Topic: [IICO] Kleros’ IICO Analysis (Read 109 times)

Kleros’ IICO Analysis

Understanding user behavior in the first Interactive Coin Offering…

Detailed article with pictures here.

We at Kleros recently concluded the world’s first Interactive Initial Coin Offering (IICO). This model, proposed by Jason Teutsch and Vitalik Buterin, attempts to improve on the token sale models previously available, particularly by giving would-be contributors a more fined tuned control over how they want to handle the trade-off between the certainty with which they participate in the sale and the certainty they have over the valuation determined by the sale (so that contributors can refuse to have their contribution diluted beyond a certain point).

This is done by allowing bids that are automatically withdrawn if the sale raises more than a personal cap that the contributor specifies in advance. Additionally, this model allows for bids to be manually withdrawn (subject to partial lockups that increase over time to avoid blackout attacks).

So, now we can see how the Kleros IICO unfolded, what effects this new model seemed to have in practice, and whether the IICO model achieved its goals in our case.

To begin with a broad overview: 5793.5880977813 ETH was raised in exchange for 160,000,000 pinakion tokens (PNK) — which will ultimately represent 16% of the total supply of PNK. There were 307 total bids from 228 distinct addresses including bids manually and automatic withdrawn.


Certainty of Participation
One of the advantages of the IICO model is that the sale can stay open for a long period while a price is being discovered while still providing contributors guarantees on how much dilution they are willing to accept. This option is not available in traditional uncapped sales. Thus, the gas wars to be included in a capped sale that winds up only lasting for a block or two can be avoided.

The Kleros IICO was open for two months; during the first month contributors were offered a 20% bonus and could manually withdraw without any partial lock-up. After the first month, the percentage of the bonus linearly declined until the end of the sale and the percentage of partially withdrawn bids subject to partial lock-ups increased. Ideally, this model gave the time for people to leisurely determine if they wanted to contribute.

In practice, contributions were significantly concentrated around 1) the first day of the sale 2) the last day of the full bonus and 3) the last day of the sale.

On the first day of the sale, 53 bids were made that ultimately contributed 218 ETH. On the last day of the full bonus, 68 bids were made that ultimately contributed 452 ETH. On the last day of the sale, 27 bids were made, but some of the largest bids were placed in this time and these bids contributed a total of 2209 ETH. Included in these 27 bids are 11 successful bids made in the last hour of the IICO.

Note that there were two addresses that attempted to make contributions totalling 108 ETH on the last day of the sale and were unable to get their transactions included before the close despite paying gas prices of up to 121 Gwei (however, due to the relatively early point in the contract at which their bids failed and were reverted when they finally made it into a block, these two failed contributors only collectively wasted about $3.55 USD in gas). Note the surge of contributions on the last day of the sale coincided with an apparent DDOS/spam transaction attack on the Ethereum network, which likely contributed to the failure of these last bids to be included.

It is perhaps not surprising that many contributors waited to the end of the full bonus period before making their bids. This gave them the time to observe how the sale was progressing before making their bid. In this way, they could still get the entire bonus without having to pay the gas costs that would be required to adjust or withdraw their bids. The surge of bids at the very end of the sale, when the bonus was near zero, was more surprising to us. Nonetheless, only a small number of would-be participants did not manage to get their bids in on time.

Now, let’s take a closer look at how our contributors used the interactive features of the IICO.

Use of Manual Withdrawals
There were 20 bids that were manually withdrawn totalling 267 ETH; 18 of these bids were withdrawn during the full bonus/no withdraw penalty phase and 2 were partially withdrawn after the end of the full bonus phase.

However, it seems that many of these withdrawn bids were from addresses that subsequently rebid. Only 6 of the 20 manually withdrawn bids correspond to addresses that did not have some other non-withdrawn bid. These 6 bids correspond to a withdrawn value of 58.715 ETH. Further considering the timing of the withdrawals and the other bids of these addresses, 10 bids were withdrawn and not later replaced with a withdrawn value of 226.82 ETH. The other 10 withdrawn bids seem to mostly correspond to increased personal caps, often going from a capped bid to an uncapped bid.

Indeed, when we examine the graphs of personal caps with and without the manually withdrawn bids, we see that including the manually withdrawn bids creates a longer tail of small caps:


The distribution of personal caps including manually withdrawn bids
The median personal cap of the 49 bids with a personal cap, including the manually withdrawn bids, is 10,000 ETH, and the average personal cap is 26,579 ETH (though this is a situation where a few outliers can very easily skew the average).


The distribution of personal caps with manually withdrawn bids removed
Excluding the manually withdrawn bids, the median personal cap of the remaining 35 bids with a personal cap is still 10,000 ETH, and the average is 30,611 ETH.

Use of Personal Caps
Of the 307 bids, only 49 bids included a (finite) personal cap. Namely, 15.96% of bids were capped and 84.04% were uncapped.

