Decentralized Finance, Centralized Ownership? An Iterative Mapping Process to Measure Protocol Token Distribution Matthias Nadler Fabian Schär Center for Innovative Finance Center for Innovative Finance Faculty of Business and Economics Faculty of Business and Economics University of Basel University of Basel Basel, Switzerland Basel, Switzerland matthias.nadler@unibas.ch f.schaer@unibas.ch Abstract—In this paper, we analyze various Decentralized exogenous developments, upcoming interface changes, Finance (DeFi) protocols in terms of their token distributions. and potential bugs. We propose an iterative mapping process that allows us to split aggregate token holdings from custodial and escrow contracts Economics: Most tokens have some form of implicit andassignthemtotheireconomicbeneficiaries.Thismethodac- or explicitvalue-capturethat allowsthe tokenholder to countsforliquidity-,lending-,andstaking-pools,aswellastoken participate economically in the growth of the protocol. wrappers, and can be used to break down token holdings, even Value is usually distributed through a utility and burn forhighnestinglevels.Wecomputeindividualaddressbalances mechanism (deflationary pressure) or some form of for several snapshots and analyze intertemporal distribution changes. In addition, we study reallocation and protocol usage dividend-like payments. In many cases, initial token data, and propose a proxy for measuring token dependencies sales are used to fund protocol development and con- and ecosystem integration. The paper offers new insights on tinuous release schedules to incentivize protocol usage. DeFi interoperability as well as token ownership distribution and may serve as a foundation for further research. Considering the two main reasons for the existence of Index Terms—Blockchain Governance, Ethereum, Decentral- these tokens, it becomes apparent that token distribution is ized Finance, DeFi, Token Economy a critical factor in the protocols’ decentralization efforts. Heavily centralized token allocations may result in situations I. INTRODUCTION where a small set of super-users can unilaterally change Decentralized Finance (DeFi) refers to a composable the protocol – potentially at the expense of everyone else. and trust-minimized protocol stack that is built on public Moreover, a heavily concentrated distribution may create an Blockchain networks and uses smart contracts to create a ecosystem where much of the value is captured by a small largevarietyofpubliclyaccessibleandinteroperablefinancial number of actors. services. In contrast to traditional financial infrastructure, The authors are unaware of previous academic research on these services are mostly non-custodial and can mitigate this subject. In August 2020, an analysis was circulated on counterpartyriskwithouttheneedforacentralizedthirdparty. social media, [2]. Simone Conti analyzed token contracts for Fundsarelockedinsmartcontractsandhandledinaccordance their top holders and used this data to compute ownership with predefined rules, as specified by the contract code. concentration measures. However, the study was based on Some examples of DeFi protocols include constant function questionable assumptions and fails to account for the large market makers, lending-platforms, prediction markets, on- variety of contract accounts. In particular, liquidity-, lending- chain investment funds, and synthetic assets, [1]. and staking-pools, as well as token wrappers, had been Most of these protocols issue corresponding tokens that counted as individual entities. As these contract accounts are represent some form of partial protocol ownership. Although mere custodians and usually hold significant token amounts the exact implementations, the feature sets, and the token on behalf of a large set of economic agents, this approach holder rights vary greatly among these tokens, the reason for clearly leads to spurious results. their existence can usually be traced back to two motives: There are previous studies that tackle similar research Protocol Governance and Protocol Economics. questions in the context of the