Decentralized finance (DeFi) aims to build an open financial system and enable token holders to vote on protocol changes. Compound, a leading DeFi lending protocol, uses the governance token COMP for on-chain voting. While COMP holders can vote directly, Compound also allows delegation where token holders assign their votes to other addresses.

Delegation is intended to enhance the collective participation of the community in the governance process by letting users delegate voting rights if they don't want to actively participate. However, critics argue delegation could lead to vote concentration if large token holders accrue a disproportionate amount of delegated votes. Understanding the dynamics of delegation is important for assessing the decentralization and governance health of protocols like Compound.

This analysis evaluates the impact of delegation on the distribution of voting power within Compound governance. It examines the delegation behavior across different COMP balance tiers and compares decentralization metrics before and after delegation. The results provide data-driven insights into the role of delegation in Compound's governance decentralization. By presenting a detailed study of delegation, this analysis aims to further the understanding of governance dynamics in leading DeFi protocols.

KEY FINDINGS

Data as of July 31, 2023.

1. While 83% of addresses self-delegate, 95% of the total COMP balance entered into the governance process is delegated to other addresses. This shows large holders tend to delegate while smaller holders prefer self-delegating.  See Table 1.

2. Segmenting addresses by balance tiers indicates delegation is far less common among higher balance holders. Only 5% of High Balance and 0% of Very High Balance addresses self-delegate, versus 84% of Low Balance addresses. See Table 2.

3. Despite delegation from large holders, an evaluation of balance category mix pre- and post-delegation shows that their COMP still flows to other large holders. See Tables 3-4.

4. Delegation has no material impact on the balance proportion of lower COMP holders.  Delegation does impact the relative share between Medium, High, and Very High balance holders; the Very High balance holders increase from 29% of the total balance pre-delegation to 32% post-delegation.  There is a corresponding drop in the Medium and High balance holders’ share post-delegation.  See Table 4.

5. Delegation has the potential to be a significant factor in calculating metrics such as Nakomoto coefficient, Herfindahl-Hirschman Index (HHI), and the Gini coefficient for assessing the level of concentration/decentralization in governance voting.  Based on the findings of this post and scores depicted in Table 5, changes observed in common decentralization metrics before and after delegation are moderate. This implies factors like participation rates and base COMP distribution may be greater drivers of decentralization than delegation alone.

6. The post-delegation scores in Table 5 indicate room for improvement in on-chain voting even with the current address distribution and participation.  While there are over 4,900 addresses with a positive balance post-delegation, the highest voting participation on a specific proposal was only 619 participants in Proposal 111: Risk Parameter Updates for 2 Collateral Assets executed on July 4, 2022.

7. Technical improvements in the delegation process such as fractional delegation (Franchiser Spec) could also have a significant positive impact.  This would open new opportunities for higher COMP holders to not have to solely delegate to other high balance holders; allowing flexibility to allocate a mix to a greater variety of interested persons across the spectrum of COMP holdings. While fractional delegation alone is insufficient to fully decentralize protocol control from major holders, its implementation can distribute tangible governance feedback not achievable under the current model. By allocating even modest vote percentages to smaller holders, their collective input can incorporate a genuinely wider range of opinions and insights that strengthens governance decisions.

This analysis and accompanying governance analytics dashboard examines trends in voting participation and decentralization in Compound governance from 2020-2023. Key insights include:

• Participation, measured by the number of unique voting addresses, spiked in 2022Q2 but subsequently declined in 2023 back to prior levels, signaling a need for ongoing voter engagement over time.

• Despite temporary participation growth, overall voting power exhibited minimal change, indicating new voters had limited voting power.

• Concentration of voting power rose steadily over the 2020-2023 as measured under the Herfindahl-Hirschman Index. However, the minimum voters needed for majority control was relatively stable, suggesting persistent defenses against adversarial coordination.

• Compound’s level of concentration in vote distribution has historically been between those observed for Uniswap and Aave, with Compound typically more concentrated than Uniswap and less concentrated than Aave.

Introduction

Decentralized governance is a defining feature of DeFi protocols, allowing community control through on-chain voting systems. Governance is enacted through participatory mechanisms like proposal voting, enabling users to shape policies and future development.

