Curation Earnings Analysis (with stats and thoughts on 50 / 50 curation system proposal)
Would you support a move to a 50 / 50 split for author and curation rewards on Steem? It's an old Steem argument but one that has come to the fore once again in recent weeks.
In this analysis I look at curation earnings from across the Steem blockchain. My aim is to provide some high level statistics that increase understanding of curation earnings and also provide some background information to aid in the current 50 / 50 debate.
From an analysis of a single week at the start of April, I look to answer the following questions:
- Does everyone earn curation rewards? What proportion of voters earn no curation rewards?
- How much do users typically earn through curation on the Steem blockchain? How skewed is the distribution of curation earnings?
- Do most users earn close to the headline 25% curation rate of their vote values?
- Do the top curation earners already gain over 100% of their vote value - the amount available from self-voting or vote-selling? How scalable is this practice?
- What about the largest accounts? Are they all clustered around 25%? Or do they actually earn less? Are any of these large accounts already gaining 50% curation?
I have based the curation analysis on data from the first week of April 2019 (April 1 - April 7 inclusive). There's nothing special about this choice. It's a typical week chosen at random.
The value of rewards on the blockchain is heavily influenced by the price of Steem. For this analysis the main metrics I have used are:
- Curation rewards in STU: These will vary considerably as the Steem price changes, although as I am mainly looking at the distribution of the earnings with this metric, the overall pattern between accounts should be similar over time.
- Curation reward in STU / vote value in STU: This should be more stable as the Steem price changes, although distortions can still arise due to timing effects.
As such it is worth bearing in mind this variability of results over time when drawing any hard conclusions.
All earnings amounts are expressed in STU throughout this analysis. Payouts in Vests, Steem and SBD are converted to STU using factors derived empirically for each hour of the analysis period.
1. Does everyone earn curation rewards? What proportion of voters earn no curation rewards?
Over the first week of April there were 61,467 accounts that voted on the Steem blockchain. Of those, 36,356 accounts, or 59%, earned no curation rewards.
The chart below shows the distribution of curation earnings for all accounts. The actual distribution is a very high peak at zero and a very long tail and is difficult to illustrate with any meaning. So instead I have separated the earnings into useful buckets.
In summary we have:
- 59% of users earning zero curation rewards
- 37% of users earning less than $1 STU in total over the week.
- 3% of users earning up to $1 STU per day, i.e. $7 STU for the week.
- 0.7% of users earning between $1 STU and $5 STU per day.
- 0.3% of users (170 accounts) earning more than $5 STU per day.
2. How much do users typically earn through curation on the Steem blockchain? How skewed is the distribution of curation earnings?
The first question is largely answered by the chart in section 1 above. For the second question, the chart below illustrates that most curation rewards are earned by a small number of accounts.
There are 429 users in the $1 STU - $5 STU bucket earning 15% of all curation rewards and 170 accounts in the $5 STU bucket earning 70% of all curation rewards. Looking at the top bucket in more detail, the top 47 accounts earn 50% of all curation rewards.
There is nothing untoward in these figures. They mainly reflect the distribution of "post-delegation" Steem Power. "Pre-delegation" SP is concentrated in the hands of a small percentage of accounts (whales and orcas). The process of delegation further concentrates Steem Power into an even smaller number of accounts (dApps and bid-bots). These latter accounts garner the majority of curation rewards but will (most likely) be passing some back to their delegators.
3. Do most users earn close to the headline 25% curation rate of their vote values?
The metric I am using in the following sections is "Curation reward in STU / vote value in STU".
- The curation reward in STU is derived from the curation reward in SP and converted to STU using factors derived empirically for each hour of the analysis period.
- The vote value in STU is derived from the rshares for each vote made and converted to STU using the same process as above.
Note that this approach takes no account of whether accounts are using their full ten votes per day efficiently. A user that only makes one vote per day and receives 90% of that vote value in curation rewards will score 90%, even though they would only score 9% if their full potential vote values were taken into account. I have used this approach as I am more interested in how reward/vote percentages for users compare to the 25% curation level than I am in the efficiency of voting.
To generate the chart below I have taken all posts and comments made in the first week of April and determined for each user (a) the aggregate curation rewards from these posts and (b) the aggregate value of votes made on these posts.
There are other ways to collect such data, e.g. by using all votes made over the week, but using posts and comments ties the vote and curation rewards together neatly (as one post means one vote and one curation reward per voter).
