtain. In such an event, separate selling dominates as
it maximizes profits from the regular items (e.g., see our result in Section 3.2).
Although our result in this section is straightforward, it demonstrates the dual role of rarity. On
the one hand, rarity is essential to boost the valuation of the blind box. On the other hand, too
much rarity (i.e., q → 0) does not guarantee success as it erodes the contribution of the special item
to the blind box. Therefore, when considering the design of blind boxes, sellers need to balance the
trade-off between rarity and availability.

4. Homogeneous Consumers
In this section, we present our main results for the base model with homogeneous consumers under
the no rationing assumption. Particularly, we compare the profits under separate selling and blind
boxes and identify the optimal selling mechanism as a function of two parameters: the status
parameter α and the unit-manufacturing cost c. At a high level, the results provide sellers with
clear guidelines on the type of product (i.e., based on cost) and the market conditions (i.e., how
status-seeking consumers are) under which blind boxes are likely to outperform more traditional
selling methods.
We now state our main proposition, which defines the regions (as a function of α and c) under
which separate selling dominates blind boxes and vice versa.

Proposition 3 (Profit Comparison). Suppose that β = c/v. The relationship between the
seller’s optimal profit under the blind box model Π∗B and optimal profit under separate selling Π∗S
is given by:
(i) (Effect of Status) For any given β ∈ [0, 1], there exists a threshold ᾱ such that Π∗B > Π∗S
if and only if α > ᾱ. Moreover, Π∗B (α, β)/Π∗S (α, β) is non-decreasing in α.
(ii) (Effect of Cost) For any given α > 0, there exist two thresholds β L ≤ β H ∈ [0, 1] such that
Π∗B > Π∗S if and only if β < β L or β > β H .

Effect of Status (α). Proposition 3 part (i) identifies that larger values of α (i.e., consumers value
status more) benefit blind boxes, whereas smaller values of α lead to separate sales outperforming.
Figure 3 depicts this pictorially and further adds to the proposition by highlighting that the cutoff
ᾱ above which blind boxes dominate separate selling is small under low- and high-cost regimes. In
fact, as the ratio β = c/v approaches one, the cutoff ᾱ approaches zero, indicating that blind boxes
are always better.
 21

60

0.8
50

0.6 40

30
0.4

20

0.2
10

0 0
0 20 40 60 80 100

Figure 3 The ratio of (Π∗B (α, β) − Π∗S (α, β))/Π∗S (α, β) at different status seeking levels α and ratios of c/v

(n = 12, v = 10).

Notes. For any fixed c/v, with the increase of α, separate selling dominates first, followed by blind box selling. For

any fixed α, separate selling dominates when costs are intermediate, while blind box selling dominates in the low-

and high-cost regimes. Moreover, as α increases, the intermediate cost region in which separate selling dominates

shrinks. Finally, when c and v are very close, the profit from separate selling approaches zero, consequently making

the denominator approaches zero, which results in an extremely large ratio.

To understand the mechanism underlying this result, we first note that in the case of separate
selling, the seller cannot restrict the supply of the special item under the no rationing assumption.
As a result, the only option is to sell the special item to all buyers due to their homogeneity. Since
separate selling cannot make use of consumers’ status-seeking nature, it follows that larger values
of α do not lead to improved profits under this pricing strategy. When it comes to blind boxes, an
increase in α enhances the valuation of the special item as vs = (1 + (1 − r)α)v. Therefore, even if
one were to hold the probability q of drawing the special constant, this would lead to an increase
in the expected value of the blind box, implying either a) more pricing power for the seller or b)
more purchases for the consumer. In scenarios where consumers are highly status-seeking, blind
box selling dominates across all cost levels. Finally, based on this result, one may be tempted to
conclude that in an alternative model where the seller can leverage status under separate selling,
blind boxes may no longer outperform. In Section 5.1, we, in fact, consider such a model and show
that blind boxes continue to be more profitable under a variety of cases, in particular, when the
status-seeking level α is high enough.
Effect of Cost (c). In comparison to part (i), Proposition 3 part (ii) points to an intriguing
22

non-monotonicity, namely, blind box selling is more profitable when costs are either low or high.11
Once again, Figure 3 provides a visual representation of this result across different values of α.
More specifically, in low-cost regimes, it is optimal for the seller to offer the blind box at a low
margin. As described in Section 3.2, even a small decrease in the price of the blind box pb leads to
a disproportionate or non-linear increase in the expected number of times a consumer purchases a
blind box. As an illustrative example, if one only considers regular items and decreases the price
of the blind box from pb = (1 − q)(n − k + 1)v/n to pb = (1 − q)(n − k)v/n, the expected number of
purchases by a consumer can increase by n/k. However, in separate selling, reducing the price does
not encourage consumers to buy more items due to valuation homogeneity. Conversely, in high-cost
regimes, by choosing an appropriate low probability of drawing the special item, the seller can sell
the blind box at a price pb > v. On the other hand, under separate selling, the price of both the
regular and special items is capped at v, which results in poor margins as the cost approaches v.
In conclusion, the two cases reflect the two levers available to blind boxes to boost profits, namely
the volume of purchases and the profit margin.

Consumer Surplus. Our previous results illustrate that sellers can profit from blind boxes,
e.g., via the gains in purchase volume, particularly if costs are small. However, such an outcome
may not necessarily benefit consumers, especially if they end up obtaining duplicate copies of the
same item, which provide zero marginal value. To address this issue, we examine the effect of
purchasing blind boxes on consumer utility, i.e., surplus. Surprisingly, our analysis reveals that
blind boxes can achieve strictly positive surplus (and thereby improve on separate sales) even in
regimes where the expected purchase volume is high.

Lemma 4. For any given blind box (pb , q), the expected consumer surplus is non-negative. Fur-
ther, there exists β̄s such that the expected consumer surplus under optimal blind box selling is
strictly higher than the surplus under optimal separate selling, which is 0, as long as β < β̄s .

While consumer utility is always zero under optimal separate selling (as the seller can extract
full surplus via the optimal prices), the blind box strategy can permit consumers to retain a non-
negative surplus. We note that although some sample paths may result in a negative surplus owing
to an abnormally large number of purchases, such events are rare, and the consumer’s expected
surplus is always non-negative and, in fact, positive under the optimal blind box as long as the
manufacturing cost is not too high. The possible positive surplus left to consumers occurs for

11
Although the result in Proposition 3 is stated in terms of β = c/v, we simply use costs for parameterization as the
denominator v is a fixed quantity.
 23

the following reason: the inherent randomness of the blind box strategy results in heterogeneous
utilities for consumers across different sample paths. Consequently, the seller cannot necessarily
capture the entire expected consumer surplus with a uniform price for each blind box, particularly
when the manufacturing cost is low, which results in a low optimal blind box price.
In summary, combining Proposition 3 and Lemma 4, we see that blind boxes benefit the market
as a whole by simultaneously strictly improving both seller profit and consumer utility over separate
selling when the manufacturing cost is sufficiently low. Interestingly, blind boxes achieve this via
two distinct mechanisms. First, blind boxes can better price discriminate along different sample
paths. Second, the special item is rarer under blind box selling than in separate selling. The former
increases the overall purchase quantity of regular items in the market, and the latter increases the
overall value of the special item by boosting its rarity, both increasing social welfare and consumer
surplus when the seller cannot fully extract consumer surplus.

4.1. Designing the Optimal Blind Box

We now complement our findings in Proposition 3 by investigating the optimal price and rarity
levels for the blind box as well as its impact on consumer’s stopping behavior. The purpose of the
following corollary as well as Figures 4, 5, and 6 is to provide insights on how the seller should set
the blind box parameters in different regimes.

Corollary 1 (Properties of the Optimal Blind Box). For any given instance with
β L < β H , suppose that (p∗b , q ∗ ) represents the parameters of the optimal blind box. Then, the follow-
ing properties hold in regimes where blind box selling is more profitable than separate selling:
(i) (Margins under High Cost) If c/v > β H , then p∗b > v.
(ii) (Sales under Low Cost) If c/v < β L , then p∗b ≤ v. Further, the expected number of blind
boxes purchased by the consumer is at least n + 1.
Here, β L and β H are as defined in Proposition 3.

In simple terms, the above corollary shows that under high manufacturing costs, it is optimal
for the seller to price the blind box above the price of the items under separate selling that is at
the customer valuation. In other words, in high-cost regimes, the profitability of the blind box is
driven by its considerable price and unit margins. On the other hand, in low-cost regimes, we see
that the aggregate sales of blind boxes exceed that of the items under separate selling, indicating a
different mechanism driving the outperformance of blind boxes. We now use numerics to spotlight
these findings and develop deeper insights into the properties of the optimal blind box.
24

300 14

250 Blind Box 12

Separate Selling
200 10

150 8

100 6 Blind Box

Separate Selling
50 4

0 2
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

(a) Profit (b) Price

120 0.04

100 Blind Boxes

Separate Selling Blind Box

80 0.03

60

40 0.02

20

0 0.01
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

(c) Expected Purchases (d) Probability of Special Item

Figure 4 The profit, price, expected purchases, and probability of drawing the special item under the optimal

blind box and separate selling strategies at different ratios of c/v (n = 12, v = 10, α = 32).

Notes. As the cost increases, the optimal profit decreases (Panel a) under both mechanisms. Around c = 4, one can

observe a shift in the blind box policy as price increases rapidly with cost (Panel b). This leads to a decrease in the

expected quantity of blind boxes purchased by consumers (Panel c). Additionally, Panel d depicts an increase in the

probability of obtaining the special item at higher cost levels. This denotes a shift in the blind box strategy from

incentivizing more purchases (low q ∗ at low costs) to maximizing margins (higher q ∗ at high costs).

Comparative Statics. Figure 4 illustrates the optimal blind box parameters (price pb and proba-
bility of drawing special item q) at different values of c ∈ [0, v]. The figure provides an intriguing
perspective on how the seller’s optimal blind box strategy changes with respect to the cost change
(recall that the optimal price under separate selling is always pr = ps = v). Specifically, at low costs,
the seller’s pricing strategy optimizes for volume. In other words, the price of the blind box is
considerably lower than that of the items under separate selling, which is also supported by making
the special item extremely rare (small q). Together, these strategies ensure that the consumer pur-
chases the blind box a sufficiently large number of times before reaching their stopping condition
(e.g., until they find the special item), which leads to large profits. On the contrary, as manufac-
turing becomes more costly, this strategy is no longer viable as its margins shrink and eventually
 25

become negative. Therefore, at higher costs, the seller is forced to raise the blind box prices, even
possibly higher than the customer valuation for each unique regular item. This is accompanied by
an increase in q as the seller doubles down on maximizing margins (q ∗ is chosen to maximize the
expected value of the special item, i.e., qvs ). As expected, this results in a sharp drop in purchases,
but the seller is able to compensate with the increased profits per unit sale; this is particularly
pronounced when pb > v.

0.045 0.4
Probability Density

0.035

Probability Density
0.3

0.025
0.2

0.015
0.1

0.005

0 0
0 20 40 60 80 100 120 140 160 180 200 0 5 10 15
Number of Purchases Number of Purchases

(a) pb = 4.9 (b) pb = 11.4

Figure 5 The distribution of purchases under the blind box for different choices of the blind box price pb

(n = 12, v = 10, α = 35, q = 0.02).

Notes. Panel (a) illustrates a heavy tail in the expected number of purchases, which leads to a significant transaction

volume for blind boxes when the price is small (pb = 4.9). Panel (b) depicts that when the price of the blind is

higher than that of the items under separate selling, the purchase volume is smaller on average and exhibits lower

variance. Note that despite homogeneous consumers and consistent pricing, the inherent randomness of the blind box

mechanism results in heterogeneous sample paths, as the exact set of items acquired by consumers is random.

Purchase Behavior and Stopping Conditions. On the consumer side, Figures 5 and 6 present a
more detailed understanding of how the seller’s optimal policy impacts consumer decision-making.
Figure 5 illustrates that the main advantage of employing a blind box is that of incentivizing
buyers to engage in repeat purchases, e.g., in Figure 5(a), each consumer purchases the blind
box approximately 51 times in expectation when there are only twelve unique regular items. The
considerable purchase quantity is enabled by a low price—the price of the blind box is less than
half that of each regular item under separate sales. In contrast, in Figure 5(b), the blind box price
exceeds the consumer’s valuation for the regular item (i.e., pb > v), and thus (in expectation), each
consumer only purchases the blind box approximately 5.8 times before their stopping condition is
met. Figure 6 presents a more nuanced view of the stopping conditions at different costs, which
26

No Stopping With Special Item

12 Without Without Special Item
Special Item
10

Regular Items
8

0
0 0.2 0.4 0.6 0.8 1

Figure 6 The consumer’s stopping condition (i.e., the number of unique regular items obtained before the

consumer stops purchasing the blind box) with and without the special item as a function of cost

(n = 12, v = 10, α = 32).

Notes. As is apparent from Figure 4, the price of the blind box and the probability of obtaining the special increases

with c/v. Together, these shifts result in a sharp decrease in the number of regular items the consumer needs before

their marginal utility for the blind box turns negative, e.g., see Lemma 1.

indirectly influence the price (note that consumers’ decisions do not depend on the cost). As detailed
in Lemma 1, the stopping condition can be classified into two categories. If the consumer has
obtained the special item in a previous draw, then fewer unique regular items are required before
the stopping condition is met. In fact, in the extreme case when costs are high enough (therefore,
prices are high), the consumer stops purchasing immediately after drawing the special item. On
the contrary, in the low-cost regime, the consumer will not stop purchasing until they obtain the
special item (even if they draw all unique regular items), which explains the high average purchase
volume from Figure 5.

5. Extensions
In this section, we investigate the robustness of our findings from Section 4 by extending the
base model along three dimensions via (i) relaxing the no-rationing assumption, (ii) modeling
heterogeneous consumer valuations, and (iii) incorporating a lump-sum bonus for collecting all
items. Our results in this section reveal insights into how each of these mechanisms influences
the performance of blind boxes compared to separate selling and identify additional factors that
explain the prevalence of blind box selling in practice.

5.1. Separate Selling with Rationing

In our base model from Section 2, we imposed a no-rationing assumption that prevented the seller
from artificially limiting the supply of the special item, i.e., the seller could not deny the item to
 27

any buyer given their identical valuations. In this section, we relax this assumption and allow the
seller to limit the supply of the special item, capturing scenarios where manufacturers can impose
artificial scarcity, e.g., by releasing special editions of toys. Given this setup, the seller’s decisions
and consumer purchase outcomes remain unchanged (from Section 2) for blind boxes as there is
no need to limit supply under this mechanism. As a result, we only go over the additional details
required for separate selling below.
Separate Selling. The seller determines the prices for both the regular item, p̃r , and special item,
p̃s , as well as the supply of special item12 , xs ∈ [0, 1]. Based on Equation (1), the valuation of the
special item under separate selling is given by

ṽs (xs ) = (1 + (1 − xs )α) · v. (8)

Recall from Section 2 that the rarity r of the special item is defined as the ratio of the quantity
of this item sold to that of any regular item. Given that each regular item is purchased by the
entire (unit) mass of consumers, it follows that r = xs in this case. The seller’s profit under separate
selling can be expanded as

max Π̃S = (p̃r − c) · n + (p̃s − c) · xs . (9)

p̃r ,p̃s ,xs

Let p̃∗r , p̃∗s , and x∗s denote the optimal prices for the regular and special items, and the optimal supply
of the special item as per problem (9). The following lemma provides a closed-form expression for
these terms.

Lemma 5 (Separate Selling with Rationing). When the seller can optimize the supply of
the special item, the optimal price and quantities for the regular and special items are given by p̃∗r =
v, p̃∗s = ṽs (x∗s ) and x∗s = ((1+α)v − c)/(2αv), respectively, where ṽs (x∗s ) is as defined in Equation (8).