Monitoring governance health is crucial for protocols like Compound that rely on community control. While some power concentration is expected, excessive centralization and consolidation could compromise the promise of decentralized governance. This analysis examines the state of governance for Compound, a leading DeFi lending protocol. It utilizes data on voting and power distribution from 2020 through mid-2023 to quantify engagement and decentralization over time.

Specifically, the analysis looks at:

• Trends in voter participation - How many unique addresses are voting on proposals? Is participation consistent or variable over time?

• Concentration of voting power - Are proposals being controlled by a wide group or narrow subset of voters? Has influence consolidated or spread?

• Minimum voters for majority control - How has the minimum number of voters needed for 51% control changed? Does this indicate structural defenses against adversarial coordination have strengthened or weakened? Note: assessment of a protocol’s collusion resistance is complex and cannot be directly measured using the simple measures presented in this report, as it requires an understanding of the relationships, expectations, etc. of the various participants. However, we believe that simple measures such as the Nakamoto coefficient can still be informative.

The analysis also aims to quantify Compound's decentralization standing both longitudinally and relative to peers. Therefore, Compound's governance health is benchmarked against fellow DeFi leaders Uniswap and Aave using the same metrics.

Data Source and Definitions

Underlying voting and delegation data is sourced from Flipside. Please note that data source freshness may vary and is subject to updates based on the underlying data pipeline extraction schedules.

Voting Power: Sum of votes cast. Voting weight is determined by the number of votes the account had delegated to it at the time the proposal state became active.

Unique Voting Addresses: Number of addresses that cast a vote on a proposal. Note that if a delegate has multiple addresses delegating to it then it will still only count as one unique address.

Proposal Quarter: A proposal's assignment to a Proposal Quarter is determined by the minimum date of a vote cast for that proposal. Note that the decentralization metrics' quarterly values are weighted averages based on voting power by proposal.

Nakamoto Coefficient: based on metric originally defined in Srinivasan and Lee (2017). For this application, it is calculated as the minimum number of entities (i.e. unique voter addresses) required to control 51% or greater voting share. (Reference)

Herfindahl-Hirschman Index (HHI): measure of market concentration, most notably used by the United States Department of Justice (DOJ) & Federal Trade Commission (FTC). It is calculated by squaring the market share of each firm competing in the market and then summing the resulting numbers. For example, for a market consisting of four firms with shares of 30, 30, 20, and 20 percent, the HHI is 2,600 (30^2 + 30^2 + 20^2 + 20^2 = 2,600). The HHI can range from 1 (least concentrated) to 10,000 (most concentrated). The DOJ and other governmental agencies historically consider markets under 1,500 to be low concentration, markets between 1,500 to 2,500 to be moderately concentrated, and markets in excess of 2,500 to be highly concentrated.(Reference)

Gini Coefficient: a measure of equality within a community, commonly used to quantify income inequality. A low/high value represents a more equal/unequal community. It is calculated as the ratio of the area between the perfect equality (dotted) line and the Lorenz Curve divided by total area under the perfect equality (dotted) line. (Reference)

Key Findings

The accompanying dashboard presents summaries on both an individual proposal and quarterly basis. This analysis focuses on the quarterly aggregated governance metrics to evaluate overall trends and developments in Compound's governance voting.

Unique Voters

Unique voter address participation increased sharply in 2022Q2 compared to prior periods, indicating broader voter engagement. However, participation subsequently dropped back to previous levels in 2023.

New vs Returning Voters

The 2022Q2 spike was driven by an influx of nearly 1,100 new voters, making up over 90% of participants that quarter. Short-lived participation growth highlights the need for ongoing engagement. The temporary surge in participation that was not sustained highlights an opportunity to promote more consistent voter involvement over time. The drop in governance participation from its spike starting in 2022Q2 follows the broader bear market that accelerated in 2022 and through 2023. Some decline in engagement is expected given weaker market conditions and sentiment. However, there remains an opportunity to engage a broader base of users in governance to promote consistent participation through market cycles.

Voting Power Exercised

The participation spike did not translate to voting power. Despite more unique addresses participating in 2022Q2, the total voting power exercised remained stable over the same period.

