ADBI Working Paper Series

DESIGNING CENTRAL BANK DIGITAL

CURRENCIES

Itai Agur, Anil Ari, and Giovanni

Dell’Ariccia

No. 1065
December 2019

Asian Development Bank Institute

 Itai Agur is a senior economist at the Research Department of the International Monetary
Fund (IMF). Anil Ari is an economist at the European Department of the IMF. Giovanni
Dell’Ariccia is deputy director of the Research Department of the IMF.
The views expressed in this paper are the views of the author and do not necessarily reflect
the views or policies of ADBI, ADB, its Board of Directors, or the governments they represent.
ADBI does not guarantee the accuracy of the data included in this paper and accepts no
responsibility for any consequences of their use. Terminology used may not necessarily be
consistent with ADB official terms.
Working papers are subject to formal revision and correction before they are finalized and
considered published.

The Working Paper series is a continuation of the formerly named Discussion Paper series; the
numbering of the papers continued without interruption or change. ADBI’s working papers reflect
initial ideas on a topic and are posted online for discussion. Some working papers may develop
into other forms of publication.

Suggested citation:

Agur, I., A. Ari, and G. Dell’Ariccia. 2019. Designing Central Bank Digital Currencies. ADBI
Working Paper 1065. Tokyo: Asian Development Bank Institute. Available: https://
www.adb.org/publications/designing-central-bank-digital-currencies
Please contact the authors for information about this paper.
Emails: iagur@imf.org, aari@imf.org, gdellariccia@imf.org

This paper is being prepared for the Carnegie-Rochester Conference Series on Public Policy,
“Central Banking in the 2020s and Beyond”. We would like to thank Todd Keister, Morten
Bech, Maria Soledad Martinez Peria, Tommaso Mancini-Griffoli, Marcello Miccoli, Dirk
Niepelt, Beat Weber, Baozhong Yang, Jacky So, Garth Baughman, and audiences at the
IMF, the Federal Reserve Board, the Bank of England, the European Central Bank, the Bank
of Israel, Cambridge University, the 12th Annual Paul Woolley Centre Conference, the 12th
Swiss Winter Conference on Financial Intermediation, the 19th Annual FDIC/JFSR Bank
Research Conference, the ONB/BIS/CEBRA Conference on Digital Currencies, Central
Banks and the Blockchain, the Atlanta Fed 2019 Conference on the Financial System of the
Future, the Cleveland Fed and OFR 2019 Financial Stability Conference, [the 2020 AEA
Annual Meetings], the ADBI Conference on Fintech, Social Finance, and Financial Stability,
the IMF’s 2nd Annual Macro-Finance Conference, and the RESMF-FRBIF Workshop on
Financial Cycles and Central Banking for helpful comments.

The views expressed are those of the authors only and do not represent the views of
the IMF, its Executive Board or IMF management.

Asian Development Bank Institute

Kasumigaseki Building, 8th Floor
3-2-5 Kasumigaseki, Chiyoda-ku
Tokyo 100-6008, Japan

Tel: +81-3-3593-5500
Fax: +81-3-3593-5571
URL: www.adbi.org
E-mail: info@adbi.org

© 2019 Asian Development Bank Institute

Abstract

We study the optimal design of a central bank digital currency (CBDC) in an

environment where agents sort into cash, CBDC, and bank deposits according to their
preferences over anonymity and security; and where network effects make the
convenience of a payment instrument depend on the number of its users. A CBDC can
be designed with attributes similar to cash or deposits, and can be interest bearing: a
CBDC that closely competes with deposits depresses bank credit and output, while a
cash-like CBDC may lead to the disappearance of cash. Then, the optimal CBDC
design trades off bank intermediation against the social value of maintaining diverse
payment instruments. When network effects matter, an interest-bearing CBDC
alleviates the central bank’s tradeoff.

Keywords: CBDC, fintech, digital currency, financial intermediation, network effects

JEL Classification: E41, E58, G2

1
1 Introduction
Payment systems and, more fundamentally, money are evolving rapidly. Developments in
digital networks and information technology and the increasing share of internet-based re-
tailing have created the demand and technological space for peer-to-peer digital transactions
that have the potential to radically change payment and …nancial intermediation systems.
Central banks have been pondering whether and how to adapt. Many are exploring the idea
of issuing central bank digital currency (CBDC) - a new type of …at money that expands
digital access to central bank reserves to the public at large, instead of restricting it to com-
mercial banks.1 A CBDC would combine the digital nature of deposits with the usability in
peer-to-peer transactions that is characteristic of cash. But would it also resemble deposits
by coming in the form of an account at the central bank, or would it come closer to cash,
materializing as a digital token? Would it pay interest rates like a bank deposit, or would its
nominal return be …xed at naught, like cash?2 In this paper, we build a theoretical frame-
work geared at analyzing the relation between the design of a CBDC, the demand for types
of money, and …nancial intermediation.
Swings in the usage of payment instruments become particularly disruptive in the pres-
ence of network e¤ects. For example, with a decline in the use of cash, banks may cut back on
ATMs or shops may refuse to accept cash, a process currently underway in Sweden (Sveriges
Riksbank, 2018b). Because of such network e¤ects, payment instruments may disappear
when their use falls below a critical threshold, and the successful introduction of a CBDC
could risk tipping the balance. Network e¤ects are a critical feature of the model in this
paper.
Our starting point is a (static) economy with banks, …rms and households. In this
economy, banks collect deposits, extend credit to …rms, and create social value in doing
so: …rms’projects are worth less if they cannot receive bank loans.3 Both banks and …rms
engage in perfect competition.
Households face a Hotelling linear city, where they aim to minimize the distance between
1 For an overview of ongoing CBDC initiatives, see Mancini-Griffoli et al. (2018), Bank for International
Settlements (2018) and Prasad (2018). In a survey of 63 central banks, one-third of them perceived CBDC as a
possibility in the medium term (Barontini and Holden, 2019). Notably, the central banks of the PRC, Norway,
Sweden, Ukraine, and Uruguay are actively investigating the possibility of introducing a CBDC for domestic retail
payments. The Sveriges Riksbank is expected to decide on the introduction of an eKrona in the near term, while the
central banks of Ukraine and Uruguay have run successful pilots (Bergara and Ponce, 2018; National Bank of Ukraine,
2019; Norges Bank, 2018; Sveriges Riksbank, 2018a).
2 See Mancini-Griffoli et al. (2018) for other design aspects of CBDCs, which are mostly of an operational nature,

such as the means to disseminate, secure and clear CBDCs.

3 We parameterize and vary the degree to which bank financing of firms provides efficiency gains. On the special

role of depository institutions in intermediation, see Diamond and Rajan (2001) and Donaldson et al.(2018), as well as
Merrouche and Nier (2012) for supporting empirical evidence.
1
the available forms of money and their preferences. In particular, households have heteroge-
neous preferences over anonymity and security in payments. We represent these preferences
by an interval with cash and deposits at opposite ends: cash provides anonymity in transac-
tions, while bank deposits are more secure.4 A CBDC can take any point on this interval,
depending on its design. For instance, a central bank could provide partial anonymity (e.g.,
towards third parties but not the authorities), set transaction limits below which anonymity
is retained, or make anonymity conditional, only to be lifted by a court order - possibilities
under consideration in central banks studies of CBDCs (Mancini-Gri¤oli et al., 2018).
Overall, taking into account the design of the CBDC, households optimally sort into dif-
ferent types of money according to three considerations: their (heterogeneous) preferences,
network e¤ects deriving from the relation between the convenience of using a payment in-
strument and the number of its users, and the interest rates o¤ered on deposits and possibly
on the CBDC.
Our framework provides novel and policy-relevant insights for welfare analysis and opti-
mal CBDC design. In our model, variety in payment instruments increases welfare because
of the heterogeneity in household preferences. The CBDC then has social value due to its
ability to blend features of cash and deposits. As emphasized by Lagarde (2018), there is a
potential demand for partially anonymous means of payment that can, for example, protect
consumers from the use of personal transactions data for credit assessments.5
At the same time, introducing a CBDC has welfare costs to the extent that it crowds out
demand for cash and deposits.6 Speci…cally, a cash-like CBDC can reduce the demand for
cash beyond the point where network e¤ects cause the disappearance of cash. But a deposit-
like CBDC causes an increase in deposit and loan rates, and a contraction in bank lending
to …rms. Because of relationship lending frictions, this decline in bank intermediation also
curtails investment and output.7
We show that the welfare-optimal CBDC design hinges on whether the CBDC is interest
bearing, and whether network e¤ects matter. When the CBDC is not interest bearing, its
4
Empirical research on the choice of payment instruments attributes a central role to heterogeneous
preferences (Wakamori and Welte, 2017). For empirical work measuring preferences for anonymity and the
potential demand for CBDC, see Athey et al. (2017), Borgonovo et al. (2018) and Masciandaro (2018).
5
This possibility is increasingly enabled by technological developments, as for instance discussed by Yao
(2018) in the Chinese context, and forms the basis for the microfoundations that we develop in Appendix D.
6
Nevertheless, a CBDC is certain to raise aggregate welfare in our framework, but only if it is optimally
designed. Moreover, even when aggregate welfare rises, there are distributional e¤ects, and some households
are worse o¤ due to CBDC availability. We analyze these distributional e¤ects in Section 3.3.
7
A central bank could attempt to mitigate the decline in bank lending by providing banks with cheap
liquidity to replace lost deposits. However, this may not be feasible for two reasons. First, banks’ability to
intermediate funds may depend on their reliance on deposits (see, e.g., Diamond and Rajan, 2001; Donaldson
et al., 2018). Second, this policy would permanently expose the central bank to credit risk.