Lemma 5 demonstrates that the seller does indeed benefit from rationing the special item (x∗s < 1)
and leveraging its rarity to raise prices. Given that this model is a relaxation of our base model
from Section 2, it is natural that the introduction of rationing can only improve the seller’s profit
under separate selling. This, in turn, raises the question of whether or not the blind box mechanism
still holds value in this environment. Surprisingly, the following proposition answers this question
in the affirmative.

12
The seller’s profit-maximizing decision is to sell as many units of the regular items as possible as this maximizes
direct revenue and also boosts the rarity of the special item. As such, we do not explicitly include the decision variable
that denotes the supply of regular items.
28

Proposition 4 (Profit Comparison Under Endogenous Supply). Suppose that β =

c/v. When the seller is allowed to ration the supply of the special item, the relationship between
the optimal profit from blind box selling, Π̃∗B , and that from separate selling, Π̃∗S is given by:
(i) For any given β ∈ [0, 1], there exist thresholds 0 < αL ≤ αH such that Π̃∗B > Π̃∗S if α > α̃H , and
Π̃∗B ≤ Π̃∗S if α ≤ α̃L .
(ii) For any given α > 0, there exist thresholds 0 ≤ β L ≤ β H ≤ 1 such that Π̃∗B > Π̃∗S if β < β̃ L , and
Π̃∗B ≤ Π̃∗S if β ≥ β̃ H .

At a qualitative level, the above proposition is almost identical to our original result under
rationing (Proposition 3) with one key difference, namely that separate selling now outperforms
blind boxes in the high-cost regime. On the other hand, the blind box mechanism continues to be
advantageous when costs are low. Figure 7 provides a visual depiction of this result and additionally
illustrates that as consumers become more status conscious (larger values of α), the region in which
blind box selling achieves higher profits expands for a larger range of (low enough) manufacturing
costs.

1 2.6

2
0.8

0.6
1

0.4

0
0.2

0 -0.9
0 20 40 60 80 100

Figure 7 The ratio of (Π̃∗B (α, β) − Π̃∗S (α, β))/Π̃∗S (α, β) at different values of the status parameter α and

normalized costs c/v (n = 12, v = 10).

Notes. At small values of α, separate selling can be advantageous to the seller, particularly when costs are high.

However, as consumers become more status-seeking, the benefits of separate selling tend to be more limited. Moreover,

compared with Figure 3, separate selling is able to extract higher profits by enhancing the rarity of the special item.

Before explaining the drivers of Proposition 4, we first note that in our base model, blind box
selling was more profitable when costs were high due to higher margins, i.e., the seller was able to
 29

significantly raise the price of the blind box by exploiting the rarity of the special item. However, in
the current extension, it is also possible for the seller to effectively utilize rationing under separate
sales to raise the special item’s price when costs are high. Nevertheless, in the low-cost regime, the
blind box mechanism still remains optimal due to the volume effect described in Section 4, i.e., the
seller can set a low price to ensure that consumers purchase the blind box many times before their
stopping condition is met.
Finally, as α increases (e.g., in Figure 7), the advantage of blind box selling grows and eventually
becomes optimal at all cost levels when consumers are sufficiently status-seeking. This result is
particularly noteworthy due to the scale invariance of this setting. Put differently, larger values of
α now benefit both separate selling and blind boxes. However, blind box selling is able to utilize
the (special) item rarity more effectively for both the regular and special items (due to the items
being bundled together), whereas under separate selling, only the special item benefits from rarity.

5.2. Consumer Heterogeneity

In the base model, we assume that consumers are homogeneous in their valuation of the regular
item. In this section, we generalize the base model and consider an extension where consumers
exhibit heterogeneity in their valuation of the regular item. Formally, we model consumer valuation
for the regular item via a random variable V (with realization v̂) that is drawn from a Bernoulli
distribution, taking the value vL = v − ϵ and vH = v + ϵ with equal probability 1/2, where ϵ ∈ (0, v)
for some constant v ≥ c. In other words, half the consumers have a high-type (with value vH )
whereas the other half have a low-type (with value vL ). The distribution of valuations is common
knowledge. Further, the parameter ϵ represents the level of heterogeneity in the system with larger
values of ϵ denoting higher dispersion between the consumer types. We now describe the two selling
mechanisms as follows.
Separate Selling. Suppose that p̂r and p̂s denote the price of a single regular and special item,
respectively. Under this model, it is optimal for the seller to select a price p̂r ∈ {vL , vH } for the
regular items, i.e., either the seller sets a low price so that the entire population purchases the
regular item, or the seller excludes the low-value consumers. We note that even in cases where
vL < c, it may be optimal for the seller to price the regular item below cost to take advantage of
the rarity effect. Next, the consumer’s valuation for the special item v̂s depends on the type of
consumer, i.e., whether their realized value for the regular item is vL or vH and the quantities sold
for both the regular and the special items under the given prices. Formally, this can be written as
30

v̂s = (1 + α(1 − r))v̂, where v̂ ∈ {vL , vH } denotes their value for the regular item, and the rarity r
is defined as
1 − P(v̂s < p̂s )
r= ,
1 − P(v̂ < p̂r )
where the numerator and denominator capture the quantity of the average regular and special
items sold (in equilibrium), respectively. Despite the endogeneity apparent in the definition of v̂s ,
we note that from a practical perspective, there are only three potential policies that can be availed
by the seller, namely: a) (LL) offering regular and special items to both types of consumers, b)
(LH) selling regular items to both consumer types and the special item to high-type consumers,
and c) (HH) targeting both regular and special items exclusively at high-type consumers.13
Blind Box. Given a blind box (p̂, q̂), the consumer’s valuation of the special item given type
v̂ ∈ {vL , vH } is:
 
q̂
v̂s = 1 + (1 − )α · v̂.
(1 − q̂)/n
Given this valuation, the rest of the dynamics under the blind box mechanism proceed similarly
to our descriptions in Section 2, with the key difference being that the low-type and high-type
consumers experience different stopping conditions. More precisely, after a certain number of pur-
chases, suppose that nr ∈ {0, 1, 2, . . . , n} and ns ∈ {0, 1} represent the number of unique regular and
special items, respectively, that a consumer of type v̂ ∈ {vL , vH } possesses. The consumer’s marginal
valuation for the blind box at this point is given by (1 − q̂) · n−n
n
r
v̂ + q̂(1 − ns )v̂s . Since vL < vH , the
low-type consumer’s valuation for the blind box is always weakly smaller than that of the high-type
consumer. Consequently, the low-type consumer reaches their stopping condition earlier—i.e., for
every sample path, the low-type consumer purchases fewer items than the high-type consumer.

Analysis of Optimal Policy. At a high level, consumer heterogeneity can impact separate
selling and blind box selling differently. First, under separate selling, the seller faces the important
trade-off of whether or not to subsidize the regular item (e.g., by selling it at a low price) in order
to enhance the rarity of the special item. Second, even though the seller can effectively target
high-value consumers, the no-rationing assumption prevents the seller from optimizing rarity any
further, e.g., rationing the item among just high-type consumers. The following lemma characterizes
the optimal policy under separate selling as a function of the level of heterogeneity in the system,
which is captured by the parameter ϵ > 0.

13
It is straightforward to show that the (HL) case where the seller targets the regular item to high-type consumers
and the special item to both types is sub-optimal.
 31

Lemma 6 (Optimal Separate Selling Policy Under Heterogeneity). Given any ϵ >
0, the optimal policy for separate selling can be characterized as follows:
(i) (LL) If 0 < ϵ ≤ ϵS , the optimal policy is to offer the regular and special items to both types of
consumers with p̂r = p̂s = vL , where ϵS = ((2 − α)v − 2c)/(α + 6).
(ii) (LH) If ϵS < ϵ < ϵ̄S , the optimal policy involves selling regular items to both consumer types
and the special item to high-type consumers p̂r = vL and p̂s = (1 + α/2)vH , where ϵ̄S = (2n(v −
c) + vα)/(6n − α).
(iii) (HH) If ϵ̄S ≤ ϵ < v, the optimal policy involves targeting both the regular and special items
exclusively at high-type consumers p̂r = p̂s = vH .

Lemma 6 demonstrates that the optimal policy for separate selling depends on the level of
heterogeneity, denoted by ϵ. When the valuation difference between the two consumer types is
minimal, it is optimal to offer both regular and special items to all consumers. When heterogeneity
is in an intermediate region, the seller benefits from selling regular items to both consumer types
while targeting the special item exclusively to high-type consumers. In this case, the seller can set
a high price for the special item by leveraging its rarity relative to the regular items, which are sold
to all consumers. However, when a significant valuation gap exists between the two types, selling
both item types solely to high-type consumers becomes optimal.
Now that we have a deeper understanding of how the seller’s optimal strategy varies with the
level of heterogeneity, it is natural to question whether increasing heterogeneity benefits or hurts
the seller under either mechanism. Figure 8 presents the seller’s revenue under both separate selling
and blind boxes as the level of heterogeneity (ϵ) varies. Under the given choice of parameters, the
profit under separate selling strictly decreases with ϵ, whereas the effect on blind boxes is mixed.
Lemma 7 formalizes these observations via conditions under which the seller’s profit increases or
decreases with ϵ.

Lemma 7 (Comparative Statics on ϵ).

(i) (Separate Selling) There exist thresholds αS and ϵ̄S such that the seller’s optimal profit under
separate selling is:
• decreasing in ϵ when α ≤ αS and increasing in ϵ when α > αS for ϵ ∈ [0, ϵ̄S ),
• always increasing in ϵ for ϵ ∈ [ϵ̄S , v].
(ii) (Blind Box) There exists ϵ̄B ≥ 0 such that the seller’s optimal profit under blind box selling is
increasing in ϵ for ϵ ∈ [ϵ̄B , v].
32

240

Blind Box
220 Separate Selling

200

180

160

140
0 2 4 6 8 10

Figure 8 The optimal profit under blind box selling and separate selling at different heterogeneity levels

(n = 12, v = 10, c = 3, α = 40).

Notes. The optimal profit under separate selling decreases with ϵ because the valuation of the low-type consumer

decreases as ϵ increases, forcing the seller to set the regular item at a low price to maintain the rarity of the special

item. On the other hand, the optimal profit under blind box selling initially decreases with ϵ as the reduction in

the purchase quantity by the low-type consumer outweighs any corresponding increase by the high-type consumer.

However, beyond a certain level of heterogeneity, the seller increases prices to extract more surplus from the high-type,

and therefore, the contribution of the low-type consumer becomes insignificant.

Figure 8 depicts the thresholds ϵ̄S and ϵ̄B for the specific choice of parameters α = 40 and
c/v = 0.3. To understand the rationale behind Lemma 7 for separate selling, we note that under the
LH policy, the regular item acts as a loss leader, which serves the purpose of improving the rarity
of the special item. Consequently, further heterogeneity undermines the seller’s revenue from the
regular items (as vL = v − ϵ decreases) and hurts the overall profit. This is particularly stark when
consumers are not status-seeking (i.e., α ≤ αS ) as the increased revenue from selling the special
item to high-type consumers does not balance the loss in revenue from underpricing the regular
items. When α is large (i.e., α > αS ), this effect is reversed, and the seller’s optimal benefit from
targeting the high-type consumer with the special item outweighs any loss in revenue from selling
the regular items to low-type customers.
For blind box selling, the platform’s profit increases in ϵ as long as the two consumer types are
sufficiently heterogeneous, i.e., ϵ ≥ ϵ̄B . When ϵ exceeds this threshold, the low-type consumers do
not contribute much to the overall revenue as their valuations become smaller and smaller. In this
regime, the seller designs the blind box primarily aimed at the high-type consumer, and in turn,
benefits as ϵ and vH = v + ϵ increases.
Given that the profits under both separate selling and blind boxes increase with ϵ when the level
 33

of heterogeneity is sufficiently large (and in the case of separate selling when α is also large), an
important consideration is whether heterogeneity benefits separate selling more than blind boxes.
In Lemma C7 in Online Appendix C, we address this issue and show that the difference between
the seller’s optimal profit under blind boxes and that under separate selling is strictly increasing
in ϵ when the level of heterogeneity is large enough. This suggests that blind boxes are able to
better leverage heterogeneity by collecting revenue from both consumer types—even though the
contribution of the low-type is not as significant. In other words, the seller is forced to target one
or both types under separate sales, whereas it is possible to mimic personalized pricing through
self-selected purchase quantities (with high- or low-valuation consumers purchasing more or less
on average) and extract most of the surplus from both types via a blind box as both consumer
types continue to purchase until their respective stopping conditions are reached.
As a natural follow-up, Proposition 5 presents conditions under which it is optimal for the seller
to switch from separate selling to blind box selling for a given level of heterogeneity ϵ.

Proposition 5. (When Is Blind Box Selling Optimal?) Suppose that β = c/v. For any
given ϵ ∈ [0, v], there exist thresholds α̂ and β̂ such that the seller’s optimal profit under blind box
selling is strictly larger than that under separate selling for α ≥ α̂ and β ≤ β̂.

Proposition 5 states that under valuation heterogeneity, blind box selling remains a superior
option when consumers are sufficiently status-seeking, and the manufacturing cost is low enough,
which is aligned with our original intuition from Proposition 3. In particular, this insight is valid
regardless of the level of heterogeneity. While Lemma 7 provides comparative statics on how varying
ϵ impacts blind box selling relative to separate sales, Proposition 5 complements this analysis by
comparing the optimal profits under both mechanisms for a fixed value of ϵ.

5.3. Lump-Sum Bonus

In this section, we consider a straightforward extension of our base model where the consumer
receives bonus utility of v + —i.e., a lump-sum bonus—upon collecting all items, both regular and
special. The notion of a lump-sum bonus is motivated by the so-called incompleteness effect, where
consumers are driven to collect a complete set, and has been studied in other works on blind
boxes (Feng et al. 2022). In particular, we assume that the same bonus is applicable to both separate
selling and blind box selling and investigate how this addition impacts both these mechanisms. In
the case that the consumer imposes a higher bonus on collecting a full set under the blind box
selling than under separate sales due to that it would be harder to achieve in the former than the
34

latter, the comparison results obtained in this extension still qualitatively hold with the scale tilted
more towards the blind box.
Separate Selling. Suppose that p̌r and p̌s denote the price of the regular and special items,
respectively. As with the base model, we assume that the seller cannot ration the special item.
Therefore, it is always profitable to sell the special item to all consumers, who in turn receive the
lump-sum bonus. Further, the addition of the bonus strengthens the seller’s pricing power, and
the optimal price for both items is given by p̌∗r = p̌∗s = v + v + /(n + 1)—this is strictly higher than
a price of v under the base model. As a result, the seller’s optimal profit under separate selling
increases when compared to the base model. Formally, the optimal profit under separate selling,
denoted by Π̌∗S , can be written as

Π̌∗S = n(p̌∗r − c) + (p̌∗s − c) = (n + 1)(v − c) + v + .

Blind Box. We assume that consumers are forward-looking and incorporate the lump-sum bonus
into their expected utility and stopping conditions. We define the consumer’s purchase decision
and stopping condition in terms of their state (nr , ns ), which represents the number of unique
regular items and the special item that they possess from prior purchases. Specifically, for any
given blind box (p̌b , q̌), suppose that U (nr , ns ) denotes the consumer’s total expected utility from
future purchases given the current state (nr , ns ). Then, we have:

(n − nr )
· v + I{nr =n−1,ns =1} v + + U (nr + 1, ns )


U (nr , ns ) = max 0, (1 − q̌)
n
 (10)
+


+ q̌ 1 − ns )(vs + I{nr =n} v + U (nr , ns + 1) + ω(nr , ns )U (nr , ns ) − p̌b ,

where ω(nr , ns ) = (1 − q̌)nr /n + q̌ns denotes the probability that the consumer did not receive a
new item upon purchase upon which case the expected future utility reverts to U (nr , ns ). Note
that the bonus of v + is only received once the consumer collects all n regular items along with
the special item. The boundary case is defined as U (nr , ns ) = 0 when nr = n and ns = 1. As an
example, when nr = n − 1 and ns = 1 (i.e., the consumer needs one additional regular item to
1−q̌ 1−q̌
complete the collection), the utility becomes: U (nr , ns ) = n
(v + v + ) + (1 − n
)U (nr , ns ) − p̌b ,
n
which can be simplified as U (nr , ns ) = v + v + − 1−q̌ p̌b . The higher value of the expected payment,
n
p̌ ,
1−q̌ b
represents the fact that the user may need to make more purchases before they receive the
final unique item.
Based on the above expression, a consumer will continue purchasing the blind box at price p̌b
until they reach a state (nr , ns ) where U (nr , ns ) < 0. However, it is possible to show that there
 35

exists a cutoff price p̄ such that if p̌b ≤ p̄, the consumer will always purchase the blind box until
they collect all items. Finally, let us denote the expected number of purchases a consumer makes in
this setting by m̌(p̌b , q̌). The seller’s optimal profit under the blind box mechanism can be written
as:

max Π̌B = max m̌(p̌b , q̌) · (p̌b − c). (11)

p̌b ,q̌ p̌b ,q̌

Given this setup, the following proposition delineates clear conditions under which it is optimal
for the seller to utilize blind boxes or separate selling respectively. For clarity, we use Π̌∗S and Π̌∗B
to denote the seller’s optimal profit under separate selling and blind box selling, respectively, in
the model where the consumer receives a lump-sum bonus.