Herfindahl-Hirschman Index (HHI)

The Herfindahl-Hirschman Index (HHI) shows voting power becoming increasingly concentrated from 2020-2023. The HHI rose from ~1,400s in 2020 to exceed 2,000 by 2022. This has continued to increase in 2023. Continued monitoring is warranted.

Nakamoto Coefficient

The Nakamoto coefficient has remained relatively stable, hovering between 2.5-3.5 from 2020-2023. This range indicates that while some consolidation of voting power has occurred, it has not been as marked of a change as in HHI. Ongoing tracking of the Nakamoto Coefficient is important to quickly identify any significant shifts that could impact  governance integrity.

Gini Coefficient

The Gini coefficient observed in governance voting has oscillated considerably throughout 2020-2023 within a range of 0.66 - 0.96. Such high Gini coefficient scores often indicate significant levels of concentration. Yet, there are also limitations from relying too heavily on the use of a Gini coefficient as the main signal of the level of inequality amongst a community, as further described in a July 2021 article by Vitalik Buterin. Further, given the way the Gini coefficient is calculated, an increase in voter participation amongst token holders with relatively smaller vote share will lead to an increase in the Gini coefficient. Therefore, a positive engagement with the broader community may actually appear in the data as an increase in the Gini score. Overall, the Gini coefficient alone is insufficient as a sole measure of decentralization---it should continue to be used but complemented with other measures.

Recommendations

Based on these findings, the following steps are recommended:

• Broaden engagement through educational campaigns and governance tutorials to ease onboarding barriers for new users.

• Analyze voter drop-off and churn to optimize retention after initial participation, promoting an engaged core community.

• Assess outreach, UX, and understandability of the governance portal. Benchmark against high-participation protocols like Uniswap.

• Monitor changes in inequality metrics quickly and respond to concerning consolidations in voting power.

• Increase transparency through regular reporting on governance health metrics for the community.

• Evaluate perceptions of the community through polls and surveys to identify areas of concern or disconnect from current governance.

• Explore incentives like rewards programs to encourage long-term participation and decentralization-promoting behaviors.

• Review delegator models and explore possibility of a fractional delegation mechanism, eg Franchiser.

Ongoing quantification of governance metrics will enable Compound to track progress over time. But most importantly, these indicators should inform action to uphold Compound's commitment to decentralized, community-driven governance.

Conclusion

This analysis explored the evolving state of decentralization in Compound governance using data-driven assessment of participation and power distribution.While some trends such as declining engagement and increasing concentration warrant increased monitoring, Compound has generally preserved resistance to centralization thus far. With targeted initiatives to broaden involvement and voting power equality, Compound can reinforce its standing as a leading example of community-governed DeFi protocols.

Grant Acknowledgement

This dashboard and associated analysis was made possible by support from Compound Grants Program 2.0

• August 2022 (initial article), including an introduction and definition of the metrics analyzed.

• September 2022 (update).  Review of Proposal 2.24 Create the Uniswap Foundation

• October 2022 (not applicable).  There were no on-chain proposals completed during the prior month.

• November 2022 (update). Review of Proposal 2.25 Deploy Uniswap V3 on zkSync

• December 2022 (not applicable).  There were no on-chain proposals completed during the prior month.

We also encourage readers to view the accompanying dashboard that can be accessed at https://signalcorps.shinyapps.io/Uniswap_On_Off_Chain_Governance/ .

JANUARY 2023 UPDATE

Since our last on-/off-chain analysis released in November, there has been one governance proposal that has had an on-chain vote:  Proposal 2.29: Fix the Cross Chain Messaging Bridge on Arbitrum.  Proposal 2.29 was executed with an overwhelming and near unanimous vote with 55,568,689 UNI in favor and only 36 UNI against.

Please see Tables 2 and 3 below for the respective scores and rankings for each of the six metrics analyzed, included two engagement-based metrics (voting participants and voting power) and decentralization-based metrics (Nakamoto coefficient, HHI, Gini coefficient, Shannon Diversity Index).

Proposal 2.29’s on-chain vote had the fourth highest number of unique voting participant addresses (1,821 participants) reported in an Uniswap on-chain proposal. Proposal 2.29 posted the highest voting participant counts in the Temperature and Consensus off-chain phases (5,238 and 6,741 participants respectively).  This is building on the previous all-time high voter participants counts achieved in Proposal 2.25 (2,457 and 3,807 participants respectively).