2
similarity to cash becomes the sole design instrument. The more important the role of banks
in alleviating lending frictions, the more cash-like the optimal CBDC design becomes. But
network e¤ects twist the optimal design problem, as the variety of payment instruments that
households value becomes challenging to sustain. A nonlinear optimal design pattern then
emerges: increasing the value added of banks translates into a more cash-like optimal design
up to a point, after which CBDC design is constrained to preserve cash. This is so unless
bank-based intermediation is su¢ ciently precious that the CBDC is introduced with a design
that eliminates cash, and replaces it with a less than fully anonymous CBDC.
As long as network e¤ects do not constrain policy, the CBDC interest rate is best kept
at zero, because it brings about price distortions in the households’choice of payment in-
struments. Unlike the interest rates on bank deposits, there is no production underlying the
payment of interest on the CBDC, which is funded with a lump-sum tax. Hence, the CBDC
interest rate is a suboptimal tool compared to the design of CBDC payment attributes, which
optimally center on meeting some households’demand.
However, access to an adjustable CBDC interest rate makes a palpable di¤erence to the
central bank when network e¤ects come to the fore. If the introduction of a CBDC threat-
ens cash with extinction, a negative CBDC interest rate can compensate.8 Indeed, when
households care enough about payment instrument variety, an interest-bearing CBDC will
optimally always keep cash alive, while limiting the CBDC’s impact on bank intermediation.
This is a …nding of policy relevance, since all ongoing central bank CBDC initiatives center
on CBDCs that do not bear interest.
In several extensions, we investigate whether considerations other than network e¤ects
can cause optimal CBDC rates to diverge from zero. Alternate production functions make no
di¤erence. In contrast, bank market power in lending matters, as it entices the central bank
to make the CBDC compete harder with deposits, leading optimal CBDC rates to diverge
from zero, regardless of network e¤ects. The same is true in another extension, in which
households dislike anonymity in other households’ payments. That is, there are negative
externalities associated with anonymity in payments, because this may spur illicit activities.
Once more, giving the central bank an additional ball to juggle - here, counteracting the
negative externality - leads to the conclusion that welfare is strictly higher when the CBDC
is interest bearing. In the last extension, we provide a model where households choose
between cash and deposit-based payment services that are bundled with the provision of
other services (e.g., credit provision related to transaction data), to microfound a linear city
8
Beyond satisfying household preferences, the disappearance of cash may reduce economic activity when
a portion of the population is unable or unwilling to transact with digital payment methods because of digital
illiteracy or informality. See Chodorow-Reich et al. (2018) for an empirical assessment of such costs.

3
of payment preferences, and to highlight the potential demand for a CBDC whose attributes
straddle those of existing payment systems.
Our paper is closely related to a recent and growing literature on CBDCs. One strand
of this literature focuses on the impact of introducing a CBDC on the banking sector. An-
dolfatto (2018) and Chiu et al. (2019) develop models where a CBDC increases welfare by
reducing banks’ deposit market power, while Brunnermeier and Niepelt (2019), Kim and
Kwon (2019) and Miccoli (2019) model the relation between banking panics and the avail-
ability of a CBDC.9 In Keister and Sanches (2019), a CBDC contributes to e¢ ciency in
exchange at the expense of crowding out deposits.
Compared to this literature, our …rst contribution is to highlight a tradeo¤ between
preserving variety in payment instruments in the face of network e¤ects, and mitigating the
adverse e¤ects of a CBDC on …nancial intermediation. Our second contribution is to show
when and why this tradeo¤ is harder to overcome with a non interest-bearing CBDC than
with a CBDC that o¤ers an adjustable interest rate.10
Our paper also relates to the literature on payment systems as well as that on network
e¤ects. Our modeling of network e¤ects follows closely on the seminal work of Katz and
Shapiro (1985). While Katz and Shapiro (1985) study …rms’decisions to introduce mutually
compatible products, we focus on a social planner’s decision to introduce and design a new
payment instrument. In the literature on payment systems, the analysis and measurement of
network e¤ects centers on credit and debit card networks (Bounie et al., 2017; Chakravorti,
2010; Rochet and Tirole, 2006).11 Instead, we study the impact of a new form of money on
the demand for the existing payment instruments, and on welfare. Lastly, the value of variety
in payment instruments has a similarity to the value of product variety in international trade
(Krugman, 1979). However, our model does not build on an assumed love of variety, but
rather on a heterogeneity in preferences that is best served by variety.
The remainder of this paper is organized as follows. Section 2 presents the model. Section
3 analyzes the optimal design of a CBDC. Section 4 concludes. Extensions of the model can
be found in the appendices.
9
There is also a sizeable policy literature discussing the e¤ects on …nancial stability of a CBDC (see, e.g.,
Bech and Garratt, 2017; Fung and Halaburda, 2016; He et al., 2017; Kahn et al., 2019).
10
In our framework, CBDC interest rates represent any type of subsidy or cost associated with holding a
CBDC. For example, the pilot conducted by the central bank of Uruguay o¤ered subsidies to CBDC holders
(Bergara and Ponce, 2018). Moreover, we focus on the steady-state e¤ects of CBDC rates on …nancial
intermediation and cash use, rather than their implications for monetary policy over the business cycle. On
the relation between CBDC and monetary transmission, see Agarwal and Kimball (2015, 2019), Assenmacher
and Krogstrup (2018), Barrdear and Kumhof (2016), Bordo and Levin (2017), Bjerg (2017), Davoodalhosseini
(2018), Goodfriend (2016), Meaning et al. (2018), and Niepelt (2019).
11
The role of strategic coordination and adoption equilibria has also been considered in the literature on
cryptocurrencies (Biais et al., 2019, 2018; Bolt and Van Oordt, 2019).

4
2 A model of payment instruments
We consider a …nancial economy populated by households, banks, …rms, and a central bank
that aims to maximize welfare. Events unfold over two stages. In the …rst stage, the central
bank decides whether and in what form to introduce a CBDC. In the second stage, households
choose between holding cash, bank deposits and (if introduced) CBDC for their transactions,
and banks use deposits collected from households to extend loans to …rms, which in turn
produce a consumption good.
Along with the rate of interest o¤ered (if any), households value two attributes of a
payment instrument – anonymity and security – with heterogeneous preferences over their
relative importance. At the core of our model lies a tension between these two attributes,
because delinking transactions from personal identity leads to a loss of traceability that
creates risks for the holder. For example, while depository accounts are relatively safe and
traceable, cash is vulnerable to accidental loss and theft.12 It is precisely the fact that cash
can be lost without any claim that also makes it perfectly anonymous.
This intrinsic link between the degree of anonymity of a means of payment and the
di¢ culty of keeping it safe extends to the realm of digital money. A CBDC can only approach
the anonymity of cash if it takes the form of a token, such as a cryptocurrency, which is
accessible through user accounts that are not independently veri…ed, or a nameless payment
card that can be purchased at stores or online. These forms of CBDC would also su¤er from
the risks of loss and theft associated with cash, either physically (e.g., card loss) or digitally
(e.g., the untraceable loss of account information). At the other extreme, an account at the
central bank that can be opened only using o¢ cial identi…cation would mimic the security
and traceability of bank deposits. Unlike cash and deposits, however, CBDC can be designed
to blend intermediate amounts of anonymity and security.13 For example, anonymity may
be preserved vis-à-vis third parties only, and transactions can be recorded but not accessed
by the central bank unless a transaction size limit is breached and/or there is suspicion of
wrongdoing.
We formalize these considerations in an anonymity-security scale [0; 1], where higher
values denote a greater extent of anonymity and, equivalently, a lesser degree of security.
If we let xj denote the place of each money type j in the anonymity-security scale, cash
(denoted by c) is placed at the top of the scale xc = 1, deposits (denoted by d) are at the
12
We abstract from default risk on bank deposits, which is negligible in normal times due to deposit
insurance and implicit bailout guarantees.
13
While some legal jurisdictions allow for deposit accounts that o¤er a degree of anonymity, these accounts
are typically incompatible with payment services. Moreover, providing anonymity in deposits may undermine
their complementarity with relationship lending (see, e.g., Donaldson et al., 2018).

5
bottom xd = 0, and CBDC is placed at xcbdc = , where 2 [0; 1] is a design parameter
determined by the central bank. In addition, the central bank determines the (net) interest
rate o¤ered on the CBDC, rcbdc , which we allow to take any (positive, zero or negative) value.
The combination ( ; rcbdc ) thus describes CBDC design in our framework.
There are two important frictions in the model economy: relationship lending frictions
and network externalities in payment transactions. Relationship lending frictions take the
form of information asymmetries that bar households from lending directly to …rms, which
are endowed with positive net present value projects that require …nancing. Firms may then
either rely on intermediation by banks or liquidate their projects. The importance of bank
intermediation is proportionate to the gap between …rm productivity A and the liquidation
value of …rm projects . When (A ) is higher, a given decline in bank deposits and credit
leads to a sharper reduction in output and consumption.
Network externalities give rise to a disutility cost of relying on a payment instrument
that is not commonly used. We denote by j the disutility cost of using money type j and
adopt the functional form

j = max f0; g (sj )g 8j 2 fc; cbdc; dg (1)

g (1) < 0 ; g (0) > 0 ; g 0 (:) < 1

where sj stands for the share of households holding a money type j. The functional form
for j is restrictive in two ways. First, we implicitly assume that, for a given share of users
sj , network externalities are equivalent across di¤erent money types. Second, the maximum
operator and the restrictions on g (:) imply that network externalities only take hold when
the share of households using a money type falls short of a threshold. We …nd it convenient
to de…ne this threshold as s g 1 (0), where 0 < s < 1. Once network externalities come
into play, the restriction g 0 (:) < 1 ensures that they lead to a cascade: each household that
switches to a di¤erent money type incentivizes another to switch until that type of money
falls out of use. This setup helps preserve tractability by eliminating unstable equilibria and
ensuring that the equilibrium shares of money types that remain in use are not a¤ected by
network externalities.
Below, we explain the activities of the households, banks, …rms and the central bank in
further detail. We then proceed to characterize the equilibria and conduct welfare analysis.

6
2.1 Agents and their optimal strategies
2.1.1 Households

There is a unit continuum of households with preferences i 2 [0; 1] uniformly distributed

over the anonymity-security scale.14 Households start with identical endowments in terms
of (atomistic) shares in …rms and liquid funds, which are normalized to 1. Liquid funds
are stored in a money type j 2 fc; d; cbdcg and used to purchase the consumption good
at the end of the stage.15 Most importantly, households cannot attain their preference i
by mixing di¤erent forms of money in their transactions because a transaction is only as
anonymous as the least anonymous payment instrument used. In other words, anonymity is
undiversi…able.16
The household budget constraint is given by

Cj = 1 + rj T+ (2)

where Cj is consumption, represents dividends from …rm pro…ts, rj is the (net) interest
earned on money holdings, so that
8
>
< 0 if j = c
rj = rcbdc if j = cbdc (3)
>
:
rd if j = d

and T = rcbdc scbdc is a lump-sum tax used to fund interest rates on the CBDC (or equivalently
a lump-sum transfer to redistribute revenues when rcbdc < 0).17
The households’utility maximization problem can then be written as

max Ui (j) = Cj jxj ij j (4)

j2fc;d;cbdcg

subject to the budget constraint (2). Here, jxj ij represents the utility cost of selecting
a payment method that di¤ers from the household’s anonymity-security preference i while
14
We adopt a uniform distribution for the sake of tractability. Our qualitative results generalize to any
single peaked distribution with continuous support and su¢ cient weight in the tails to ensure that, absent
a CBDC, both deposits and cash are sustained as payment instruments.
15
We assume that all forms of money are traded at par.
16
This notion is further explored in Appendix D, which provides an example of how a linear-city setup, in
the manner of Hotelling (1929), can be microfounded for the case of payment preferences.
17
This can be interpreted as a zero-capital central bank: any revenue that the central bank makes is
immediately paid out to households, and any capital shortfall leads directly to a recapitalization through a
lump-sum tax.