Proposition 6 (Profit Comparison Under Lump-Sum Bonus). Suppose that β = c/v.

(i) For any given β, there exists a threshold α̌ such that the optimal profit under blind box selling
dominates separate selling if and only if α > α̌.
(ii) For any given α, there exists a threshold β̌ such that if β < β̌, then Π̌∗B > Π̌∗S .
Furthermore, α̌ ≥ ᾱ and β̌ ≤ β L , where ᾱ and β L denote the constants from Proposition 3.

Consistent with Proposition 3, the above result indicates that blind box selling is optimal for the
seller when consumers are sufficiently status-seeking or, alternatively, when costs are low. However,
surprisingly, under the extended model with lump-sum bonuses, the threshold α̌ that denotes the
minimum status level at which blind box selling outperforms is larger than the corresponding status
threshold from Proposition 3 implying that separate selling is optimal under a larger range of
status-seeking parameters. Analogously, the same is true for the cost ratio β = c/v, which implies
that there exists a cost regime (low-to-intermediate) under which blind box selling is optimal for the
base model but is dominated by separate selling with the addition of a lump-sum bonus. Overall,
this suggests that compared to the base model, the addition of a lump-sum bonus favors separate
selling.
To understand the intuition behind this result, recall that under separate selling, the addition
of the lump-sum bonus allows the seller to set a higher price and extract the entire bonus v + as
surplus. However, in the blind box model, due to the random consumption process, there is no
way to guarantee customers will collect all items, which requires the price to be low enough while
extracting a high enough surplus. In other words, the blind box policy is ill-suited to the notion of
a lump-sum bonus due to the sequential and random nature of consumption.
36

6. Conclusion
In recent years, blind boxes have emerged as a popular model for selling toys14 and collectibles by
tapping into consumers’ penchant for surprise. This unprecedented success has spurred sellers to
launch similar randomized mechanisms in other industries, notably books15 , fashion16 , and food
and beverage17 . The present work attempts to systematically analyze the trade-offs faced by sellers
in choosing whether or not to employ blind box selling and, simultaneously, uncover the underlying
principles that contribute to the profitability of blind boxes.
More concretely, our work provides insights into the circumstances that make blind boxes more
advantageous than selling items separately. A necessary condition for blind boxes to outperform
separate sales is that the underlying assortment is both finite and heterogeneous, i.e., regular items
and special items whose value is proportional to their rarity. Under these conditions, our findings
suggest that blind box selling is the optimal mechanism when consumers value rarity (i.e., are
status-seeking) or manufacturing costs are either low or high. The latter non-monotonicity in terms
of cost is of particular interest and points to two distinct mechanisms that blind boxes can take
advantage of, namely: i) volume effect, where a low price can incentivize consumers to purchase
the blind box multiple times until they receive the special item, and ii) price effect where the seller
can leverage the rarity of the special item to enhance the price of the blind box. Given that blind
boxes are associated with selling physical goods, the characterization based on cost is particularly
crucial to determine which types of markets sellers can effectively implement this strategy in.
In practice, the performance of blind boxes may be influenced by a variety of secondary factors.
In Section 5, we extend our model to accommodate three such features, namely seller rationing,
consumer heterogeneity, and lump-sum bonuses. Broadly speaking, our core findings continue to
hold under these three models, albeit with some caveats. For example, when the seller is allowed
to ration (i.e., induce artificial scarcity) the special item under separate sales, blind boxes may no
longer be optimal in high-cost regimes if consumers are not status-seeking. On the other hand, blind
boxes are able to exploit heterogeneity more effectively than separate sales, leading to improved
profits. Finally, it is critical to note that there may be factors beyond the scope of this work that also
drive consumers towards blind boxes such as secondary markets Feng et al. (2022), signalling (Stock
and Balachander 2005), or bundling (Figure 1). It would be interesting to understand better how
these factors influence the design of blind boxes.
14
See https://www.nytimes.com/2020/04/16/parenting/lol-surprise-doll-isaac-larian.html
15
See https://bookgrocer.com/collections/mystery-book-boxes
16
See https://www.nytimes.com/wirecutter/reviews/stitch-fix/
17
See https://www.nytimes.com/2022/09/20/climate/food-waste-app.html
 37

In summary, our results complement the growing body of works in operations management that
studies how sellers can leverage uncertainty and probabilistic mechanisms to improve profitability.
At the same time, blind boxes reveal a new pathway through which randomness can benefit sellers,
namely by incentivizing buyers to engage in multiple purchases. In this regard, blind boxes can
provide sellers with secondary benefits, such as a higher volume of sales and the absence of rationing,
which can improve brand loyalty in the long run. Finally, our study also illustrates that blind boxes
are not a panacea, and the seller has to deploy them carefully, particularly in settings where the
manufacturing costs are moderate, where rationing is feasible, or where buyers value a lot more
for collecting all items. In such scenarios, it may be preferable to rely on traditional mechanisms
such as separate selling.

References
Amaldoss W, Jain S (2005) Conspicuous consumption and sophisticated thinking. Management Sci.

51(10):1449–1466.

Anderson ET, Fitzsimons GJ, Simester D (2006) Measuring and mitigating the costs of stockouts. Manage-

ment Sci. 52(11):1751–1763.

Bernstein F, Martı́nez-de Albéniz V (2017) Dynamic product rotation in the presence of strategic customers.

Management Sci. 63(7):2092–2107.

Biraglia A, Usrey B, Ulqinaku A (2021) The downside of scarcity: Scarcity appeals can trigger consumer

anger and brand switching intentions. Psychol. Market 38(8):1314–1322.

Chen C, Fang Z (2023) Gacha game analysis and design. Proc. ACM Meas. Anal. Comput. Syst. 7(1):1–45.

Chen N, Elmachtoub AN, Hamilton ML, Lei X (2021) Loot box pricing and design. Management Sci.

67(8):4809–4825.

Chen RR, Gal-Or E, Roma P (2014) Opaque distribution channels for competing service providers: Posted

price vs. name-your-own-price mechanisms. Oper. Res. 62(4):733–750.

Elmachtoub AN, Hamilton ML (2021) The power of opaque products in pricing. Management Sci. 67(8):4686–

4702.

Elmachtoub AN, Yao D, Zhou Y (2019) The value of flexibility from opaque selling. Preprint, submitted

November 19, https://ssrn.com/abstract=3483872.

38

Fay S, Xie J (2008) Probabilistic goods: A creative way of selling products and services. Marketing Sci.

27(4):674–690.

Fay S, Xie J (2010) The economics of buyer uncertainty: Advance selling vs. probabilistic selling. Marketing

Sci. 29(6):1040–1057.

Fay S, Xie J (2015) Timing of product allocation: Using probabilistic selling to enhance inventory manage-

ment. Management Sci. 61(2):474–484.

Feng Y, Sun C, Yang C (2022) The impact of secondary markets on selling blind boxes with set bonuses.

Preprint, submitted December 13, https://ssrn.com/abstract=4284733.

Festinger L (1954) A theory of social comparison processes. Human Relations 7(2):117–140.

Feylessoufi A, Kavadias S, Ralph D (2023) Behavioral microfoundations of new practice adoption: The effects

of rewards, training and population dynamics. Management Sci. Forthcoming.

Han J, Ryan CT, Tong XT (2023) Algorithms for loot box design. Preprint, submitted January 19, https:

//ssrn.com/abstract=4326724.

Huang T, Yin Z (2021) Dynamic probabilistic selling when customers have boundedly rational expectations.

Manufacturing & Service Oper. Management 23(6):1597–1615.

Huang T, Yu Y (2014) Sell probabilistic goods? A behavioral explanation for opaque selling. Marketing Sci.

33(5):743–759.

Jerath K, Netessine S, Veeraraghavan SK (2010) Revenue management with strategic customers: Last-minute

selling and opaque selling. Management Sci. 56(3):430–448.

Koford K, Tschoegl AE (1998) The market value of rarity. J. Econ. Behav. Organ. 34(3):445–457.

Lei X, Wan B, Wang S (2023) Content rotation in the presence of satiation effects. Available at SSRN

4593945 .

Mai Y, Hu B (2023) Optimizing free-to-play multiplayer games with premium subscription. Management

Sci. 69(6):3437–3456.

Miao J, Jain S (2024) Designing loot boxes: Implications for profits and welfare. Marketing Sci. Forthcoming.

Momot R, Belavina E, Girotra K (2020) The use and value of social information in selective selling of

exclusive products. Management Sci. 66(6):2610–2627.

 39

Rao RS, Schaefer R (2013) Conspicuous consumption and dynamic pricing. Marketing Sci. 32(5):786–804.

Roels G, Su X (2014) Optimal design of social comparison effects: Setting reference groups and reference

points. Management Sci. 60(3):606–627.

Sheng L, Zhao X, Ryan CT (2024) Selling bonus actions in video games. Management Sci. Forthcoming.

Stock A, Balachander S (2005) The making of a “hot product”: A signaling explanation of marketers’ scarcity

strategy. Management Sci. 51(8):1181–1192.

Tereyağoğlu N, Veeraraghavan S (2012) Selling to conspicuous consumers: Pricing, production, and sourcing

decisions. Management Sci. 58(12):2168–2189.

Yang L, Jin C, Shao Z (2023) Help-and-haggle: Social commerce through randomized, all-or-nothing dis-

counts. Management Sci. Forthcoming.

Yang L, Yu M (2023) Too good to go: Combating food waste with surprise clearance. Preprint, submitted

October 13, https://ssrn.com/abstract=4573386.

 1

Online Appendix to
“Pricing and Rarity Design of Blind Boxes with Random Items”

A. Notation

Table A.1 Notation

Notation Definition
Base Model
v, vs Valuation of the regular item and special item
c Cost of manufacturing
r Rarity of the special item
xr , xs Rarity of the regular and special item under separate selling
α The degree of status-seeking
pr , ps Price of the regular item and special item
pb Price of the blind box
q Drawing probability of the special item
nr , ns The number of unique regular items and the special item a consumer
already collected
κ The relative value of a special item in the blind box compared to that
of a regular item
ΠB , ΠS Profit under blind box selling and separate selling
Endogenous Supply
ṽs Valuation of the special item
r̃ Rarity of the special item
p̃r , p̃s Price of the regular item and special item
Π̃B , Π̃S Profit under blind box selling and separate selling
Consumer Heterogeneity
vL , vH Valuation of the low-type and high-type consumer to the regular item
v The mean of the valuation
ϵ The degree of heterogeneity
v̂s Valuation of the special item
p̂r , p̂s Price of the regular item and special item
Π̂B , Π̂S Profit under blind box selling and separate selling
Lump-sum Bonus
v̌s Valuation of the special item
v+ Lump-sum bonus
ř Rarity of the special item
p̌r , p̌s Price of the regular item and special item
p̌b Price of the blind box
Π̌B , Π̌S Profit under blind box selling and separate selling

B. Data
Table B.1 presents data from the blind box trading app Qiandao, where individual items from
different blind boxes are traded. Each series below represents a distinct collection of items, such
as the Harry Potter series shown in Figure 1. The table provides key details regarding the price at
which each item is traded.
2

Table B.1 Item Details Across Various Series

Series Number pb (¥) n q psb (¥) prb (¥) prb /psb r = nq/(1 − q)
1 69 12 1/144 679 58.3 0.086 0.084
2 69 12 1/144 390 49.5 0.127 0.084
3 69 12 1/144 650 55.8 0.086 0.084
4 69 9 1/108 690 53.3 0.077 0.084
5 69 12 1/144 378 53.0 0.140 0.084
6 69 12 1/144 635 56.1 0.088 0.084
7 69 12 1/144 665 53.8 0.081 0.084
8 69 16 1/128 328 36.5 0.111 0.126
9 69 12 1/144 683 60.7 0.089 0.084
10 69 12 1/144 495 53.1 0.107 0.084
11 69 12 1/144 845 47.3 0.056 0.084
12 69 12 1/144 650 44.3 0.068 0.084
13 69 12 1/144 480 35.8 0.075 0.084
14 69 12 1/144 620 43.7 0.070 0.084
15 59 12 1/144 1127 52.8 0.047 0.084
16 74 8 1/144 950 93.6 0.099 0.056
17 69 12 1/144 295 43.3 0.147 0.084
18 69 12 1/72 253 43.3 0.171 0.169
19 69 12 1/144 256 28.8 0.113 0.084
20 69 12 1/144 854 56.7 0.066 0.084
21 69 12 1/144 329 48.6 0.148 0.084

Notes. Each row corresponds to a different series of items. pb , n and q represent the retail price of a single blind box

in the primary market, the number of regular items, and the probability of drawing a special item, respectively. psb

and prb denote the secondary market prices of the special and regular items, respectively. The mean value of psb /prb is

11.4. Data collected on April 1, 2023.

C. Auxiliary Results
In this Appendix, we present additional technical lemmas (not stated in the main body) that
will serve as key ingredients to prove our main results. We begin by characterizing the consumer’s
stop condition under blind boxes.

Lemma C1. For any given instance and blind box package (pb , q) satisfying qvs < (1 − q)v, the
consumer’s stop condition under different pricing policies can be summarized as follows:
(i) (Low Price) Suppose pb = (1 − q) k+1
n
v ≤ qvs . Then, consumers stop their purchases after
obtaining the special item and at least n − ⌊k ⌋ unique regular items.
(ii) (Moderate Price) Suppose qvs < pb = (1 − q) k+1
n
v ≤ (1 − q)v. Then, consumers stop pur-
chasing either upon obtaining the special item and at least n − ⌊k ⌋ unique regular items or
alternatively, upon obtaining ⌊n − k + κ⌋ unique regular items.
(iii) (High Price) Suppose pb = (1 − q) k+1
n
v > (1 − q)v. Then, consumers stop their purchases
 3

after obtaining either the special item or ⌊n − k + κ⌋ unique regular items.

Here κ = nqvs /(1 − q)v as defined in Section 3.1.

Lemma C2. For any given instance and blind box package (pb , q) satisfying qvs ≥ (1 − q)v, the
consumer’s stop condition under different pricing policies can be summarized as follows:
(i) (Low Price) Suppose pb = (1 − q) k+1
n
v ≤ (1 − q)v. Then, consumers stop their purchases after
obtaining the special item and at least n − ⌊k ⌋ unique regular items.
(ii) (Moderate Price) Suppose (1 − q)v < pb = (1 − q) k+1
n
v ≤ qvs . Then, consumers reach their
stop condition after acquiring the special item.
(iii) (High Price) Suppose pb = (1 − q) k+1
n
v > qvs . Then, consumers end their purchases after
obtaining either the special item or ⌊n − k + κ⌋ unique regular items.
Here κ = nqvs /(1 − q)v as defined in Section 3.1.

The proof of both Lemma C1 and Lemma C2 follow immediately from our characterizations in
Section 3.1.