Footer 1:  Proposal 2.29 posted a high Gini coefficient at 0.995, similar to previous proposals that have generally ranged between 0.995 – 0.998.  Counterintuitively, part of this is due to the high participation, as the Gini coefficient is a measure of equality.  Therefore, as more relatively low UNI holders participate, then this will increase the Gini coefficient.  A contrasting example is Proposal 2.09 that had the lowest on-chain Gini coefficient of 0.766 but only had 41 voting participants.  This further highlights how the Gini coefficient is important for understanding the underlying base distribution among all UNI token holders, but it can be a poor signal when examining vote distribution and engagement.

• August 2022 (initial article), including an introduction and definition of the metrics analyzed.

• September 2022 (update).  Review of Proposal 2.24 Create the Uniswap Foundation

• October 2022 (not applicable).  There were no on-chain proposals completed during the prior month.

We also encourage readers to view the accompanying dashboard that can be accessed at https://signalcorps.shinyapps.io/Uniswap_On_Off_Chain_Governance/ .

November 2022 UPDATE

Since our last on-/off-chain analysis released in September, there has been one governance proposal that has had an on-chain vote:  Proposal 2.25: Deploy Uniswap V3 on zkSync.  Proposal 2.25 was executed with an overwhelming and near unanimous vote with 72,421,473 UNI in favor and only 120 UNI against.  Proposal 2.25’s on-chain vote had the highest number of unique voting participant addresses (5,832 participants) reported in an Uniswap on-chain proposal; building on the previous all-time high voter participant count (2,450) achieved in the previous proposal, Proposal 2.24: Create the Uniswap Foundation.

Please see Tables 2 and 3 below for the respective scores and rankings for each of the six metrics analyzed, included two engagement-based metrics (voting participants and voting power) and decentralization-based metrics (Nakamoto coefficient, HHI, Gini coefficient, Shannon Diversity Index).  

In addition to the high on-chain participation described above, Proposal 2.25 also posted the highest voting participant counts in the Temperature and Consensus off-chain phases (2,457 and 3,807 participants respectively).  Again, this is building on the previous all-time high voter participants counts achieved in Proposal 2.24 (1,860 and 2,199 participants respectively).  Therefore, Uniswap has managed to sustain the recent surge in governance engagement.

With respect to the decentralization-based metrics, Proposal 2.25 performed especially well in the off-chain phases conducted on Snapshot relative to previous Uniswap proposals. Proposal 2.25 ranked in the top four best scores for each of the metrics reported except the Gini coefficient (see footer 1).  Proposal 2.25 performed modestly on-chain in terms of absolute rank, but its scores were comparable to previous proposals. For example, Proposal 2.25’s on-chain Herfindahl-Hirschman Index (HHI) score was 994.  The HHI is on a scale from 1 to 10,000 with lower scores signifying less concentration.  Although Proposal 2.25’s on-chain HHI of 994 ranked twelve of eighteen amongst on-chain proposals, it should be noted that its HHI was lower than the on-chain average-to-date (1,271) and only slightly above the on-chain median (975.5). 

Note: lower rankings (e.g. 1, 2, 3) are assigned to scores indicating greater decentralization, such as lower values for HHI and Gini Coefficient and higher values for Voting Participants, Nakamoto Coefficient, and Shannon Diversity Index.  Voting Power is neutral but is ranked in the table above in descending order.

Footer 1:  Proposal 2.25 posted a high Gini coefficient at 0.999, similar to previous proposals that have generally ranged between 0.995 – 0.998.  Counterintuitively, part of this is due to the high participation, as the Gini coefficient is a measure of equality.  Therefore, as more relatively lower balance UNI holders participate, then this will increase the Gini coefficient.  A contrasting example is Proposal 2.09 that had the lowest on-chain Gini coefficient of 0.766 but only had 41 voting participants.  This further highlights how the Gini coefficient is important for understanding the underlying base distribution among all UNI token holders, but it can be a poor signal when examining vote distribution and engagement.  

This post explores Uniswap delegation data to assess whether delegation is a contributing factor to the levels of concentration/decentralization observed in on-/off-chain voting on Uniswap governance proposals.

