Embedded supervision: how to build regulation into blockchain financeRaphael AuerBank for International SettlementsAny views expressed in this presentation are those of the presenter and not necessarily those of the Bank for International Settlements

2

Can DLT help bring down the cost of financial services?

Prices of financial services have not come down…1

…despite availability of low-cost IT2 Cross-border payments are particularly costly3

1995 = 100 1995 = 100 US dollars

1 Gross output price index normalised to equal 100 in 1995. 2 Simple average of the prices of computer components, software and communication equipment. For DE, price of software. 3 Average total cost for sending $200 with all remittance service providers worldwide. For CN and IN, receiving country average total cost; for G20, SA and US, sending country average total cost. Sources: EU KLEMS; Eurostat; US Bureau of Economic Analysis (BEA); World Bank, Remittance Prices Worldwide, remittanceprices.worldbank.org; World Bank; BIS calculations; author’s calculations.

3

Regulation of distributed ledger technology (DLT) in finance

Two contrasting visions for DLT:A. Permissionless digital commodities controlled by no oneB. Novel financial technology controlled by entities (ie permissioned)

My conclusions thus far:A. Today’s permissionless DLT too limited to be relevant at large R Auer “Beyond the doomsday economics of proof-of-work in cryptocurrencies”

BIS WP 765 Implies: focus on AML/KYC & consumer protection, but not much more

B. However, permissioned & decentralised exchange, asset-backed tokens, stablecoins, smart contracts, etc. may evolve operational setup of financial markets Regulators can complement technological progress by embedding supervision in

the consensus of DLT-based markets

Potential applications of DLT

Improve credibility of information Liquefy small markets May have a role in payments (but AML/KYC is key) Wholesale financial engineering in the absence of a middlemen:

Automation flow of funds and the updating of security registers, replace central securities depositories, options and futures clearing houses (Ruttenberg and Pinna (2016) and Chiu and Koeppl (2019), Malinova and Park (2017), Lee (2015), ECB (2019), Mills et al (2016), and Benos et al (2017)

Need to be careful to disentangle use of automation from value of DLT

4

5

Same risk, same regulation and the case for “Embedded supervision”

“Same risk, same regulation”: DLT does not change underlying risks (see FINMA (2018), FCA (2018),…)

ICO=IPO, Payment=payment, etc. But: DLT improves information about underlying risks. “Same regulation, but evolving supervision!”

Embedded supervision is a regulatory framework that provides for compliance with regulatory standards in DLT-based markets to be automatically monitored by reading the market’s ledger. It would reduce the administrative burden for firms, while increasing the quality of data available to the supervisor.

6

Embedded supervision replaces part of today’s supervision process …

7

…with automated aggregation of information in DLs

8

The aim is to bring down compliance costs & fill costly data gaps

Firm surveys point to high compliance costs1, 2 Per cent

9

The principles of Embedded supervision

Key question: to what extent can the superviser trust the distributed ledger?

This paper develops four principles. Embedded supervision:I. can only function as part of an overall regulatory framework that is backed up

by an effective legal system and supporting institutions.II. can be applied to decentralised markets that achieve economic finality.III. needs to be designed within the context of economic market consensus, taking

into account how the market will react to being automatically supervised.IV. should promote low-cost compliance and a level playing field for small and large

firms.

10

I. ES can only function as part of an overall regulatory framework that is backed up by an effective legal system and supporting institutions

For intermediary-free financial engineering based on asset-backed tokens, legal system remains paramount backstop: Asset-backed tokens only as good as underlying real asset External reference points (“oracles”) can manipulate payoffs of smart contracts Host of other issues (faulty code, obfuscation, illegality, …) need to be resolved via legal

processes Also, management not relived of responsibility for truthfully reporting: can always trade on

other markets

11

II. Embedded supervision applies to markets that achieve economic finality

For a regulator to accept information of a ledger, there must be a notion of transaction finality:

“a transfer of funds [or] a transfer of securities that have become irrevocable and unconditional” (see CPSS (2003, p. 496))

I build on Chiu and Koeppl (2019), Budish (2018), and Auer (2019) and focus on finality via economic incentives DLT achieves consensus via the incentives of individual actors (miners, stakers,

validators) If the cost of an attack is larger than the gain, a market is economically final

12

A permissioned decentralised market

Users write financial contracts into a blockchain Contracts generate transfers, losers have incentives to “undo” the chain There is no central authority, so agents need to coordinate via incentives

Third parties - “validators” - loose their verification capital s should a blockchainreversal ever occur

Technological interpretation: Permissioned DLT with external deposits or cost of reputation Proof-of-stake with institutional underpinning to deter long range attacks

Note: do not investigate whether DLT is more efficient that centralised exchange

13

Setup – contracts and payoffs

𝑁𝑁𝑏𝑏 financial contracts in block b Life of contract i:

