Analyze the token sale model

Author: Vitalik Buterin

Original Title: 《Analyzing Token Sale Models》

Published on: June 9, 2017

Note: The names of several projects mentioned below are for comparative purposes regarding their token sale mechanisms and do not represent an endorsement or criticism of any project. Well-designed token sale models can also be complete garbage projects.

In recent months, there has been an increase in the number of innovations in token sale models. Two years ago, the field was quite simple: there were sales with a capped maximum amount, selling a fixed number of tokens at a fixed price, thus the valuation was also fixed and sold out quickly; there were also sales without a cap, where people bought as much as they wanted. Now, whether in theoretical exploration or in many practical deployment cases, we have seen a surge in the popularity of these sales projects, including hybrid capped sales, reverse Dutch auctions, Vickrey auctions, proportional refunds, and various other sales mechanisms.

The emergence of these mechanisms is a response to previous design failures. Almost every significant token sale project, including Brave's Basic Attention Tokens (BAT), Gnosis, the upcoming Bancor token sale, and the completed Maidsafe and Ether sales, has faced substantial criticism, all pointing to a simple fact: so far, we have not found a mechanism that possesses all or most of the features we like.

Let’s review some examples.

Maidsafe

This decentralized internet platform raised $7 million in five hours. However, they mistakenly accepted payments in two currencies (Bitcoin and MSC), offering a more favorable exchange rate for MSC buyers. This mistake caused the MSC price to double temporarily, as users rushed to buy MSC at a better rate to participate in the sale, but after the sale ended, the MSC price plummeted sharply in a similar manner. Many users exchanged their Bitcoin for MSC to participate in the sale, but the sale ended too quickly, resulting in about a 30% loss.

This project and several others that followed (WeTrust, TokenCard) taught us an undisputed lesson: accepting multiple currencies at a fixed exchange rate is dangerous and bad. So don’t do that.

Ethereum

The Ethereum token sale had no maximum cap and lasted for 42 days. The sale price for the first 14 days was 2000 Ether (ETH) for one Bitcoin, after which it began to increase linearly, capped at 1337 Ether for one Bitcoin.

Almost every project without a sales cap has faced criticism for being too "greedy" (I personally reserve my opinion on this criticism, but we will return to this issue later), and these sales also faced another interesting accusation: they brought a high degree of uncertainty regarding the valuation of the products purchased by participants. For example, in projects that have not yet started sales, it is likely that many people would be willing to pay $10,000 for a bunch of Bancor tokens, as long as they understand that this bunch represents 1% of the total supply of Bancor tokens; however, if they purchased 5000 Bancor tokens without knowing whether the total supply is 50,000, 500,000, or 500 million, many would feel anxious.

In the Ether sale, buyers who genuinely cared about valuation predictability typically chose to buy in on the 14th day of the project, inferring that since this was the last day for discounts, the predictability on that day was the highest, while also enjoying the discount; however, this model is hard to say is economically optimal, as the equilibrium resembles everyone buying in at the last moment on the 14th day, making a trade-off between valuation certainty and obtaining a maximum profit of 1.5% (or, if certainty is really important, possibly delaying purchases to the 15th, 16th day, or even later). Thus, it is evident that this model has quite strange economic characteristics, so if there is a convenient way, we would still prefer to avoid doing so.

BAT

In 2016 and early 2017, designs with a maximum cap were the most popular. Their characteristic was that excessive enthusiasm could lead to oversubscription, thus creating an incentive to rush. Initially, sales could be completed in a few hours. However, the speed quickly increased. First Blood completed a $5.5 million sales target in two minutes, attracting considerable media attention; meanwhile, the Ethereum blockchain suffered a denial-of-service attack. However, it was not until last month’s BAT token sale that there was an excessive faith and hype in Nash equilibrium; at that time, the high interest in the project led to a sales target of $35 million being completed in less than 30 seconds.

This project not only completed within two blocks but also:

1. The total transaction fee paid was 70.15 Ether (over $15,000), with the highest single fee being around $6,600.
2. 185 purchases were successful, while over 10,000 failed.
3. The capacity of the Ethereum blockchain remained saturated within three hours after the sale began.

Thus, we began to see capped sales approaching a natural equilibrium: people competed to outbid each other with transaction fees, even leading to millions of dollars spent on miners. Then came the next stage: at the start of the project, large mining pools intervened, buying all the tokens before anyone else.