BALANCE CATEGORY CLASSIFICATION

The following categorical definitions are used in this post to describe addresses with different UNI balances. For example, Whales are defined as those addresses with one million or greater UNI, Dolphins as those with balances between 100k – 1M, etc.

KEY FINDINGS

Data as of September 10, 2022.

1. A super majority (95%) of addresses self-delegate. Yet, the vast majority (91%) of balances are delegated to other addresses. See Table 1.

2. Self-delegation percentages segmented by balance category demonstrate that large UNI holders (e.g. dolphins, whales) often delegate to others while lower UNI holders typically self-delegate. See Table 2.

3. An evaluation of balance category mix pre- and post-delegation shows that although large UNI holders delegated to others, it is to other large UNI holders. See Tables 3-4.

4. Delegation has no material impact on the balance proportion of lower UNI holders.

   1. Delegation does impact the relative share between dolphins and whales, with whales increasing from 89% of the total balance pre-delegation to 95% post-delegation. See Table 4.

   2. There is a corresponding drop in dolphins’ share from 10% pre-delegation to 4.4% post-delegation.

5. In our prior analysis, delegation was hypothesized to be a potential significant factor in calculating metrics such as Nakomoto coefficient, Herfindahl-Hirschman Index (HHI), and the Gini coefficient for assessing the level of concentration/decentralization in Uniswap’s on-/chain governance voting.

   1. Based on the findings of this post and scores depicted in Table 5, delegation is deemed to only have a moderate impact on observed decentralization metrics.

   2. Depending on the metric analyzed, the primary contributing factors to the observed decentralized scores in on-/off-chain voting is due to the base underlying distribution and/or relatively low governance participation, not delegation.

6. The post-delegation scores in Table 5 show an opportunity to improve on-/off-chain voting decentralization scores even if assuming the existing distribution of addresses that have entered the governance process by either delegating or self-delegating remains constant.

   1. While there are over 10,000 addresses with a positive balance post-delegation, the highest vote participant count in on-chain governance has been the recently successful Proposal 2.24: Create the Uniswap Foundation with 2,450 unique addresses.

   2. The relatively high engagement in Proposal 2.24 is encouraging, but there is an opportunity for even further engagement which would be expected to further improve the Nakamoto coefficient and HHI scores.

7. Technical improvements in the delegation process such as fractional delegation (e.g. Franchiser) could also have a significant positive impact. This would open new opportunities for higher UNI holders to not have to solely delegate to other dolphins and whales; allowing flexibility to allocate a mix to a greater variety of interested persons across the spectrum of UNI holdings.

We also encourage readers to view the accompanying dashboard that can be accessed at https://signalcorps.shinyapps.io/Uniswap_On_Off_Chain_Governance/ .

September 2022 Update

Since our initial on-/off-chain analysis released in August, there has been one governance proposal that has had an on-chain vote: Proposal 2.24: Create the Uniswap Foundation.

Proposal 2.24 was executed with an overwhelming and near unanimous vote with 86,298,595 UNI in favor and only 770 UNI against. This is an impressive accomplishment considering Proposal 2.24’s on-chain vote had the highest number of unique voting participant addresses (2,450 participants) reported in an Uniswap on-chain proposal, and the third largest on-chain voting power exercised.

In addition to the high on-chain participation, Proposal 2.24 also posted the highest voting participant counts in the Temperature and Consensus off-chain phases. Thus, making it one of the most engaged proposals to date.

Overall, Proposal 2.24 had the best (i.e. most distributed) scores reported for Nakamoto coefficient, HHI, and Shannon diversity index in both the on-chain and temperature phases, and it had relatively good scores in the consensus phase. Please see Tables 2 and 3 for the respective scores and rankings for each metric analyzed.

The Gini coefficient continued to be high for Proposal 2.24, ranging from 0.995 – 0.998. Counterintuitively, part of this is due to the high participation, as the Gini coefficient is a measure of equality. Therefore, as more relatively low UNI holders participate, then this will increase the Gini coefficient.

A contrasting example is Proposal 2.09 that had the lowest on-chain Gini coefficient of 0.766 but only had 41 voting participants. This further highlights how the Gini coefficient is important for understanding the underlying base distribution among all UNI token holders, but it can be a poor signal when examining vote distribution and engagement.

