He himself has said he hasn't taken a wage/salary from LRM. He never said that. He said he didn't take a fee out of the first two weeks of dividends he withheld from new hardware, and he explicitly said he takes a fee out of his 25%. But has never said what that fee is. Also, how do you know he just didn't give himself shares at IPO?
|
|
|
I'm a little disappointed that after finally changing my mind about the best way to collect commission, none of you investors like it and want to keep things the same.
I'm an investor that understands the change. I wouldn't object to keeping it the same or implementing the new system. If it eases doog's ability to expand the site via referrals, then that sounds good to me.
|
|
|
The value of a feeless account is worth 10% of the maximum feeless amount. In the case of 1BTC it would be 0.1BTC. If you were to buy a feeless account at less than that 10% then you would technically profit as compared to a brand new JD account.
Yes, but does that affect your calculations? (I know how to calculate what it is worth, I'm interested in second order affects)
|
|
|
@zipmaster: what do you think of a secondary market for accounts? This way people can monetize their losses and other accounts can start with a negative balance so doog doesn't make anything on them at first. How valuable is it to have the first 1BTC of profit fee-free?
|
|
|
I think it's only fair to warn you guys that someone has discovered a way of emptying the Doge-Dice bankroll. It's only a matter of time until this strategy becomes widely known. So in the interest of full disclosure, here it is:
I bid 6 million doges for much exploit very strong
|
|
|
If you want to sell miners, there is a mechanism for that. Sell your shares.
I think there is no reason to liquidate unless tyrion can find a better deal than the market rate for shares.
Edit: you can sell for $12000 usd or $18/ GH/s our shares trade at 14.23 per GH/s. So selling hardware could be a good deal or shares are undervalued
|
|
|
Order book got out of order, I'm ordering the bids now from low to high, this makes the spread a little easier to see
|
|
|
You haven't answered my question, you keep explaining that 1% Kelly is better than 0.5% Kelly. Let me rephrase it: Suppose my investment setting is 0.5% Kelly, and yours is 1% Kelly, and we have invested the equal amount of BTC. The bettor bets 1BTC and looses. Does that mean that you get 2X the profit from that bet than I get? Or are you proposing something much more complicated? Once again - please give us one good example, it's far from obvious what exactly is your idea.
This has been discussed, with much detail, a few months ago over several pages of this thread. Here's a summary. There are two competing philosophies: 1) distribution proportional to risk, and 2) classify bets into tiers and only expose those that volunteered for that level of risk. Here are two implementations of (1) 1a) The most straight forward system is that every user contributes to the overall bank roll, UserContribution = UserBankroll*K, where K is the percentage a user sets as the maximum they can lose and can be set to any value. The overall bankroll is the sum of UserContributions, and each user then gets UserContribution/Bankroll times the amount won or loss per bet. This approach, however, has been shown to be too computationally difficult as it requires constant calculation of every user's contribution. Right now the site only calculates it when money is invested or divested. 1b) An improved idea (and the one Doog was favoring) was to ask users to select one of a few options for maximum bet. For example, you would chose if you want 0.5% maximum wagered per bet or 1%. The site then just has to calculate the percentage of the bank roll owned collectively by those two bins after every roll since the percentage a single investor owns of either bins remains a constant unless there is an invest or divest event. You get a return proportional to what you choose for maximum wager (1% gets 2x the losses and gains as 0.5%). This, for now, seems to be the way it may be implemented. Here is the implementation of (2) The argument for (2) above is that some people that wanted lower variance felt like they would get a much lower return than they are currently getting if distribution is proportional to risk since max bets are placed very rarely. They proposed a system where people choose their risk and are placed in bins like 1b. However, they want to implement betting "tiers." In this example, the two bins are 0.5% and 1%. If a bet would win <=0.5% of the bankroll, then everyone shares the winnings just like it is now, which is proportional to the bank roll. If a bet is placed that would win > 0.5% of the bankroll, everyone shares like they currently do for the first 0.5%, but all winnings/losses between 0.5% and 1% are only shared amongst those that are are in the 1% bin. They want effectively 2 bankrolls: one that covers from 0-0.5% of the bankroll and the other that covers 0.5-1%.
|
|
|
The latest whale used a unusual martingale strategy where the betting probability he used went as low as 6.33%. I think that this is the reason he managed to profit so much from his luck.
Looking at the kelly criterion again, I realized that it depends on the probability chosen by the player. This means that the optimum maximum winnings are only 1% of wagered amount if the player uses a 49.5% bet. For bets with lower chances of success the kelly bet is much lower.