Bitcoin network, [3], [4], Governance: Tokens may entitle the holder to vote [5]. However, due to Bitcoin’s relatively static nature and the on contract upgrades or parameter changes. A token- separation of token ownership and protocol voting rights, the basedgovernancesystemallowsfortheimplementation question is less pressing. Moreover, the fact that Bitcoin’s of new features. Moreover, the protocol can react to standard client discourages address reuse makes these anal- 0202 ceD 61 ]NG.noce[ 1v60390.2102:viXra yses much harder to perform. In a similar vein, a recent TableI working paper conducted an analysis for the evolution of TOKENSELECTION shares in proof-of-stake based cryptocurrencies, [6]. The remainder of this paper is structured as follows: In Token MC VL SC Deployment Section II we describe how the token and snapshot samples BAL (cid:55) (cid:51) (cid:51) 2020-06-20 have been selected. Sections III and IV explore the data BNT (cid:55) (cid:55) (cid:51) 2017-06-10 preparationandanalysisrespectively.InSectionVwediscuss COMP (cid:51) (cid:51) (cid:51) 2020-03-04 the results, limitations and further research. In Section VI we CREAM (cid:55) (cid:51) (cid:55) 2020-08-04 CRV (cid:55) (cid:51) (cid:55) 2020-08-13 briefly summarize our findings and the contribution of this KNC (cid:51) NA (cid:51) 2017-09-12 paper. LEND (cid:51) (cid:51) (cid:51) 2017-09-17 LINK (cid:51) NA (cid:55) 2017-09-16 LRC (cid:51) (cid:55) (cid:55) 2019-04-11 II. SAMPLESELECTION MKR (cid:51) (cid:51) (cid:51) 2017-11-25 MTA (cid:55) (cid:55) (cid:51) 2020-07-13 In this section, we describe the scope of our analysis. In NXM (cid:51) (cid:55) (cid:55) 2019-05-23 particular, we discuss how tokens and snapshots have been REN (cid:51) (cid:55) (cid:51) 2017-12-31 selected. The token selection determines which assets we SUSHI (cid:51) (cid:51) (cid:55) 2020-08-26 UMA (cid:51) (cid:55) (cid:55) 2020-01-09 observe. The snapshot selection determines at which point YFI (cid:51) (cid:51) (cid:51) 2020-07-17 in time the blockchain state is observed. YFII (cid:51) (cid:55) (cid:55) 2020-07-26 ZRX (cid:51) NA (cid:55) 2017-08-11 A. Token Selection TableII To qualify for selection, tokens had to fulfill the following SNAPSHOTSELECTION criteria: 1) Thetokenmustbeaprotocoltoken.Itmustincorporate Nr. BlockHeight Date some form of governance and/or utility mechanism. 1 7962629 2019-06-15 Purestablecoins,tokenwrappers,ortokenbasketshave 2 8155117 2019-07-15 not been considered.1 3 8354625 2019-08-15 2) The token must be ERC-20 compliant and contribute 4 8553607 2019-09-15 towards decentralized financial infrastructure. 5 8745378 2019-10-15 6 8938208 2019-11-15 3) As of September 15th, 2020, the token must fulfill at 7 9110216 2019-12-15 least one of the following three conditions: 8 9285458 2020-01-15 a) Relevantsupplywithmarketcap≥200mm(MC). 9 9487426 2020-02-15 b) Totalvaluelockedintheprotocol’scontracts(vest- 10 9676110 2020-03-15 11 9877036 2020-04-15 ing not included) ≥ 300 mm (VL). 12 10070789 2020-05-15 c) Inclusion in Simone Conti’s table (SC). 13 10270349 2020-06-15 Market cap and value locked serve as objective and quan- 14 10467362 2020-07-15 15 10664157 2020-08-15 titative inclusion criteria. Tokens from Simone Conti’s table 16 10866666 2020-09-15 have mainly been included to allow for comparisons. Applyingthesecriteria,wegetasampleof18DeFitokens. The tokens and the reason for their selection are summarized III. DATAPREPARATION in Table I. Please note that we have decided to exclude SNX since some of its features are not in line with standard WeuseourtokenandsnapshotselectionfromIItoanalyze conventions and make it particularly difficult to analyze. the allocation characteristics and observe how they change over time. All the necessary transaction- and event data was directlyextractedfromaGo-EthereumnodeusingEthereum- B. Snapshot Selection ETL,[7].Toconstructaccuratesnapshotsoftokenownership, Toanalyzehowtheallocationmetricschangeovertime,we we must map each token holding to the address that actually decidedtoconducttheanalysisforvarioussnapshots.Thefirst owns and may ultimately