Gnosis

Gnosis also attempted to alleviate this issue with a new mechanism: reverse Dutch auction. Simply put, this model works like this. The sale sets a cap of $12.5 million, but the portion of tokens that buyers can obtain depends on how long the sale lasts. If it ends on the first day, only about 5% of the tokens are allocated to buyers, while the rest are held by the Gnosis team; if it ends on the second day, it’s about 10%, and so on.

Its purpose is to create a plan where, assuming you buy at time T, you will definitely buy at a valuation close to 1/T.

The goal is to create a mechanism that provides a simple optimal strategy. First, you decide the maximum valuation you are willing to accept (let's call it V). Then the sale starts, and you do not need to buy immediately; instead, wait until the valuation drops below V, and then you can send a transaction request.

It can have two outcomes:

1. If the valuation has not dropped below V, the sale ends. Then you can be happy because you avoided what you considered a bad trade.
2. If the valuation drops below V and the sale ends, you can submit a transaction request, and the outcome is also what you like because you participated in what you considered a good trade.

However, many predict that due to the "fear of missing out" (FOMO), many will "irrationally" buy on the first day without even looking at the valuation. The reality is indeed so: the sale ended within a few hours, resulting in only 5% of the total supply (valued at over $300 million) being sold, reaching the cap of $12.5 million.

These certainly demonstrate the market's irrationality, as people did not think carefully before investing large amounts of money (the implication being that the entire field needs some degree of control to prevent the market from overheating further); unless an undeniable fact arises: the traders buying in are correct.

Despite the significant rise in Ether prices, the price of GNO has increased from about 0.6 Ether to about 0.8 Ether.

What happened? In the weeks leading up to the sale, due to public criticism that if they held the majority of tokens, they could manipulate GNO prices like a central bank; the Gnosis team agreed to hold 90% of the unsold tokens for a year. From the traders' perspective, long-term frozen tokens would not affect the market, and based on this, they analyzed that such an impact would not exist in the short term. This was also a factor that supported Steem's high valuation when it started last July, and similarly for Zcash, where the price of individual tokens exceeded $1,000 at the beginning.

Now, a year is not that long. Locking tokens for a year is not the same as permanently locking them. However, people continue to reason anew. Even when the one-year lock period ends, you can still think that if the Gnosis team believes that releasing the locked tokens could raise the price, then this move would benefit the team. So if you believe in the Gnosis team's judgment, it means that what they intend to do would benefit GNO prices at least equivalently to permanently locked tokens. Therefore, in reality, the GNO sale looks more like setting a cap of $12.5 million and a valuation of $37.5 million. Traders participating in the sale also reacted correctly, leading many network commentators to speculate about what happened.

Cryptographic assets indeed have strange bubbles, with various obscure assets reaching market capitalizations of $1 to $100 million (including BitBean's $12 million market cap during the writing, $22 million PotCoin, $13 million PepeCash, and $14.7 million SmileyCoin). However, there may be a good case here: in many instances, participants in the sales phase were not wrong, at least not for themselves; while the buying traders accurately predicted that a bubble had been brewing since 2015 (and possibly even since 2010).

More importantly, aside from the bubble issue, there is another reasonable criticism of the Gnosis token sale: despite promising not to sell tokens within a year, they would eventually become part of those tokens, thus being able to manipulate GNO prices to some extent like a central bank. Traders would also have to deal with all the monetary policy uncertainties arising from this.

The Certainty Problem

So what does a good token sale mechanism look like? One starting point is to observe the criticisms currently faced by sales models and find a list of desirable features.

Let’s do that together. Some inherent characteristics include:

1. Certainty of valuation: If you participate in a sale project, you must at least be certain about the upper limit of the valuation (or the maximum proportion of tokens you can obtain).
2. Certainty of participation: If you want to participate in such a sale project, you must have the intention to succeed.
3. Cap on financing amount: To avoid greedy criticism (or to eliminate regulatory concerns), the sale must limit the amount of financing.
4. No central bank: The initiators of the token sale cannot hold an excessive number of tokens to avoid controlling the market.
5. Efficiency: The sale should not lead to economic inefficiency or unnecessary losses.

Does that sound reasonable?

Well, here’s the less interesting part.

