BitNames - DNS Auction

[cunicula]

So, you buy a domain name, spend time and money building up a business on it, and have to be afraid that a competitor will outbid you and take the domain from under your feet at some point.

You have a 20 to 1 advantage in the auction.  You could make it 100 to 1 by decreasing the "rental" to 1%.

Have it as 2 steps

- auction (owner doesn't participate)
- If the owner doesn't pay 1% of the winner's bid within 60 days, the domain is transferred and the owner gets 99% of the bid (rest goes to support the network)

I don't see any other way to figure out market price without actually attempting to sell the domain.

Has to be a form of all-pay auction. Otherwise you will see competitors strategically bidding up name prices to drain (or extort) money from the name holder. You should pay a % of your bid even if you lose.

[TierNolan]

Has to be a form of all-pay auction. Otherwise you will see competitors strategically bidding up name prices to drain (or extort) money from the name holder. You should pay a % of your bid even if you lose.

That only works if they bid less than the value of the name.  The point of the auction is to establish the value of the name.

The issue is that the value of the name is high to the current "brand" owner but much lower to everyone else.

If the rent is 1% of the value, then it is a big risk for firms who overbid.  If they actually win, then they get a name that has low value to them.

[cunicula]

Has to be a form of all-pay auction. Otherwise you will see competitors strategically bidding up name prices to drain (or extort) money from the name holder. You should pay a % of your bid even if you lose.

That only works if they bid less than the value of the name.  

Yes, you never bid the subjective value of the thing you are trying to buy in an all-pay auction. If you win, you always profit. If you lose, you always lose. Such is the nature of all-pay auctions.

The point of the auction is to establish the value of the name.

Revenue equivalence theorem. Under some pretty general assumptions, game theory tells us "total revenue in an all-pay auction = total revenue in a traditional only the winner pays auction."
In other words, (at least to the extent that people obey game theory)
value of the name = price paid by winning bidder in traditional auction = total revenue in all pay auction.
  
http://www.cs.uiuc.edu/~chekuri/teaching/spring2008/Lectures/scribed/Notes20.pdf

I could look up empirical tests, but these tend to be kind of useless due to extreme context dependence.
People slightly alter their bidding behavior if you do stuff like dim the lights or offer them sugary drinks.

Aside: Wow the revenue equivalence theorem pops up everywhere in computer science notes. Funny how computer scientists study auction theory intensively, maybe some discrete choice models too, but not other kinds of economic theory. It is very random. AI in computer science seems to be all about dynamic programming. Macroeconomics theory research (for the last 30 years at least) is also about dynamic programming. But how often do computer scientists looking at theoretical work in macroeconomics. I think it has more to do with the needs of tech companies right now than the potential set of applications of economics to computer science. Tech companies have no use for macroeconomists, but do hire microeconomists for auction design and the design of discrete choice algorithms.  
 

The issue is that the value of the name is high to the current "brand" owner but much lower to everyone else.
If the rent is 1% of the value, then it is a big risk for firms who overbid.  If they actually win, then they get a name that has low value to them.

Unclear of your meaning here.

[TierNolan]

Revenue equivalence theorem.

Hmm, do the assumptions actually hold?  It assumes that all bidders have the same valuations.

Unclear of your meaning here.

The name is mostly of value to the original owners.  If the name is worth $500k to the other bidders, but worth $10 million to the winner, then bidding $5 million is risky, if you actually win.  The other firm gets a name worth $500k to them but pays $5 million.  This could happen if they over-estimate the bid amount.

[cunicula]

Revenue equivalence theorem.

Hmm, do the assumptions actually hold?  It assumes that all bidders have the same valuations.

Assumption are "Suppose bidders have independent and identically distributed valuations and are risk neutral." That is all.

Independent and identically distributed means that bidders valuations are drawn from the same random distribution. Not that the realizatoins are actually the same.
i.e a skewed distribution of participant valuations can be perfectly consistent with the assumption.

