Author Topic: New attack that the consensus might be vulnerable to ("P + e Attack")  (Read 18054 times)

psztorc

  • Administrator
  • Sr. Member
  • *****
  • Posts: 464
    • View Profile
Re: New attack that the consensus might be vulnerable to
« Reply #45 on: February 11, 2015, 07:07:26 pm »
I don't know why I'm bothering...

You used the sheet without understanding it. The price reaction is the same for both if their betas are the same. Take all of the "overpayments" in LS-LMSR, and add them up, if that quantity of money was used to amp beta in the vanilla LMSR (where these overpayments were never made), the vanilla LMSR would have a beta which precisely equalized the price-reaction.

The LS-LMSR is better because it forces "who pays for this liquidity" in the blockchain to more-resemble "who pays for this liquidity" in real life. It doesn't magically produce liquidity somehow.
Nullius In Verba

psztorc

  • Administrator
  • Sr. Member
  • *****
  • Posts: 464
    • View Profile
Re: New attack that the consensus might be vulnerable to
« Reply #46 on: February 11, 2015, 07:16:54 pm »
If you want to do something that actually would help me, why don't you try to derive the formula for transforming LMSR prices into their LS-LMSR counterparts. They are both 0 and 1 at their extremes, and at uniform ( 50%-50% in the basic two State case) they are each multiplied by v (from the paper, alpha = v / (n log(n)) ). Last weekend I tried transformations for two hours, trying to cheat and grab it easily using statistics.

Later this week, I will resign myself to do the...algebra...
Nullius In Verba

ceci

  • Newbie
  • *
  • Posts: 5
    • View Profile
Re: New attack that the consensus might be vulnerable to
« Reply #47 on: February 11, 2015, 07:41:11 pm »
I'm trying to follow this conversation, but can't see what you are talking about.  Could someone tell me where can I find the LS-LMSR?

psztorc

  • Administrator
  • Sr. Member
  • *****
  • Posts: 464
    • View Profile
Re: New attack that the consensus might be vulnerable to
« Reply #48 on: February 12, 2015, 12:57:10 am »
http://www.eecs.harvard.edu/cs286r/courses/fall12/papers/OPRS10.pdf

LMSR requires one to set an initial "beta" parameter, but in LS-LMSR the beta is beta = alpha * sum(outstanding shares). Choice of alpha is actually very stable at  ( .05 / (n log n ) ) which produces slightly more tractable markets. Traders suffer under LS-LMSR (they are always overcharged).

Feel free to email me, I can send you an in-progress Excel sheet which tries to clarifies the differences.
« Last Edit: February 12, 2015, 01:00:16 am by psztorc »
Nullius In Verba