7
 > 0 denotes the marginal utility of consumption relative to payment preferences.18 Network
e¤ects are captured by j , as de…ned by (1). The solution to the household’s problem yields
the following cut-o¤ conditions for a household with preferences i to choose

cash over CBDC : 1 i+ c <j ij rcbdc + cbdc (5)

cash over deposits : 1 i+ c <i rd + d (6)
CBDC over deposits : j ij rcbdc + cbdc <i rd + d (7)

2.1.2 Banks

Banks collect deposits d from households at net deposit rate rd and extend loans l to …rms
at net loan rate R, with the budget constraint

l=d (8)

The representative bank is risk neutral and a price taker in both deposit and loan markets,
so that its pro…t maximization problem yields the …rst-order condition

rd = R (9)

which simply equates interest rates on deposits and loans.19

2.1.3 Firms

Firms are perfectly competitive and begin with an identical endowment of productive projects
k0 which require …nancing to implement.20 The representative …rm uses bank loans l to
…nance a portion k = l of projects. Once implemented, these projects yield a payo¤ in terms
of consumption goods with the technology

k
Y = A k (10)
2
18
The manner in which we combine consumption with payment preferences bears similarity to the utility
function adopted in Gopinath and Stein (2018).
19
See Appendix C.3 for an extension where we allow for market power in the bank loan market.
20
We impose the restriction k0 > 1 to ensure that lending frictions always bind such that k < k0 .

8
where A > 1 denotes productivity.21 The remaining projects (k0 k) are liquidated at a
constant rate of return 2 (0; 1).22
The representative …rm’s pro…t maximization problem can then be written as

max Y + k0 k (1 + R) l (11)
k;l

subject to (10) and k = l. This yields the …rst-order condition

1+R=A l (12)

which can be interpreted as a downward sloping loan demand curve. Notably, a decline
in the liquidation value raises …rm demand for loans, re‡ecting the increased importance
of bank intermediation. More generally, we can refer to (A ) as ‘the value added from
bank intermediation,’since it captures the value generated by channeling funds to …rms with
productivity A, instead of having the …rms sell o¤ their projects at value . One can think of
this from a cross-country perspective, where some countries are more reliant on bank-based
intermediation and others less so, since in some countries relationship lending frictions are
easier to overcome by alternate (i.e., nonbank) means.

2.1.4 The central bank

The central bank aims to maximize social welfare, de…ned as the sum of household utilities
Z
W = Ui (j (i)) di (13)
i2[0;1]

where j (i) denotes the payment instrument selected by household i. In doing so, the central
bank decides whether to introduce a CBDC, and if introduced, its design characteristics
21
We adopt a quadratic functional form in the interest of tractability. Appendix C.1 considers a constant
returns to scale technology as an alternative. In a derivation available upon request, we also generalize the
k
quadratic technology to the form Y = A 2 k and show that results are robust to varying .
22
The liquidation value is also in terms of consumption goods. The liquidation of projects can be mi-
crofounded in a framework similar to Stein (2012) where projects are sold to outside buyers with a lower
marginal valuation. While we do not explicitly incorporate outside buyers into our model, doing so would
have no impact on welfare provided these buyers are non-residents and/or projects are priced at their op-
portunity cost to outside buyers. In the interest of tractability, we also assume that funds from liquidated
projects cannot be used for …nancing other projects. This could be due to a combination of information
asymmetries and timing. For example, the time required for outside buyers to verify and pay for a project
may exhaust the time for implementation by …rms.

9
( ; rcbdc ).23 If a CBDC is introduced, the central bank’s design problem is given by
Z
max Ui (j (i)) di (14)
2[0;1];rcbdc i2[0;1]

subject to a design constraint

scbdc s (15)

which ensures that there is su¢ cient uptake of the CBDC to overcome network e¤ects.24

2.2 Equilibrium and welfare

A competitive equilibrium where deposit, loan and capital markets clear, is given by

sd = d = l = k (16)

The interaction of network externalities with the three money types in our framework leads
to a rich set of equilibrium types, as shown in Table 1. For the sake of brevity, we introduce
the parameter restrictions

5 3 3 1
1 < (A ) ; ; s (17)
2 4 2 17

which allow us to focus our discussions on “well-behaved”equilibria that give rise to plausible
outcomes. For instance, when there is no CBDC, we should observe that deposits and cash
23
An implicit assumption in our model is that the central bank does not allow any agent to take a short
position in CBDC (i.e., the central bank does not grant CBDC credit to other parties). This precludes
arbitrage opportunities by entities without payment preferences, such as banks, which might prefer funding
themselves with CBDC rather than deposits. Based on CBDC studies currently underway at central banks,
we consider this a realistic assumption.
24
The design constraint subsumes two conditions, rcbdc (1 ) 1 and > (rd rcbdc ), which
respectively rule out the strict dominance of CBDC by cash and deposits (i.e., ensure that neither cash
nor deposits o¤er all households a strictly better utility than CBDC) as per (5) and (7). For example, a
completely cash-like CBDC ( = 1) that pays negative rates (rcbdc < 0) would violate the …rst condition,
such that all households have a strict preference for cash over CBDC. Because of network externalities, these
conditions are necessary, but not su¢ cient, for positive CBDC take-up.

10
are able to coexist, as they do in most countries.25;26 We also bring forward a number of
results based on optimal CBDC designs that are formally derived later in the paper. Lemma
1 shows that an optimally designed CBDC always increases welfare. Therefore, the central
bank always prefers to introduce a CBDC. Lemma 2 shows that the CBDC design constraint
does not bind, while deposits remain in use under an optimal CBDC design.

Lemma 1 Under optimal policies derived in Section 3, the introduction of an optimally-

designed CBDC always raises social welfare.

Proof. Provided in Appendix A.

Lemma 2 Under optimal policies derived in Section 3, the parameter restrictions in (17)
imply that sd s and scbdc s.

Proof. Provided in Appendix A.

As indicated by Table 1, the parameter restrictions and optimality results allow us to
narrow the set of possible equilibria to just two well-behaved equilibria: a ‘cash equilibrium’
(denoted by ce) where all three money types are in use, and a ‘cashless equilibrium’(denoted
by nce) where cash disappears and only deposits and CBDC remain in use.27

Cash equilibrium In the cash equilibrium, cash use remains high enough to prevent
network externalities from causing its disappearance. Using the properties of the uniform
distribution, (5)-(7) can be solved to obtain the shares of households holding each money
25
While our model is not quantitative in nature, empirical evidence suggests that network e¤ects only
begin to play a signi…cant role when the use of a payment instrument becomes very small, as represented
1
by s 17 . For instance, in Canada, cash is widely accepted although only about 10 percent of transactions
in value terms are conducted with cash (Engert et al., 2018). In contrast, in Sweden, where network e¤ects
on cash are becoming a source of concern, cash use stands near 1 percent of transactions value (Sveriges
Riksbank, 2017). We discuss the outcome when cash demand is too low to sustain cash, even absent the
introduction of CBDC, at the end of Section 3.1.
26
The restriction (A ) > 1 ensures that aggregate output (and hence consumption) increases in …nancial
intermediation in equilibrium. This follows directly from the derivative dY dk , which, given k 1, is always
positive for (A ) > 1.
27
The three equilibria referred to as never occurring under optimal policy are further discussed in Appendix
C.4, which considers outcomes of suboptimal CBDC design. The equilibria referred to as “impossible under
any policy”are ruled out by the parameter restrictions which imply that, when there is no CBDC, the lowest
7 3
possible shares of deposits and cash, respectively, are inf sd = 22 and inf sc = 22 , both of which are above s.
The derivations of these results are available upon request.

11
 Table 1: Possible equilibria and role in discussion
Equilibrium Role in discussion
Deposits & Cash & CBDC Referred to as ce
Deposits & CBDC Referred to as nce
Deposits & Cash Never occurs under optimal policy
CBDC & Cash Never occurs under optimal policy
CBDC only Never occurs under optimal policy
Cash only Impossible under any policy
Deposits only Impossible under any policy

type, which are

1 rcbdc
sce
c = (18)
2
(rd rcbdc ) +
sce
d = (19)
2
1 rd
sce
cbdc = + rcbdc (20)
2

Cashless equilibrium In the cashless equilibrium, network externalities lead to the

disappearance of cash such that snce
c = 0, in which case the shares of households holding
CBDC and deposits become

(rd rcbdc ) +
snce
d = (21)
2
(rd rcbdc ) +
snce
cbdc = 1 (22)
2

Observe from (19) and (21) that the expressions for the shares of deposit holders are same
in the two equilibrium types (sce nce
d = sd ). This is because, when there is a CBDC, deposits
do not directly compete with cash. By (16), the expressions for deposit (and loan) interest
rates, …rm production and aggregate consumption are also the same in both equilibrium
types and given by

2 (A 1
) ( rcbdc )
rd = (23)
2+
(A 1 rcbdc ) +
sd = (24)
2+
sd
Y = A sd (25)
Z 2
sd
Cj (i) di = 1 + k0 + A 1 sd (26)
i2[0;1] 2

12
where sd represents the share of deposits after substituting for rd . Notably, regardless of the
equilibrium type, the introduction of a CBDC that competes closely with deposits through
a deposit-like design (i.e., low ) and/or by o¤ering high interest rates rcbdc crowds out
bank deposits. Although banks partially o¤set this by o¤ering higher deposit rates rd , in
equilibrium this brings about a decline in bank intermediation, which also reduces …rm
production and aggregate consumption as per (25) and (26). The extent of the decline
in aggregate consumption depends on relationship lending frictions. When these frictions
are stronger (i.e., A is larger), a given decline in sd leads to a larger fall in aggregate
consumption. This is precisely why we refer to (A ) as the value added from bank
intermediation.
Lastly, it is important to note that network externalities lead to strategic complemen-
tarities in households’payment decisions, thus bringing about the possibility of multiplicity
between cash and cashless equilibria. However, both of these equilibrium types may also
arise due to fundamentals, and multiplicity does not lead to insights that are interesting, or
that we observe in reality. Therefore, we rule out multiplicity by assuming that it is always
resolved in favor of a cash equilibrium, which we consider to be similar to the pre-digital
currency economy. A cashless equilibrium then arises only when fundamentals are such that
the cash equilibrium is not self-con…rming, which is the case when the boundary condition

sce
c s (27)

is violated.28 Using (18), we can also write this condition in terms of CBDC design parame-
ters as
+ rcbdc 1 2s (28)

which indicates that a CBDC that competes strongly with cash through a cash-like design
(i.e., high ) and/or by o¤ering a su¢ ciently high interest rate rcbdc may eliminate cash and
give rise to a cashless equilibrium. We proceed with a discussion of the resulting comparative
statics.

2.2.1 Comparative statics

Figure 1 depicts the comparative statics of money shares in terms of CBDC design para-
meters ( ; rcbdc ). The unshaded part of Panel A shows that cash holdings decline and bank
deposits rise as the CBDC becomes more cash-like with higher . Notably, the share of
CBDC holders rises as the CBDC becomes more cash-like. This is because banks respond
28
Resolving multiplicity in favor of the cashless equilibrium shifts the boundary condition to + rcbdc >
1 2s g (0) without any qualitative impact on our analysis.