claim the funds. snapshotisfromJune15th,2019.Wehadthentakenmonthly A simple example is the YFI/wETH Uniswap V2 liquidity snapshots. The snapshots’ block heights and timestamps are pool: A naïve analysis would lead to the conclusion that listed in Table II. the tokens are owned by the Uniswap exchange contract. However, this contract is just a liquidity pool with very limitedcontroloverthetokensitholds.Fullcontrol,andthus 1Althoughwrappersandbasketswillbeconsideredforfundreallocation, asdescribedinSectionIII. ownershipofthetokens,remainswiththeliquidityproviders. To account for this and to correctly reflect the state of token and withdrawing stakes and rewards. For Sushi-like ownership, the tokens must be mapped proportionally from staking pools, we also account for a possible migration the exchange contract to the liquidity providers. of staked liquidity pool tokens. A more complex example illustrates the need for an itera- UniqueContracts:Thesecontractsdonotfitanyofthe tive mapping process: YFI is deposited into a Cream lending above categories, but the tokens can still be mapped to pool, minting crYFI for the owner. This crYFI together with theirowners.Eachcontractistreatedindividually,using crCREAM is then deposited in a crYFI/crCREAM Balancer- contract-specificeventsandarchivalcallswhereneeded. likeliquiditypool,mintingCRPT(Creampooltokens)forthe AfewexamplesincludeMKRgovernancevoting,REN depositor. Finally, these CRPT are staked in a Cream staking darknode staking, or LRC long-term holdings. pool, which periodically rewards the staker with CREAM tokens but does not mint any ownership tokens. The actual Smart contracts which hold funds that are not owned by YFI tokens, in this case, are held by the Cream lending pool. individual actors or where no on-chain mapping exists are Trying to map them to their owners via the lending pool excluded from the analysis. Most commonly, this applies to tokens(crYFI)willleadustotheliquiditypoolandfinallyto contracts that hold and manage funds directly owned by a the staking pool, where we can map the YFI to the accounts protocol with no obvious distribution mechanism. that staked the CRPT tokens. Each of these steps needs to B. Iterative Mapping Process for Tokens be approached differently, as the underlying contracts have For each token and snapshot, we construct a token holder distinctformsoftrackingtokenownership.Andfurther,these table listing the initial token endowments per address. We steps must also be performed in the correct order. then proceed with an iterative mapping process as follows: A. Identifying and Categorizing Addresses Algorithm 1 Iterative Mapping Process Addresses that do not have bytecode deployed on them - also called externally owned accounts or EOAs - can- 1: H ← initial token holder table not be analyzed further with on-chain data. To determine 2: repeat whether to include or exclude an EOA from our analysis, we 3: sort H by token value, descending use a combination of tags from etherscan.io, nansen.ai, and 4: for all h∈ top 1,000 rows of H do coingecko.com,[8],[9],[10].AnEOAqualifiesforexclusion 5: identify and categorize h if it is a known burner address, owned by a centralized, off- 6: apply inclusion logic to h chain exchange (CEX) or if the tokens on the account are 7: if h is mappable then disclosed by the developer team as FTIA (foundation, team, 8: map h according to its category investor, and advisor) vesting. Every other EOA is assumed 9: end if to be a single actor and is included in the analysis. 