It is hard to say if the identical distribution assumption holds in practice. Game theory fails empirically in so many ways. If you assign valuations in a lab, then you control the distribution. If you go into the real world, true valuations are unknown. You have to use bidding behavior to estimate the valuations. But these valuation estimates rely on the assumption that game theory accurately predicts behavior. We know that it doesn't in most cases (e.g. chess grandmasters behave according to game theory, real people don't). So you can't really credibly test the assumption.

The name is mostly of value to the original owners.  If the name is worth $500k to the other bidders, but worth $10 million to the winner, then bidding $5 million is risky, if you actually win.  The other firm gets a name worth $500k to them but pays $5 million.  This could happen if they over-estimate the bid amount.

Sure, but I still don't see what you are driving at. I've overpaid for food at restaurants too. Shit happens when you buy stuff without being sure of its real market value.
Maybe, I'll just have to ponder this.

[TierNolan]

Assumption are "Suppose bidders have independent and identically distributed valuations and are risk neutral." That is all.

Independent and identically distributed means that bidders valuations are drawn from the same random distribution. Not that the realizatoins are actually the same.
i.e a skewed distribution of participant valuations can be perfectly consistent with the assumption.

Doesn't identically distributed go against the fact that one "bidder", i.e. the owner, has a much higher valuation.

Sure, but I still don't see what you are driving at. I've overpaid for food at restaurants too. Shit happens when you buy stuff without being sure of its real market value.
Maybe, I'll just have to ponder this.

The attack was that competitors bid much higher than their valuation in order to harm the owner.  However, if they estimate the value wrongly, it backfires.

The owner pays 1% (or maybe 0.5%) of the value per year.  If the name is actually worth $10 million to them and you push the value up from $5 million to $8 million, then you are costing them $3 million extra.  That adds 30k per year extra in rent for them.  However, if you estimate wrong, and they only valued it at $7.5 million, then you end up paying $7 million more for the name than it is worth to you.

Maybe they end up buying it back from you for $5 million and make $2 million profit from the deal.

The point is that they are risking $7 million in order to inflict 30k of harm to their competitor.

In addition, if it is 2 stage, then the owner has even more of an advantage.  The auction happens and then the owner can take the money or pay 1% of the winning bid.

They don't even have to say what they would have paid.

[bytemaster]

Roots?   Where we're going, we don't need roots.  

Would you ban the current official roots at least?

ICANN is adding lots more top level domains, so you increase the chances of a clash. 

At least if the system only uses one TLD, you can argue that others should not use that.

Using .bt2 seems a little derivative though .  Maybe, .bns (for bit name server or just BitNames).

A DNS system that isn't widely accepted is pointless.

If namecoin becomes sufficiently popular, then people might be willing to manually add it and/or large DNS providers might support it.

If your system is incompatible with the main DNS system, then it is much less likely to gain much traction.

You could make it so that registering

my.toplevel.domain

also inherently registers

my.toplevel.domain.<standard TLD>

With sufficient traction, browsers might even swap my.toplevel.domain with my.toplevel.domain.bt2

I like the .bns extension.    Clearly you could map the names to any TLD.   

[cunicula]

Assumption are "Suppose bidders have independent and identically distributed valuations and are risk neutral." That is all.

Independent and identically distributed means that bidders valuations are drawn from the same random distribution. Not that the realizatoins are actually the same.
i.e a skewed distribution of participant valuations can be perfectly consistent with the assumption.

Doesn't identically distributed go against the fact that one "bidder", i.e. the owner, has a much higher valuation.

No. As a practical demonstration consider the following Independent and identical distribution (iid is an abbreviation for this) .
Draw a random variable X from the uniform distribution on [0,1]. Define a new random variable Z
Z= 10^10*X^100

Here is are 11 iid draws of Z:
1.04111E-57
1.90948E-51
5.64052E-06
38551506.91
8.70704E-12
8.29249E-21
2.57879E-11
0.011592097
9.91236E-11
7.99028E-08
2.91875E-74

As you can see, one guy will bid 38,551,506.91 BTC for the name. The next highest bidder offers 0.01 BTC. The third highest bidder will pay a few satoshis.
The remaining bidders are not willing to offer even a satoshi.
 