13
 Figure 1: Comparative statics of money shares

Note: Shares and equilibrium boundaries are drawn according to (18)-(28).

to reduced competition from CBDC by lowering deposit rates rd , whereas cash o¤ers no in-
terest. However, when CBDC becomes so cash-like that cash use declines below a threshold
s, network externalities lead to a cashless equilibrium, depicted by the shaded area at the
right end of Panel A. This leads to a jump in the use of the CBDC, because cash holders
switch to it. As the CBDC becomes even more cash-like, households with preferences on the
margin between CBDC and bank deposits switch to the latter, thereby raising deposits and
reducing CBDC use. Panel B shows that a higher CBDC rate rcbdc reduces the shares of
both cash and deposits, while raising that of CBDC. However, as banks raise deposit rates
in response to higher CBDC rates, deposits decline less than cash. A su¢ ciently high rcbdc
leads to a cashless equilibrium, which is depicted by the shaded area at the right end of the
panel. Furthermore, the striped areas at the left ends of Panel A and B indicate domains
where the CBDC design constraint is violated, and CBDC falls out of use.
Finally, note that we can analyze the impact of introducing a CBDC with a given design
( ; rcbdc ) by comparing it with the equilibrium under a CBDC that is completely cash-like
( = 1) and o¤ers no interest (rcbdc = 0). With this design, the CBDC is identical to cash
and becomes completely innocuous in our model. In other words, introducing a CBDC
is like moving from 1 to a lower value and/or changing rcbdc away from 0. Once the
CBDC moves away from cash mimicry, the combination ( ; rcbdc ) needs to be competitive
enough if the CBDC is to have a positive uptake (as represented by the CBDC design
constraint). Moreover, as compared to a world without CBDC, any such positive uptake of
CBDC necessarily derives from both cash and deposits in our model. That is, introducing
a CBDC always brings about some decline in banks’deposit base and consequently in bank
intermediation to …rms, although the extent of this e¤ect depends on how closely the CBDC
competes with deposits. To the extent that the CBDC reduces bank intermediation, it also
causes a decline in aggregate output and consumption.

14
2.2.2 Welfare analysis

In equilibrium, social welfare can be split into two terms

Z
W = Cj (i) xj (i) i di (29)
i2[0;1]

The …rst term represents aggregate consumption as given by (26). Because of the role of
banks in providing …rm …nancing, aggregate consumption relates closely to the extent of
bank intermediation, and therefore to bank deposits sd . The second term xj (i) i repre-
sents welfare losses due to the distance between households’payment preferences and their
preferred instrument. This term embodies the social value of variety in payment instruments,
as increased variety provides heterogeneous households with greater opportunity to minimize
the distance to their payment preferences.
How these two terms a¤ect welfare, particularly in relation to CBDC design instruments
and rcbdc , becomes clearer in the closed form expressions for welfare provided in Lemma 3.

Lemma 3 Social welfare in the cash and cashless equilibria are respectively given by
1 2 2 2
ce 4 A 2
+ 4 (1
) 3 (4 + ) rcbdc + !1
W ( ; rcbdc ) = (30)
4 (2 + )
2 (A ) + (4 3 ) 2 2 2 2
rcbdc + !2
W nce ( ; rcbdc ) = (31)
2 (2 + )

where ! 1 and ! 2 are collections of constants

!1 7 + 4k0 (2 + ) + 6 + 2A2 4A (1 + ) + 2 (2 + ) 2 (32)

2
!2 3 + A2 2A (1 + ) + (2 + 2k0 + ) + (3 + 4k0 ) 2 (33)

Proof. Provided in Appendix A.

To provide intuition, we focus on the breakdown of the terms in the …rst expression (30)
which pertains to welfare under the cash equilibrium, although a similar breakdown applies
to (31) and the cashless equilibrium as well.
1
The …rst term 4 A 2
captures the relation between the value of bank inter-
mediation (A ) and CBDC design characteristic . When (A ) is greater, aggregate
consumption and social welfare depend more strongly on the intermediation derived from
bank deposits. This, in turn, calls for a CBDC that competes less intensely with banks,
through a more cash-like CBDC design (i.e., high ).
The second and third terms, 4 (1 ) 3 2 , pertain to the relation between variety

15
in payment instruments and CBDC design. Notably, these terms are linear-quadratic in ,
meaning they have an interior maximum, and this captures the fact that variety in payment
instruments is best served by an intermediate CBDC design that is di¤erentiated from both
cash and deposits.
The …nal term (4 + ) 2 rcbdc 2
, is negative and quadratic in rcbdc , with the implication
that a non interest-bearing CBDC maximizes welfare within a given equilibrium type. This
is because an interest-bearing CBDC distorts households’ payment choices away from the
instrument closest to their payment preferences. As this a¤ects households on the margin
between CBDC and another instrument, these distortions rise at an increasing rate as rcbdc
moves further from zero. Moreover, unlike deposit rates, payment choice distortions caused
by rcbdc are not o¤set by a contribution to …nancial intermediation. While deposit rates
re‡ect the surplus from increased bank lending and the resulting rise in …rm production,
CBDC rates are funded by lump-sum transfers that have no direct productive impact.
Lastly, it is important to note that by focusing on welfare within a given type of equilib-
rium, our discussion has so far abstracted from network e¤ects and the associated equilibrium
determination condition (28). Accounting for these e¤ects, social welfare is given by
(
W ce ( ; rcbdc ) if + rcbdc 1 2s
W ( ; rcbdc ) = (34)
W nce ( ; rcbdc ) otherwise

and CBDC design parameters ( ; rcbdc ) may lead to a switch from one equilibrium type to
another. In the next section, we shed more light on optimal CBDC design, including the
role of the CBDC interest rate, in the presence of network e¤ects.

3 Optimal CBDC design

In this section, we analyze the optimal CBDC design which maximizes social welfare, pro-
ceeding in two steps. First, we investigate the optimal design of a non interest-bearing
CBDC, and then how optimization of an interest-bearing CBDC di¤ers from this.
This two-step approach is more than a matter of analytical convenience. In practice, most
central banks appear to be constraining themselves to non interest-bearing CBDCs. This may
be due to political economy considerations, such as concerns about a central bank liability
that is held by the general public and can be made to pay negative interest. Alternatively,
there could be legal hurdles to an interest-bearing CBDC. For instance, the need to tax
positive interest earnings may interfere with a desire to o¤er a degree of anonymity on
the CBDC in certain jurisdictions (Engert and Fung, 2017). Using our framework, we can

16
analyze the economic rami…cations of constraining the CBDC to a non interest-bearing form.

3.1 Non interest-bearing CBDC

When a CBDC is required to be non interest-bearing (rcbdc = 0), the design optimization
problem is given by
max fW ( ; 0)g (35)
2[0;1]

where W ( ; 0) is de…ned according to (34) and Lemma 3.29 Accordingly, the optimal CBDC
design in the cash and cashless equilibria can be solved to

ce 2 + (2 (A ) 1)
= (36)
4+3
nce 2 + (A )
= (37)
3+2

where the parameter restrictions (17) ensure that ce and nce are well-de…ned on [0; 1].30
The implication is that optimal policy leads to an interior CBDC design where CBDC’s
cash-likeness equates the marginal bene…t on bank intermediation with marginal losses to
payments system variety from moving too close to cash. Furthermore, combining these
expressions with (30) and (31) shows that as long as network e¤ects play no role, the cash
equilibrium welfare dominates the cashless equilibrium, so that

ce ce nce nce ( (A 2) 1)2

W ( ; 0) W ( ; 0) = >0 (38)
2 (3 + 2 ) (4 + 3 )

where the inequality is strict because (A 2) 6= 1 given our parameter assumptions.

This dominance of the cash equilibrium derives from the fact that payment instrument
variety creates social value for heterogeneous households. If sustaining that variety is costless,
welfare is best served by having all three payment instruments in use.
However, network e¤ects impose costs on maintaining cash use by “constraining” the
CBDC design optimization. In particular, when applied to a non interest-bearing CBDC,
the condition (28) can be written as an upper bound on the degree to which the CBDC
approximates cash,
1 2s (39)

above which the economy moves to a cashless equilibrium. That is, whenever ce > , the
choice is no longer between W ce ( ce ; 0) and W nce ( nce ; 0), since the latter is no longer im-
29
The design constraint (15) is slack under optimal policies as per Lemma 2.
30
Given (17), these optimal policies can range between ce 2 17 ; 17 and nce 2
7 16 7 23
12 ; 24 .

17
 Figure 2: Optimal non interest-bearing CBDC design

plementable: A CBDC with design ce is too cash-like and would reduce cash use below the
threshold s where network e¤ects cause cash demand to spiral down to zero. Instead, opti-
mization centers on W ce ; 0 versus W nce ( nce ; 0), that is, preserving the cash equilibrium
under a CBDC design constrained at versus allowing cash to vanish and having uncon-
strained optimal policy with only CBDC and deposits in existence. De…ning as optimal
policy when taking into account the constraint imposed by network e¤ects, we obtain Figure
2.31
Figure 2 brings together several key aspects of our model.32 First, when network e¤ects
do not constrain policy, the optimal similarity of the CBDC to cash depends on the extent
to which banks have an advantage at alleviating …nancial frictions. On the one hand, lo-
cating the CBDC “centrally” relative to the attraction points of deposits and cash serves
the payment needs of households with diverse preferences. On the other hand, when bank
intermediation has more value, the CBDC is optimally made more cash-like, so as to limit its
adverse impact on banks’deposit base, and thereby on aggregate output and consumption.
Second, as the value of bank intermediation (A ) rises, a threshold A is eventually
reached, beyond which optimal design freezes in relation to (A ). This is because optimal
policy prevents the disappearance of cash in order to protect payment instrument variety.
As long as the welfare gains from this variety outweigh the welfare costs from lost bank inter-
31
This is formally derived in Proposition 1 below.
32
In addition to optimal policy derived in the Proof of Proposition 1, the exact shape of Figure 2 relies on
@ ce @ nce
two more properties from (36) and (37): …rst, nce > ce ; second, @(A ) < @(A ) < 0 and therefore the
slope of nce is ‡atter.