1
Lemma C3. For any given probability q ≤ n+1
of drawing the special item under the blind box,
the seller’s optimal price for the blind box pb ∈ {vB (n − 1, 1) or vB (n̄ − 1, 0)} for some n̄ ≥ n, where
vB (nr , ns ) = (1 − q) n−n
n
r
v + q(1 − ns )vs denotes the marginal valuation from purchasing a blind box
from state (nr , ns ). The terms n and n̄ are as defined in Lemma 1.

Proof of Lemma C3. As discussed in the main paper, the consumer stops purchasing the blind
box if and only if vB (nr , ns ) < pb . By Lemma 1, the consumers’ stopping condition is met upon
acquiring either one special item and at least n regular items or n̄ regular items without the special
item. This implies that the price of the blind box must satisfy the following two conditions:

vB (n, 1) < pb ≤ vB (n − 1, 1) and vB (n̄, 0) < pb ≤ vB (n̄ − 1, 0).

Suppose that for any given q, the optimal price under the blind box model meets the conditions
above for some given n ≤ n̄. As a result, the optimal price must be either vB (n − 1, 1) or vB (n̄ − 1, 0).
To see why, consider any price pb = vB (n − 1, 1) − ϵ for some sufficiently small ϵ > 0 so that this
price meets both the conditions above. We note that under this new price, the consumer’s stopping
condition remains unchanged, and therefore, the expected number of purchases will not increase.
Consequently, the seller can strictly increase profits by increasing the price by ϵ (reverting to pb =
vB (n − 1, 1)). In other words, the price vB (n − 1, 1) dominates any price pb satisfying vB (n, 1) < pb <
vB (n − 1, 1). One can also apply a similar argument to show the sub-optimality of any price strictly
4

smaller than vB (n̄ − 1, 0) − ϵ. Note that any price larger than either vB (n − 1, 1) or vB (n̄ − 1, 0)
would violate the stop conditions defined above.

Lemma C4. Consider an instance with n regular items and no special item. Given a blind box
with price pb = (k + 1)v/n, where the consumer has an equal probability of obtaining each regular
item, the expected purchase quantity can be expressed as:
n
X
m(k) = (1/n + 1/(n − 1) · · · + 1/(k + 2) + 1/(k + 1)) = n 1/i.
i=k+1

Proof of Lemma C4. Define the quantity m(nr , pb ) as the expected number of blind boxes pur-
chased by consumers before reaching their stop condition, if they initially possess nr ∈ {0, 1, . . . , n}
unique regular items. Furthermore, given price pb = (k + 1)v/n, it is clear that the consumer would
stop purchasing blind box if and only if they have obtained n − k unique regular; this is because
k
the marginal value from any subsequent purchase is exactly n
v, which is strictly smaller than the
price.
For each purchase, there are two possible outcomes: (i) drawing a new regular item, in which
case the state transitions to nr + 1; (ii) drawing a repeated item, in which case the state remains
unchanged at nr . Hence, we can write the recursion for the expected purchase volume as follows:

n − nr nr
m(nr , pb ) = 1 + m(nr + 1, pb ) + m(nr , pb ),
n n

with the boundary condition m(nr , pb ) = 0 if nr ≥ n − k. In the above expression, the fractions (n −
nr )/n and nr /n represent the probability of drawing a new regular and repeated item respectively.
Solving the above recursion, we see that:

n
m(nr , pb ) = + m(nr + 1, pb ).
n − nr
k+1
Using the fact that pb = n
v and working backwards from nr = k, we get that:

n
n n n n X 1
m(0, pb ) = + +···+ + =n .
n n−1 k+2 k+1 i=k+1
i

Note that m(0, pb ) denotes the expected number of purchases a consumer will make from an initial
state with no items.

Lemma C5. There exists a threshold α′ such that for every instance satisfying α ≤ α′ , it must
qvs
be the case that κ = (1−q)v/n
≤ n − 1 for every choice of (pb , q).
 5

Proof of Lemma C5 We can derive the proof via simple algebra using the insight that κ is
strictly increasing in α. Specifically, we have that:

qvs
κ=n·
(1 − q)v
q q
=n · (1 + α(1 − )
1−q (1 − q)/n
n
≤ · (1 + α)
n−1
n (n − 1)2
≤
n−1 n
= n − 1.

q 1
In the first inequality, we used the observation that 1−q
achieves its largest value when q = n
(note
that it is not possible for q to take on a larger value according to our model). In the penultimate
(n−1)2
inequality, we set the threshold α′ = n
− 1.

Lemma C6. There exists a threshold α0 such that for any given blind box (pb , q) where pb >
max(v, qvs ), the following inequality holds as long as α < α0 :

mpb < (n + 1)v,

where m is the expected number of purchases of the blind box for a given consumer. Further α0 > 0
for all n ≥ 4.

Proof of Lemma C6. As stated in Lemma 3, suppose that pb = (1 − q) k+1

n
v. Since pb > v, it is
(n−1)2
clear that k > n − 1. Further as long as 1 + α0 ≤ n
, we can infer from Lemma C5 that κ ≤ n − 1,
and therefore, κ < k. Later on, we will verify that our choice of α0 meets this requirement. We
begin the proof by deriving an upper bound on m, which denotes the expected number of blind
boxes purchased by the consumer before stopping.
Under the given blind box parameters (pb , q), the consumer always stops purchasing when they
obtain the (n − (k − κ))-th unique regular item. Using the same rationale as Lemma C4, we can
bound the expected number of purchases m as follows:

 
n 1 1 n n
m< ( +·+ )≤ ln( ) .
(1 − q) n k−κ+1 (1 − q) k−κ
The lemma requires us to prove an upper bound on mpb . Using the above formula, we expand
on the required term below:

k+1 n
mpb = m · (1 − q) v < (k + 1)v ln( ).
n k−κ
6

n
Define w(k) = (k + 1) ln( k−κ ). The first-order derivative of w(k) with respect to k is given by

n k+1 n n
w′ (k) = ln( )− < ln( )− < 0.
k−κ k−κ k−κ k−κ

Thus, for any fixed κ, w(k) achieves its maximum value when k = n − 1. Therefore, we have

n
mpb < v · n ln( ).
n−1−κ

To complete the proof of the lemma, it is sufficient to show that when α is small, the right-hand
side of the above inequality is at most vn, which in turn is at most v(n + 1). In what follows, we
n
prove that ln( n−1−κ ) ≤ 1 when α is not too large.
q
First, observe that when α = 0, there is no special item and κ = (1−q)/n
= 1, and therefore,

n
ln( ) < 1,
n−2

as long as n ≥ 4. Moreover, since κ is strictly increasing in α, there must exist some α0 at which
n
ln( n−2 ) = 1. Clearly at this threshold α0 , κ < n − 1 holds, so we still meet the conditions in
Lemma C5.

Lemma C7. (Relative Benefits of Heterogeneity)

(i) If α > α̂(ϵ) and c/v < β̂(ϵ), Π̂∗B (ϵ) > Π̂∗S (ϵ), and Π̂∗B (ϵ) − Π̂∗S (ϵ) is strictly increasing in ϵ ∈
[ϵ̄B , v].
(ii) If α ≤ αS , Π∗B (ϵ) − Π∗S (ϵ) is strictly increasing in ϵ ∈ [ϵ̄B , ϵ̄S ] .

Proof of Lemma C7. (i) In the proof of Proposition 5, we have shown that if α > α̂(ϵ) and
c/v < β̂(ϵ), Π̂∗B (ϵ) > Π̂∗S (ϵ). Then, we demonstrate that Π̂∗B (ϵ) − Π̂∗S (ϵ) is increasing in ϵ ∈ [ϵ̄B , v]. For
ϵ ≥ ϵ̄B , the low-type consumer in the blind box selling scenario will not purchase any items, and
Π̂∗B (ϵ) increases with ϵ. For separate selling, the seller has three potential pricing policies that can
be implemented: LL, LH, and HH (see the definition of these policies in Lemma 6.). For policy LL,
the profit decreases with ϵ, and Π̂∗B (ϵ) − Π̂∗S (ϵ) increases with ϵ. Then, we will separately analyze
the cases in which the seller implements policies LH and HH.
Under policy LH, Π̂∗B (ϵ) − Π̂∗S (ϵ) can be represented as
! 
∗

∗ ∗ ∗ ∗ k̂ + 1 α 1
ΠB (ϵ) − ΠS (ϵ) = m̂ ((1 − q̂ ) (v + ϵ) − c) − (v − ϵ − c)n + ((v + ϵ)(1 + ) − c)
n 2 2
 
α 1
= m̂∗ ((1 − q̂ ∗ )(k̂ ∗ + 1)/n + n − (1 + ) ϵ
2 2
 
∗ ∗ ∗ α 1 1
+ m̂ ((1 − q̂ )(k̂ + 1)/n − n − (1 + ) v − (m̂∗ − n − )c.
2 2 2
 7

If Π̂∗B (ϵ) > Π̂∗S (ϵ), we have m̂∗ ((1 − q̂ ∗ )(k̂ ∗ + 1)/n + n − (1 + α/2)/2 > 0 for the following reason. If
Π̂∗B (ϵ) > Π̂∗S (ϵ), then

k̂ ∗ + 1  α 1
m̂∗ ((1 − q̂ ∗ ) (v + ϵ) − c) ≥ (v − ϵ)n − cn + (v + ϵ)(1 + ) − c
n 2 2
 α 1
≥ −cn + (v + ϵ)(1 + ) − c ,
2 2
where the second inequality holds since v − ϵ > 0. Upon further transformation, we have:

k̂ ∗ + 1 α 1 1
(m̂∗ (1 − q̂ ∗ ) − (1 + ) )(v + ϵ) ≥ (m̂∗ − n − )c. (C.1)
n 2 2 2

If (m̂∗ − n − 1/2) > 0, for (C.1) to hold, the left-hand side of (C.1) must be positive, indicating
that
k̂ ∗ + 1 α 1 α 1
m̂∗ (1 − q̂ ∗ ) > (1 + ) > (1 + ) − n.
n 2 2 2 2
If (m̂∗ − n − 1/2) ≤ 0, for (C.1) to hold, we must have

k̂ ∗ + 1 α 1 1
m̂∗ (1 − q̂ ∗ ) − (1 + ) ≥ m̂∗ − n − ,
n 2 2 2

because v + ϵ > c. Thus, we further obtain

k̂ ∗ + 1 α 1 1
m̂∗ (1 − q̂ ∗ ) ≥ (1 + ) − n + m̂∗ − > (1 + α/2)/2 − n,
n 2 2 2

where the second inequality holds due to the fact that the consumer will purchase at least once,
 
implying that m̂∗ > 1. Therefore, m̂∗ (1 − q̂ ∗ )(k̂ ∗ + 1)/n > 1+α/2
2
− n holds. This indicates that as
ϵ increases, the profit under the blind box increases at a faster rate if the seller maintains a fixed m̂.
Additionally, the seller can further optimize profit by adjusting the price and drawing probability.
Consequently, the ratio Π̂∗B (ϵ) − Π̂∗S (ϵ) strictly increases as ϵ ranges from ϵ̄B to v.
Under policy HH, Π̂∗B (ϵ) − Π̂∗S (ϵ) can be expressed as
!
∗
k̂ + 1
Π∗B (ϵ) − Π∗S (ϵ) = m̂∗ (1 − q̂ ∗ ) (v + ϵ) − c − (n + 1)(v + ϵ − c).
n

If Π̂∗B (ϵ) > Π̂∗S (ϵ), then

k̂ ∗ + 1
(m̂∗ (1 − q̂ ∗ ) − (n + 1))(v + ϵ) ≥ (m̂∗ − (n + 1))c.
n

In Proposition 5, we show that for ϵ̄S ≤ ϵ < v, if α > αL (c/(v + ϵ)) and c < (v + ϵ)β L , we have
Π̂∗B > Π̂∗S . Combining this with the results of Corollary 1, we obtain (m̂∗ − (n + 1)) > 0, indicating
∗
that (m̂∗ (1 − q̂ ∗ ) k̂ n+1 − (n + 1)) > 0. Consequently, the profit under the blind box increases at a
faster rate.
8

Therefore, we can infer that Π∗B (ϵ) − Π∗S (ϵ) is strictly increasing in ϵ ∈ [ϵ̄B , v].
(ii) By Lemma 7, Π∗B (ϵ) − Π∗S (ϵ) is strictly increasing in ϵ ∈ [ϵ̄B , ϵ̄S ] if α ≤ αS = 4n − 2, where
ϵ̄S = (2n(v − c) + vα)/(6n − α).

1
Lemma C8. Suppose that q̌ ≤ n+1
denotes the probability of drawing the special item for a given
blind box and let m denote the expected number of purchases of the blind box before the consumer
obtains all n + 1 unique items. Then, under blind box parameters (p̌b , q̌) where p̌b = (1 − q) min{1,κ}v
n
+
v+
m
, the consumer continues to purchase the blind box until they obtain all n + 1 items. Here, v +
denotes the lump-sum bonus for collecting all items.

Proof of Lemma C8. Recall that U (nr , ns ) denotes the consumer’s expected forward utility
starting from state (nr , ns )—i.e., when the consumer currently possesses nr unique regular times
and ns special items. To prove the claim, we first abuse notation and consider the blind box package
v+
(p̌b , q̌) where q̌ is as given in the lemma statement and p̌b = (1 − q) min{1,κ}v
n
+ m
− ϵ for some
sufficiently small ϵ > 0. Given this choice, we show that U (nr , ns ) > 0 for all nr , ns , where either
nr < n or ns < 1. This, in turn, implies that under any given state, it is optimal for the consumer
to purchase the blind box. Consequently, the consumer purchases the blind box till they receive all
n + 1 items and the lump-sum bonus. Since this holds for all ϵ → 0, we conclude that the consumer
will continue to follow the same purchasing behavior at ϵ = 0 (although they may be indifferent
between purchasing and not-purchasing at some states).
We begin by writing the recursion for U (nr , ns ), following which we prove that U (nr , ns ) > 0
under two cases—when ns = 1 and ns = 0.

(n − nr )
· v + I{nr =n−1,ns =1} v + + U (nr + 1, ns )


U (nr , ns ) = max 0, (1 − q̌)
n

+


+ q̌ 1 − ns )(vs + I{nr =n} v + U (nr , ns + 1) + ω(nr , ns )U (nr , ns ) − p̌b ,

where ω(nr , ns ) = (1 − q̌)nr /n + q̌ns denotes the probability that the consumer did not receive a
new item upon purchase, in which case the expected future utility reverts to U (nr , ns ). Before
proceeding with the two cases, we first define m(nr , ns ) to be the expected number of blind boxes
purchased before the consumer collects n + 1 unique item starting from state (nr , ns ). Note that
m(0, 0) = m, where m is the expected purchase number for given blind box (p̌b , q̌).
Case (1) ns = 1.
In this case, the consumer has already obtained the special item, and therefore, it suffices to
only look at (future) states with ns = 1. We claim and prove the following hypothesis by induction,
namely for all nr < n:
U (nr , 1) = (n − nr )v + v + − m(nr , 1)p̌b > 0 (C.2)
 9

n
In the base case where nr = n − 1, the expression U (n − 1, 1) = v + v + − 1−q̌ p̌b follows from our
+ +
argument in Section 5.3. Further, substituting p̌b = (1 − q̌) min{1,κ}v
n
+ vm − ϵ ≤ (1 − q̌) nv + vm − ϵ, it
m n
follows that U (n − 1, 1) ≥ 1−q̌
ϵ > 0. Here, we used the fact that m ≥ 1−q̌
. Working backward from
nr = n − 1, suppose that the claim holds up to nr + 1 for some arbitrary nr < n − 1. Then, the proof
of the inductive claim for the general case is as follows:
(n − nr )
U (nr , 1) = (1 − q̌) · (v + U (nr + 1, 1)) + ω(nr , ns )U (nr , ns ) − p̌b
n
p̌b
=⇒ U (nr , 1) = v + U (nr + 1, 1) −
1 − ω(nr , ns )
p̌b
= (n − nr )v + v + − m(nr + 1, 1)p̌b −
1 − ω(nr , ns )
+
= (n − nr )v + v − m(nr , 1)p̌b .