Each account thus has a single statistic:
"Aggregated STU curation rewards from those posts / aggregated STU values of votes made on those posts"
The chart below shows the range of these statistics across all voting users:
- The blue line shows the user count at each percentage value of the statistic.
- The orange bars show the curation rewards for users at each percentage value of the statistic.
i.e. It's effectively the same statistics weighted by user count in blue and by reward value in orange.
Reward Value (orange bars):
As would be expected, the majority of curation rewards are earned by users with "reward/vote value" statistics close to the headline curation rate of 25%. However the actual peak is at 20%. This is due to:
- Curation rewards being passed back to the reward pool from voting prior to 15 minutes - so the actual headline curation rate is lower than 25%.
- The distribution has a long tail, with a fair number of users earning in the 25%-50% range and some users earning much more. With some users capturing much more than 25% curation rewards, the bulk of rewards will be earned at a level less than 25% curation rewards.
User count (blue line):
Looking at user count there is a large collection of users with very low "reward/vote value" stats in the 0% - 5% band, although with almost no actual rewards. I assume that this is either:
- Newer users who are not worried about / informed about curation rewards and so vote early on posts and pass their curation rewards back to the reward pool.
- Small users affected by dust value thresholds (i.e. they are only getting one lucky curation reward for every ten posts voted).
- Bots doing bot things.
As an interesting aside, the continuation of the above chart shows some "waves" out past 100%:
I'm guessing that these are curation-bot curation trails, i.e. groups of users voting on content determined by high-curation reward seeking bots. However the actual volume of curation rewards out there is small. An investigation for another day.
- Although the current curation rewards system is billed as 75% / 25% author-curator, in fact the distribution of users (by curation rewards) centres around the 20% of vote value level. This will be partly from curation rewards being passed back to the reward pool from voting prior to 15 minutes - so the actual headline curation rate is lower than 25% - and partly from those users earning much more than 25% curation rewards.
4. Do the top curation earners already gain over 100% of their vote value - the amount available from self-voting or vote-selling? How scalable is this practice?
Segueing neatly into section four, we can see from the chart above (section 3) that there are users with "reward/vote value" statistics out past 100%, i.e. users that are doing better through curation than is possible through self-voting or vote selling. The chart above suggests that the volume of rewards at these levels are low, so how much is being earned by these high performers?
The chart below looks at the top 100 users ranked by the "rewards/vote values" ratio statistic. I have limited the field to users earning over $1 STU for the week because:
(a) The stats are volatile at low vote levels, partly due to edge cases.
(b) I'm looking for scalability, which requires a certain volume of rewards.
The chart shows the "rewards/vote values" statistic on the x-axis and the curation rewards for the week on the y-axis.
The chart suggests that a 100% "rewards/vote values" statistic is achievable for users looking to maximise curation rewards, with a grouping of users earning around $1-10 STU per week in curation rewards at this level.
On the right hand side of the chart there are a few users (six) earning up over 200% of vote value - but all with comparatively low total curation rewards earned.
At the top of the chart there are two users reaching $50 STU per week in the 100%-120% of vote value range. These users are at the 20k-30k SP holdings level, so mid-size dolphins (partly / largely through delegation).
- There are users earning over 100% of vote value. There are in fact thousands of accounts at lower vote levels reaching this level but less than 100 at higher curation reward levels.
- The practice of achieving 100% of vote value is scalable up to the level of individual mid-size dolphins, but only a couple of users achieve this.
- Overall the volume of curation rewards earned by users up past 100% of vote value is small. Mathematically only a small volume of rewards will be earned by users past this point in the distribution.
5. What about the largest curation earners? Are they all clustered around 25%? Or do they actually earn less? Are any of these large accounts already gaining 50% curation?
Most of the 75/25 vs 50/50 author/curation debate centres around the behaviour of large SP holders. One of the main lines the argument follows is that orcas and whales can currently earn 90% or 100% of their vote values through delegation to bid-bots, so why would they manually curate authors for a reward of close to 25% of vote value (3-4 times less reward).
Looking at these accounts, the first question is whether large accounts actually earn 25% of vote value, or do they actually earn less?
The chart below shows the ratio v rewards scatter for the top 100 accounts ranked by total curation rewards for the week. As in section 4, the chart shows the "rewards/vote values" statistic on the x-axis and the curation rewards for the week on the y-axis.
The chart shows that the largest accounts typically earn closer to 21% of vote value rather than the 25% headline curation rate. For information, the top ten accounts (those above $600 STU for the week) include six voting bots, three dApps, and a solitary whale.
The reason these accounts earn less than 25% is likely to be that they are front-run by other users who use the predictable voting patterns of the large accounts to gain high curation rewards for themselves.