18
mediation, optimal policy maintains all three payment instruments, rather than tipping cash
over the disappearance point induced by network e¤ects. However, when preserving bank
intermediation becomes the dominant concern (namely when A exceeds the threshold
A ), optimal policy forgoes variety, allowing cash to disappear, in exchange for a larger
deposit base for the banks.
Third, once cash vanishes, the CBDC bears the brunt of servicing former cash users,
and therefore optimally moves further towards cash than it would have if all three forms of
money were still in existence. In Figure 2, this is seen from the portion of the blue line to
the right of A , which is above the dashed gray line.
This last portion of the blue line also demonstrates the outcome when fundamentals are
such that cash disappears prior to the introduction of a CBDC. If the CBDC is introduced
after cash has disappeared, optimal policy is simply depicted by extending the portion of
the blue line to right of A all the way to the vertical axis. In this case, network e¤ects
no longer play a role since cash would already be in disuse. As such, in a country that starts
o¤ cashless, optimal CBDC policy is quite straightforward, and the CBDC is always more
cash-like than when cash exists.

3.2 Interest-bearing CBDC

This section considers an interest-bearing CBDC, where the CBDC rate can be varied as
desired to maximize welfare. Unconstrained n optimal policy is now giveno by the solution to
the system of two …rst-order conditions, @W ( @;rcbdc ) = 0; @W@r( cbdc
;rcbdc )
= 0 , which yields the
ce nce
same expressions for and as in (36) and (37) and for CBDC rates

ce nce
rcbdc = rcbdc =0 (40)

with the implication that the optimal CBDC rate is always zero in the absence of network
e¤ects. This outcome is consistent with the discussion in Section 2.2.2, which suggest that
the CBDC interest rate is a suboptimal tool compared to . As such, our model indicates
that in a world without network e¤ects, central banks would be right to focus their attention
on the issuance of non interest-bearing CBDCs.33
However, as discussed in Section 3.1, such an optimal policy pro…le is not always imple-
mentable, due to network e¤ects. For an interest-bearing CBDC, the equilibrium determi-
nation condition (28) a¤ects both and rcbdc . When this condition binds, and the central
33
Appendix C investigates the robustness of this key result. We …nd that the optimality of zero CBDC
rates (absent network e¤ects) is robust to the speci…cation of the production function. However, when
banks have market power (Appendix C.3), or when anonymous payment instruments create negative social
externalities (Appendix C.2), the optimal CBDC rate can deviate from zero.

19
 Figure 3: Optimal interest-bearing CBDC design

bank chooses to satisfy it in order to preserve cash, optimal CBDC design becomes

e = 2 + (2 (A ) 1 (4 + ) (2s 1))
(41)
4 + (1 + ) (3 + )
((A ) + 3s 2) + 4s 1
recbdc = 2 (42)
4 + (1 + ) (3 + )

Hence, when network e¤ects come into play, the optimal CBDC rate diverges from zero.
Indeed, under the parameter restrictions in (17), the optimal CBDC rate always turns neg-
ative. This in turn allows CBDC design to become more cash-like than in the non-interest
bearing case (e > ), since the value of bank intermediation (A ) rises.
Note that the constrained non interest-bearing optimal policy ( ; rcbdc ) = ; 0 is within
the feasible set of policies delineated by (28), but is found to be sub-optimal. Therefore,
access to a second policy tool in the CBDC interest rate strictly raises welfare when network
e¤ects bind. Proposition 1 records our key results on optimal CBDC design, which are
depicted in Figure 3.

Proposition 1 There is a cuto¤ 2 34 ; 32 , such that when < , cash never vanishes
under an optimally designed interest-bearing CBDC, where optimal design is given by
(
e; recbdc if ce ce
+ rcbdc <1 2s
(43)
ce ce
( ; rcbdc ) otherwise

For > , cash can vanish when the value of bank intermediation exceeds the threshold
^ . However, this threshold is higher when the CBDC bears interest than when it does
A
^>A
not, that is, A

20
Proof. Provided in Appendix A.
When the relative weight of payment preferences in household utility is large enough
( < ), the presence of a variable CBDC interest rate as a second tool fundamentally alters
the outcomes under optimal policy, as compared to a non interest-bearing CBDC, depicted
in Figure 2. With a non interest-bearing CBDC, the only means to safeguard deposits is to
make the CBDC eat into cash demand. But with a variable CBDC interest rate, optimal
policy simultaneously reaps the welfare bene…ts of sustaining variety in payment instruments
and limits bank disintermediation. In particular, when network e¤ects bind, optimal policy
combines a (more) cash-like CBDC with a negative CBDC interest rate, thereby avoiding
adverse network e¤ects on cash use by making the CBDC less attractive, while simultaneously
limiting the CBDC’s impact on …nancial intermediation and production. This optimal policy
is portrayed in the unshaded part of Figure 3. For < , the shaded part of this …gure is
never reached, which means that cash never vanishes under optimal policy.
However, the deeper the CBDC interest rate moves into negative territory, the larger its
costs in terms of payment choice distortions become. If the weight on payment preferences
is relatively small ( > ), then a point is reached where the value of bank intermediation
(A ) is large enough that welfare is best served by letting go of cash. This case, portrayed
by the shaded area in Figure 3, is similar to the jump seen in Figure 2: optimal policy switches
to ( nce ; rcbdc
nce
), which implies a more cash-like CBDC to better accommodate the preferences
of previous cash users, and a return to zero CBDC rates. Nevertheless, even when > , the
availability of CBDC interest rates serves a purpose. In particular, raising (A ) from lower
to higher values, the possibility of varying the CBDC interest rates "delays" the jump to a
cashless equilibrium where households lose access to three di¤erentiated means of payment
^>A
(i.e., A )

3.3 Distributional e¤ects

So far, our welfare analysis has centered on aggregate welfare, which represents the total
utility of all households. Introducing an optimally designed CBDC always raises aggregate
welfare in our framework, but this is far from a Pareto improvement: some households gain
while others lose. Figure 4 shows the welfare impact of introducing a CBDC across the
distribution of household preferences i 2 [0; 1].34 The blue line depicts the impact of a non
interest-bearing CBDC.
To begin with, households with payment preferences closest to deposits (i.e., low i) remain
as deposit users after the introduction of a CBDC. These households are impacted by the
34
See Appendix B for the underlying derivations.

21
 Figure 4: Distributional e¤ects of CBDC

Note: See Appendix B for the underlying derivations.

introduction of a CBDC through its negative e¤ects on …nancial intermediation, as well

as its positive e¤ects on bank deposit rates. On the one hand, the decline in …nancial
intermediation reduces total production and therefore pro…t transfers, , from …rms. On the
other hand, CBDC competition with bank deposits drives up deposit rates, rd . Overall, the
latter e¤ect dominates and the introduction of a CBDC raises the consumption, and hence
the welfare, of all deposit users.
At the other end of the spectrum, households with a strong preference for anonymity
(i.e., high i) continue to use cash. CBDC impacts the welfare of these households through
consumption. Since cash does not pay interest, the decline in …rm pro…ts brings about
a decline in consumption and welfare for these households. If instead the CBDC instigates
network e¤ects on cash and drives it out of use, these households su¤er from a further decline
in welfare due to the loss of their preferred payment instrument. The extent of their welfare
loss then becomes proportionate to their preference for anonymity, as depicted by the dashed
blue line.
Households that switch from deposits to CBDC experience a net welfare gain from the
introduction of the CBDC, since otherwise they would have continued to use deposits. In
other words, by virtue of optimality, these households only switch to CBDC if the gains in
terms of payment preferences outweighs the loss of interest payments rd . The household with
preference i = experiences the greatest increase in welfare. For households that marginally
prefer CBDC to cash, the net welfare e¤ect is negative, since CBDC holders also su¤er from
a fall in consumption due to reduced …rm pro…ts.
Overall, we can de…ne a boundary household, i, such that households i 2 0; i gain from

22
the introduction of a CBDC and households i 2 i; 1 lose, where the gains of the former
group more than o¤set the losses of the latter in the aggregate. The fact that depositors
emerge as winners and cash holders as losers hints at a potentially regressive impact of a
CBDC. In our analysis, all households have identical endowments. In practice, however,
households that primarily conduct their payments with cash tend to have lower income,
while higher income households more often rely on deposit-based payments.
The red line in Figure 4 shows the welfare impact of an interest-bearing CBDC with
slightly negative CBDC rates as per the optimal design prescribed in Section 3.2. Three
factors determine the impact of negative CBDC rates here. First, the revenues from nega-
tive CBDC rates are transferred lump-sum to all households, which e¤ectively redistributes
welfare gains from CBDC users to cash and deposit users. Second, since negative CBDC
rates increase deposits and …nancial intermediation, …rm pro…ts rise, which bene…ts all
households, while deposit rates rd decrease, hurting depositors. However, the second e¤ect
is dominated by the …rst, in that overall CBDC users lose out and deposit and cash users
gain from the CBDC rate cut. Third, when CBDC rates prevent cash from going out of
use, they stave o¤ large welfare losses for cash holders from the loss of a preferred payment
instrument.

4 Conclusion
As central banks across the world weigh the introduction of a digital currency, the impli-
cations of a CBDC for money demand and …nancial intermediation are coming to the fore.
This paper relates the e¤ects on cash, deposits and bank intermediation to two key design
choices involved in developing a CBDC: the degree to which the CBDC resembles cash, and
whether it bears interest. In our framework, the social value of the CBDC comes from the
fact that it can bring some of the anonymity of cash into the digital realm. The demand
for digital payment privacy is already a major issue in some jurisdictions, and is likely to
gain increased prominence globally with the spread of …ntech and companies’ability to parse
transactions data for their own gain.
The CBDCs currently under consideration are mostly of a non interest-bearing type.
Analyzing the optimal design of a non interest-bearing CBDC exposes a challenging welfare
tradeo¤ for the central bank. On the one hand, a cash-like CBDC risks reducing the demand
for cash below the critical mass where ATMs become sparser and fewer shops accept cash,
placing at risk the variety of payment instruments that is valuable to households with diverse
needs. On the other hand, if the central bank makes a CBDC more similar to deposits, the
banks’deposit base can come under threat, with negative implications for credit provision

23
and output, especially if banks have a signi…cant role in alleviating lending frictions.
Overall, in an economy where the banks’role is limited, a CBDC is best designed in a
manner that is as distinct from existing payment instruments as possible. Greater focus on
preserving bank intermediation instead drives the optimal design of a CBDC to be more
cash-like, but only up to a point: concerns that cash may fall prey to network e¤ects gives
the central bank cause to limit the extent to which CBDC competes against cash. Only
when conserving banks’deposit base becomes the overarching concern does the central bank
give up on cash, and optimal policy then jumps towards a more cash-like CBDC.
When network e¤ects matter, an interest-bearing CBDC helps the central bank alleviate
these tradeo¤s. Moving the CBDC interest rate away from zero causes welfare losses as
it creates price distortions in households’ choice between payment instruments. As long
as network e¤ects do not hold sway, the central bank thus shies away from varying the
CBDC interest rate. Therefore, in a world where network e¤ects have no material impact,
nothing is lost by limiting CBDC design to non interest-bearing CBDCs. However, when
network e¤ects pose a threat to the variety of payment instruments, an interest-bearing
CBDC becomes optimal. Notably, provided households care enough about payment variety,
the CBDC interest rate can be used to ensure that cash remains in use. That is, an optimally
designed interest-bearing CBDC meets the aims of safeguarding bank intermediation and
protecting the trio of payment instruments against network e¤ects, irrespective of the role
of …nancial frictions in the economy.
This …nding provides an economic counterweight to the political economy considerations
that may otherwise drive central banks to opt for a non interest-bearing CBDC, such as
concerns about the possibility of negative rates on publicly accessible central bank liabilities.
At this early stage, when CBDCs are still in the laboratory, central banks may want to at
least keep an eye on the inclusion of an adjustable CBDC interest rate, weighing its bene…ts
against possible political economy costs.