The coefficient for pb in the final line is a result of observing that m(nr , 1) − m(nr + 1, 1) = 1/(1 −
ω(nr , 1)). That is the expected number of purchases from state (nr , 1) to (nr + 1, 1) is equal to the
derivative of the probability of drawing the (nr + 1)-th unique item. To complete the proof of the
inductive hypothesis, it is sufficient to show that the expression in the final line is strictly positive.
To that end, we have

U (nr , 1) = (n − nr )v + v + − m(nr , 1)p̌b

v v+
 
≥ (n − nr )v + v + − m(nr , 1) (1 − q̌) + −ϵ
n m
 
1 1
≥ v (n − nr ) − (1 + + . . . + ) + v + (m − m(nr , 1)) + m(nr , 1)ϵ
2 n − nr
≥ m(nr , 1)ϵ > 0.

In the penultimate expression, we used the result from (the proof of) Lemma C4 that m(nr , 1) =
n
1−q̌
(1 + 12 1
+ . . . + n−n r
). This completes the proof that U (nr , ns ) > 0 when ns = 1.
Case (2) ns = 0.
In this case, we claim and prove by induction that for all nr ∈ {0} ∪ [n]:

U (nr , 0) = (n − nr )v + vs + v + − m(nr , 0)p̌b > 0, (C.3)

Base Case: If nr = n, we have

1
U (n, 0) = vs + v + − p̌b ,
q̌
which satisfies the inductive claim. Note that the above expression is positive since 1q̌ p̌b ≤ 1q̌ (q̌vs +
v+
m
− ϵ) and m ≥ 1q̌ . Working backwards from nr = n, suppose that the inductive claim in (C.3)
holds up to nr + 1 for some arbitrary nr < n. Then, by (10), we have
 
1 (n − nr )
U (nr , 0) = (1 − q̌) (v + U (nr + 1, 0)) + q̌(vs + U (nr , 1)) − p̌b ,
1 − ω(nr , 0) n
10

where 1 − ω(nr , 0) = (1 − q̌) (n−n

n
r)
+ q̌ denotes the probability that the consumer obtains a new
item. Expanding on the expressions for U (nr + 1, 0) and U (nr , 1) from (C.3) and (C.2) respectively,
we have

1 (n − nr )
U (nr , 0) = (1 − q̌) ((n − nr )v + vs + v + − m(nr + 1, 0)p̌b )
1 − ω(nr , 0) n
!
+ q̌(vs + (n − nr )v + v + − m(nr , 1)pb ) − p̌b

= (n − nr )v + vs + v + − m(nr , 0)p̌b ,

where the coefficient for pb holds since

1 (1 − q̌) (n−n
n
r)
q̌
m(nr , 0) = + m(nr + 1, 0) + m(nr , 1).
1 − ω(nr , 0) 1 − ω(nr , 0) 1 − ω(nr , 0)
| {z } | {z } | {z }
I II III

Here, term I represents the expected number of purchases required to obtain one more unique
item from state (nr , 0), and terms II and III denote the probability of transitioning from (nr , 0) to
(nr + 1, 0) and (nr , 1), respectively. Thus, we have U (nr , 0) = (n − nr )v + vs + v + − m(nr , 0)pb .
To complete the proof of the inductive claim in (C.3), we need to show that U (nr , 0) > 0 for all
nr < n, i.e.,

min{1, κ}v v +
 
+
(n − nr )v + vs + v > m(nr , 0) (1 − q̌) + −ϵ .
n m

Note that the right-hand side is at most m(nr , 0) · (1 − q̌) min{1,κ}v

n
+ v + − m(nr , 0)ϵ, since m(nr , 0) ≤
m = m(0, 0) by definition. Therefore, it suffices for us to prove that m(nr , 0) · (1 − q̌) min{1,κ}v
n
≤
(n − nr )v + vs .
Using the same rationale as Lemma C4 and the fact that the expected number of purchases
before the consumer receives the special item is 1q̌ , we have
 
n 1 1 1
m(nr , 0) ≤ 1+ +...+ + .
1 − q̌ 2 n − nr q̌

Therefore, we have
 
1 1
m(nr , 0)p̌b ≤ v 1 + + . . . + + vs ≤ (n − nr )v + vs ,
2 n − nr

which completes the proof that U (nr , 0) > 0.

+
In conclusion, we have U (nr , ns ) > 0 for any nr and ns if p̌b = (1 − q̌) min{1,κ}v
n
+ vm − ϵ, indicating
+
that the consumer will collect all n + 1 unique items even when p̌b = (1 − q̌) min{1,κ}v
n
+ vm .
 11

D. Proofs of Results from the Main Body

D.1. Proofs from Section 2
Proof of Lemma 1. To delineate the stopping conditions, first recall that for any given blind box
(pb , q), the consumer will continue to purchase the blind box as long as
n − nr
(1 − q) v + q(1 − ns )vs ≥ pb ,
n
where nr and ns denote the number of unique regular and special items currently in the consumer’s
possession. This can also be interpreted as the consumer’s current state based on past purchases.
Define vB (nr , ns ) = (1 − q)(n − nr )v/n + q(1 − ns )vs to be the consumer’s marginal value from
purchasing the blind box at state (nr , ns ). Clearly, vB (nr , ns ) is decreasing in both nr and ns . Next,
we formally define the two cutoffs n̄ and n as the smallest values in [n] satisfying:

vB (n̄, 0) < pb ≤ vB (n̄ − 1, 0) and vB (n, 1) < pb ≤ vB (n − 1, 1),

respectively. Intuitively, n̄ is the maximum number of unique regular items the consumer can obtain
without the special item before the stop condition is reached. Analogously, n denotes the maximum
number of unique regular items that can be collected along with the special item. Additionally, we
also account for the boundary cases. If pb ≤ vB (n, 0), the consumer would never stop purchasing
even after receiving all n unique regular, in which case, we set n̄ = n + 1 as convention. We also
set n̄ = 0 if pb > vB (0, 0) as no purchases would occur when the price is too high. Analogously,
if pb ≤ vB (n, 1), we set n = n since the consumer would stop purchasing after collecting all n + 1
items. We set n̄ = 0 if pb > vB (0, 1).
First, it is clear that n̄ ≥ n since vB (nr , 0) > vB (nr , 1) for every nr . We now go over each of the
stopping conditions in the lemma statement:
(i) ns = 0, nr = n̄: Suppose the consumer does not acquire the special item before receiving at
least n̄ unique regular items. Prior to obtaining the n̄-th unique regular item, the consumer’s
marginal valuation for the blind box is vB (n̄ − 1, 0). By definition, this term is at least pb .
Moreover, since vB is decreasing in nr , this implies that for all nr ≤ n̄ − 1, the consumer would
not stop purchasing if they have nr unique regular items and no special item, i.e., pb ≤ vB (nr , 0)
for all nr ≤ n̄ − 1. Consequently, the consumer would stop purchasing upon receiving the n̄-th
regular item since pb > vB (n̄, 0).
(ii) ns = 1, nr = n̄: In a similar vein, suppose that the consumer has already obtained the special
item before purchasing the n-th unique regular item. Once again, by definition of n and since
vB (nr , 1) is decreasing in nr , it is clear that the consumer would not stop when nr ≤ n − 1.
Subsequently, since pb > vB (n, 1), this state—i.e., (n, 1)—serves as a stopping point.
12

(iii) ns = 1, nr ∈ [n, n̄ − 1]: Suppose that the consumer obtains the special item between receiv-
ing the n-th and n̄-th unique regular item. In this case, the consumer would not stop after
receiving the n-th unique regular item (or prior to this) as pb ≤ vB (n, 0) by definition. Further,
define n+ ∈ [n, n̄ − 1] to be the number of regular items possessed by the consumer when the
special item is drawn for the first time. In this case, we know that vB (n+ , 1) < pb ≤ vB (n+ , 0).
Therefore, the consumer stops purchasing after drawing the special item for the first time.

Proof of Lemma 2. In this proof, we demonstrate that at equilibrium—once all consumers finish
their purchases—the fraction of special items collected by consumers equals the probability of
drawing the special item from the blind box q. In other words, the consumers’ beliefs are consistent
with the equilibrium state, in expectation.
First, we present an alternative interpretation of consumers purchasing blind boxes as a stochastic
process and introduce pertinent notation. For the sake of this proof, we assume that the purchase
of blind boxes takes place over rounds, where in each round τ , a single consumer purchases18 the
blind box if its marginal value not lower than the price. With some probability, this consumer
reaches the stopping condition (outlined in Lemma 1) and ceases purchasing, i.e., drops out. With
the remaining probability, the consumer moves on to round τ + 1, and the process continues until
they meet their stopping condition. Formally, at the beginning of round τ = 1, the consumers
purchase the blind box as long as pb ≤ qvs + (1 − q)v. Suppose that xτ denotes the probability that
the consumer does not reach their stopping condition before round τ with x1 = 1. Further, let msτ
and mτ be random variables denoting the number of special items and the total number of items
collected by the consumer at the beginning of round τ > 1. Since the consumer receives one item
in every round before dropping out, one can derive the expected value of mτ as follows:
τ −1
X
E[mτ ] = xt ,
t=1

where xt − xt+1 is the probability that the consumer reaches their stopping condition at the end of
round t.
To prove the required lemma, it is sufficient for us to show E[msτ ] = qE[mτ ] in the limit τ → ∞
as the consumer reaches their stop condition almost certainly. In fact, we show a stronger result,
namely that for every τ > 1, E[msτ ] = qE[mτ ]. We proceed by induction. For the base case, consider
τ = 2. Since the consumer always purchases the blind box in the first round, we have m2 = 1.
Moreover, E[ms2 ] = q by definition of the blind box, which completes the base case.
18
Since there is a unit mass of homogeneous consumers in our original model, we will equivalently consider all possible
sample paths of a single consumer in this proof.
 13

Next, suppose that the inductive claim holds up to round τ and consider round τ + 1, where
we need to prove that E[msτ +1 ] = qE[mτ +1 ]. First, note that E[mτ +1 ] = E[mτ ] + xτ , where xτ is
the probability that the consumer receives an item in round τ , i.e., has not reached their stopping
condition. In a similar vein, we have that E[msτ +1 ] = E[msτ ] + qxτ . Since xτ is the probability that
the consumer purchase the blind box, qxτ is the probability of receiving a special item. To prove
the inductive hypothesis, we have

E[msτ +1 ] = E[msτ ] + qxτ

= qE[mτ ] + qxτ

= q(E[mτ ] + xτ )

= qE[mτ +1 ].

The second equation above stems from the inductive hypothesis. This completes our proof.

D.2. Proofs from Section 3

Proof of Lemma 3. Recall the formulation of vB (nr , ns ) from the proof of Lemma 1 as the
consumer’s marginal value from purchasing the blind box given state (nr , ns ). This can be rewritten
as
(n − nr ) κ(1 − ns ) n − nr + κ(1 − ns )
vB (nr , ns ) = (1 − q) v + (1 − q) v = (1 − q) v,
n n n
nqvs
where κ = (1−q)v
as defined in Section 3. Note that since ns ∈ {0, 1}, it must be the case that
n − nr + κ(1 − ns ) ∈ {0, 1, · · · , n} ∪ {κ, κ + 1, · · · , κ + n}. Lemma C3 shows that the optimal price
satisfies one of the following forms: p∗b = vB (n − 1, 1) or p∗b = vB (n̄ − 1, 0). Combining this insight
with our expression for vB (nr , ns ), it follows that the optimal price satisfies

p∗b = (1 − q)(k + 1)/n,

where k ∈ {0, 1, · · · , n − 1} ∪ {κ − 1, κ, · · · , κ + n − 1}. Note that we exclude the case where the
optimal price is zero, and so k ̸= −1.

Proof of Proposition 1. We begin by deriving closed-form expressions for the seller’s revenue
under separate sales (Π∗S ) and the revenue under blind box selling (Π∗B ). From Section 3, we know
that in the absence of the special item:

Π∗S = n(v − c).

14

Further, from Lemma C4, we know that when the price of the blind box is given by pb (k) =
(k + 1)v/n, the expected number of purchases is given by m(k) = n(1/n + 1/(n − 1) · · · + 1/(k +
Pn
2) + 1/(k + 1)) = n i=k+1 1/i. Consequently, the seller’s profit under blind box selling is:
n
X 1 k+1
Π∗B = max (pb (k) − c) · m(k) = max n ( v − c).
k∈{0,1,...,n−1} k∈{0,1,...,n−1}
i=k+1
i n

We prove that for every feasible k ∈ {0, 1, . . . , n − 1}, the above expression is smaller than Π∗S ,
i.e., the profit under blind boxes is dominated by the seller’s profit under separate selling. For
simplicity, define ΠB (k) as the seller’s profit when the price is pb (k). Our goal is to show that for
all k ∈ {0} ∪ [n − 1], Π∗S > ΠB (k) or more formally, using the above closed-form expressions (and
canceling the common term), we need to prove:
n
X 1 k+1
v−c> ( v − c).
i=k+1
i n

One can rearrange this to show the equivalent inequality:

n
v−c X 1
k+1
> .
n
v − c i=k+1 i

We now prove this using the sequence of steps below:

n n
X 1 X 1
≤ (D.1)
i=k+1
i i=k+1 k + 1
n
≤ (D.2)
k+1
v
= k+1
n
v
v−c
≤ k+1 .
n
v−c
′
The final inequality comes from the algebraic identity that vv−c v
′ −c ≥ v ′ for v ≤ v and c ∈ [0, v).
Pn
Additionally, the inequality i=k+1 1i < k+1
v−c
v−c
is strict. This is because for every choice of k ∈
n

{0, 1, . . . , n − 1} at least one of (D.1) or (D.2) is strict. This completes our proof.

Proof of Proposition 2. Recall that q < 1/(n+1). As n goes to infinity, q approaches 0. Therefore,
the consumer’s expected value for receiving the special item alone can be bounded as given below:
 
q
lim qvs = lim q · 1 + (1 − )α v = 0.
n→∞ n→∞ (1 − q)/n

Informally, this implies that the special item does not contribute much to the blind box’s revenue
and therefore, this case boils down to the setting without the special item. In Proposition 1, we
 15

showed that in the absence of the special item, separate selling always dominates. We provide a
similar argument below by proving that for any fixed price pb > 0, the profit from blind box selling
is strictly smaller than that from separate selling.
First, consider any fixed pb ∈ (0, v]. We will show the existence of an integral threshold nt such
that as long as the number of unique regular items is greater than n̄ (i.e., n > nt ), blind box selling
1
cannot be profitable than separate sales for all q < n+1
. Following this, we will extend this argument
to the case where pb > v as well.
1
Given pb ∈ (0, v] and any q < n+1
, let k be the smallest number in {0, 1, . . . , n − 1} such that
pb ≤ (1 − q)v(k + 1)/n. We will assume that n is sufficiently large and so pb ≤ (1 − q)v—i.e., the
existence of a suitable k is guaranteed. We know from Lemma 1 that the consumer purchases the
blind box until they receive at most n̄ unique regular items (i.e., this is an upper bound) where n̄
is the smallest integer in the range [n] that satisfies:

n − n̄
pb > (1 − q)v + qvs .
n

Recall that as n → ∞, qvs → 0. Therefore, there exists some sufficiently large n (say nt ) above
which n − n̄ = k. Put differently, as qvs → 0, a sufficient condition for the consumer to stop pur-
chasing is when they receive n̄ = n − k unique regular items. This is because at this point, we have
pb > (1 − q)v n−n̄
n
= (1 − q)vk/n, by definition of k. Consequently, using the same argument as in
Lemma C4, the expected number of blind box purchases by a given consumer before they obtain
n − k unique regular items is given by:
n
n X
m(k) = 1/i.
1 − q i=k+1

Therefore, the seller’s expected profit under the blind box for pb for any n > nt can be written
as:

ΠB (pb , q) = (pb − c)m(k)

n
n X
≤ ((1 − q)v(k + 1)/n − c) · 1/i
1 − q i=k+1
  X n
v(k + 1)
≤ −c · 1/i
n i=k+1

By using the exact same algebra as in the proof of Proposition 1, we can conclude that the
above expression is smaller than Π∗S . For the sake of completeness, consider pb > v. As n → ∞ and
qvs → 0, the consumer would not even purchase a single blind box as the expected marginal value
16

from the first purchase equals (1 − q)v + qvs , which is strictly smaller than v as qvs → 0. Any such
price, therefore, results in zero profit.
In conclusion, we have proved that for any given pb , there exists a threshold nt such that when
n > nt , the platform’s expected profit under blind box selling is smaller than the profit under
separate sales. Since this is true for any pb > 0, this must also be true for the optimal blind box
price as the number of regular items becomes infinitely large.