1. (1) and (2) cannot be satisfied simultaneously.
2. At least if no tricks can be employed, (3), (4), and (5) cannot be satisfied simultaneously.

These can be referred to as "the first token sale dilemma" and "the second token sale trilemma."

The proof of the first dilemma is simple: suppose you offer users a certain valuation of $100 million in the sale. Now assume users want to invest $101 million. At least some people will fail. Simple supply and demand can prove the second trilemma. If you satisfy item (4), you must sell all or a fixed share of the total tokens, thus the valuation you sell is proportional to the price you sell at. If you satisfy item (3), you set a cap on the price. However, the equilibrium price corresponding to the quantity you sold may exceed the price cap you set, leading to a situation of excess demand, thus unavoidable: (i) the digital equivalent of waiting four hours in line at a busy restaurant; (ii) the digital equivalent of scalping tickets, resulting in significant unnecessary losses, conflicting with (5).

The first dilemma cannot be overcome; certainty of valuation and participation is unavoidable, although in possible cases, choosing certainty of participation is better than choosing certainty of valuation. The closest result we can derive is to sacrifice full participation to guarantee partial participation. This can be achieved through proportional refunds (for example, buying tokens valued at $1 billion for $1.01 billion, with everyone receiving a 1% refund). We can also consider this mechanism as an unlimited sale, where part of the payment is presented in the form of locked funds instead of consuming it; however, from this perspective, it can be determined that the requirement for locked funds is a loss of efficiency, thus this mechanism cannot satisfy (5). If Ether shares are not allocated well, it may favor wealthy shareholders, harming fairness.

The second trilemma is difficult to overcome. Many attempts to overcome it may lead to failures or unpredictable bad outcomes. For example, the Bancor token sale considered limiting the gas price for purchase transactions to 50 entropy (about 12 times the regular gas price). However, this means that the optimal strategy for buyers is to set up multiple accounts, sending one transaction from each account to trigger the contract and then buy in (circumventing the risk of buyers accidentally exceeding their expected purchase, reducing capital requirements). The more accounts buyers set up, the higher their potential purchases. Thus, in equilibrium, this could exacerbate congestion on the Ethereum blockchain, even exceeding BAT-type sales; the single transaction fees for such sales are at least $6,600, rather than being subjected to denial-of-service attacks. Moreover, any type of on-chain transaction competition severely harms fairness, as the cost of participating in the competition is constant, while the rewards are related to your capital amount, thus the outcome will disproportionately favor wealthy shareholders.

Next Steps

There are three wise things you can do. First, you can conduct a reverse Dutch auction like Gnosis, but with an adjustment: do not hold unsold tokens, but use them for the public good. Simple examples include: (i) airdrops (redistributing to Ether holders); (ii) donations to the Ethereum Foundation; (iii) donations to Parity, Brainbot, Smartpool, or other companies and individuals building infrastructure independently for the Ethereum space; (iv) combining the first three, possibly according to the ratio voted by token buyers.

Second, you can retain unsold tokens but resolve the "central bank" issue by fully automating the way tokens are spent. The reasoning here is similar to why many economists are interested in rule-based monetary policy: even if a centralized institution can significantly control powerful resources, it can eliminate the resulting political uncertainty by following certain usage rules through that institution. For example, unsold tokens could be given to market makers to maintain the stability of token prices.

Third, you can conduct capped sales. Limit the amount everyone can buy. Effectively doing this requires a KYC process, but fortunately, KYC institutions can complete this task in one go; after confirming that an address represents a specific individual, a whitelist of user addresses can be created and reused for each token sale, as well as for other applications that may benefit from this model, such as voting in the Akasha game. There is still unnecessary loss (efficiency), which may lead uninterested individuals to participate in the sale, knowing they can quickly profit in the market. However, this may not be so bad: it creates a basic income for everyone in cryptocurrency, and if behavioral economics assumptions like the "endowment effect" have even a little accuracy, it will achieve the goal of ensuring the widespread distribution of tokens.

Is Single Round Sale Good?