Signed into block 𝑏𝑏𝑖𝑖, posting fee 𝜋𝜋 Netting:

- Each block, chance 1 − 𝛽𝛽 that contract expires and is netted - Maximum contract length L

From t = 𝑏𝑏𝑖𝑖 + 1 onwards, contracts generate transfers. Assume NPV of payouts is:

𝑐𝑐𝑖𝑖,𝑡𝑡>𝑏𝑏𝑖𝑖 �~𝑈𝑈 −𝑐𝑐, 𝑐𝑐 𝑖𝑖𝑖𝑖 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑐𝑐𝑟𝑟𝑟𝑟𝑐𝑐𝑖𝑖𝑐𝑐𝑟𝑟 𝑐𝑐𝑐𝑐 𝑐𝑐𝑡𝑟𝑟 𝑙𝑙𝑟𝑟𝑙𝑙𝑙𝑙𝑟𝑟𝑐𝑐

0 𝑖𝑖𝑖𝑖 𝑐𝑐𝑡𝑟𝑟 𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐𝑐 𝑡𝑐𝑐𝑟𝑟 𝑏𝑏𝑟𝑟𝑟𝑟𝑐𝑐 𝑐𝑐𝑟𝑟𝑐𝑐𝑐𝑐𝑟𝑟𝑙𝑙

No settlement risk (users pre-commit sufficient funds on-ledger)

14

II. Embedded supervision applies to markets that achieve economic payment finality

No legal review process for financial contracts, so transaction must be “irreversible by incentives” (as in Auer (2019)): it is unprofitable to reverse the blockchain now & at any time in the future.

Block verifiers 𝒗𝒗𝒗𝒗𝒗𝒗, each having a stake of 𝒔𝒔 𝒗𝒗 looses 𝒔𝒔 if a block verified by 𝒗𝒗 is no longer in longest chain Stake 𝒔𝒔 freed after L periods (per unit fee 𝑖𝑖 > 0)

In this model, consensus concept is highest stake/most costly to forge Interpretation:

𝒔𝒔 could be stake deposited with external institution, or (legal) fine Can view it as a delegated/permissioned Proof-of-stake (not Proof-of-work)

15

Verification

𝑣𝑣𝑏𝑏 verifiers stake verification capital 𝑟𝑟 Receive a fee f for their services

What happens in case of misbehaviour? Verification means "agree to lose stake & fee in histories that do not include b“

Consensus algorithm:1. If conflicting blockchains are ever observed, one with higher amount of total

verification capital continues to be used2. Users only sign transactions into blocks that are economically final

(Given 2., 1. does not occur, but still need to specify)

16

Economic finality – avoiding a history reversion attack

Original chain

Original chain

Original chain

Original chain

I. Contracts in block b generate loosers

Attacker’s chain

Attacker’s chain

Attacker’s chain

II. Loosers bribe verifiers to make a new chain

III. Original chain becomes a dead

end and

Original chain

17

Maximal incentives to undo ONE block

Define expected maximum possible loss in block b at time t by 𝑪𝑪𝒃𝒃,𝒕𝒕:

𝐶𝐶𝑏𝑏,𝑡𝑡 ≡ 𝑟𝑟𝑐𝑐𝑚𝑚 �𝑖𝑖∈𝑏𝑏

Π𝑖𝑖,𝑡𝑡𝑟𝑟𝑐𝑐𝑚𝑚 𝑐𝑐𝑖𝑖,𝑡𝑡;−𝑐𝑐𝑖𝑖,𝑡𝑡

Π𝑖𝑖,𝑡𝑡 denotes indicator function=1 if contract is active. Solves to:

𝐶𝐶𝑏𝑏,𝑡𝑡 = �βt+1 −bNb ̅𝑐𝑐, 𝑐𝑐 − 𝑏𝑏 < 𝐿𝐿

0, 𝑐𝑐 − 𝑏𝑏 ≥ 𝐿𝐿 Note that this does allow for contract payoffs to be correlated (aggregate risk), but

contract expiration is idiosyncratic

18

Economic finality in a blockchain

Need to show: it is not profitable to undo the last block, neither the last 2 blocks, … For example, no 1-block attack requires:

𝜷𝜷𝑪𝑪𝒃𝒃,𝒕𝒕 = 𝜷𝜷𝜷𝜷𝒃𝒃 ̅𝑐𝑐 ≤ 𝒗𝒗𝒃𝒃𝐬𝐬

Generally:

max𝑥𝑥<𝐿𝐿

�𝑘𝑘=𝑜𝑜

𝑥𝑥𝐶𝐶𝑏𝑏−𝑘𝑘,𝑡𝑡 − 𝑣𝑣𝑏𝑏−𝑘𝑘s ≤ 0

In the paper, I solve this problem and show how high the verifier’s skin in the game has to be to ensure that the blockchain will never be reversed.