10: end for Addresses with deployed bytecode are smart contracts or 11: until no mappable rows found in last iteration contract accounts. These contracts are analyzed and catego- 12: asserteveryrowwithmorethan0.1%ofthetotalrelevant rizedbasedontheirABI,bytecode,returnvalues,andmanual supply is properly identified and categorized code review. Most implementations of multisig wallets are detected and treated equivalent to EOAs. Mappable smart It is possible that tokens must be mapped from an address contracts are described by the following categories: onto themselves. For most mappable contracts, these tokens Liquidity Pools: Decentralized exchanges, converters, are permanently lost2 and are thus treated as burned and are token baskets, or similar contracts that implement one excluded from the analysis. For contracts where the tokens or more ERC-20 liquidity pool tokens. The funds are are not lost in this way, we implemented contract-specific mapped proportionally to the relevant liquidity pool solutions to avoid potential infinite recursion. tokens. Everyinstanceofaremappingfromoneaddresstoanother, called an adjustment, is tracked and assigned to one of Lending Pools: Aave, Compound, and Cream offer five adjustment categories. There is no distinction between lending and borrowing of tokens. Both the debts and situations where the protocol token or a wrapped version deposits are mapped to their owners using protocol- thereof is remapped. The five adjustment categories are: specific events and archival calls to the contracts. Internal Staking: Depositing the token into a contract Staking Contracts: Staking contracts differ from liq- that is part of the same protocol. This includes liquid- uidity pools in the sense that they usually do not ity provision incentives, protocol stability staking, and implement an ERC-20 token to track the stakes of some forms of governance voting. the owners. We further differentiate if the token in External Staking: Depositing the token into a contract questionisusedasareward,asastake,orboth.Future that is not part of the same protocol. This is most staking rewards are excluded as they cannot be reliably mapped to future owners. The remaining tokens are 2For example, if Uniswap liquidity pool tokens are directly sent to their mapped using contract-specific events for depositing liquiditypooladdress,theycanneverberetrieved. prominent for Sushi-like liquidity pool token staking to an OLS regression line; the standard deviation shows the withtheintentionofmigratingtheliquiditypooltokens, volatility of the trend. butitalsoincludesavarietyofother,externalincentive B. Ecosystem Integration programs. Table IV presents key metrics of the tokens’ integration AMM Liquidity: Depositing the token into a liquidity into the DeFi ecosystem. The table is described below. pool run by a decentralized exchange with some form of an automated market maker. Inclusion %: Relevant token supply divided by total token supply, excluding burned tokens. Lending / Borrowing: Depositing the token into a liquidity pool run by a decentralized lending platform Wrapping Complexity: Relevant adjustments divided or borrowing tokens from such a pool. by relevant supply. This includes only adjustments to non-excluded addresses3 and may (in extreme cases) Other: Derivatives, 1:1 token wrappers with no added reach values above 1. The Wrapping complexity is functionality, token migrations, and investment fund- formalized in (2), where ωωω :=(ω ,...,ω ) represents like token baskets. 1 N the vector of all relevant adjustments for a given token and S¯ represents relevant supply. IV. DATAANALYSIS In this section, we will use our data set to analyze two (cid:80)N |ω | i=1 i (2) questions: First, we study the token ownership concentration S¯ and use our remapping approach to compute more accurate Multi-Token Holdings: Number of addresses with a ownershiptablesandintroducenewallocationmetrics.These minimumallocationof0.1%ofthistokenand0.1%for metricsareofparticularinterest,ashighlyconcentratedtoken at least n∈(1,2,3,4) other tokens from our sample. allocations could potentially undermine any decentralization efforts. Second, we use our remapping and protocol usage Shorted: Negative token balances in relation to rele- data to introduce wrapping complexity, shortage, and token vant supply; i.e. value on addresses that used lending interaction measures. These measures essentially serve as a marketstoborrowandresellthetoken,toobtainashort proxy and indicate the degree of integration into the DeFi exposure, divided by S¯. ecosystem.Moreover,theymayserveasanimportantmeasure It is important to note that the inclusion ratio is pre- for potential