Returning to the topic of "greed." I want to say that, in principle, not many people oppose a development team that can spend $500 million to create a great project that earns $500 million. What people oppose is: (i) a brand new untested development team receiving $50 million all at once; (ii) more importantly, the time lag between developer rewards and token buyer interests. In a single round sale, developers only have one chance to obtain funding for project creation, which is close to the time when the development process begins. There is no such feedback mechanism here: the team first receives a small amount of funding to prove themselves, and then, after proving their reliability and ability to succeed, they receive more funding over time. During the sale, there is relatively little information available to distinguish between excellent and poor development teams, and once the sale is completed, the developers' motivation to continue working is lower compared to traditional companies. "Greed" does not mean obtaining a lot of money, but rather wanting to receive large amounts of funding without making efforts to prove that they can spend this money wisely.

If you want to hit the nail on the head, how should we solve this? I would say the answer is simple: focus on the mechanism, not on single round sales.

I can provide several examples for reference:

Angelshares, this project sold in 2014, selling a fixed proportion of AGS daily over several months. Every day, people could provide unlimited funds to the crowdfunding project, and the daily AGS allocation would be divided among several contributors. Essentially, this was like maintaining hundreds of unlimited "small round trades" for nearly a year; I would say the duration of this sale could be extended even longer.

Mysterium conducted six months of small sales before a large-scale sale, which almost no one noticed.

Bancor recently agreed to hand over all financing to market makers, who would maintain price stability, while setting the lowest limit at 0.01 Ether. These funds cannot be withdrawn for two years.

It seems difficult to see the connection between Bancor's strategy and solving this time lag, but there is an element of the solution here. To understand, consider two scenarios. In the first case, suppose a sale raised $30 million, with a cap of $10 million, but a year later everyone agrees the project is a failure. Then the token price may drop below 0.01 Ether, and the market maker may lose all funds trying to maintain the minimum price, leaving the team with only $10 million in operating funds. In the second case, suppose the sale raised $30 million, with a cap of $10 million, and two years later everyone is satisfied with the project. In this case, the market maker would not be triggered, and the team could receive all $30 million in funds.

A related proposal comes from Vlad Zamfir's "safe token sale mechanism." This concept is broad and can be parameterized in various ways, but one way is to sell tokens at the highest price, then obtain a slightly lower minimum price, and gradually widen the distance between the two prices over time. If the price can maintain itself, development funds are gradually released.

The above three may not be sufficient. We want to extend the duration of sales projects, allowing us more time to understand which development teams are the most trustworthy before providing large amounts of funding. However, this seems to be the most productive direction to explore.

Breaking Out of the Dilemma

In summary of the above analysis, it can be clearly concluded that although the above dilemmas and trilemmas cannot be directly overcome, we can break through the margins little by little through innovative thinking and overcoming less severe similar issues. We can reach an agreement on participation, using time as the third dimension to eliminate the impact: if you did not participate in the Nth round sale, you can wait for the N+1 round a week later, when the price may not be significantly different.

We can conduct unlimited sales, but must include different cycles, without setting a cap on sales within a cycle; thus, without proving the ability to handle small amount sales, the team will not demand large amounts of funding. We can sell a small portion of tokens at once, stipulating the remaining supply through a contract and automatically selling according to a pre-set formula to eliminate the political uncertainty it brings.

Here are some feasible mechanisms following these ideas:

Conduct a reverse Dutch auction similar to Gnosis, setting a lower cap (for example, $1 million). If the auction sells less than 100% of the total supply of tokens, automatically transfer the remaining funds to another auction two months later, setting a cap 30% higher. Repeat this process until all issued tokens are sold.

Sell tokens at a price of X dollars without limit, depositing 90% of the proceeds into a smart contract, guaranteeing a minimum price of 0.9 times X for five years, while the maximum price can increase indefinitely, and the minimum price can approach zero.

Imitate AngelShares' approach, but extend the time to five years instead of several months.

Conduct a reverse Dutch auction similar to Gnosis; if the auction sells less than 100% of the total supply of tokens, transfer the remaining funds to an automated market maker to maintain price stability (if the price continues to rise, the market maker can sell tokens, with part of the proceeds going to the development team).

Immediately transfer all tokens to market makers, setting parameters and variables X (minimum price), s (share of sold tokens), t (time the sale has been ongoing), T (expected duration of the sale, for example, five years), to sell tokens at a price of k / (t/T - s) (this is a bit strange and may require economic analysis).

Note that other mechanisms to solve token sale issues must also be attempted, such as entrusting proceeds to multiple custodians, only releasing funds upon reaching a certain threshold; this idea is interesting and worth exploring further. However, the design space is multidimensional, and there are many ways to explore.