Topic: Satoshi Dice -- Statistical Analysis - page 33. (Read 192914 times)

dooglus, quick q if you would oblige...

Two days ago Erik implemented a change which returns 0.5% on a losing bet. The significant benefit of this to Bitcoin is that unspendable bitdust utxo is nearly eliminated. Only bets of <=0.06 BTC create such outputs.

For information can you advise what percentage of SD bets are 0.06 or less?
(the last two days is probably fine enough for a sampling).

Edit. Don't bother. I saw your file above and sampled that and about 54% of recent bets are 0.06 or less. Wow. Most of these are really small!
In any case, an fx rate of $200 will mean that all utxo since the change will be spendable. This is not a permanent problem then...

OK, here it is.

  http://dooglus.com/allbets.txt.bz2

It's 4.8MB compressed,  69.5 MB uncompressed, one bet per line.  Time (seconds since 1970) and bet size (in satoshis):


The timestamp is the timestamp of the block which paid the bet out.

I don't have a list of all bets but can generate it.  I'm a little busy at the moment getting ready to give back $82k that I've become kind of attached to tomorrow.  But I'll prepare the bet list when I get a chance.

Is it possible paste a list of all bets, rather than the average size per time period (I'm guessing it's per block)?

Well, if I send my money to satoshidice with a public note saying "send my winnings back to 1MyOtherSecretAddress so they don't know it's me" then that's not a very useful laundry either.

I've seen this "SDICE launders money" argument before, but I think it's bogus.  Everything they do is public, so is absolutely not laundering money.

Thanks for the stats folks
It also might be interesting to see what fraction of users are selecting to have winnings routed to another address. 
If the money goes back to the same address, it's hardly a very useful launderer.

http://dooglus.com/sd-data.txt.gz has 2 numbers per line.  Time (seconds since 1970) and average bet size in the block with that timestamp:

This is where boxplots or violin plots might be more useful. If you're not sure what I mean (and tbh I haven't exactly explained myself well) if you post the data (or pm me) I'll post some examples.

... but I do now.

This graph shows the moving average bet size (in BTC) over the previous 7 days:



I also tried using a 1-day window, but it's very spiky:



and here's the average over the previous 28 days:

I don't think anything I said assumes anything about the average bet size.

I don't have data readily available about the average bet size, but I do have the total amount bet over time.  Here's a graph showing the total bet volume over time since September last year when the price was down around $10/BTC.  Notice how the slope of the line (indicating bet volume per unit time) is about the same now as it was then.  There are a few very steep parts from when the whale was playing, but the overall volume doesn't seem to be affected by the dollar value of Bitcoins, odd as that seems.  Maybe the price goes up because more people are using it, so more people are betting, which offsets the small size per bet, but that's just a guess.

Thanks.

But this assumes that the average bet will remain in the same range.

My contention is that the average bet will keep on reducing as the BTC / USD rate keeps on increasing. The amount of total BTC bet on the first 100,000 bets will be higher than the amount of BTC made on the next 100,000. (I have not looked at the numbers and I don't know if I'm right or wrong, it just feels that way to me intuitively.)

If I'm right, then the gap between the black line and the red line won't go away as more time goes by.

The house keeps 1.9% of all bets, on average.

If a million people bet 1 BTC each, or 100k people bet 10 BTC each, the total bet volume is 1 million BTC either way, and the house will keep 19k BTC on average in both cases.

Does that make sense?

All bets * (1-.981).  That is roughly how much the house is expected to rake.

Think of SatoshiDice as a laundrer, it returns you 98% (roughly) of your money over time.  Now some people play wierdly trying to "beat the numbers" and they might actually lose 100% of their money.  That sweetens Satoshi's profits.  But as Dooglus has shown, pretty much it's just laundering and taking a 1.9% fee.

I thought the house edge was based on the number of bets. And not on the amount of betting.

The red and black lines are pretty close together (61k vs 66k).  Notice that zero is a few miles off the bottom of the graph, and the graph is auto-scaled to make the best use of space.  House profits are about 5k higher than they should be, or less than 10% higher than they should be.  A few big losses will correct for that.  If you look at the graph for January you'll see the red and black lines cross each other at least 17 times, meaning the house was doing exactly as expected.  In general the slope of the two lines should roughly match each other.  The law of large numbers says that the two lines will get arbitrarily close to each other in the long run, but we're not at the "long run" yet, when a single bet can change the total house profits by more than 1%.

The whale lost big in the beginning of February, which put the black line well above the red line, where it has stayed since.

December and January saw a lot of betting activity from one user, which made the profit graph jump up and down erratically and made the red line go steeper than usual.  It was almost certainly a bot, just because I doubt if a person would have the stamina to keep up such a level of play for so long.  Whether it was a "house bot" or a real player it's impossible to prove.  Erik says it's a real player and I believe him.

If the cost of running a collusion bot (increased transaction fees plus payouts to shareholders) is less than the expected gain in value from share price appreciation, then running the bot would be profitable.

How do we interpret this graph? It looks like SatoshiDICE is consistently beating expectations. Over time shouldn't the actual revenue be very close to the red line (house edge)? Why is it so much higher? And why is it that in two particular months there was a large jump in the activity (i.e. change in slope)? Is this an indicator of botting?

- (3 million shares) * (temporary price increase due to 'whale' business)...

This guy uses address "1Lucky..." and lived up to his name:

+ 13% of 1.9% (0.25%) of each bets, theorically, that goes to the shareholders.

Yeah, and there's no reason it can't collect the SD statistics as it does that first blockchain load.  Then it'd be done before you run any code after "setOnlineMode".  Unfortunately, it's not as simple as just adding the addresses to the initial scan wallet, because you need to ignore non-bet transactions.  It would require altering the C++ scan loops, but I bet it wouldn't be too terrible...

But other priorities call.  If it is running okay, then why fix it if it ain't broken!

It would be like a casino handing a stack of chips to an employee to go play at the casino so that the casino doesn't appear empty. Whether they win or lose, the employee hands whatever chips he has left back to the casino. The only cost to inflate the activity graphs would be the transaction fee, a minimal .0005 BTC at this time.

There is no way to tell if there is an insider placing SD bets.