D.3. Proofs from Section 4

Proof of Proposition 3: We begin with the proof of statement (i) of the proposition, namely
that for any given β = c/v ∈ [0, 1], there exists a threshold ᾱ such that Π∗B (α, β) > Π∗S (α, β) if
and only if α > ᾱ. Additionally, we also prove that the ratio of the seller’s profit under blind box
selling to separate sales—i.e., Π∗B (α, β)/Π∗S (α, β)—weakly increases with α. For clarity, we make
the dependence of the seller’s profit on α (the status parameter) and β explicit in this proof.
Proof of Statement (i): First, we prove that for any given β, the ratio Π∗B (α, β)/Π∗S (α, β) weakly
increases with α. From our characterizations in Section 2, the seller’s profit under separate selling
can be written as Π∗S (α, β) = (n + 1)(v − c) (recall the no rationing assumption). The seller’s profit
under the blind box strategy is given by Π∗B (α, β) = m∗ (α, β)(p∗ (α, β) − c), where p∗ (α, β) and
m∗ (α, β) denote the optimal price and expected volume (sales) of a blind box. The ratio of the
profits can now be expressed as follows:

Π∗B (α, β) m∗ (α, β)(p∗ (α, β) − c)

= , (D.3)
Π∗S (α, β) (n + 1)(v − c)
The monotonicity argument follows almost directly. First, we note that the denominator is
independent of α. To show that Π∗B (α, β) is increasing in α, consider any α and α′ > α. Suppose
that p∗ (α, β) is the optimal price under status parameter α. If the seller adopts the same price
and probability q of drawing the special item, then, under α′ , we get a lower bound on the profit,
namely:
Π∗B (α′ , β) ≥ m(α′ , β)(p∗ (α, β) − c),

where m(α′ , β) is the expected number of blind boxes purchased by a consumer under price p∗ (α, β)
and status parameter α′ . However, one can argue in a straightforward manner that m(α′ , β) ≥
m∗ (α, β)—since vs is strictly increasing in α, the stopping condition under α′ would involve more
items (or at least the same as α). Thus, for any α′ > α, Π∗B (α′ , β) ≥ Π∗B (α, β). It then follows that
the ratio Π∗B (α, β)/Π∗S (α, β) weakly increases with α, holding β fixed.
 17

Armed with the monotonicity result, we now aim to prove the existence of a finite threshold
ᾱ > 0 such that blind box selling dominates separate sales if and only if α > ᾱ. In order to prove
this claim, it is sufficient to show the following:
1. When α = 0, Π∗S (α, β) > Π∗B (α, β) for any β ∈ [0, 1).
2. There exists some αt > 0 such that when α = αt , Π∗S (α, β) < Π∗B (α, β).
Π∗
B (α,β)
The second statement above combined with the monotonicity of Π∗ (α,β)
immediately implies
S
Π∗
B (α,β)
the existence of a threshold ᾱ ≤ αt such that Π∗ (α,β)
> 1 for all α > ᾱ. To fulfill this checklist,
S

first consider α = 0. This scenario is equivalent to a setting where there are n + 1 regular items
and no special item, i.e., vs = v always. Consequently, we can invoke Proposition 1 to show that
Π∗S (α, β) > Π∗B (α, β) for α = 0. By continuity, this also implies the threshold ᾱ at which blind box
selling dominates must be strictly positive, i.e., ᾱ > 0.
It remains for us to identify a parameter αt at which Π∗S (α, β) < Π∗B (α, β). First, let us consider a
1
solution with pb = qvs with q = 2(n+1)
denoting the probability of drawing the special item. We will
prove that when α is sufficiently large, blind box selling is more profitable under this parameter
choice. By Lemma C2 (statement (ii)), it is clear that when pb = qvs , the consumer purchases the
blind box at least until they receive the special item. Therefore, the expected number of purchases
m ≥ 1q . The seller’s profit under this specific strategy can be expanded as follows:
1 c q
ΠB = m(pb − c) ≥ (qvs − c) = vs − = v(1 + α(1 − )) − 2c(n + 1).
q q (1 − q)/n
Fixing n and, therefore, q, it is clear that the above expression is strictly increasingly in α. Let
α′ be the smallest value of α at which ΠB = Π∗S = (n + 1)(v − c). Then, define αt = 2α′ . Since the
above expression ΠB (α) is strictly increasing in α, then it follows that ΠB (αt ) > Π∗s . Since ΠB (αt )
is a lower bound on the seller’s optimal revenue, it follows that Π∗B (α, β) > Π∗S (α, β) at α = αt .
To conclude, we have met both conditions in the checklist above—i.e., Π∗S (α, β) > Π∗B (α, β) at
α = 0 and Π∗S (α, β) < Π∗B (α, β) at α = αt . Given that ΠB (α, β) is weakly increasing in α, there must
exist some cutoff ᾱ such that Π∗S (α, β) < Π∗B (α, β) if and only if α < ᾱ. This completes our proof
of the first statement.

Proof of Statement (ii). Next, we show that for any given α > 0, there exist two thresholds
0 ≤ β L ≤ β H ≤ 1 such that Π∗B > Π∗S if and only if β < β L or β > β H . Following this, we argue that
the thresholds are not vacuous—i.e., β L , β H ∈ (0, 1)—when the status parameter α is sufficiently
large.
We now outline the proof approach. First, we prove that if the seller adopts a high price—i.e.,
pb > v—there exists a cutoff β H (pb ) ∈ [0, 1] such that blind box selling strictly dominates separate
18

selling at this price if and only if β > β H (pb ). Alternatively, if the seller pursues a low or moderate
pricing policy where pb ≤ v, there exists a cutoff β L (pb ) ∈ [0, 1] such that blind boxes strictly
dominate at this price if and only if β < β L (pb ). Combining these two scenarios, we conclude that
blind boxes are more profitable if and only if β > β H or β < β L . In particular, one can define the
two thresholds by endogenizing the dependence on price19 :

β L = sup β L (pb ) and β H = inf β H (pb ) (D.4)

pb ≤v pb >v

To see why the condition is both necessary and sufficient, note the following: if there existed
a value β ∈ [β L , β H ] where blind boxes obtained more profit (e.g., under a price pb ≤ v), then by
definition of β L , it must be true that β L ≥ β L (pb ), which is a contradiction. The same argument
holds for pb > v and β < β H .
We first write out the seller’s profit under separate selling Π∗S and blind boxes given (pb , q) where
pb = (1 − q) k+1
n
v as per the representation in Lemma 3.

ΠS = (n + 1)(v − c),

k+1
ΠB = m((1 − q) v − c),
n
where m is the expected purchase volume given (pb , q).
(a) Suppose pb > v. Our goal here is to identify a threshold β H ∈ [0, 1] such that blind box selling
strictly dominates if and only if β > β H .
First, note that if blind box selling is to outperform separate selling (i.e., ΠB > ΠS ), then it must
be the case that
k+1
m((1 − q) v − c) > (n + 1)(v − c).
n
Rearranging the terms, it is (necessary and) sufficient to show that the following expression holds:
 
k+1
m(1 − q) − (n + 1) v > (m − (n + 1)) c. (D.5)
n

Recall that since pb = (1 − q) k+1

n
v > v by definition, it must be the case that (1 − q) k+1
n
> 1. This in
turn implies that m(1 − q) k+1
n
− (n + 1) > m − (n + 1). In what follows, we predicate the arguments
based on whether or not the expression on the right-hand side of (D.5) is positive or negative. We
also ignore the dependence of β L and β H on pb when the meaning is clear from the context.

19
If β L > β H as per the definitions, then we can simply set β L = β H without loss of generality.
 19

1. m − (n + 1) ≥ 0: In this case, the required inequality in (D.5) follows immediately since we

have already proved that m(1 − q) k+1
n
− (n + 1) > m − (n + 1) and given that v ≥ c. Therefore,
when m ≥ n + 1 under the given (pb , q), we can infer that β H = 0, i.e., blind box selling is
always more profitable regardless of cost.20
2. m − (n + 1) < 0: Under this condition, (D.5) can be equivalently transformed as follows:

c (n + 1) − m(1 − q) k+1
n
β= > ,
v (n + 1) − m

where the denominator is strictly positive. If the numerator is also non-negative, we can
conclude that β H < 1 since m < m(1 − q)(k + 1)/n. Alternatively, if the numerator is negative,
then the above inequality is trivially true, implying that β H = 0.
In conclusion, for any given policy (pb , q) with pb > v and given α, one can identify a threshold
β H such that blind boxes strictly dominate separate selling if and only if β > β H . Note that β H < 1
in all of these cases.

(b) Suppose pb ≤ v. Our goal here is to identify a threshold β L ∈ [0, 1] such that blind box selling
strictly dominates if and only if β < β L .
Considering (D.5) once again, we now have that m(1 − q) k+1
n
− (n + 1) ≤ m − (n + 1) since
pb = (1 − q) k+1
n
v ≤ v. Before proving the required claim (namely (D.5)), we partition the parameter
space based on the sign of the left and right-hand sides of (D.5) and rule out the infeasible cases.
(1) m(1 − q) k+1
n
− (n + 1) ≤ 0 and m − (n + 1) > 0: we discuss this case below.
(2) m(1 − q) k+1
n
− (n + 1) > 0 and m − (n + 1) > 0: we discuss this case below.
(3) m(1 − q) k+1
n
− (n + 1) ≤ 0 and m − (n + 1) ≤ 0: under this case, the only possibility is β L = 0.
To see why, one can rearrange (D.5) to get

c (n + 1) − m(1 − q) k+1
n
> .
v (n + 1) − m

The numerator is always larger than the denominator, and therefore, the condition never holds
since β = c/v ≤ 1.
(4) m(1 − q) k+1
n
− (n + 1) > 0 and m − (n + 1) ≤ 0: this region is infeasible as it violates the
requirement outlined earlier, i.e., that m(1 − q) k+1
n
− (n + 1) ≤ m − (n + 1).
We now go over the first and second cases. In the first case, where m(1 − q)(k + 1)/n − (n + 1) ≤ 0,
it is straightforward to see that (D.5) never holds as the left-hand side is negative (or zero) and the
right-hand side is positive. In other words, the profit derived from blind box selling is no more than
20
Note that for a small value of α, m(1 − q) k+1n
− (n + 1) > 0 does not exist as shown in Lemma C6. Therefore,
β H = 0 does not violate the first statement of this proposition.
20

that of separate selling. Therefore, β L = 0 in this case. Under Case (2), one can rearrange (D.5) so
that blind box selling is more profitable than separate selling if and only if:

c m(1 − q) k+1
n
− (n + 1)
< ,
v m − (n + 1)

The right-hand side of this inequality is less than one due to the fact that 0 < m(1 − q)(k + 1)/n −
m(1−q) k+1
n −(n+1)
(n + 1) < m − (n + 1). It follows that β L = m−(n+1)
for the given pb .
We conclude the proof by commenting on when β L and β H are not vacuous. First, note that
β H < 1 for any given instance. Second, when α is large, case (2) holds for suitable pb and, therefore,
1
β L > 0. For example, as described earlier, suppose that pb = qvs and q = 2(n+1)
. Further, let α be
sufficiently large so that
1 pb q
· = (1 + α(1 − )) = n + 2.
q v (1 − q)/n
Then, we argue that case (2) is applicable here. To see why, note that:

k+1 pb 1 q
m(1 − q) = m · ≥ · qv(1 + α(1 − )) > n + 1,
n v q (1 − q)/n
q 1
due to our choice of α. Further pb ≤ v since q · (1 + α(1 − (1−q)/n )) = 2(n+1)
· n + 2 ≤ v. This concludes
our argument that β L is strictly positive under large α as there exist prices that fall under case (2).
Conversely, when α is small, it is possible that β L = 0 which can also be observed in Figure 3.

Proof of Lemma 4. We begin with some notation, following which we outline the key ideas
underlying this proof. Define CS(nr , ns ) to be the consumer’s expected surplus resulting from any
future purchases of the blind box, given that the consumer currently possesses nr ∈ {0, 1, 2, . . . , n}
unique regular items and ns ∈ {0, 1} special items. We prove by induction that for any given state
(nr , ns ), the expected future surplus is non-negative and identify conditions under which it is strictly
positive. In essence, this follows from the observation that given any state (nr , ns ), the consumer
only purchases the product if the expected value from purchase (1 − q)(n − nr )v/n + (1 − ns )qvs is
at least the price of the blind box pb . We show that the expected value can be decomposed into
two components—the actual value for receiving a new item and the expected number of purchases
before receiving a unique item. The consumer’s expected payment equals the price pb multiplied
by the expected number of purchases. As a result, the consumer’s expected surplus as they move
from state (nr , ns ) to the next state (nr + 1, ns ) or (nr , ns + 1) is non-negative. Upon proving the
inductive claim, the main lemma statement follows immediately since the surplus under separate
selling is zero.
 21

In the following segment, we will verify CS(nr , ns ) ≥ 0 if (1 − q)(n − nr )v/n + (1 − ns )qvs ≥ pb

under two cases: (1) ns = 0 and (2) ns = 1. Note that the inductive claim follows trivially when
(1 − q)(n − nr )v/n + (1 − ns )qvs < pb since the consumer would simply not purchase any product.
We treat any such state with (1 − q)(n − nr )v/n + (1 − ns )qvs < pb as our base case and proceed
with the inductive claim. Specifically, given (nr , ns ), we assume that CS(n′r , n′s ) ≥ 0 holds as long
as either n′r > nr or n′s > ns .
Case I: ns = 0.
Given state (nr , ns ), let us first define qr = (1 − q) (n−n
n
r)
to be the probability of drawing a new
regular item from a subsequent blind box purchase. Also, without loss of generality, suppose that
nr < n. If the consumer is in state (nr , ns ), then upon purchasing the blind box at price pb , one of
three events could occur, namely: (i) the consumer receives a new regular item with probability
qr and moves to state (nr + 1, ns ); (ii) the consumer receives the special item with probability q
and moves to state (nr , ns + 1); (iii) the consumer does not receive any new item with probability
(1 − q − qr ). Based on this, one can write an expression for CS(nr , ns ) as follows.
∞
X  
CS(nr , ns ) = (1 − q − qr )t−1 q(vs + CS(nr , ns + 1)) + qr (v + CS(nr + 1, ns )) − pb
t=1
∞
X  
≥ (1 − q − qr )t−1 qvs + qr v − pb (D.6)
t=1
qvs + qr v − pb
=
q + qr
≥ 0. (D.7)

In (D.6), we used the inductive hypothesis, namely that CS(nr + 1, ns ) ≥ 0 and CS(nr , ns + 1) ≥
0. The final expression follows from our assumption that (1 − q)(n − nr ) nv + qvs = qr v + qvs ≥ pb .
Case II: ns = 1.
Once again, suppose that nr < n. In this case, the consumer already possesses the special item.
Therefore, if the blind box is purchased, it must be the case that (1 − q)(n − nr ) nv ≥ pb . The proof
then follows from the following expression along the lines of our arguments for Case I.
∞
X  
CS(nr , ns ) ≥ (1 − qr )t−1 qr v − pb ≥ 0.
t=1

Finally, we address the question of when the surplus is strictly positive. In this regard, it is easy
to observe that consumer surplus cannot be positive when pb = vB (0, 0) = (1 − q)v + qvs as the
consumer would only purchase the blind box once before receiving a unique item and reaching the
stop condition. On the contrary, when pb < (1 − q)v + qvs , the consumer always receives a positive
22

surplus from the first purchase and, therefore, from the blind box in general as all subsequent
purchases result in non-negative surplus as per our inductive hypothesis. Therefore, it suffices
for us to identify conditions under which the optimal blind box parameters (p∗b , q ∗ ) satisfy p∗b <
(1 − q ∗ )v + q ∗ vs . Consider any q. Then the seller’s profit (π1 ) under the blind box strategy for
pb = (1 − q)v + qvs can be written as:

π1 = (1 − q)v + qvs − c.

