As a rule of thumb, the optimal voting time is before 5 minutes
First, let's quickly review the different aspects of voting. (As I understand them. Please correct me if there are any misunderstandings.)
Time-based decays
| Time | Rewards |
|---|---|
| 0-5 minutes | Some curation rewards are returned to the reward pool. At time 0, 100% of rewards are returned, at time 5 minutes, 100% of rewards are kept. In between, the percentage that the curator keeps increases in a straight line. Basically, for every three seconds that pass, the curator keeps 1 more percentage point in rewards. This is referred to as a "reverse auction", and it was intended to level the playing field between human and bot voters. |
| 5 minutes to 61/2 days | The curator keeps 100% of rewards. Let's call this "normal voting". |
| 61/2 days - 7 days | The full value of the vote starts to degrade to 0 at a rate of 81/3% per hour. This is intended to give curators time to guard against payouts to abusive content (downvotes retain full power). Let's call it late voting. |
rshare damping
Votes are damped by the rewards curve, according to the rule: ( ( rshares + s ) * ( rshares + s ) - s * s ) / ( rshares + 4 * s ), where "s" is 2,000,000,000,000. Near the left side of the curve, the vote is reduced to a fraction of its starting value. Near the right side of the curve, as the stake gets bigger, the value is nearly (but not quite) unchanged:
Here it is before converging:
And here it is after (nearly) converging:
Here is the shape of the marginal change curve:
Thus, the marginal benefits of increasing account size increase quickly near the beginning of the curve, but level off as the account size grows.
The goal here was to discourage large stakeholders from voting with multiple accounts, since vote-splitting for large accounts would be deboosted. This was intended to make it easier to find and counter the worst forms of abuse.
| This section was corrected. See the comments for details. Thanks to @danmaruschak. Fortunately, this section wasn't needed for the argument about voting time. |
Boosts for content discovery
Rewards are distributed using a square-root rule, so that earlier votes receive a larger share of the final rewards pool than later votes. The idea here is that earlier voters are taking a bigger risk, since the post might not get additional votes, so the higher payout encourages them to discover content. That curve is shaped like this (early votes on the left, late votes on the right):
So here's the argument
This argument comes from @biophil, If you're voting open-loop, it's optimal to vote earlier than 30 minutes. The rules have changed, but the logic is the same.
If you combine the reverse auction with the boost for content discovery, you arrive at the following reasoning:
- The optimal time is no later than 5 minutes, when I get to keep 100% of my rewards. Voting at 5 minutes assures that no other voters will get in line in front of me.
- On some posts, there will be an optimal time before 5 minutes, where I give up some rewards to the reverse auction, but I get them back - and more - from the content discovery boost.
- If we average them all together, the optimal voting time (knowing nothing else about the post) must be earlier than its max value - 5 minutes.
The hard question is to decide how much earlier😉.
Of course, this assumes that we know nothing else about the post. As soon as we start considering factors like post quality and the strength of someone's follower network, the reasoning becomes more complicated.
FTR, this post was prompted by a post from @philhughes, here and a comment thread with SC02 and steem.botto, here.
Thank you for your time and attention.
As a general rule, I up-vote comments that demonstrate "proof of reading".
Steve Palmer is an IT professional with three decades of professional experience in data communications and information systems. He holds a bachelor's degree in mathematics, a master's degree in computer science, and a master's degree in information systems and technology management. He has been awarded 3 US patents.
Pixabay license, source
Reminder
Visit the /promoted page and #burnsteem25 to support the inflation-fighters who are helping to enable decentralized regulation of Steem token supply growth.
I think it makes more sense to think if it as scaling down rather than amplifying (the amplification is generally less than 1.0).
This was the stated goal, but 1) it wasn't at all clear that that kind of abuse was happening, and 2) it happened as part of a hardfork when people were trying to convince some big accounts to get out of the voting-bot business and switch to a "curation" business model, so the real goal may have just been to lower the impact of small votes and cause bigger votes to dominate even more than they already did so that that big accounts delivering big votes would have a more viable business model.
When I first read your post (the first version), I wondered why you were using the reward formulas to determine the ideal time to vote.
Your dialogue with danmaruschak basically confirmed this. I have already tried to visualise the reward calculation in some way. Unfortunately, this is not very clear because of the large constant
STEEM_CONTENT_CONSTANT_HF21.Showing the changes of rshares after each vote would be very interesting. I see you tried that in your comment. I can't quite understand it yet.
To the question in your post, however, I would say that in principle it doesn't matter when you vote, but who votes after you. Above all, it is an advantage if very strong accounts vote after you. If nobody voted after you, your reward share would always be the same regardless of the voting time.
Maybe we can find a way to visualise this somehow.
For purposes of the reverse auction, the "time" of voting matters. For purposes of the curation reward curve, the "order" of voting matters. What @biophil demonstrated, and I parroted here, is that when you combine those two factors the optimal time (on average) has to be some time before the reverse auction ends.
It's been a good clarifying conversation with @danmaruschak, and I'm glad that we had it, but most of that isn't really relevant to the argument about optimal voting time (and neither is the late voting time, which I also included). I only included the 2nd section so that the rules would all be documented in the same place. (and I got it wrong at first, so maybe it would've been better to leave it out😉.)
Interesting analysis! I suppose I "sort of knew" some of this, and experimented (manually) with the timing of front running certain bidbots, back when they were widely used.
The perfectly timed vote is really more of an art than a science, at least in the fine tuning of it. Each author, their major followers and the type of content witin their blog becomes a variable...
Exactly. In the end, the effective curator is trying to predict human behavior. That's always going to be challenging and error-prone. But, I think it's important to understand that we don't necessarily have to wait for the 5-minute mark in order to get good results.
The balance between early voting and content discovery rewards is fascinating, especially the balance between reverse auction mechanics and content discovery boosts, provides valuable clarity. It's interesting how early voting can outweigh the risks when combined with these factors. The challenge is fine-tuning timing, considering post quality and potential engagement.
Upvoted. Thank You for sending some of your rewards to @null. It will make Steem stronger.
This post has been featured in the latest edition of Steem News...