24
References
Agarwal, R. and Kimball, M. (2015). Breaking Through the Zero Lower Bound. IMF
Working Papers 15/224, International Monetary Fund.

Agarwal, R. and Kimball, M. (2019). Enabling Deep Negative Rates to Fight Recessions: A
Guide. IMF Working Papers 19/84, International Monetary Fund.

Andolfatto, D. (2018). Assessing the Impact of Central Bank Digital Currency on Private
Banks. Working Papers 2018-25, Federal Reserve Bank of St. Louis.

Assenmacher, K. and Krogstrup, S. (2018). Monetary Policy with Negative Interest Rates:
Decoupling Cash from Electronic Money. IMF Working Papers 18/191, International
Monetary Fund.

Athey, S., Catalini, C., and Tucker, C. (2017). The Digital Privacy Paradox: Small Money,
Small Costs, Small Talk. NBER Working Papers 23488, National Bureau of Economic
Research.

Bank for International Settlements (2018). Central Bank Digital Currencies. Technical
report, Basel Committee on Payments and Market Infrastructures.

Barontini, C. and Holden, H. (2019). Proceeding with Caution - a Survey on Central Bank
Digital Currency. BIS Papers 101, Bank for International Settlements.

Barrdear, J. and Kumhof, M. (2016). The Macroeconomics of Central Bank Issued Digital
Currencies. Bank of England working papers 605, Bank of England.

Bech, M. L. and Garratt, R. (2017). Central Bank Cryptocurrencies. BIS Quarterly Re-
view Q3, Bank for International Settlements.

Bergara, M. and Ponce, J. (2018). Central Bank Digital Currencies: the Uruguayan e-
Peso Case. In Masciandaro, D. and Gnan, E., editors, Do We Need Central Bank Digital
Currency? Economics, Technology and Institutions. SUERF Conference Volume.

Biais, B., Bisière, C., Bouvard, M., and Casamatta, C. (2019). The Blockchain Folk Theorem.
Review of Financial Studies, 32(5):1662–1715.

Biais, B., Bisière, C., Bouvard, M., Casamatta, C., and Menkveld, A. J. (2018). Equilibrium
Bitcoin Pricing. TSE Working Papers 18-973, Toulouse School of Economics (TSE).

25
Bjerg, O. (2017). Designing New Money - The Policy Trilemma of Central Bank Digital
Currency. CBS Working Paper, June 2017.

Bolt, W. and Van Oordt, M. R. (2019). On the Value of Virtual Currencies. Journal of
Money, Credit and Banking, forthcoming.

Bordo, M. D. and Levin, A. T. (2017). Central Bank Digital Currency and the Future of
Monetary Policy. NBER Working Papers 23711, National Bureau of Economic Research.

Borgonovo, E., Caselli, S., Cillo, A., and Masciandaro, D. (2018). Between Cash, Deposit
and Bitcoin: Would We Like a Central Bank Digital Currency? Money Demand and
Experimental Economics. BAFFI CAREFIN Working Papers 1875.

Bounie, D., François, A., and Van Hove, L. (2017). Consumer Payment Preferences, Network
Externalities, and Merchant Card Acceptance: An Empirical Investigation. Review of
Industrial Organization, 51(3):257–290.

Brunnermeier, M. K. and Niepelt, D. (2019). On the Equivalence of Private and Public

Money. NBER Working Papers 25877, National Bureau of Economic Research.

Chakravorti, S. (2010). Externalities in Payment Card Networks: Theory and Evidence.

Review of Network Economics, 9(2):1–28.

Chiu, J., Davoodalhosseini, M., Jiang, J. H., and Zhu, Y. (2019). Central Bank Digital
Currency and Banking. Sta¤ Working Papers 19-20, Bank of Canada.

Chodorow-Reich, G., Gopinath, G., Mishra, P., and Narayanan, A. (2018). Cash and the
Economy: Evidence from India’s Demonetization. NBER Working Papers 25370, National
Bureau of Economic Research.

Davoodalhosseini, M. (2018). Central Bank Digital Currency and Monetary Policy. Sta¤
Working Papers 18-36, Bank of Canada.

Diamond, D. W. and Rajan, R. G. (2001). Liquidity Risk, Liquidity Creation, and Financial
Fragility: A Theory of Banking. Journal of Political Economy, 109(2):287–327.

Donaldson, J. R., Piacentino, G., and Thakor, A. (2018). Warehouse Banking. Journal of
Financial Economics, 129(2):250 –267.

Engert, W. and Fung, B. (2017). Central Bank Digital Currency: Motivations and Implica-
tions. Discussion Papers 17-16, Bank of Canada.

26
Engert, W., Fung, B., and Hendry, S. (2018). Is a Cashless Society Problematic? Discussion
Papers 18-12, Bank of Canada.

Fung, B. and Halaburda, H. (2016). Central Bank Digital Currencies: A Framework for
Assessing Why and How. Discussion Papers 16-22, Bank of Canada.

Garratt, R. and van Oordt, M. (2019). Privacy as a Public Good: A Case for Electronic
Cash. Sta¤ Working Papers 19-24, Bank of Canada.

Goodfriend, M. (2016). The Case for Unencumbering Interest Rate Policy at the Zero Lower
Bound. Paper presented at the August 2016, Jackson Hole Conference.

Gopinath, G. and Stein, J. C. (2018). Banking, Trade, and the Making of a Dominant
Currency. Working Paper 24485, National Bureau of Economic Research.

He, D., Leckow, R. B., Haksar, V., Mancini-Gri¤oli, T., Jenkinson, N., Kashima, M.,
Khiaonarong, T., Rochon, C., and Tourpe, H. (2017). Fintech and Financial Services;
Initial Considerations. IMF Sta¤ Discussion Notes 17/05, International Monetary Fund.

Hotelling, H. (1929). Stability in Competition. Economic Journal, 39(153):41–57.

Kahn, C. M., Rivadeneyra, F., and Wong, T.-N. (2019). Should the Central Bank Issue
E-Money? Working Papers 2019-3, Federal Reserve Bank of St. Louis.

Katz, M. L. and Shapiro, C. (1985). Network Externalities, Competition, and Compatibility.

The American Economic Review, 75(3):424–440.

Keister, T. and Sanches, D. R. (2019). Should Central Banks Issue Digital Currency? Work-
ing Papers 19-26, Federal Reserve Bank of Philadelphia.

Kim, Y. S. and Kwon, O. (2019). Central Bank Digital Currency and Financial Stability.
Working Papers 2019-6, Economic Research Institute, Bank of Korea.

Krugman, P. (1979). Increasing Returns, Monopolistic Competition, and International

Trade. Journal of International Economics, 9(4):469–479.

Lagarde, C. (2018). Winds of Change: The Case for a New Digital Currency. Remarks at
the Singapore Fintech Festival.

Mancini-Gri¤oli, T., Martinez Peria, M. S., Agur, I., Ari, A., Ki¤, J., Popescu, A., and Ro-
chon, C. (2018). Casting Light on Central Bank Digital Currencies. IMF Sta¤ Discussion
Notes 18/08, International Monetary Fund.

27
Masciandaro, D. (2018). The Demand for a Central Bank Digital Currency: Liquidity, Return
and Anonymity. In Masciandaro, D. and Gnan, E., editors, Do We Need Central Bank
Digital Currency? Economics, Technology and Institutions. SUERF Conference Volume.

McAndrews, J. (2017). The Case for Cash. ADBI Working Papers 679, Asian Development
Bank Institute.

Meaning, J., Dyson, B., Barker, J., and Clayton, E. (2018). Broadening Narrow Money:
Monetary Policy with a Central Bank Digital Currency. Bank of England working papers
724, Bank of England.

Merrouche, O. and Nier, E. (2012). Payment Systems, Inside Money and Financial Interme-
diation. Journal of Financial Intermediation, 21(3):359–382.

Miccoli, M. (2019). Central Bank Digital Currencies, Bank Runs and Welfare. mimeo.

National Bank of Ukraine (2019). Analytical Report on the E-hryvnia Pilot Project. Tech-
nical report, National Bank of Ukraine.

Niepelt, D. (2019). Reserves for All? Central Bank Digital Currency, Deposits, and their
(Non)-Equivalence. International Journal of Central Banking, forthcoming.

Norges Bank (2018). Central Bank Digital Currencies. Norges Bank Papers No 1/2018.

Prasad, E. (2018). Central Banking in the Digital Age: Stock-Taking and Preliminary
Thoughts. Discussion paper, Hutchins Center on Fiscal and Monetary Policy at Brookings.

Rochet, J.-C. and Tirole, J. (2006). Externalities and Regulation in Card Payment Systems.
Review of Network Economics, 5(1):1–14.

Rogo¤, K. (2016). The Curse of Cash. Princeton University Press, 1st edition.

Stein, J. C. (2012). Monetary Policy as Financial Stability Regulation. The Quarterly

Journal of Economics, 127(1):57.

Sveriges Riksbank (2017). Central Bank Digital Currencies. Technical report, Sveriges
Riksbank.

Sveriges Riksbank (2018a). Special Issue on the e-Krona. Sveriges Riksbank Economic
Review, Sveriges Riksbank.

Sveriges Riksbank (2018b). The Riksbank’s E-Krona Project. Technical report, Sveriges
Riksbank.

28
Wakamori, N. and Welte, A. (2017). Why Do Shoppers Use Cash? Evidence from Shopping
Diary Data. Journal of Money, Credit and Banking, 49(1):115–169.

Wright, R., Tekin, E., Topalli, V., McClellan, C., Dickinson, T., and Rosenfeld, R. (2017).
Less Cash, Less Crime: Evidence from the Electronic Bene…t Transfer Program. Journal
of Law and Economics, 60(2):361–383.

Yao, Q. (2018). A Systematic Framework to Understand Central Bank Digital Currency.

Science China Information Sciences, 61(3):033101.

29
5 Appendix
A Proofs
Proof of Lemma 1. A CBDC can be designed in a manner that mimics cash: ( ; rcbdc ) =
(1; 0). From this, it directly follows that welfare in both ce and nce is higher than in an
equilibrium without CBDC: in both ce and nce the central bank could attain the same
welfare as in the equilibrium without CBDC, by setting = 1 and rcbdc = 0, but this policy
combination is never optimal, as seen from (36) and (37) where ce < nce < 1. Hence,
W (1; 0) < W ( nce ; rcbdc
nce
) < W ( ce ; rcbdc
ce
), where the last inequality follows from (38).