dependencies and the general stability of the dominantly dictated by the tokens’ emission schemes. In system. some cases, the total supply is created with the ERC-20 token deployment but held in escrow and only released over A. Concentration of Token Ownership the following years. Consequently, we excluded this non- Table III shows key metrics to illustrate the concentration circulating supply. of adjusted token ownership for the most recent snapshot, Figure 1 shows the development of the tokens’ wrapping September 15th, 2020. The table is described below. Please complexities by adjustment category in a stacked time series. note that relevant supply refers to the sum of all adjusted Note that the limits of the y-axis for the CREAM graph and included token holdings, taking into account outstanding are adjusted to accommodate for the higher total wrapping debts. Excluded token holdings are described in detail in complexity. We have not included a graph for the SUSHI section III-A. token,asthereisonlyonesnapshotavailablesinceitslaunch4. Owner #: Totalnumber ofaddressesowning apositive A wrapping complexity > 1 means that the same tokens amount or fraction of the token. are wrapped several times. If, for example, a token is added Top n: Percentage of the relevant supply held by the to a lending pool, borrowed by another person, subsequently top n addresses. addedtoanAMMliquiditypool,andtheresultingLPtokens staked in a staking pool, the wrapping complexity would Top n%: Minimum number of addresses owning a amount to 4. Similarly, a single token could be used multiple combined n% of the relevant supply. times in a lending pool and thereby significantly increase the Gini 500: The Gini coefficient, [11], is used to show wrapping complexity. the wealth distribution inequality among the top 500 Notethatmosttokenshaveexperiencedasharpincreasein holders of each token. It can be formalized as (1). wrapping complexity in mid-2020. The extent to which each (cid:80)500 (cid:80)500|x −x | G = i=1 j=1 i j (1) 3Some of the excluded addresses still deposit their tokens in mappable 500 2·5002x¯ contracts; e.g. a CEX depositing their users’ tokens in a staking pool. To preventdistortion,weexcludethesefundsfromboththerelevantsupplyand For tokens with historical data of at least 12 months, we therelevantadjustments. 4OnSeptember15th,2020,the109.9%wrappingcomplexityofSUSHIis include the trend and standard deviation over this period. composed of 28.2% internal staking, 49.3% external staking, 30.1% AMM Thetrendrepresentsthemonthlychangeinpercentaccording liquidity,and2.2%lending/borrowing. TableIII TOKENOWNERSHIPSTRUCTURE Token Owner# Top5 Top10 Top50 Top100 Top500 Top50% Top99% Gini500 BAL† Sep20 16,661 27.6% 36.71% 77.3% 85.01% 94.86% 18 2,157 83.77% Sep20 49,294 15.69% 24.71% 49.5% 61.77% 80.95% 52 10,010 69.82% BNT Trend +1.64% -5.43% -4.43% -2.94% -2.14% -1.06% +49.45% +7.52% -1.5% σ 12m 2,882.0 0.0712 0.0764 0.0827 0.0669 0.0378 15.7 1,481.9 0.0487 COMP† Sep20 36,033 31.23% 43.79% 86.75% 96.15% 98.91% 14 564 90.36% CREAM† Sep20 4,426 48.44% 57.11% 74.32% 81.77% 94.16% 6 1,549 83.04% CRV† Sep20 11,076 56.92% 61.09% 73.23% 79.07% 90.27% 2 3,549 84.64% Sep20 92,780 24.93% 35.63% 57.73% 64.62% 77.99% 26 19,922 77.6% KNC Trend +6.51% +3.36% +5.01% +2.14% +0.98% +0.04% -5.39% +15.74% +1.21% σ 12m 12,589.4 0.0302 0.0594 0.0489 0.0336 0.0171 13.9 3,971.3 0.0374 Sep20 174,861 36.67% 43.64% 61.44% 67.42% 80.05% 16 57,534 79.97% LEND Trend +0.23% +33.26% +22.23% +11.35% +8.26% +3.74% -9.77% -4.7% +3.98% σ 12m 3,066.9 0.1294 0.1389 0.1358 0.1258 0.0878 82.2 21,962.9 0.0933 Sep20 233,128 7.18% 13.46% 37.0% 44.99% 61.23% 166 61,910 65.27% LINK Trend +31.34% -0.5% -0.62% +1.72% +1.24% +0.08% -2.73% +16.99% +1.24% σ 12m 52,004.9 0.0029 0.004 0.0221 0.0204 0.0067 25.0 12,158.7 0.0279 Sep20 66,382 13.75% 20.06% 43.44% 62.11% 87.9% 66 5,251 66.36% LRC Trend +1.49% -2.3% -1.68% -1.26% -1.14% -0.41% +3.23% +7.95% -0.74% σ 12m 3,392.5 0.0236 0.0232 0.0261 0.0313 0.0163 6.1 811.7 0.0205 Sep20 29,765 24.43% 36.49% 67.71% 79.49% 93.72% 20 3,918 79.26% MKR Trend +8.31% -3.45% -2.12% -0.45% -0.19% -0.12% +4.5% +7.17% -0.22% σ 