It is worth pointing out that in order to use the interactive features of the sale — both the personal caps and the manual withdrawals — contributors had to make their bids from a web3 wallet such as Metamask. We provided a “simple interface” with which contributors could make uncapped bids, so it is possible that there was a portion of the participants would have taken advantage of the interactive features but did not have Metamask installed.

After the 18 fully manually withdrawn bids are excluded, there were 289 bids from 223 distinct addresses. Of these, 35 included a personal cap, and only 5 of these 35 capped bids were ultimately not included due to the final amount raised exceeding their cap. These 5 bids would have contributed a combined 11 ETH had they been included.

On the other hand, 3410.833 of the 5793.588 ETH contributed (after the removal of the automatically withdrawn bids) came from the remaining 30 bids that had a personal cap. Namely, 58.873% of the ether contributed was contributed as part of a bid with a personal cap. Moreover, breaking down median and mean contributions by whether bids are capped or uncapped we see :


So we observe that the capped bids are disproportionately likely to have been the larger bids. This observation is reinforced when we consider the graph of contributions and personal caps for each bid (excluding the manual withdrawals):


Contributions and personal caps for each bid (ordered by personal cap). Note the two separate scales for inclusion on the same graph. The end of the red dotted line indicates the beginning of uncapped bids. Most of the bids make up a “long tail” of small uncapped bids. On the other hand most of the capped bids are relatively large, including a few very large bids.
We also can get a sense of the relative weight of bids by personal cap to the ultimate amount raised by the following graph that shows the cumulative contribution as one moves across personal caps:


Individual and cumulative contributions removing automatic withdraws as caps reached (bids in ascending order by personal cap).
So apparently contributors were still willing to accept a certain amount of dilution and still participate in the sale. As we noted above only 5 bids totaling 11 ETH were automatically withdrawn. However, the amount raised was beginning to reach levels where it triggered a larger number of personal caps.

This is already visible in the graphs about where we looked at the distribution of personal caps with and without the manually withdrawn bids — we see that there were a number of bids with personal caps slightly to moderately above the amount ultimately raised. These contributions that were narrowly included turn out to be relatively large as we can see in the following graph:


Contributions in ETH ordered by the size of the personal cap. All bids on the left side of the green line had finite personal caps wheras bids on the right did not include personal caps. The red line indicates the cutoff for which bids had high enough personal caps to be accepted in the sale; the orange line indicates the point at which the contributions with lower personal caps and the contributions with higher personal caps each constitute 50% of the total contributions.
For example, if the sale had raised 25% more it would have had to do so despite triggering an additional 8 personal caps worth 253 ETH.


A look at by how much the amount raised would have needed to change (assuming that the difference was due entirely to uncapped bids) for bids to be automatically withdrawn. One sees that there is a group of bids worth about 250 ETH that would have triggered with a modest increase in the amount raised and larger amounts with personal caps that would only have activated as a safeguard against much greater levels of dilution.

Conclusions
Conducting the first IICO ever has taught us a number of lessons about user behavior.

We have seen that larger bids were more likely to provide personal caps; indeed the majority of the ether raised by the sale was submitted as part of a capped bid. On the other hand, small bidders seem particularly willing to accept dilution. This may partially be due to small bidders being less likely to have Metamask installed and thus obliged to use the simple interface to contribute. Moreover, many of the smaller bids that didn’t use the interactive features of the IICO probably came from contributors who just wanted to participate no matter what, with motivations similar to someone who issues a market order on an exchange.

However, considering the tokenomics of PNK, from the perspective of a small contributor that hopes to obtain enough PNK to be able to arbitrate a given quantity of disputes, it is perhaps not unreasonable to be relatively indifferent to dilution.

In the long-term, holding things like arbitration fees and interest rates constant, one expects that the valuation of PNK to scale proportionally with the amount of arbitration that Kleros is being used for. Of course, the arbitration fees have yet to be set, and in fact in the long run will be determined by liquid democracy governance mechanisms that may respond to the number of arbitration cases available and introduce non-linearity into the relationship between the value of PNK and the amount of arbitration work.

However, as a first approximation, if you believe in the efficient market hypothesis, you might think that the amount contributed in the IICO is already a rough proxy for how much arbitration work the market expects Kleros to attract. In this approximation, for however much arbitration a contributor wants to do, if the the valuation of the sale doubled, then the total abitration work a contributor expects would also double and the contributor would still expect to have the appropriate number of cases to arbitrate even though she only got half as many tokens.

As other projects use the IICO model, it will be interesting to see if the token model of the token being sold has an impact on the behavior of contributors.

Moreover, estimating the amount of arbitration cases available in advance and how much PNK will be required to be drawn as a juror with a sufficient frequency to be worth a given contribution is difficult, and it may not have been worth it to small contributors to expend the effort to make this estimate.

Nonetheless, many larger contributors seem to have taken advantage of the controls offered to them by the IICO model. Additionally, this model has mostly delivered on the promise of avoiding gas wars and allowing certainty of participation while still having controls on dilution.


Join Kleros!
Join the community chat on Telegram.

Visit our website.

Follow us on Twitter.

Join our Slack for developer conversations.

Contribute on Github.