Proof of Lemma 2. Substituting (36), (40) and (23) into (18)-(20) gives the expressions
for the shares of money, sce ce ce
c , sd , and scbdc , when all forms of money exist (ce), in terms of
parameters only. We can then calculate the in…ma of sce ce ce
c , sd , and scbdc over the parameter
space de…ned by (17). This yields

1
inf sce
c =
34
2
inf sce
d =
17
1
inf sce
cbdc =
17
1
and therefore, given s 17 in (17), it follows that d = cbdc = 0.35
Moreover, using (36) and (40), as well as (23), we can also verify that two necessary
conditions for positive uptake of the CBDC, which are subsumed by the CBDC design
constraint (15), are also satis…ed. These conditions are

1
rcbdc (1 ) (44)
> (rd rcbdc ) (45)

which respectively rule out the strict dominance of CBDC by cash and deposits (i.e., ensure
that neither cash nor deposits o¤er all households a strictly better utility than CBDC) as
per (5) and (7). First, since sup ce = 1617
< sup nce = 2324
< 1, while rcbdc = 0, condition (44)
ce 1
cannot be violated. Second, as inf ( rd ) = 17 over the parameter space in (17), (45) is
never violated either (and this necessarily also holds for nce , since nce > ce ).

35 1 1
This also remains valid in nce where inf sce
d = 6 and inf sce
cbdc = 12 .

30
Proof of Lemma 3. W ce ( ; rcbdc ) can be determined by solving the following system of
11 equations in 11 unknowns, which gives the expression (30):

sd
W = 1 + k0 + A 1 sd sd E d scbdc1 Ecbdc1 scbdc2 Ecbdc2 sc E c
2
+ (rd rcbdc )
sd =
2
scbdc = scbdc1 + scbdc2
(rd rcbdc )
scbdc1 =
2
1 + rcbdc
scbdc2 =
2
1 rcbdc
sc =
2
1 sd
Ed = 0 sd =
2 2
1 1
Ecbdc1 = sd + scbdc1 = sd scbdc1
2 2
1 1
Ecbdc2 = sd + scbdc1 + scbdc2 = sd + scbdc1 + scbdc2
2 2
1 1
Ec = 1 1 sc = sc
2 2
2 (A 1 ) + rcbdc
rd =
2+

Similarly, the solution for W nce ( ; rcbdc ) is found by setting sc = 0 in the above, and solving.
This yields the expression in (31).

Proof of Proposition 1. First, we note that W ce ( ce ; rcbdc ce

) > W nce ( nce ; rcbdc
nce
), per (38).
Moreover, W ce ( ce ; rcbdc
ce
) > W ce e; recbdc per de…nition, as welfare under unconstrained
optimal policies exceeds welfare under constrained optimal policies within a given equilibrium
(namely, ce). Hence, as long as the unconstrained ce is feasible, it is optimal. Therefore,
the relevant comparison centers on W nce ( nce ; rcbdc
nce
) versus W ce e; recbdc when the network
e¤ects constraint matters, that is, when ce + rcbdc
ce
= ce > 1 2s.
Second,
h i
inf W nce ( nce nce
; rcbdc ) W ce e; recbdc >0, > 1:431 = (46)
A; ;s

which means that for < , the policy combination e; recbdc always (i.e., for any values
of other parameters) welfare dominates ( nce ; rcbdc
nce
), and hence cash never vanishes under

31
optimal policies. Instead, for > , there exist parameterizations, including the extremes of
(A ) = 52 and s = 17 1
, such that ( nce ; rcbdc
nce
) welfare dominates e; recbdc . That is, when
> , cash can optimally be allowed to vanish, when network e¤ects are strong enough (s)
and the value of bank intermediation (A ) is large enough.
ce
Third, whenever > = 1 2s, it is necessarily true that

W ce e; recbdc W ce ;0 > 0 (47)

since e; recbdc = ; 0 is within the possibility set of e; recbdc but is not optimally chosen,
as seen from (41) and (42). Hence, the range of parameter values where W ce e; recbdc >
nce nce nce ce nce nce nce
W ( is broader than the range where W
; rcbdc ) ;0 > W ( To put this ; rcbdc ).
1
in more concrete terms, consider > and s = 17 . Then, the value of (A ) that is large
enough to induce a switch from ce to nce is higher when policies are set at e; recbdc than
when they are set at ;0 .

B Derivation of distributional e¤ects

The foundations for Figure 4 are found by considering the impact of a CBDC on, respectively,
deposit, cash and CBDC users. We use the term "after the introduction of a CBDC" to
indicate the comparison between a world with cash and deposits only, and one where CBDC
is available as an additional payment instrument.

B.1 Depositors

For a household that continues using deposits after the introduction of a CBDC, such as
i = 0, nothing changes in terms of the payment preference aspect of utility through the
introduction of a CBDC. Hence, tradeo¤ of that household centers on consumption, as
represented by

Cd = 1 + rd + T (48)
sd
= 1 + rd (1 s d ) + k0 + A 1 sd rcbdc scbdc (49)
2

where T = rcbdc scbdc and has been replaced using (10), (11), and (16). Further substituting
for sd , scbdc , and rd with the expressions as shown in the proof of Lemma 3, this gives a
closed-form expression for Cd . From this expression, we obtain

@Cd (A 2) + 2
= <0 (50)
@ rcbdc =0 (2 + )2

32
which means that the introduction of a non interest-bearing CBDC always raises welfare for
households that continue using deposits, because the introduction of a CBDC is equivalent
to lowering from = 1 (cash equivalence) to a lower value.36 Put di¤erently, the more
intensely the CBDC competes with bank deposits (lower ) the more it pushes up deposit
rates, and the larger the welfare gains to depositors.
Moreover,

@Cd 4 (4 (A 1) + 2rcdbc (8 + (5 + )) + (4 + ))
=
@rcbdc 2 (2 + )2

where we …nd that at rcdbc = 0, this term is negative overall, given the parameter space in
(17) and 1. Hence, a marginal CBDC interest rate cut from rcdbc = 0 to rcdbc < 0 always
raises the welfare of depositors.

B.2 Cash holders

For a household that continues using cash after the introduction of a CBDC (provided cash
remains in use), such as i = 1, welfare e¤ects similarly center on consumption only, as that
household’s preferences for payment instruments are una¤ected. Contrary to depositors,
however, the impact of a non interest-bearing CBDC on cash holders is straightforward:
while depositors see gains from increased deposit rates that (more than) compensate for
lost …rm pro…t transfers, cash holders only experience those lost pro…t transfers, and are
therefore necessarily worse o¤: @C @
c
rcbdc =0
> 0. Those cash holders would be even worse o¤
if network e¤ects push cash out of use and they are forced to take solace in a CBDC that is
more distant from their payment preferences.
The impact of negative CBDC rates is also straightforward for cash holders. As cash pays
no interest, the only channels through which cash holders are a¤ected are , which rises as
the CBDC rate declines (increased …nancial intermediation), and T , which is positive when
CBDC interest rates are negative (CBDC holders are taxed, and the proceeds accrue to all
households). That is, @r@C c
cbdc
< 0, as shown in Figure 4.

B.3 CBDC users

For households that switch to CBDC after it has been introduced, the key question is whether
their gains in payment preferences outweigh lost consumption arising from bank disinterme-
36 @Cd
Formally, we can verify that @ rcbdc =0
< 0 by noting that in (50) when A ! 1 the expression
1
2 + 2
becomes (2+ )2
and when A ! 52 it becomes 2
(2+ )2
, both of which are smaller than 0 given 3
2 and
@Cd
1. Hence, @ rcbdc =0
< 0 always holds over the parameter space in (17).

33
diation. Former depositors switching to CBDC, always see a welfare improvement overall. If
they did not, they would have remained depositors, since depositors see welfare gains from
the introduction of a CBDC, as per (50). The i = household experiences the largest welfare
gain from the availability of a CBDC, because the CBDC precisely meets that household’s
payment preferences.
However, some of the i > CBDC holders would have been better o¤ had CBDC not
existed. After all, the household that is exactly indi¤erent between holding cash and holding
CBDC experiences a welfare loss, since all cash holders lose welfare, and this household is
indi¤erent between the welfare loss of continuing to hold cash, and the welfare loss from
holding CBDC. CBDC holders with i marginally below this indi¤erent household would
also certainly see an overall welfare loss. CBDC does not o¤er them enough of an attractive
payment option to compensate for the loss in …rm pro…t transfers. Finally, a negative CBDC
rate acts as a tax on CBDC holders, and therefore reduces their welfare, as shown in Figure
4.

C Extensions
C.1 Constant returns to scale production function

The baseline model considers a decreasing returns to scale (quadratic) …rm production func-
tion. Here, we show that central components of the optimal policy pro…les we derived, as
represented by equations (36), (37) and (40), are robust to the use of a constant returns to
scale production function. Instead of Y = A k2 k, we now replace (10) with

Y = Ak (51)

Following the same steps as in the main text, we obtain the following outcomes for optimal
policies in ce

ce 1 + (A 1)
= (52)
2
ce
rcbdc = 0 (53)

and in nce

nce 2 + (A 1)
= (54)
3
nce
rcbdc = 0 (55)

34
Thus, the optimal unconstrained CBDC interest rate remains zero, in both ce and nce.
Moreover, the CBDC is optimally made more similar to cash (i.e., to help preserve bank
deposits) when the value of bank intermediation, (A ), rises.37

C.2 Anonymity externalities

In this extension, we consider the possibility that anonymous means of payment, like cash,
are associated with negative externalities, due to the potential for illicit activities. There can
be legitimate reasons that households desire anonymous forms of money, but by providing
for that demand, the illicit uses of anonymity are also bolstered.38 In particular, we now let
the utility of household i be given by
Z
Ui (j) = Cj jxj ij j xj(n) dn (56)
n6=i

R
where x dn captures the notion of negative externalities from anonymous means of
n6=i j(n)
payment. Here, n 2 [0; 1] represents “all other households”.39 While every household with
i > 0 likes anonymity in her own means of payments, every household also dislikes anonymity
in other households’transactions. The weight 2 [0; 1] represents the extent to which the
household dislikes others’anonymity in payment transactions.
Following the same steps as before, we derive unconstrained optimal policies as

ce 2 + (2 (A ) 1) (2 + )
= (57)
4+3 (4 + )
ce (A ) (4 + ) 2 (3 + )
rcbdc = 2 (58)
4+ 4+3 (4 + )

which nest the solutions in (36) and (40) for = 0.40 The most interesting aspect of these
solutions is that, for any > 0, rcbdc 6= 0 is now optimal, even when network e¤ects play no
ce
role. Depending on parameter values, rcbdc can be either positive or negative. In particular,
ce
in relation to the value of bank intermediation, rcbdc moves inversely with ce : A higher value
of bank intermediation leads to a more cash-like optimal CBDC design and lower (including
37
Decreasing and constant returns to scale production functions do lead to a di¤erent bank response to
CBDC competition. Under decreasing returns to scale, banks push back against the competition through
higher deposit rates (and also lending rates in Appendix C.3). Instead, in the constant returns to scale setup,
rd = A 1 and therefore the deposit rate is irresponsive to and rcbdc
38
The magnitude of negative externalities from cash is a topic of intense debate (Engert et al., 2018;
McAndrews, 2017; Rogo¤, 2016; Wright et al., 2017).
39
Given that each individual agent is atomistic, the space of all agents excluding one agent remains de…ned
on [0; 1].
40
The same holds for the nce solutions. These are not shown here in the interest of brevity, but are
available on request.