However these large accounts are not too far from 25%. I would assume that this is because their votes are large enough to form the major volume of reward on many of the posts for which they vote. For example a Utopian post may have $3 STU of votes prior to the Utopian vote, and $30 STU afterwards. The $3 has a small impact on the overall curation level earned by Utopian on the post.
So are any of these large accounts already gaining 50% curation?
On the right hand side of the chart there are a few of the "smaller" top 100 accounts already earning over 50% curation. Two of these are actually voting bots. Another investigation for another day!
Overall, as with the high curation ratio earners in section 4, only a few accounts achieve these levels. The majority of the largest curation earners are clustered around 21% in the 15%-25% band.
- Large accounts are typically earning between 15% and 25% of vote value, with the average at 21% rather than the headline 25% curation rate. A doubling of the curation rewards to a 50 / 50 approach would (all other things being equal) increase this to 42%, still well short of 50% and a long way from the 90%-100% that could be obtained by large accounts from vote-selling / delegation.
6. My own personal thoughts on a change to 50/50 author / curation rewards
Reading through the debates, everyone seems to have their own view on what should be done with curation rewards. So here (briefly) is mine:
Whilst I think a change to the current curation mechanism would be beneficial, I do not think a change to 50/50 is the right approach:
- As the analysis above shows, doubling curation rewards will only bring large accounts to around 42% curation - still a long way short of the rewards available from vote-selling. Although the proposal can be considered a step in the right direction, it is a considerable (and for many, controversial) change that does not look likely to solve the problem.
- The proposed approach of 50 / 50 curation (like the current 75 / 25 system) is too inflexible. Different Steem users have different priorities and any new system should reflect this, allowing Steem to cater to all needs.
For example on the second point, some accounts may want to provide 100% of their upvote to the author, i.e. with no curation, or at a lower level of curation than the current 25%. Examples could include dApps that are trying to build their community (particularly those with delegation from Steemit, or those with funds from external parties that are dedicated to rewarding users in return for services) or "basic income" style systems.
Clearly some accounts (those currently delegating to bid-bots) would prefer much higher curation even than the proposed 50% but perhaps could be comfortable manually curating at higher curation reward levels, say 80%.
I would propose:
- Removing the non-linear element of curation - a move to "flat curation". The idea of "content discovery" does not really exist on Steem and the incentives of the current system are misaligned, discouraging voting on content as it accumulates votes.
- A completely flexible curation slider (0 - 100%) at the discretion of the voter.
As I see it, the goal for the "content creation" side of Steem is to get people to vote on the content they like best, irrespective of the timing of the vote or the existing rewards on the post. This will get the best content onto trending. Hence "flat curation".
As for the rewards, you cannot force unwilling people to give vote rewards to others - there are, and have always been, many ways to avoid this on Steem. You can only encourage them back to manual curation by providing the flexibility to distribute the level of rewards they are comfortable with. Manual, unincentivised, non-gamifiable voting is the best chance for getting the best content into the light and getting the "eyes-on" that Steem needs to support revenues and growth.
Whilst this could mean accounts using 80%, 90% or even 100% curation for their votes, this is better than the current system (or 50/50) under which the same users gain 90% or 100% of vote value through vote selling, the remaining margins flow to bid-bots and vote-buyers, and bad content ends up on trending.
Would the approach encourage current manual voters to distribute less to authors? Probably in some cases but this would be the case for all voters by a change to fixed 50/50. Under the system above there would be flexibility for the large dApps to remain at 25% curation, or even go lower, and there would potentially be more for authors from high SP voters switching from bid-bots delegations back to manual curation - or to a system of voting trails run by "trusted curators" that would mean minimal effort for passive investors.
Anyway, this is a long debate and everyone has their own view! Maybe I will write a fuller proposal at some point. Thanks for reading!
This analysis is of data from the Steem blockchain which is an open source project. In particular the analysis aims to add value to the current debate on whether the Steem blockchain should be hard forked for a change in curation rewards structure.
Tools and scripts:
I used the block.ops analysis system to produce this study. Block.ops is an open-source analysis tool designed for heavy-duty analyses of the Steem blockchain data.
You can find the repository for block.ops here:
The analysis used all the Steem blocks from the week analysed.
The study can be recreated by (once I upload the new analysis to github!):
- Loading the data for the relevant time period into block.ops.
- Using the newcuration command from the command line, for example:
$ node blockOps newcuration "2019-04-01" "2019-04-08"
Thanks for reading!