Next, consider an alternative strategy for the same probability q of drawing the special item, but
where pb = (1 − q) n−1
n
v + qvs . Under this price, the seller’s profit (π2 ) can be written as:

n−1
π2 ≥ (1 + (1 − q))((1 − q) v + qvs − c),
n

where the inequality holds due to the fact that the expected purchase number is higher than
1 + (1 − q). Further transforming, we get that π2 ≥ π1 as long as the β is smaller than some suitably
defined threshold β̄s .

Proof of Corollary 1. The corollary is a consequence of the arguments made in the proof of
Proposition 3 as well as the definitions of β L and β H . First, note that since we only consider
instances where β L is strictly smaller than β H , there is no overlap between the two parameters as
per the definitions in (D.4).
To prove the first statement, consider c/v > β H . According to the definition of β H , the only way
to outperform separate selling in this scenario is by selecting a price pb > v. Therefore, the optimal
price must also satisfy p∗b > v.
Next, in order to prove the second statement, consider c/v < β L . Following the same argument
as the previous case, it is clear that the optimal price for the blind box must satisfy p∗b ≤ v. Further,
among the four cases outlined in the proof of Proposition 3, the only feasible case where β L ∈ (0, 1)
is case (2), in which m > n + 1. The second statement of the corollary then follows.

D.4. Proofs from Section 5

Proof of Lemma 5. First, regardless of any other choice, it is optimal for the seller to extract
full surplus for the regular items, i.e., p̃r = v. Second, if the supply of the special item is given by
xs ∈ [0, 1], then the optimal price for this item under separate sales must be p̃s = ṽs (xs ). Fixing the
two prices, one can now solve for the optimal rarity level by reformulating (9) as follows:

ΠS (xs ) = (v − c) · n + (v(1 + (1 − xs )α) − c) · xs . (D.8)

 23

Note that it is always optimal to sell each of the n regular items to all of the buyers, as this
would also boost the rarity of the special item. Subsequently, it can be demonstrated that the
quadratic function in (D.8) reaches its maximum value, (v − c) · n + (((1 + α)v − c)2 )/(4αv), when
x∗s = ((1 + α)v − c)/(2αv).

Proof of Proposition 4. We break down the proof into the following four steps: Step 1: there
exists a threshold α̃L such that if α ≤ α̃L , then Π̃∗B ≤ Π̃∗S . Step 2: there exists a threshold α̃H
such that if α > α̃H , then Π̃∗B > Π̃∗S . Step 3: there exists a threshold β̃ H such that if β ≥ β̃ H , then
Π̃∗B ≤ Π̃∗S . Step 4: there exists a threshold β̃ L such that if β < β̃ L , then Π̃∗B > Π̃∗S .
Step 1. From Proposition 1, we know that in the absence of the special item, separate selling
strictly dominates blind box selling for all β = c/v. Since the absence of the special item corresponds
to the α = 0 case, one can then apply continuity arguments to infer the existence of a threshold
α̃L such that if α ≤ α̃L , separate selling dominates blind box selling for any given β.
Step 2. In this step, we construct a blind box package with price p̃b under which blind box selling
outperforms separate selling when the value of α is sufficiently large. We then use this construction
to derive a cutoff α̃H such that blind box selling is optimal for all α > α̃H .
Consider a price p̃b = qvs for the blind box for some appropriate value of q. By Lemma C2(ii), a
consumer will continue purchasing the blind box until they acquire the special item. This implies
that the total purchase quantity m is 1/q. Consequently, the profit function under the blind box
strategy can expressed as:
1 q c
Π̃B (q) = m(p̃b − c) = (qvs − c) = (1 + α(1 − )v − .
q (1 − q)/n q
p
The above function Π̃B (q) is maximized at q ∗ = 1/(1 + nαv/c). Under this choice of q, the
√
maximum profit is given by (1 + α)v − c − 2 nαvc. We now compare this profit to the optimal
profit under separate selling.
Specifically, the optimal profit under separate selling is (v − c) · n + (((1 + α)v − c)2 )/(4αv) as
shown in the proof of Lemma 5. If Π̃B > Π̃∗S , then the following inequality holds:
√ ((1 + α)v − c)2
(1 + α)v − c − 2 nαvc > (v − c) · n +
4αv
Upon rearranging, we get the equivalent inequality:
3 √ (1 + 2α)v 2 + c2 − 2c(1 + α)v
(1 + α)v − c − 2 nαvc > (v − c) · n + (D.9)
4 4αv
It is evident that the right-hand side of (D.9) is decreasing in α whereas the left-hand side must be
√
increasing in α as long as α > 16cn
9v
, i.e., the derivative of 43 αv − 2 nαvc is non-negative. Therefore,
24

there must exist a threshold α̃H such that (D.9) holds for all α > α̃H , which completes our proof
of step 2. Since blind box selling is optimal under our specific price (i.e., p̃b = qvs ), this must also
be the case under the optimal parameter set.
Step 3. To prove this result, we consider three scenarios based on the optimum price of the
blind box as outlined in Lemmas C1 and C2:
(1) A low price scenario satisfying p̃b ≤ v (without making any assumptions on how p̃b relates to
qvs ) or a moderate price scenario where qvs < pb ≤ v;
(2) A moderate price scenario where v < pb ≤ qvs ;
(3) A high price scenario where p̃b > qvs and p̃b > v.
In each of these three scenarios, we demonstrate that when β is sufficiently large, separate selling
is more profitable than the blind box strategy. Consequently, one can define the final threshold β̃ H
as the maximum of the thresholds obtained for each of the three cases.
(1) Low price with p̃b ≤ v:
Suppose that p̃b = (1 − q) k+1
n
as defined in Lemma 3 and m denotes the total volume of blind
boxes sold at this price. In order to prove Π̃B ≤ Π̃∗S , we must show that:

k+1
m((1 − q) v − c) ≤ (v − c) · n + (v(1 + (1 − x∗s )α) − c) · x∗s ,
n

where we used the fact that Π̃B = m(p̃b − c) and the expression for Π̃∗S is from Lemma 5. At
c = (1 − q) k+1
n
v, the left-hand of the above expression equals zero, whereas the right-hand side is
strictly positive for any x∗s ∈ (0, 1) when α > 0. By continuity arguments, we can infer the existence
of a threshold β1 < 1 (recall β = c/v) such that Π̃∗S > Π̃B for all β > β1 .
(2) Moderate price with v < p̃b ≤ qvs :
Recall from statement (ii) of Lemma C2 that in the regime where v < p̃b ≤ qvs , the consumer’s
expected number of purchases always equals m = 1/q. As a result, in this regime, the seller’s
optimal price p̃b = qvs . Upon inverting the inequality in (D.9), we obtain the following (necessary
and) sufficient condition to prove Π̃B < Π̃∗S :
√ ((1 + α)v − c)2
(1 + α)v − c − 2 nαvc < (v − c) · n + (D.10)
4αv

When c = v (i.e., β = 1), the above inequality reduces to:

√ αv
αv − 2v nα < ,
4
64 64 64
or equivalently α < 9
n. For α ≥ 9
n, define β2 = 1. Moreover, for any fixed α < 9
n, there must
exist some β2 < 1 such that (D.10) holds for all β ∈ (β2 , 1]. The existence of such a threshold
 25

follows from continuity arguments (applied at c = v or β = 1). In other words, for any given α, we
have identified a threshold β2 such that for any β > β2 , when the seller’s price p̃b = qvs , separate
selling outperforms blind box selling. Finally, we remark that setting β2 = 1 for large values of α
is unavoidable as blind boxes are always more profitable under large values of α; see step (2).
(3) High price with p̃b > qvs and p̃b > v:
Our goal is to identify a cutoff β3 ∈ [0, 1] such that separate selling is more profitable for all
β > β3 . The high-level approach is as follows. We first consider the case of c = v (i.e., β = 1) and
prove that separate selling is more profitable than blind boxes for smaller values of α. Following
this, we once again invoke continuity to argue about the existence of a threshold β3 < 1 (for small
values of α) so that separate selling is more profitable for all β ≥ β3 . On the other hand, for larger
values of α, we set β3 = 1. This choice is unavoidable since blind boxes always outperform when
α > α̃H ; see step 2.
To begin, we derive a lower bound on the seller’s profit from adopting blind boxes. Given (p̃b , q),
the seller’s profit under the blind box model can be written as Π̃B = m(p̃b − c), where m denotes
the expected quantity sold. We note that p̃b ≤ (1 − q)v + qvs as any price higher than this level
would result in zero sales. Further, if we represent p̃b = (1 − q) k+1
n
v, then we know from the proof
n n qvs
of Lemma C6 that m ≤ 1−q
log( k−κ ) where κ = (1−q)v/n
as defined in Section 3. In what follows,
we will assume that the value of α is sufficiently small so that κ ≤ n − 1, as per Lemma C5. Since
n
k > n − 1, this ensures that the denominator of the term log( k−κ ) is always positive. Later on, we
will ensure that our choice of β3 validates the κ ≤ n − 1 claim (by setting β3 = 1 when α is large).
First, consider the case where c = v (i.e., β = 1). In this regime, we can derive an upper bound
on the seller’s profit as follows:

Π̃B = m(p̃b − v)
n n
≤ log( ) ((1 − q)v + qvs − v)
1−q k−κ
 
n n q
= log( ) qvα(1 − )
1−q k−κ (1 − q)/n
 
n q q
= log( ) · vα(1 − )
k − κ (1 − q)/n (1 − q)/n
n
= log( )χv.
k−κ
q q
In the final step, we use χ to represent (1−q)/n
· α(1 − (1−q)/n ). Next, for the case of separate sales
at c = v, we can invoke Lemma 5 to derive

Π̃∗S = (v − c) · n + max (v(1 + (1 − xs )α) − c) · xs = v max (xs (1 − xs )α)) ≥ χv.

xs ∈[0,1] xs ∈[0,1]
26

q q
In the final inequality above, we used the observation that χ = (1−q)/n
· α(1 − (1−q)/n
) ≤
maxxs ∈[0,1] (xs (1 − xs )α)).
Now, we are ready to complete the proof. In order to prove Π̃B ≤ Π̃∗S , it is sufficient to show
that:
n n
log( )χv ≤ χv =⇒ log( ) ≤ 1.
k−κ k−κ
n n
Towards this objective, define h(κ) = log( n−1−κ ). Note that h(κ) is an upper bound on log( k−κ )
since k > n − 1 (recall p̃b = (1 − q) k+1
n
v > v). When α = 0, we have κ ≤ 1 and therefore, h(κ) ≤
n
log( n−2 ), which is strictly smaller than one for n ≥ 4. Given that κ is strictly increasing in α, let
α1 denote the smallest value of α at which h(κ) = 1.21 Then, we can conclude that for all α ≤ α1 ,
n
log( k−κ ) < h(κ) ≤ 1 and therefore, Π̃B ≤ Π∗S at c = v. In other words, when β = 1 and n ≥ 4,
separate selling is more profitable for all α ≤ α1 .
We now use a continuity argument to derive the threshold β3 that we alluded to at the beginning
of this case. First, set β3 = 1 for n ≤ 3 or α > α0 . For any given α ≤ α0 , we know that Π̃B ≤ Π̃∗S at
β = 1. Therefore, there must exist some threshold β3 < 1 such that Π̃B ≤ Π̃∗S holds for all β ≤ β3 .
This completes our proof of the high-price case.
Step 3 Conclusion:
In conclusion, let β̃ H = max{β1 , β2 , β3 } for any given value of α. Then, Π̃∗B ≤ Π̃∗S if β ≤ β̃ H . As
mentioned earlier, when α is extremely large (step 2), blind box selling dominates separate selling
and therefore β H = 1.
Step 4. If β < β̃ L , then Π̃∗B > Π̃∗S .
Invoking our derivation from step (2)—in particular (D.9)—we can show that for Π̃∗B > Π̃∗S to
hold, it is sufficient to prove the following inequality:

3 √ (1 + 2α)v 2 + c2 − 2c(1 + α)v

(1 + α)v − c − 2 nαvc > (v − c) · n + . (D.11)
4 4αv

In particular, the above inequality is the result of setting p̃b = qvs and optimizing over the value of
q. Setting c = 0, (D.11) simplifies to:

3 (1 + 2α)v v v
(1 + α)v > vn + = vn + + .
4 4α 4α 2

Let α0 denote the smallest value of α at which the above inequality holds at equality. Since the
left-hand side is increasing in α whereas the right-hand side is non-increasing, it is evident that
the inequality holds strictly for all α > α0 . Therefore, define β̃ L = 0 for α ≤ α0 .

21
Note that for α ≤ α1 , it must also be true that κ ≤ n − 1 and thus our choice earlier is valid.
 27

Next, for any α > α0 , there must exist some β̃ L > 0 (recall β = c/v) such that (D.11) holds for
all β < β̃ L . The existence of such a threshold must follow from continuity arguments since the
described inequality holds at β = 0. To conclude, we have just proved for any given value of α,
there exists a cutoff β̃ L such that blind boxes outperform separate selling for all β < β̃ L . Further,
the threshold is not vacuous for larger values of α.

Proof of Lemma 6. As mentioned earlier, when buyers’ valuations for the regular item are
heterogeneous, i.e., v̂ ∈ {v − ϵ, v + ϵ}, the seller can choose prices for either item in such a way
that the item is either sold to both buyer types (low price) or only purchased by high-type buyers
(high price). Table D.1 summarizes the three possible pricing policies that a seller may adopt under
separate selling.

Table D.1 Potential pricing policies under separate selling. In this table, “Both” indicates that the seller sells

the item to both types of consumers, whereas “High” indicates the seller targets the item to high-type consumers

only.

Policy Regular item Special item Profit

LL Both Both (v − ϵ − c)(n + 1)
LH Both High (v − ϵ − c)n + ((v + ϵ)(1 + α/2) − c)/2
HH High High (v + ϵ − c)(n + 1)/2

For the sake of clarity, we use the superscripts {LL, LH, HH } on the profit term Π̂S to differen-
tiate the profit under different policies.
Statement (i):
We prove that when ϵ ≤ ϵS , Π̂LL LH
S ≥ Π̂S and Π̂LL HH
S ≥ Π̂S , where ϵS = ((2 − α)v − 2c)/(α + 6). We

begin by comparing the LL and LH policies. Since these policies yield identical profits from the
regular item, it is sufficient to compare the profit from the special item. In particular, if Π̂LL LH
S ≥ Π̂S ,

then following inequality must hold:

(v + ϵ)(1 + α/2) − c
(v − ϵ − c) ≥ , (D.12)
2

where (v + ϵ)(1 + α/2) denotes the high-type consumers’ valuation for the special item when the
low-type is priced out. Solving, it is straightforward to see that the LL policy is preferred if and
only if ϵ ≤ ϵS . Next, we compare the LL and HH policies. If Π̂LL HH
S ≥ Π̂S , we have:

(n + 1)
(v − ϵ − c)(n + 1) ≥ (v + ϵ − c) . (D.13)
2
28

Simplifying this inequality yields ϵ ≤ ϵS , where ϵS = (v − c)/3. Further, ϵS ≥ ϵS for all α ≥ 0.