12m 4,511.7 0.0503 0.0405 0.0175 0.0107 0.0057 3.0 587.0 0.01 MTA† Sep20 5,595 13.81% 22.97% 51.18% 63.51% 88.27% 47 2,090 65.93% Sep20 7,355 32.17% 44.3% 70.42% 78.51% 91.29% 14 2,817 81.14% NXM Trend -36.69% -2.87% -2.71% -1.65% -1.12% -0.37% +18.09% -33.11% -0.24% σ 12m 1,918.2 0.0704 0.0992 0.0869 0.0619 0.0238 2.7 747.1 0.0434 Sep20 22,770 10.45% 15.29% 32.81% 41.79% 67.85% 166 8,500 55.31% REN Trend +26.0% -3.12% -2.97% -2.98% -2.64% -1.5% +42.78% +25.39% -1.56% σ 12m 4,673.4 0.0232 0.0313 0.0671 0.072 0.0579 38.4 1,718.0 0.0437 SUSHI† Sep20 22,740 25.64% 35.26% 58.31% 66.28% 83.78% 28 7,300 74.11% UMA† Sep20 5,634 56.21% 75.64% 96.87% 98.21% 99.43% 5 240 95.61% YFI† Sep20 14,296 11.52% 16.98% 37.32% 48.1% 73.75% 114 5,145 57.6% YFII† Sep20 8,513 20.8% 27.78% 53.93% 66.23% 85.15% 40 3,278 72.18% Sep20 161,285 23.71% 38.4% 59.39% 63.87% 72.91% 21 38,404 82.63% ZRX Trend +4.05% -1.15% -0.02% +0.76% +0.64% +0.22% -2.96% +6.28% +0.43% σ 12m 16,372.0 0.0133 0.0056 0.0158 0.0147 0.0082 3.6 5,233.6 0.0132 †Insufficienthistoricaldata. categoryisuseddependsonthecharacteristicsofeachtoken; some interesting findings. internal staking, in particular, can take very different forms. WhatseemstobetrueacrosstheboardisthatDeFitokens The“other”categoryismainlydrivenbytokenmigrations, have a somewhat concentrated ownership structure. This is where new tokens are held in redemption contracts, and 1:1 certainly an issue that merits monitoring, as it may poten- token wrappers. tially undermine many of the advantages this new financial infrastructure may provide. For protocols with token-based governance models, the V. DISCUSSION lowerboundnumberofaddressesneededtoreachamajority, Inthissectionwediscusstheresultsfromourdataanalysis. i.e., >50%, may be of special interest. A relatively low We revisit Table III and IV as well as Figure 1 and discuss threshold can indicate a higher likelihood of collusion and Figure1. AdjustmentGraphs Category Internal Staking External Staking AMM Liquidity Lending / Borrowing Other 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA BAL Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA BNT Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA COMP Token Adjustments 4 3 2 1 0 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA CREAM Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA CRV Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA KNC Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA LEND Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA LINK Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA LRC Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA MKR Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA MTA Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA NXM Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA REN Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA UMA Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA YFI Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA YFII Token Adjustments 1.00 0.75 0.50 0.25 0.00 Jun 19 Jul 1 A 9 ug 1 S 9 ep 1 O 9 ct 1 N 9 ov D 19 ec 1 J 9 an 2 F 0 eb 2 M 0 ar 2 A 0 pr 2 M 0 ai 2 J 0 un 20 Jul 2 A 0 ug 2 S 0 ep 20 Date ylppuS tnaveleR / tnemtsujdA ZRX Token Adjustments TableIV TOKENWRAPPINGCOMPLEXITY WrappingComplexity Multi-TokenHoldings Token Inclusion% Shorted Jun-19 Sep-19 Dec-19 Mar-20 Jun-20 Sep-20 1+ 2+ 3+ 4+ BAL 19.6% - - - - - 51.7% 17.6% 5.5% 1.1% - 0.026% BNT 56.8% 11.9% 11.9% 10.3% 20.8% 9.6% 10.2% 8.7% 1.4% 0.7% 0.7% - COMP 36.0% - - - 0.0% 0.0% 7.5% 8.4% 3.6% 2.4% - 0.004% CREAM 3.6% - - - - - 455.0% 30.1% 11.8% 5.4% - 11.971% CRV 2.2% - - - - - 43.1% 20.9% 9.9% 4.4% 2.2% 0.761% KNC 70.7% 0.2% 0.2% 0.4% 2.9% 1.8% 48.4% 17.7% 9.4% 4.2% 2.1% 0.123% LEND 69.3% 0.0% 0.0% 0.1% 28.9% 50.7% 63.1% 38.6% 19.3% 6.8% 2.3% 0.039% LINK 31.3% 0.0% 0.0% 0.0% 1.8% 2.2% 13.6% 12.9% 5.9% 4.0% 2.0% 0.383% LRC 58.8% 5.3% 4.7% 7.4% 19.0% 21.4% 23.1% 1.8% 0.6% - - - MKR 81.5% 33.6% 23.2% 31.5% 28.6% 37.3% 41.5% 7.2% 2.4% 0.8% - 0.036% MTA 3.1% - - - - - 73.8% 15.1% 4.8% 1.8% - 2.631% NXM 95.1% 0.0% 0.0% 0.0% 0.0% 0.0% 66.7% 17.0% 