Again, as the other non-Gini metrics demonstrate, Proposal 2.24 has been one of the most distributed and engaged Uniswap governance proposals at this time. While there continues to be areas of opportunity to improve these scores, this recent activity is encouraging. The topic of the proposal itself was what garnered so much attention and engagement for Proposal 2.24, but to the extent that there were new and/or additional outreach conducted, then it should be evaluated if those initiatives could be sustained for future governance activities.

This analysis is a supplement to an accompanying dashboard analyzing Uniswap on-chain and off-chain (Snapshot) voting participation and distribution data. A diverse set of metrics were calculated to examine the degree of decentralization exhibited in historical Uniswap's governance proposals across the different phases, including:

Phase 1: Temperature Check — Discourse/Snapshot (Off-chain)

Phase 2: Consensus Check — Discourse/Snapshot (Off-chain)

Phase 3: Governance Proposal — Governance Portal (On-chain)

The accompanying dashboard, including summary data tables and visualizations, can be accessed at https://signalcorps.shinyapps.io/Uniswap_On_Off_Chain_Governance/.

METRICS DEFINITIONS

Voting Participants: count of unique voting addresses.

Voting Power: sum of votes cast.

Nakamoto Coefficient: based on metric originally defined in Srinivasan and Lee (2017). For this application, it is calculated as the minimum number of entities (i.e. unique voter addresses) required to control 51% or greater voting share. (Reference)

Herfindahl-Hirschman Index (HHI): measure of market concentration, most notably used by the United States Department of Justice (DOJ) & Federal Trade Commission (FTC). It is calculated by squaring the market share of each firm competing in the market and then summing the resulting numbers. For example, for a market consisting of four firms with shares of 30, 30, 20, and 20 percent, the HHI is 2,600 (30^2 + 30^2 + 20^2 + 20^2 = 2,600). The HHI can range from 1 (least concentrated) to 10,000 (most concentrated). The DOJ and other governmental agencies historically consider markets under 1,500 to be low concentration, markets between 1,500 to 2,500 to be moderately concentrated, and markets in excess of 2,500 to be highly concentrated. (Reference).)

Gini Coefficient: a measure of equality within a community, commonly used to quantify income inequality. A low/high value represents a more equal/unequal community. It is calculated as the ratio of the area between the perfect equality (dotted) line and the Lorenz Curve divided by total area under the perfect equality (dotted) line. (Reference)

Shannon Diversity Index: a popular measure used in ecology to quantify the diversity of species, often by species' respective population count, within a community. For this application, it is used to quantify the diversity of voting weight of participants. A higher value represents a more diverse and equitable community. The Shannon diversity index is calculated by taking the sum of each voter’s proportional voting power (Pi) multiplied by the natural log of Pi and then multiplying by -1. (Reference)

KEY FINDINGS

Data as of August 6, 2022.

Readers are encouraged to view the accompanying dashboard when reviewing the key findings presented below. The dashboard is available at https://signalcorps.shinyapps.io/Uniswap_On_Off_Chain_Governance/

Voting Participation and Voting Power

1. On-chain voting participation was sparse in the initial proposals (0.1 – 2.10), with a range of voting participants from a low of 41 in Proposal 2.09 to 340 participants in Proposal 0.2. There was a reinvigoration in on-chain voting participation beginning in Proposal 2.11 with 1,965 participants that has generally sustained through to Proposal 2.23.

2. There was a marked increase in voting participation on Snapshot (i.e. Temperature Check and Consensus Check) beginning in Proposal 2.13. Prior to Proposal 2.13, the number of voting participants typically ranged from 300-500 whereas it has been consistently between 1,000 – 1,500 from Proposals 2.13 – 2.23.

3. In a vast majority of proposals, the off-chain voting on Snapshot had higher voting participation compared to the on-chain governance vote. For example, in Proposals 2.11 – 2.23, on-chain voting participation was on average 85% of Snapshot voting. This is likely due to Snapshot not requiring gas fees, unlike on-chain votes.

4. While the participant count was higher in off-chain, voting power exercised was significantly higher in on-chain voting. For Proposals 2.11 – 2.23, the average on-chain voting power exercised was 58,268,661 while Snapshot was on average 8,901,470. This implies 6.5X on-chain to Snapshot ratio.