35
possibly negative) CBDC rates.
This inverse relation between optimal CBDC rates and is intuitive, and derives from a
ranking of forms of payment according to their anonymity externalities: cash is the worst,
deposits are the best, and the CBDC is somewhere in between, depending on its design.
When CBDC design is optimally quite similar to cash, then it is also optimal to have negative
CBDC rates, to push more households into deposits, and limit the anonymity externalities
induced by the CBDC. Instead, when the CBDC is more similar to deposits, then a positive
CBDC rate is optimal, to help attract more households away from cash.

C.3 Bank market power

We now consider banks that compete à la Cournot in the loan market, taking the actions
of other banks as given. Each bank therefore internalizes that total loans and the interest
rates on those loans depend on its individual lending as follows

@L
L = l + (1 )L ! =1 (59)
@l
@R @R
= = 1 (60)
@l @L

where @R @L
= 1 comes from equation (12). Here, represents the extent of bank market
power, with the extremes of = 0 and = 1 representing, respectively, perfect competition
(i.e., our baseline model) and a monopoly.
The bank’s pro…t maximization problem is given by

max f(R (l) rd ) lg (61)

l

where the bank recognizes the dependence of loan rates on an individual bank’s lending
decision: R depends on l. This yields the …rst-order condition

@R (l)
R (l) + l = rd (62)
@l

Moreover, deposit market equilibrium is derived from D = L, where D is from sd in (19):

(rd rcbdc ) +
L= (63)
2

Together, (12), (62), and (63) provide three equations in three unknowns, L, R and rd .

36
 @R
Replacing @l
= 1 from (60), and l = L, we can solve this to obtain

(A
1 rcbdc )+
L = (64)
2 + (1 + )
(A 1 rcbdc )+
R = A 1 (65)
2 + (1 + )
2 (A 1 ) + (1 + ) ( rcbdc )
rd = (66)
2 + (1 + )

Following the same steps as before, we again derive welfare and, from there, optimal policies
2
ce 8 + 2 (2 + (A ) (4 + + 2 ) 1) (1 + (2 ) )
= 2 (67)
2
16 + 3 (1 + ) + 8 (2 + )
ce 1 + 3 (A 1 )
rcbdc = 2 2 (68)
16 + 3 2 (1 + ) + 8 (2 + )

where for = 0 we retrieve our earlier solutions for optimal policies in (36) and (40). Indeed,
by comparing the above expressions to (36) and (40), we can see the direction in which > 0
pulls optimal policies. That is, using the expressions for ce and rcbdc
ce
in (67) and (68), we
numerically obtain that, over the parameter ranges in (17):

ce ce 279
inf j =0 = , sup ce ce
j =0 =0
5372
ce ce 35 ce ce
inf rcbdc rcbdc j =0 = , sup rcbdc rcbdc j =0 =0
163

and therefore > 0 means that both ce and rcbdcce

are lower than with = 0. This emanates
from the fact that greater market power in lending helps insulate banks from the negative
impact of a CBDC. Although increased competition for retail funding still drives up banks’
deposit rates, banks with market power partly compensate by also raising loan rates. In view
of banks’increased ability to withstand the impact of a CBDC, the optimal CBDC design
moves closer to deposits (lower ), although the policy maker partly insulates the impact of
this move by also cutting CBDC rates into negative territory.

C.4 Alternate equilibria under suboptimal policies

Table 1 listed three equilibria that do not occur under optimal policies. However, these
equilibria can come about if policies are set suboptimally.

CBDC and cash Per Lemma 2, deposits never vanish under optimal policies. This is
intuitive, since without deposits, our model yields zero intermediation, and the production

37
of consumption goods shuts down. Nevertheless, it is easy to show that suboptimal policies
could yield this equilibrium. For instance, for = 0, if the CBDC rate is set such that

rcbdc > A 1 (69)

then this ensures that rcbdc > rd (by equation (23)), while the payment pro…le ( = 0)
is equivalent to deposits. Hence, the CBDC strictly dominates deposits in this case: no
household would choose to hold deposits.

CBDC only Any arbitrarily high rcbdc would kill o¤ both deposits and cash. Households
would be paying for these CBDC interest payments through the lump-sum tax T , and
therefore this scenario brings only disadvantages to households, who lose payment instrument
variety and the productive bene…ts of bank intermediation, without gaining anything in
return.

Cash and deposits There are three ways that a suboptimally designed CBDC could lead
a situation where the design constraint (15) is violated such that there is no uptake of CBDC,
and only cash and deposits are in use. First, CBDC could be designed in such a way that
it is strictly dominated by cash, and violates (44). Second, CBDC design could imply that
bank deposits are a strictly preferred form of payment, which occurs when (45) is violated.
Third, even if the CBDC is not strictly dominated by cash or deposits, its design could be
such that network e¤ects prevent the buildup of a critical mass of CBDC users (15).
To give a concrete example, we replace rd from (23) into (45). This yields

1+
+ rcbdc > A 1 (70)

which means that when the policy combination ( ; rcbdc ) is set such that the condition above
is violated, as for example for a su¢ ciently negative rcbdc , deposits strictly dominate CBDC.

D Deriving a linear city of payment preferences

This appendix provides a stylized model highlighting how a linear-city model of payment
preferences can be derived from microfoundations. The model is based on the notion that
payment privacy can have value for households, when their digital transaction data can
be used by private companies with monopoly power. We concentrate on a simple setup
with cash and deposits only, and show how a "line" between these can arise endogenously,

38
including a cuto¤ that determines household sorting. Once a spectrum of this sort is derived,
formulating the intermediate case of a CBDC is a relatively straightforward extension.41
In this model, deposit-based payments are processed by a …ntech provider (or a bank
that has a similar business model), which is capable of tracking all transactions and is
legally unencumbered to use this data to its own bene…t. The …ntech company is also the
sole provider of credit in the economy, and provides loans to households. Moreover, the only
means that the …ntech company has to assess the creditworthiness of its customers is by
parsing their transactions data. For simplicity, we abstract from explicitly modeling deposit
and lending markets and interest rates here, and instead focus purely on household choice
based on the characteristics of deposits versus cash.
There are two types of products for households to purchase in this economy: G (Good)
and B (Bad), where B can be considered a type of sin product, such as alcoholic beverages or
cigarettes. Credit quality is inferred from the share of its income that a households spends on
G. We assume identical incomes across households, and each household i determines what
fraction (i) to spend on good G. Each household has a preferred share of its income that
it would like spend on each type of product: we denote by p (i) the ideal fraction of house-
hold i’s income spent on good G. Households are heterogeneous in their ideal consumption
patterns. In particular, households are uniformly distributed on p (i) 2 [0; 1]. Moreover,
any distance between a household’s ideal and actual consumption allocation, comes at a
quadratic disutility cost to the household: ( (i) p (i))2 .
The key distinction between cash and deposits here, is that deposit transactions are
monitored, while cash transactions are not. Monitoring matters because of the credit scores
being assigned to households by the …ntech company. For households using cash, the com-
pany cannot assign individualized credit scores, but rather uses an aggregate credit score,
based on the consumption pattern of the average cash user. That is, all cash users are pooled
together, in this respect. Instead, deposit using households are di¤erentiated by the …ntech
company according to their own purchase behavior.
Importantly, once households use deposits for any fraction of their payments, they are
unable to hide their overall purchase pattern from the …ntech company. Endogenously, the
model contains full revelation, because households have known, identical incomes.42 If the
…ntech company observes a depositor using only a fraction (i) of income, and fully using it
on G, then the company infers that the household used the rest of its income to purchase B
41
See also Garratt and van Oordt (2019), who develop a payment model with privacy as a public good,
where each consumer fails to internalize that her payments data is used to price discriminate among future
consumers, and privacy in government issued electronic cash can create social value.
42
More generally, the underlying assumption can be seen as a requirement on deposit-opening households
to reveal their income to the …ntech provider.

39
using cash. It is in this sense that deposits and cash cannot be e¤ectively mixed: while the
household is technically capable of mixing, the choice for using deposits at all, immediately
implies full revelation: payment privacy is undiversi…able.
The aim of this appendix is purely qualitative, and as such we choose simple functional
forms to highlight the relevant tradeo¤. In particular, we let credit scores be a linear function
of (i) (for depositors) and assume that the utility derived from a higher credit score also
enters linearly in the household’s utility function. Household utility is given by

U (i) = E [ (i)j j (i)] ( (i) p (i))2 (71)

where j (i) is household i’s chosen form of money, namely either d (deposits) or c (cash), is
a parameter that weighs the utility value of the welfare score as compared to approximating
the household’s ideal consumption shares, and
(
(i) if j (i) = d
E [ (i)j j (i)] = (72)
b if j (i) = c

where b equals the average share of G purchased by cash holders. Since households are
atomistic, a given cash holder will always consume exactly the same as her bliss point:
(i) = p (i) when j (i) = c.
Instead, a depositor will solve the following optimization problem

max (i) ( (i) p (i))2 (73)

(i)

leading to optimal consumption share of G

(i) = + p (i) (74)

2

where 2 parameterizes the extent of overconsumption of G induced by monitored transac-

tions.
The choice between cash and deposits then boils down to a comparison of utility under
household optimal consumption. A household chooses deposits over cash if and only if utility
as a depositor (setting (i) = 2 + p (i)) is greater than utility as a cash holder (which equals
b). This becomes the following condition for choosing deposits:

+ p (i) > b (75)

4

40
which can also be written as
p (i) > b =p (76)
4
This implies a sorting of households: households with p (i) > p choose deposits, while house-
holds with p (i) < p choose cash. That is, those households whose preferences favor a rela-
tively large share of G consumption, are more eager to engage in a full revelation relationship
with the …ntech provider, in order to reap the bene…ts of an improved credit score. Instead,
households with a relatively larger preference for consuming B, choose cash, opting out of a
depositor relationship with the …ntech provider that e¤ectively "forces" them to overconsume
G in order to appear more creditworthy. Overall, then, this model shows that heterogeneity
in consumption preferences can translate into heterogeneous payment instrument choice.

41