8.0% 2.0% - - REN 61.3% 0.0% 0.0% 0.0% 0.2% 12.1% 59.9% 11.4% 4.4% 3.2% 1.3% 0.035% SUSHI 48.2% - - - - - 109.9% 28.9% 9.9% 1.7% - 0.844% UMA 53.8% - - - 0.0% 0.4% 3.0% 4.3% - - - - YFI 94.8% - - - - - 70.5% 41.0% 14.1% 2.6% - 0.307% YFII 40.1% - - - - - 54.2% 8.6% 4.3% 1.4% - - ZRX 57.9% 0.7% 1.9% 1.7% 4.5% 6.8% 32.8% 19.0% 6.3% 4.8% 3.2% 0.052% centralized decision making. In extreme cases, a few indi- degree. viduals could jointly enact protocol changes. However, since In addition to the computation of more accurate holder governance rules, the implementations of voting schemes, tables, transparency is a precondition for the analysis of and security modules (e.g., timelocks) vary greatly between protocol interconnections and dependencies. For this pur- protocols,directcomparisonsshouldonlybemadewithgreat pose, we introduce the wrapping complexity and multi-token care. holdingmetrics.Wrappingcomplexityessentiallyshowshow In addition to the decentralization and governance con- the token is used in the ecosystem. On the one hand, high cerns, the study also shows DeFi’s limitations with regard to wrapping complexities can be interpreted as an indicator for transparency.WhileitistruethattheDeFispaceisextremely a token that is deeply integrated into the DeFi ecosystem. On transparent in the sense that almost all data is available on- the other hand, high wrapping complexities may also be an chain,itisverycumbersometocollectthedataandprepareit indicatorforconvolutedandunnecessarilycomplexwrapping in a digestible form. High nesting levels with multiple proto- schemes that may introduce additional risks. cols and token wrappers involved will overwhelm most users A potential indicator for how the market feels about the and analysts and create the need for sophisticated analysis complexity is the shortage percentage, i.e., the value of all tools. The computation of accurate token ownership statistics decentralizedshortpositionsinrelationtotherelativesupply. and reliable dependency statistics is extremely challenging. Interestingly, there is a high positive correlation between the The problem becomes apparent when we compare our twomeasures,whichmayatfirstglancesuggestthatwrapping results to the results of Simone Conti’s analysis, [2]. Recall complexity is interpreted as a negative signal. However, that Conti’s analysis has not controlled for any account- this would be a problematic interpretation since wrapping specific properties. Our analysis shows that for most tokens, complexityis,infact,atleastpartiallydrivenbytheshortage the token holdings of the top 5 addresses thereby have been activity. Once we exclude the lending and borrowing, as well overestimated by approximately 100% and in some extreme as “other” categories, the effect becomes less pronounced. cases by up to 700%. The main source of these errors is The DeFi space is developing very rapidly and constantly the inclusion of token holdings from custodial- and escrow increases in complexity. Many new and exciting protocols contracts, such as liquidity-, lending-, and staking-pools, as have emerged in 2020. Novel concepts such as complex well as token wrappers, vesting contracts, migrations, burner staking schemes started to play a role in most protocols. addresses, and decentralized exchange addresses. We control We see staking, or more specifically staking rewards, as a for these accounts and split their holdings to the actual catalyst for the immense growth in the DeFi space. However, beneficiaryaddresseswherepossibleandexcludethemwhere it is somewhat questionable whether this growth will be not possible. A closer comparison of the two tables reveals sustainable. Treasury pools will eventually run out of tokens, that the differences remain high for lower holder thresholds and uncontrolled token growth leads to an increase of the (i.e., top 10, top 50, and top 100). At the top 500 threshold, relevanttokensupply,whichmaycreateinflationarypressure. the differences are still significant, although to a much lesser While we are confident that our study provides interesting