The revenue equivalence theorem can break down if some bidders are drawing values from X while other bidders draw values from Z. Really hard to establish whether this is happening empirically.
  

The attack was that competitors bid much higher than their valuation in order to harm the owner.  However, if they estimate the value wrongly, it backfires.

Okay, yes I understand now and agree with you. It won't work unless you have a reasonably good idea of what the owner valuation is.

[domob]

Assumption are "Suppose bidders have independent and identically distributed valuations and are risk neutral." That is all.

Independent and identically distributed means that bidders valuations are drawn from the same random distribution. Not that the realizatoins are actually the same.
i.e a skewed distribution of participant valuations can be perfectly consistent with the assumption.

Doesn't identically distributed go against the fact that one "bidder", i.e. the owner, has a much higher valuation.

No. As a practical demonstration consider the following Independent and identical distribution (iid is an abbreviation for this) .
Draw a random variable X from the uniform distribution on [0,1]. Define a new random variable Z
Z= 10^10*X^100

Sorry, but this is not really a fair "proof" of your assumption that the valuations are indeed IID.  Of course they need not all be the same and can still be IID, but I really believe that they are not in the case we are talking about.  The fact that one bidder owns the name and the others do not clearly expresses that they are not in the same position and thus probably won't have their valuations distributed according to the same distribution and in fact probably a very different one.  Your argument that there exist distributions for which a random draw might produce valuations that look like one of them valuing the name much more than the others is not valid - you could just as well also get a draw out of this distribution where two or none value the name much more than the rest, which clearly does not fit the situation we're assuming.

Note that it is common in statistics or mathematical modelling to just "assume" IID distributions even though it is clearly very hard to impossible to "verify" that this is fulfilled in your practical application (in particular independence of random variables is hard to ensure for a real world system).  But in particular for this example I think the assumption of IID is especially poor, to be honest.

[cunicula]

Note that it is common in statistics or mathematical modelling to just "assume" IID distributions even though it is clearly very hard to impossible to "verify" that this is fulfilled in your practical application (in particular independence of random variables is hard to ensure for a real world system).  But in particular for this example I think the assumption of IID is especially poor, to be honest.

The revenue equivalence theorem can break down if some bidders are drawing values from X while other bidders draw values from Z. Really hard to establish whether this is happening empirically.

Yeah, I think you misunderstood me. As I see it,

"Note that it is common in statistics or mathematical modelling to just "assume" IID distributions even though it is clearly very hard to impossible to "verify" that this is fulfilled in your practical application (in particular independence of random variables is hard to ensure for a real world system)." = "Really hard to establish whether this is happening empirically."

But in particular for this example I think the assumption of IID is especially poor, to be honest.

Could be. Since there is no credible way of substantiating this empirically, I prefer not to speculate.

Note: The traditional auction format (winner pays his bid) doesn't generate accurate valuations if the revenue equivalence theorem breaks down either. The gold standard  "theoretical" auction format for revealing true valuations is the second-price auction, where the winner pays the second highest bid instead of the highest bid. Here the Nash-Equilibrium is always bid your true valuation. The random process generating valuations becomes irrelevant for equilibrium analysis. Again, real people don't behave as game theory predicts. Give them a traditional auction, complex from a theoretical perspective, and they feel comfortable bidding. Give them a second-price auction, simple as anything from a theoretical perspective, and they get confused.

I actually hate auction theory.
a) limited relevance outside of a very narrow sphere of economic life
b) almost useless empirically (once you move outside of relatively simple situations)
c) a lot of complex math (facilitating the type of penis comparison contest adored in all male-dominated fields)