The kelly criterion states that optimal bet size is (1-p)/a-p = (1-p)/(0.99/p-1)-p = 0.01p/(0.99-p) where a is the payout for probability p.
When p=49.5% then kelly=1%, as we know. But when for example p=33%, kelly=0.5%, and when p=9% kelly=0.1% i.e. five times smaller that what is allowed today.
As we know, going well above kelly is risky and lowers the long term profit. I suggest that we let the maximum profit depend on probability. Alternatively, the edge could be increased for low probability bets.
Note also that the "lottery" function, where a player can win 10,000x his bet is extremely unprofitable.
Edit: fixed numbers
this is not true: I think you're mixing up player edge and house edge, using the player payout formula for computing the house expected value. from the house (investor) point of view: probability is 1-p payout b= 1/(.99/p-1) kelly = (p*(b+1)-1)/b = ((1-p)*(1/(.99/p-1)+1)-1)/(1/(.99/p-1)) = 1-.99=0.01 Here's some matlab/octave code that computes ideal kelly fraction as a function of player chosen percentage (it's always 1%, or the house edge) clear all close all edge=0.99;
for kk=2:98 player_percent=.01*kk; house_percent=1-player_percent
house_odds=1/((edge/player_percent)-1) kelly2(kk)=(house_percent*(house_odds+1)-1)/house_odds end
plot(kelly2)
Here is some matlab/octave code that demonstrates the same concept. It finds the final bankroll as a function of whatever probability the player chooses. No matter what probability they choose, 1% (i.e., the house edge) is always the correct play. clear all;
percent=.98; edge=0.99; payout=1/percent*edge;
number_of_rolls_per_trial=100000; bankroll=zeros(1,number_of_rolls_per_trial); bankroll(1)=1;
bankroll(number_of_rolls_per_trial)=0; number_of_trials=1;
for test=1:100 kelly=0.001*test; for yy=1:number_of_trials for kk=1:number_of_rolls_per_trial maxwin=kelly*bankroll(kk); wagered=maxwin/(payout-1); roll=rand; if roll > percent bankroll(kk+1)=bankroll(kk)+wagered; else bankroll(kk+1)=bankroll(kk)-maxwin; end end last(yy)=bankroll(end); end out(test)=(sum(last)/number_of_trials-bankroll(1))/bankroll(1); end semilogy([0.001:.001:.1],out) xlabel('MaxWin as Ratio of Bankroll') ylabel('Ratio of ending to Initial Bankroll')
|
|
|
Wouldnt kelly be 0.010101 period then? I mean 0.01 / (2*0.495) = 0.010101010...
Youre right... but this max win on changed chance has to be changed automatically to still meet kelly. Thats something hopefully will be set up for obvious reasons.
But what i mean is... lets assume all bets are 50%. Kelly 1% would only be the optimum for an investor when a player plays to full max profit. But most of the bids are way way lower than the amount needed for max profit. So the risks that 1% involves are never met for the single investor. He could easily use 10x kelly and still remain way under the 1% of his personal investment used in each bet since only a small portion of the house is used in each bet. The individual investor would have to set a way higher kelly to rally maximize his profits. Though of course he has to hope that its not often happening that someone really plays 1% max profit of the house since then 10% of his investment would be involved. Which is a high risk. I mean there should be a sweat spot for the real kelly value for an individual investor. And thats not 1% kelly. Its higher. I guess im not smart enough for calculating...
We can't make investors go up to 10x kelly on small bets because we we can't force users (players) to increase their bets if they bet small. A small bet is a small bet: every investor takes a share of the gains proportional to their investment; no bets are allowed which will exceed the 1% maximum optimum bet since any bets that exceed it will reduce long term gain. No need to simulate, the math isn't that bad: when you win, you win: (1+f*b)*current_bankroll where f is your kelly percentage, b is how much you win when you win (minus original bet, e.g., "1" for a 50:50 roll) when you lose, you lose: (1-f*a)*current_bankroll where a is your wager. is if you win p*N times, and lose q*N times (where p is the probability you win, q is the probability you lose, and N is the total number of games played), your ending money is: endbankroll=(1+f*b)^(p*N) * (1-f*a)^(q*N) * starting bankroll your percentage profit is then: C=(1+f*b)^(p*N) * (1-f*a)^(q*N) you can plug in any numbers you want for any type of roll, house edge, or kelly fraction and you will see what you get the optimum f to maximize C is simply the derivative of C (or log of C since log is a monotonic function) respect to f set to 0. optimum f = p/a-q/b C is negative (that is, a losing strategy when) p*log(1+f*b)+q*log(1-f*a) is negative, even if you have a positive house edge.
|
|
|
|