contributions with new metrics and processes to compute token ownership tables with unprecedented accuracy, we uals. This finding may raise important questions regarding would still like to mention some of the limitations of our protocol decentralization and build a foundation for DeFi study and point out room for further extensions. governance research. First, we perform no network analysis to potentially link We further investigated dependencies and ecosystem in- multiple addresses of the same actor. This approach has tegration. Our analysis suggests that the complexity of the likely lead to an overestimation of decentralization. In a ecosystem has drastically increased. This increase seems to further research project, one could combine our data set and be consistent among most tokens. However, the main drivers remapping method with address clustering. vary significantly, depending on the nature of the token. Second, while the automated process may remap tokens To conclude, DeFi is an exciting and rapidly growing new for all contract accounts, our manual analysis was limited to financialinfrastructure.However,thereisaparticularriskthat contract accounts with a significant amount. We decided to high ownership concentration and complex wrapping struc- set the threshold value at 0.1% of relevant supply. turesintroducegovernancerisks,underminetransparencyand Third, we used various data sources to verify the labeling createextremeinterdependenceaffectingprotocolrobustness. ofaddresses.Insomeunclearcases,weapproachedtheteams directly for more information. However, this information REFERENCES cannot be verified on-chain. Consequently, this is the only [1] F. Schär, “Decentralized finance: On blockchain-and smart contract- part of the study for which we had to rely on information basedfinancialmarkets,”AvailableatSSRN3571335,2020. [2] S. Conti. (2020) Defi token holder analysis - 6th aug provided by third parties. 2020. [Online]. Available: https://twitter.com/simoneconti_/status/ Further research may adopt the methods of this paper to 1291396627165569026/photo/1 analyze token characteristics in the context of governance [3] M. Gupta and P. Gupta, “Gini coefficient based wealth distribution inthebitcoinnetwork:Acasestudy,”inInternationalConferenceon models. The data could be used as a parameter for more Computing,AnalyticsandNetworks. Springer,2017,pp.192–202. realisticsimulationsandgame-theoreticalgovernancemodels. [4] U. W. Chohan, “Cryptocurrencies and inequality,” Notes on the 21st Novel metrics, such as the wrapping complexity, may be Century(CBRI),2019. [5] D.Kondor,M.Pósfai,I.Csabai,andG.Vattay,“Dotherichgetricher? useful for studies concerned with the interdependencies and an empirical analysis of the bitcoin transaction network,” PloS one, risk assessment of the DeFi landscape. Finally, the proposed vol.9,no.2,p.e86197,2014. readjustment categories may provide a good base for further [6] I. Rosu and F. Saleh, “Evolution of shares in a proof-of-stake cryp- tocurrency,”HECParisResearchPaperNo.FIN-2019-1339,2020. research on how DeFi tokens are being used and the reasons [7] E. Medvedev and the D5 team, “Ethereum etl,” 2018. [Online]. for their spectacular growth. Available:https://github.com/blockchain-etl/ethereum-etl [8] Etherscan.(2019)Etherscan.io.[Online].Available:https://etherscan.io VI. CONCLUSION [9] E.MedvedevandtheD5team.(2020)Nansen.ai.[Online].Available: https://nansen.ai In this paper, we analyze the holder distribution and [10] CoinGecko. (2020) Coingecko.com. [Online]. Available: https:// ecosystem integration for the most popular DeFi tokens. The coingecko.com paper introduces a novel method that allows us to split and [11] C.Gini,“Variabilitàemutabilità,”vamu,1912. iteratively reallocate contract account holdings over multiple ACKNOWLEDGEMENTS wrapping levels. Our data indicate that previous analyses severely overesti- The authors would like to thank Mitchell Goldberg, mated ownership concentration. However, in most cases, the John Orthwein and Victoria J. Block for proof-reading the majority of the tokens are still held by a handful of individ- manuscript.