Therefore, we can conclude that the LL policy is optimal for ϵ ≤ ϵS
Statement (ii):
We now prove the optimality of the LH policy when ϵS ≤ ϵ ≤ ϵ̄S , where ϵ̄S = (2n(v − c)+vα)/(6n −
α).
First, from (D.12), it is clear that the LL policy underperforms the LH policy for all ϵ ≥ ϵS .
Therefore, it suffices to compare the LH and the HH policies to prove statement (ii). In this regard,
Π̂LH HH
S ≥ Π̂S as long as:

(v + ϵ)(1 + α/2) − c (v + ϵ − c)(n + 1)

(v − ϵ − c)n + > (D.14)
2 2

Upon simplification, this leads to ϵ ≤ ϵ̄S , which completes the proof.

Statement (iii):
We prove that Π̂HH
S ≥ max(Π̂LL LH
S , ΠS ) if ϵ ≥ ϵ̄S . The proof follows almost directly from our

previous derivations. First, from (D.13), we know that Π̂HH

S ≥ Π̂LL
S for all ϵ ≥ ϵS = (v − c)/3.
Moreover, we can establish that:

v−c
ϵS =
3
2n(v − c)
=
6n
2n(v − c) + vα
≤
6n − α
= ϵ̄S .

We conclude the proof of this statement and the lemma by noting that the outperformance of
HH over LH for all ϵ ≥ ϵ̄S follows from (D.14).

Proof of Lemma 7. From Table D.1, it is evident that the profit under the LL policy decreases
with ϵ, while under policy HH, it increases with ϵ. We now turn our attention to the monotonicity
of policy LH. Recall from the proof of Lemma 6 that Π̂LH
S denotes the seller’s profit under policy
LH. Differentiating this profit with respect to ϵ gives us:

dΠ̂LH
S α 1
= −n + + .
dϵ 4 2

If α ≤ 4n − 2, then dΠ̂LH LH
S /dϵ ≤ 0; conversely, if α > 4n − 2, dΠ̂S /dϵ > 0. Setting αS = 4n − 2 and

recalling the optimal profits from Lemma 6, the first statement of the lemma follows, namely:
• if α ≤ αS , Π̂∗S is decreasing in ϵ for ϵ ∈ [0, ϵ̄S );
 29

• if α > αS , then ϵS < 0 where ϵS is as defined in Lemma 6. Therefore, the optimal profit for all
ϵ ≤ ϵ̄S is given by Π̂∗S = Π̂LH
S . The profit then increases with ϵ for all ϵ ∈ [0, ϵ̄S ];

• given that the profit under the HH policy increases with ϵ, Π̂∗S increases with ϵ for ϵ ≥ ϵ̄S .
Statement (ii) - Blind Box. Our objective is to show the existence of a threshold ϵ̄B such that the
optimal profit under blind box selling is weakly increasing in ϵ for ϵ ∈ [ϵ̄B , v]. We begin the proof by
showing that when ϵ is sufficiently large, low-type consumers do not purchase even a single blind
box under the optimal blind box policy (p̂∗ , q̂ ∗ ). Following this, we argue that increasing ϵ further
can only lead to a corresponding increase in the seller’s profit.
Under any given blind box (p̂, q̂), let mH and mL denote the expected purchase quantity for the
high-type and low-type consumers, respectively. Further, let mmax denote the expected number of
purchases from either consumer type when the price of the blind box is zero; this serves as an
upper bound on both mH and mL . First, consider any price p̂′ at which mL > 0—i.e., the low-type
consumer purchases the blind box at least once. Then, we can establish the following upper bound
on the profit (Π̂L ) from such a policy:

mH + mL ′ n+κ
Π̂L = (p̂ − c) ≤ mmax ((1 − q̂) (v − ϵ) − c),
2 n

where (1 − q)(n + κ)(v − ϵ)/n denotes the maximum price p̂ under which the low-type consumer
still purchases the blind box at least once.
Alternatively, suppose that p̂′′ = (1 − q̂) n1 (v + ϵ), which denotes the price under which the high-
type consumer continues purchasing the blind box until they collect all n regular and the special
item. For simplicity, we fix q̂ for both the policies being compared. Note that q̂vs ≥ (1 − q̂) n1 (v + ϵ)
holds always and so p̂′′ ≤ q̂vs under the proposed policy. Then, the seller’s profit (Π̂H ) under this
specific policy of ‘targeting high-type consumers’ can be written as follows:
 
mH + mL ′′ mmax 1
Π̂H = (p̂ − c) ≥ (1 − q̂) (v + ϵ) − c .
2 2 n

Note that mH = mmax by definition. We make two claims that compare and contrast Π̂L and Π̂H .
First, there exists a threshold ϵ′ such for all ϵ ≥ ϵ′ , it must be the case that Π̂H ≥ Π̂L . Since Π̂H > Π̂L
holds when ϵ = v (note that ΠL = 0 when ϵ = v), the claim follows from a continuity argument.
Second, there exists a threshold ϵ′′ such that for all ϵ > ϵ′′ , p̂′′ > p̂′ and therefore, the low-type
consumer does not purchase the blind box even once. One can obtain a closed-form expression for
ϵ′′ by solving the following equation.

1 n+κ
p̂′′ = (1 − q̂) (v + ϵ) = (1 − q̂) (v − ϵ) = p̂′ .
n n
30

Once again, we note that when the price of the blind box is strictly larger than p̂′ , the low-type
consumer will not purchase the blind box (i.e., mL = 0) since even the first purchase is not profitable
in expectation. To conclude, suppose we set ϵ̄B = max(ϵ, ϵ′′ ). Then, for all ϵ ≥ ϵ̄B , it is optimal to
price the blind box such that the low-type consumer does not make any purchases. Consequently,
for any price p̂, the number of purchases by the high-type consumer is weakly increasing in ϵ. This
is also true once we endogenize the optimal price. In other words, we conclude that the optimal
profit under the blind box selling, Π̂∗B , is increasing in ϵ for ϵ ∈ [ϵ̄B , v].

Proof of Proposition 5. First, consider the two scenarios where 0 < ϵ ≤ ϵS and ϵ̄S ≤ ϵ < v, where
ϵS and ϵ̄S are as defined in Lemma 6. Note that when 0 < ϵ ≤ ϵS , it is optimal for the seller to
adopt the LL policy under separate sales, and when ϵ̄S ≤ ϵ < v, the HH policy is optimal. In both
these scenarios, the outcome under separate selling can be viewed as arising from a system of
homogeneous consumers with the valuation of the regular item being v − ϵ when 0 < ϵ ≤ ϵS , and
v + ϵ when ϵ̄S ≤ ϵ < v. Consequently, by Proposition 3, it is evident that for small β, the optimal
profit from blind boxes, Π̂∗B exceeds the optimal profit from separate selling, Π̂∗S . In particular:
• Suppose 0 < ϵ ≤ ϵS . In this scenario, one could consider a homogeneous instance where all
consumers have the same valuation v − ϵ for the regular item. Then, from the proof of Proposi-
tion 3, there exists a threshold β L ≥ 0 such that the profit from blind boxes Π∗B is strictly larger
than that from separate selling Π∗S when β < β L . Transforming this to our original instance
with heterogeneous consumer valuations, it is clear that Π̂∗S = Π∗S and Π̂∗B ≥ Π∗B as long as
β < β L v−ϵ
v
. We then set β̂ = β L · v−ϵ
v
in the proposition statement. We note that Π̂∗B ≥ Π∗B
holds since the expected number of blind boxes purchased by consumers with valuation v + ϵ
is at least the number of blind boxes purchased by consumers with valuation v − ϵ in the
homogeneous instance.
• Suppose ϵ̄S ≤ ϵ < v. In a similar vein to our previous argument, we can compare against a
homogeneous instance (where the total population mass is one-half) where all consumers value
the regular item at v + ϵ. Once again, adding consumers with valuation v − ϵ does not alter
the optimal profit under separate selling but can only (weakly) increase the profit for blind
boxes. Further, the increase is strict as long as the low-type consumers purchase a non-zero
quantity of blind boxes. In this case, we set β̂ = β L v+ϵ
v
.
Note that in both the scenarios above, we set α̂ = 0. However, the thresholds are non-vacuous
(i.e., β L > 0) only when α is sufficiently large.
Finally, we consider the intermediate interval where ϵS ≤ ϵ < ϵ̄S , and therefore, the seller adopts
the LH policy under separate selling. In this scenario, we construct a specific policy for blind
 31

box selling that outperforms the optimal profit under separate sales. Building on the proof of
q
Proposition 4 (step 2), suppose we price the blind box at q(v + ϵ)(1 + α(1 − (1−q)/n
) for some
arbitrary q. Then, it is clear that each high-type consumer would purchase the blind at least until
they obtain the special item, and therefore, the expected purchase volume is at least 1q . Optimizing
for q (along the lines of our proof of Proposition 4), we get a lower bound on seller’s profit Π̂′B by
focusing only on the purchases by high type consumers, i.e.,
√
(1 + α)vH − c − 2 nαvH c
Π̂′B ≥ ,
2

where vH = v + ϵ. In the above expression, the denominator is 2 because we only consider the profit
from half the consumers (i.e., high-type). In the case of separate selling, we can write the optimal
profit under the LH policy from Table D.1 as follows:

Π̂∗S = (v − ϵ − c)n + ((v + ϵ)(1 + α/2) − c)/2

≤ (v + ϵ − c)n + ((v + ϵ)(1 + α/2) − c)/2

= (vH − c)n + (vH (1 + α/2) − c)/2.

Comparing Π̂′B with Π̂∗S , we conclude that blind box selling outperforms as long as
√
(1 + α)vH − c − 2 nαvH c
> (vH − c)n + (vH (1 + α/2) − c)/2
2
α √
=⇒ vH − 2 nαvH c > 2(vH − c)n
2
4nc
Given that the left-hand side is increasing in α (as long as α > vH
) and the right-hand side is
independent of α, one can infer the existence of a threshold α′ such that for all α > α′ , we have
Π̂′B > Π̂∗S . This in turn implies that Π̂∗B > Π̂∗S for all α > α′ . Finally, note that α′ is a function of
c, and hence β = vc . To ensure consistency with the proposition statement, we pick any arbitrary
β L ∈ (0, 1] and subsequently set α̂ to be the largest value of α′ such that Π̂′B > Π̂∗S holds for all
β < βL.
This concludes our proof, namely that for every ϵ ∈ [0, v], there exist thresholds α̂ and β̂ such
that Π̂∗B > Π̂∗S if α > α̂ and β < β̂.

Proof of Proposition 6. Before proving the stated claims, we first highlight some useful prop-
erties. Under separate selling, the seller can extract all the surplus from consumers, and therefore,
the optimal denoted as Π̌∗S , is (n + 1)(v − c) + v + , where v + is the lump-sum bonus. On the other
hand, given a blind box (p̌b , q̌), the consumer may only collect the bonus on some sample paths.
32

The profit from the optimal blind box can still be represented as Π̌∗B = m(p̌∗b − c), where m is the
expected number of blind boxes purchased again. Finally, suppose that Π∗S , Π∗B denote the optimal
profits from separate selling and blind box selling in the absence of the bonus. Then, the following
relationships hold:
Π̌∗S = Π∗S + v + and Π̌∗B ≤ Π∗B + v + . (D.15)

We now proceed with the two statements in the proposition.

Statement (i):
Suppose that at α = α1 , blind box selling is strictly more profitable, i.e., Π̌∗B (α1 ) > Π̌∗S (α1 ). In
order to prove the statement, it is necessary to show that for all α′ > α1 , Π̌∗B (α) > Π̌∗S (α) continues
to hold. First, let (p̌b (α1 ), q̌(α1 )) denote the optimal blind box parameters at α = α1 . Then, for
any α′ > α1 , it is straightforward to see that the consumer’s expected number of purchases cannot
decrease. Specifically, at state (nr , ns ), the consumer’s utility from purchasing a blind box remains
the same (at both values of α) if ns = 1 and strictly increases when ns = 0. Therefore:

Π̌∗B (α′ ) ≥ Π̌∗B (α1 ) > Π̌∗S (α1 ) = Π̌∗S (α′ ).

Note the revenue from separate sales Π̌∗S is independent of α. To complete the proof, we show
that the cutoff α̌ exists and is finite. For any fixed v + ≥ 0, we note that as α → ∞, the profit under
blind boxes Π̌∗B (α) → ∞. Therefore, blind boxes are more profitable at larger values of α.
Finally, we argue that α̌ ≥ ᾱ, where ᾱ is the corresponding threshold from Proposition 3. In
other words, the threshold at which blind boxes start outperforming is weakly higher under the
lump-sum bonus model. This follows from (D.15). In particular, for any α ≤ ᾱ, we have that:

Π̌∗S = Π∗S + v + ≥ Π∗B + v + ≥ Π̌∗B .

We define a partial collect pricing policy, which causes consumers to stop before collecting
all items. It is reasonable to assume that the value of v + does not approach infinity. Then, by
Proposition 3(i), we know that for any α > ᾱ, Π̌∗B > Π̌∗S − v + . Moreover, Proposition 3(i) says that
the ratio of Π̌∗B /Π̌∗S increases with α. It is clear that the optimal value under the blind box policy
approaches infinity as α goes to infinity. Thus, there exists a threshold α̌, where α̌ > ᾱ. Π̌∗B > Π̌∗S
if and only if α > α̌.
Next, we consider the scenario where the seller opts for a very low price, enabling the consumer
to collect all regular and special items. We call this the whole collect pricing policy. If, in Propo-
sition 3(i), under the optimal profit, the consumer collects all items, then in the lump-sum bonus
scenario, a larger α is not necessary.
 33

Statement (ii):
To prove this statement, it is sufficient to identify a blind box package (p̌b , q̌) under which
min{1,κ}
Π̌B > Π̌∗S holds for all β < β̌ for some suitable threshold β̌. Given q̌ = (n+1)α
, suppose that p̌ =
(1 − q̌)v · min{1, κ}/n + v + /m̌ denotes the price of the blind box, where m̌ is the expected number
of blind boxes purchased by consumers before they collect all n + 1 items. From Lemma C8, we
know under this price (for any arbitrary q̌), the consumer keeps purchasing the blind box until
they collect all n + 1 items. In what follows, we first show that under this price, the seller’s profit
for blind box selling is strictly higher than that under separate sales as long as α is not too small.
We then use a continuity argument to show the existence of a cutoff β̌ such that blind box selling
is optimal for all β < β̌ as long as α is sufficiently large (when α is small, we set β̌ = 0).
Given the blind box (p̌b , q̌) with parameters as defined earlier, suppose that c = 0. Since the
consumer purchases the blind box until they receive all n + 1 items, the expected purchase volume
is exactly m̌. Then, we can bound the profit under blind box selling as follows:

Π̌B = m̌ (1 − q̌)v · min{1, κ}/n + v + /m̌

1 min{1, κ}
≥ (1 − q̌)v · + v+
q̌ n
n+1
= (1 − q̌) · vα + v +
n
≥ vα + v + .

In the first inequality, we used the fact that m̌ ≥ 1q̌ , where the lower bound represents the expected
number of purchases before the consumer obtains the special item. In the second inequality, we
n
used the fact that 1 − q̌ ≥ n+1
. Suppose that α > (n + 1). Then, we can compare the above profit
to that under separate selling as follows:

Π̌B ≥ vα + v +

> v(n + 1) + v +

= Π̌∗S .

This completes our proof that for any α > n + 1 and c = 0, blind box selling is strictly optimal.
To complete the proof, define β̌ = 0 for α ≤ n + 1. For any α > n + 1, by continuity, there exists a
threshold β L > 0 such that blind box selling is more profitable than separate selling for all β < β̌.
To complete the proof, we need to compare the cutoffs with and without the lump-sum bonus,
i.e., β̌ and β L from Proposition 3. Note that from (D.15), we know that for any α, β, the optimal
profit under separate selling increases by exactly v + whereas the optimal profit under separate
34

selling increases by at most v + . Therefore, there cannot exist any instance—i.e., the value of β—
such that separate selling outperforms when there is no lump-sum bonus, but blind box selling is
optimal with the lump-sum bonus. This completes our proof.