The remainder of this analysis will focus on the findings for the recent Proposals 2.11 – 2.23, as there was a noticeable increase in the voting participation observed.

Nakamoto Coefficient

1. The Nakamoto coefficient has been 1-2 for the Temperature and Consensus phases on Snapshot since Proposal 2.11.

2. The Nakamoto coefficient has been 5 for the majority of on-chain votes since Proposal 2.11.

3. The lower-participation yet higher voting-power on-chain statistics compared with the Nakamoto coefficients suggests that a considerable proportion of Whales abstain from the off-chain phases on Snapshot but then participate in the formal on-chain governance vote.

Gini Coefficient

1. The Gini coefficient observed in governance voting has been between 0.980 and 0.999 for both on-chain and off-chain. Such high Gini coefficient scores often indicate significant levels of concentration. However, a limitation of this analysis is it has not adjusted for the role that delegation may play in contributing to observed levels of concentration. Even under certain variations of robust decentralized governance, there could be high observed Gini coefficients. For example, a proposal could have a Gini coefficient of 0.99 if it had 1,000 participants comprised of 10 main delegates (representing thousands or millions who have delegated their vote) and 990 individuals with smaller amounts individually voting on this particular proposal.

2. The underlying level of concentration among UNI holders must be acknowledged. Yet, there are also significant limitations from relying too heavily on the use of a Gini coefficient as the main signal of the level of inequality amongst a community, as further described in a July 2021 article by Vitalik Buterin. Overall, the Gini coefficient alone is insufficient as a sole measure of decentralization---it should continue to be used but complemented with other measures.

Herfindahl-Hirschman Index (HHI)

As stated in the Metrics Definitions section, the United States Department of Justice and other governmental agencies historically consider markets with a HHI under 1,500 to be low concentration, markets between 1,500 to 2,500 to be moderately concentrated, and markets in excess of 2,500 to be highly concentrated.

1. The Snapshot votes in the Temperature and Consensus phases have HHI scores often at or in excess of 2,500. The lone exception since Proposal 2.11 was [Temperature] Proposal 2.22 with a moderate-to-high HHI of 2,399.

2. The HHI scores for on-chain are significantly lower with the typical score being between 850 to 1,000. The highest HHI since Proposal 2.11 was 1,511 in Proposal 2.14.

3. If adopting the DOJ’s historical score interpretation, this analysis implies relatively high concentration during the Snapshot phases and low concentration in the formal on-chain governance phase.

4. These findings compliment the Nakamoto coefficient, as the Snapshot phases had a Nakamoto coefficient of 1-2 paired with HHI in excess of 2,500, while the on-chain governance votes had a Nakamoto coefficient typically at 5 with HHI between 850-1,000.

5. The HHI suggests there is potential balance of power between the largest UNI holders without one holding dominant control. Further, the HHI illustrates that the balance of power is greater than simply the Nakamoto coefficient of 5. If under perfect equality of 5 market participants, then this would imply a HHI of 2,000. Yet, the measured on-chain HHI we observe within UNI governance is approximately 1,000.

Shannon Diversity Index

1. The Shannon Diversity Index scores mirror the HHI scores. There is an inverse relationship present, so lower Shannon scores are correlated with higher HHI scores---indicative of higher levels of concentration.

2. A scatter plot presented below in Figure 1 demonstrates that the HHI and Shannon indices are redundant. While future iterations of this analysis will continue to measure the Shannon score, our general recommendation is to use the HHI in place of Shannon. There is minimal marginal signal gained from the additional measure. The HHI also has the benefit of enhanced interpretability due to the defined range of 1 – 10,000 and the DOJ’s precedent of concentration thresholds.

TECHNICAL NOTES

Only proposals with an on-chain vote were included in the scope of this analysis.  Therefore ‘canceled’ on-chain proposals were also excluded.

On-chain data sourced from the ethereum.events_emitted table on Flipside Crypto’s Velocity Platform. (Link to Flipside)

Off-chain Snapshot data sourced from the Snapshot GraphQL API (Link to GitHub).

Additional information and code can be found at GitHub.

ACKNOWLEDGEMENT

This analysis is supported by grant funding from the Uni Grants Program.