19

Free entry, fees, and equilibrium

Fee income: 𝐟𝐟 = 𝝅𝝅𝜷𝜷𝒃𝒃𝒗𝒗𝒃𝒃

, opportunity cost is forgone interest 𝜹𝜹 on stake s

Free entry:𝐬𝐬 + 𝐟𝐟 = 𝐬𝐬 𝟏𝟏 + 𝜹𝜹 𝑳𝑳

Solves to equilibrium:𝝅𝝅 = 𝜷𝜷 ̅𝑐𝑐 𝟏𝟏 − 𝟏𝟏 + 𝜹𝜹 −𝑳𝑳

𝒗𝒗𝒃𝒃 =𝜷𝜷𝜷𝜷𝒃𝒃 ̅𝑐𝑐

𝐬𝐬 𝟏𝟏 + 𝜹𝜹 𝑳𝑳

20

Properties of the equilibrium

𝜋𝜋 = ̅𝑐𝑐𝛽𝛽 1 − 1 + 𝛿𝛿 −𝐿𝐿 and 𝑣𝑣𝑏𝑏 = 𝛽𝛽𝛽𝛽𝑏𝑏 ̅𝑐𝑐s 1+𝛿𝛿 𝐿𝐿

Inexpensive only for fast-paced markets Fee 𝝅𝝅 is independent of:

s – lower s means v needs to increase. Also shows that model is isomorphic to heterogeneous verifiers

𝑁𝑁 – market is scale invariant (as long as diversification holds) Fee proportional to 𝑐𝑐, the upper bound on losses that needs to be deterred Fees increasing in opportunity cost of capital 𝛿𝛿, the length L for which such capital has to be

locked in, and decreasing in share of contracts that expire early

21

Residual stake buried in the ledger

To achieve equilibrium, market sets stake just high enough to prevent 1-block attack One period later, a fraction 1-𝛽𝛽 of the contracts expires

⇒ There is residual stake “buried” in the blockchain

At time b and for lag including up to x (ie in chain from b to b-x):

𝑹𝑹𝑹𝑹𝒔𝒔𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹𝑹 𝒔𝒔𝒕𝒕𝑹𝑹𝒔𝒔𝑹𝑹𝒃𝒃,𝒙𝒙 = 𝜷𝜷𝜷𝜷𝒃𝒃�𝒄𝒄�𝒔𝒔=𝟎𝟎

𝒙𝒙𝟏𝟏 − 𝜷𝜷𝒔𝒔

22

III. The market consensus must be strong enough to withstand embedded supervision

Have established the conditions under which market participants trust their own transactions

Does not mean that the regulator can trust the data If fulfilling the regulatory goal was free, market would not have to be regulated For example, regulator may require to finance losses with equity rather than debt

Embedded supervision creates incentives to cheat In addition to contractual payoff 𝑐𝑐𝑖𝑖,𝑡𝑡 there is a regulatory cost 𝒓𝒓𝑹𝑹,𝒕𝒕 Upper bound

for 𝑐𝑐𝑖𝑖,𝑡𝑡 is �̅�𝑐 (r, c correlated)

23

Equilibrium allowing for embedded supervision

One option is to apply compliance in real time,

Mandating 𝒗𝒗𝒃𝒃 = 𝜷𝜷𝜷𝜷𝒃𝒃 �̅�𝑟+ ̅𝑐𝑐𝒔𝒔

This will increase fees by ⁄�̅�𝑐 ̅𝑐𝑐. If cost of equity 6%, cost of debt 2%, ⁄�̅�𝑐 ̅𝑐𝑐=0.04/1.02 Another option is for regulator to wait for some time before applying regulatory measures.

Uses residual stake in the ledger Cost is that market is essentially unregulated at short horizons (akin for window

dressing)

24

IV. Design should encourage a level playing field for entrants

Smaller banks are disproportionately affected by compliance costs

In per cent Graph 6

Source: D Dahl, A Meyer and M Neely, “Scale matters: community banks and compliance costs”, Federal Reserve Bank of St Louis, The Regional Economist, July 2016, www.stlouisfed.org/~/media/publications/regional-economist/2016/july/scale_matters.pdf.

25

Conclusion

Embedded supervision is an opportunity: a new ecosystem co-created by regulators and innovators

What can generate a level playing field? Low fixed costs of compliance - regulators could develop a basic open source suite of

risk assessment and compliance tools Set standards for blockchain-interoperability Need to critically evaluate decentralisation (see Walch (2019))

Official institutions could become reference points/oracles (i.e. offer digitally signed, time-stamped information such as exchange rates, etc.)