Who on Steem is best at getting curation rewards? Big upvotes come with big curation rewards; but the nonlinear nature of the awards means that some accounts may be more efficient at converting their upvotes into rewards.
Methodology
I looked at the top 10,767 accounts by wealth: the top 3% in my earlier studies, those with at least 650 STEEM in asset value. For each of them, I summed their total curation rewards from 2018-07-23 through today (2018-08-01), and then collected the corresponding 'rshares' on each of the articles for which they had received awards.
Using this, I built two curation metrics: "dust efficiency" and "vests efficiency".
Dust efficiency is defined as number of rewards, divided by number of upvotes. An upvote too small to earn curation rewards hurts dust efficiency.
Vests efficiency is defined as total amount of rewards (in VESTS), divided by total rshares uses in upvotes. The rshares are divided by a factor of 10^9 to bring this into a convenient range, so it is technically "VESTS per billion rshares."
Of the 10767 accounts studied, only 5701 received any curation rewards during this period.
Raw Curation Value
The top 20 curators, by value, in my data set:
| user | num rewards | vests received | dust efficiency | vest efficiency |
|---|---|---|---|---|
| therising | 675 | 1918745.689 | 0.988 | 0.544 |
| kpine | 364 | 1973826.930 | 0.995 | 0.681 |
| booster | 1475 | 2019827.564 | 0.998 | 0.617 |
| sweetsssj | 136 | 2069922.727 | 0.993 | 0.584 |
| steempress-io | 1687 | 2074682.404 | 0.984 | 0.584 |
| thejohalfiles | 133 | 2127022.050 | 0.957 | 0.488 |
| minnowbooster | 294 | 2130601.613 | 0.987 | 0.541 |
| promobot | 311 | 2320016.658 | 0.969 | 0.753 |
| dsound | 756 | 2376826.386 | 0.982 | 0.631 |
| hendrikdegrote | 2450 | 2511370.069 | 0.992 | 0.541 |
| rocky1 | 945 | 2755072.409 | 0.984 | 0.508 |
| upme | 1032 | 3012232.106 | 1.005 | 0.545 |
| steemhunt | 4690 | 3152758.664 | 0.987 | 0.584 |
| buildawhale | 849 | 4092984.864 | 0.992 | 0.606 |
| dlive | 1623 | 4932400.880 | 0.996 | 0.651 |
| dtube | 542 | 5046617.935 | 0.993 | 0.661 |
| postpromoter | 1137 | 5132801.308 | 0.988 | 0.471 |
| appreciator | 1098 | 5694257.511 | 1.010 | 0.543 |
| smartsteem | 1071 | 5769486.728 | 0.970 | 0.551 |
| utopian-io | 560 | 6931529.265 | 0.981 | 0.508 |
Everybody on this list usually gets curation rewards when they upvote (though see below, there seems to be a source of error here.) But the "vest efficiency" is about 0.5-0.7; for contributing a billion rshares worth of upvote, these top accounts receive only about 0.5-0.7 VESTS back.
Curation Efficiency
Let's restrict our attention to those with at least 10 rewards. The top curators by vest efficiency are:
| user | num rewards | vests | dust efficiency | vest efficiency |
|---|---|---|---|---|
| miniature-tiger | 76 | 24391.271 | 0.938 | 3.083 |
| opaulo | 762 | 200929.279 | 0.992 | 3.118 |
| ghasemkiani | 688 | 201588.688 | 0.997 | 3.152 |
| imjohnnymills | 54 | 4974.868 | 0.857 | 3.217 |
| an0nkn0wledge | 12 | 1757.422 | 1.000 | 3.391 |
| jaff8 | 194 | 20294.389 | 0.289 | 3.424 |
| grantcardone | 10 | 1729.487 | 0.769 | 3.508 |
| stylo | 55 | 4450.000 | 1.000 | 3.549 |
| lordkingpotato | 125 | 11471.883 | 0.969 | 3.582 |
| bringolo | 135 | 8976.318 | 0.978 | 3.687 |
| teo-nyx | 299 | 29721.259 | 0.977 | 3.831 |
| mwfiae | 559 | 14449.094 | 0.944 | 3.858 |
| cryptocoinkb | 94 | 16167.039 | 0.940 | 4.010 |
| statsexpert | 143 | 1222.443 | 0.745 | 4.011 |
| dick.sledge | 1099 | 42667.580 | 0.999 | 4.020 |
| jacobts | 40 | 1226.437 | 0.952 | 4.237 |
| mys | 900 | 33216.789 | 0.865 | 4.366 |
| iabuse | 226 | 25563.027 | 0.962 | 5.948 |
| naegling11 | 823 | 8492.055 | 0.694 | 6.055 |
| cheneats | 4496 | 15250.604 | 0.851 | 13.278 |
The median for vest efficiency is 0.560 and the mean is 0.648. The distribution of vest efficiency over all accounts in my sample that received curation awards is shown below:
The users above are extracting the most curation value for their upvotes; although their total income is less, they are able to turn their Steem power into rewards in a way that is about 5x more efficient than the average account in this data set, or even the largest curators by value.
However, the total amount of rewards they earn is small. The largest reward on the list above goes to 'opaulo', who is ranked 107th by amount of curation rewards, and about 6700 by total wealth.
It is not clear whether these rewards are the result of consistent success (in which case a winning investment strategy would be to copy their upvotes) or due to a few lucky upvotes. The large number of upvotes performed by 'opaulo', 'ghasemkiani', and 'cheneats' suggests there is some skill involved.
It is striking that some of the accounts on this list achieved good efficiency even though a large percentage of their upvotes didn't pay off at all: 'cheneats' got paid only on 85% of their upvotes. (Looking at his recent curation rewards, I see that they are virtually all small, mostly 0.001 SP --- perhaps he is strategically upvoting posts that put him just over a threshold for a curation reward, but this is just speculation.)
Success by the efficiency metric does not mean these accounts are optimally using their upvotes for maximum income; particularly those with small numbers of rewards may not be upvoting enough to use all of their voting power.
Methodological problems and sources of error
I did not find all articles for which my population had "dust" upvotes that paid no curation. Also, the endpoint of the data collection was not well defined. I had underestimated the time it would take to walk all 10,000 curation reward histories, so I had assumed this effect would be small. But, it ended up running overnight so users who were visited later might get curation credit for articles that paid during that time period, while users visited earlier did not. However, when I collect the rshares (upvote worth), it covers all users who upvoted that article.
The data above shows the opposite problem, though; some users have more rewards than upvotes. It may be that the Steem API did not return all upvotes for some pieces of content.