Errata 20th June 2013
As noted in the comments below and reported to me by none other than keeper of the data, dooglus, the original dataset included bets that were refunded by SatoshiDice for being too big or too small. The fault here is mine for not performing a sanity check on the data.
I reprocessed the data, and took the opportunity to make a change to the hexbin plot based on some feedback I received, and include an extra chart showing the basic statistics of SatoshiDice bets. Let me know in the comments what you think of the changes.
19th June, 2013
0. Introduction
Last year in the 'Satoshi Dice -- Statistical Analysis" bitcointalk forum thread (started by etotheipi and continued by dooglus), I produced a few plots of various SatoshiDice bet statistics. I showed, for example, that SatoshiDice bettors prefer bets with a single significant digit, that most of the volume is contributed by a very small proportion of the bettors, and the majority of bettors contribute little to the overall bet volume.I was interested to see how the bet volume has progressed in the lead up to and aftermath of the BTC price correction in April, and dooglus very kindly provided the data. I didn't want to spam the thread, and I thought posting the results here might be a fun break from more serious analyses. Enjoy!
1. The old charts
I originally thought to produce y-axis density plots, one the density of bet size and the other weighted by bet size, as last time:
The first chart shows the density of various bet amounts, and the second chart is weighted by bet value and shows BTC bet volume.
Density plots can take a lot of memory and time. With 5,044,538 variables and 14 plots to produce I was soon running out of both RAM and patience. Furthermore I was not completely happy with the plots - I couldn't continue them on one x-axis indefinitely since I'd soon run out of room, and facetting the plots for multiple x-axis would lose the ability to compare months in different facets.
I decided that I would make a new set of plots. My criterion was that the data visualisations should:
- be able to be continued indefinitely on one x-axis;
- show the number of bets and the volume of bets in a given size range;
- show the volume in USD.
2. The new charts
Instead of using bet density (computationally expensive) I used 2D hex bins (much less computationally expensive). Each hex bin includes a weeks worth of bets at a particular value of bets, and the darker the hexagon, the more bets were at that value.
I think this chart does a much better job at showing that the number of bets at, for example, 0.001, 0.01, 0.02, 0.05 and 0.1 BTC are much higher than the bets at values in between these amounts.
To show volume, I've binned and stacked the data. The bins are shown in the legend, and the data is binned weekly. The USD volume estimate uses the daily median MTGOX USD price applied to each bet before the data is summarised for weekly binning.
The changes in numbers of bets and volume are much easier to see than using the violin plots. The BTC volume is clearly in decline, however the USD volume shows this drop is not as significant as it would seem. Even now, long after the April price correction, the last four weeks' weekly US$ volume has been significant:
3. Summary
- SatoshiDice bettors prefer bets with a single significant digit
- SatoshiDice volume in BTC is decreasing.
- SatoshiDice volume in USD is increasing.
- The largest bets were at the maximum of 500 BTC and happened several times since the start of the year, but none were the highest in terms of USD. The largest USD bet was 400 BTC at US$125.40 / BTC for a US$50160 bet.
organofcorti.blogspot.com is a reader supported blog
BTC: 12QxPHEuxDrs7mCyGSx1iVSozTwtquDB3r
LTC: LPXnETNoCBr16GduvyWRzFP83rZNeEgMuB
For notification of new posts, follow @oocBlog
<fourteenpoint one>
> The highest bet was at the end of March at 716.513 BTC
ReplyDeleteWas that a single bet? I didn't think the maximum bet was ever that high.
Perhaps link to the bet on satoshidice.com or blockchain.info?
You're right, I hadn't checked the data and it included bets that were too large and too small for SatoshiDice and so were refunded. I've fixed the post now.
Delete"that most of the volume is contributed by a very small proportion of the bettors"
ReplyDeleteI could not find on which from which data you arrived to this conclusion. Can you please submit a link?
Thanks, great job!
That what the "SatoshiDice bets, stacked and binned" charts shows - most bets are in the less than 2 btc range, yet about half of the volume is in the greater than 2 btc range. This implies that a small number of people are supplying most of the volume.
DeleteI could post tables, if exact figures are needed, but that's not really what this post is about - it's more about getting a general idea of changes to SatoshiDices' betting volume.
Glad you liked the post.
A player calling themselves "gladosscc" bet just under 500 BTC on Just-Dice.com today.
ReplyDeleteI pointed him at this article, saying that he was close to the biggest Bitcoin dice game bet record, at which he made this bet, beating the record in terms of both BTC and USD:
https://just-dice.com/roll/7093577
On the link you posted the stake was recorded as 561.29526541 BTC, which makes it larger than any single SatoshiDice bet and as far as I know the largest ever accepted bitcoin bet.
DeleteReaders might also be interested to know that over 26000btc has been wagered at Just-Dice since it started just under two weeks ago, which is already 1/4 to 1/3 of SatoshiDice's volume. Thanks to allowing 1 satoshi bets, Just-Dice has already had 7 million bets in that time. This is going to be an interesting site to keep and eye on.