2014 USD/mBTC Price Prediction Contest

[dnaleor]

If you halve the entry fee (50 mBTC) I will join. You can halve the reward if you want. I want to participate because I like forecasting, so for me it is not about the prize.
I understand that things like this take a lot of timle to calculate, organize, publish, etc.

But I guess with a 100 mBTC you will have less than 10 participants.
(I have also a guessing game running. Not much participants and the entry fee is just 10mBTC)

[Foxpup]

I'm in. Let's do this.

[rpietila]

The current bitcoin price is the wisdom of the crowd. 

Actually, no.

Think it this way: Bitcoin has been raking in consistent 1000% gains for years now. Has it done better than expected, to achieve these remarkable returns. NO! Actually I am facepalming every other day how childish the community is, how fickle the services, how early the adoption, etc. And how long everything takes. 5 years is an eternity, Rebecca Black got 100 million views in 6 months!

Information asymmetry is the thing that keeps the price depressed. The ones who are not in, don't see more than the problems mentioned above. The ones who are in, can calculate that Bitcoin is best that has happened or will happen to them, and become millionaires in 2-3 years.

If you halve the entry fee (50 mBTC) I will join. You can halve the reward if you want. I want to participate because I like forecasting, so for me it is not about the prize.
I understand that things like this take a lot of timle to calculate, organize, publish, etc.

But I guess with a 100 mBTC you will have less than 10 participants.
(I have also a guessing game running. Not much participants and the entry fee is just 10mBTC)

Okay, I'll do it (->50). The reason I want to keep it sufficiently high is also to keep the participants interested over the whole contest and put enough effort in the predictions.

Nice game you have. Was close, ha?

I will have to postpone the opening of this one over the Labor Day, because that is such a big festival here that I cannot commit myself. So the May forecast deadline will be a few days into May.

[dnaleor]

great news
I understand that having a lot of participants is not practical, but having 5 participants is maybe not sufficient

And yes, we were very close.   I think new participants will come when we are >800 USD again.

[rpietila]

Code:

30	5	100000
29	4,9	79432,82347
28	4,8	63095,73445
27	4,7	50118,72336
26	4,6	39810,71706
25	4,5	31622,7766
24	4,4	25118,86432
23	4,3	19952,62315
22	4,2	15848,93192
21	4,1	12589,25412
20	4	10000
19	3,9	7943,282347
18	3,8	6309,573445
17	3,7	5011,872336
16	3,6	3981,071706
15	3,5	3162,27766
14	3,4	2511,886432
13	3,3	1995,262315
12	3,2	1584,893192
11	3,1	1258,925412
10	3	1000
9	2,9	794,3282347
8	2,8	630,9573445
7	2,7	501,1872336
6	2,6	398,1071706
5	2,5	316,227766
4	2,4	251,1886432
3	2,3	199,5262315
2	2,2	158,4893192
1	2,1	less

It is a bit rough, but these are the classes I could agree on (serial number, LOG(price_min), price_min). There are 30 of them but most of them are not really happening in 1 month timescale. Based on last year, I estimate that going over 100,000 USD/BTC is so unlikely that it can be represented by a single class. In the lower part, cutting the range short at 10^2.2 seems a good choice.

This way the entries to the competition can be standardized. Less work for organizers and ability to forecast multiple months at once, since the classes are fixed anyway (each month guess each month, for the total of 8 rounds and 36 predictions). We will get a host of interest rate curves and everything as a result of this!!

But it is getting so mathematical that I should perhaps incentivize people by offering 1,000mBTC to the prize pot + all the 50mBTC participation fees. So in any event I will do all the work + pay 1,000mBTC OR $1,000 whichever is higher. (I knew this would happen.. )

Well what a heck - let's do it, unless we can iterate the rules to be even better before it starts!

The first event to be forecast is price in 31. May, and the deadline will be in the beginning of May, once I get back from celebrating Labor Day.

[dnaleor]

But it is getting so mathematical that I should perhaps incentivize people by offering 1,000mBTC to the prize pot + all the 50mBTC participation fees. So in any event I will do all the work + pay 1,000mBTC OR $1,000 whichever is higher. (I knew this would happen.. )

Well what a heck - let's do it, unless we can iterate the rules to be even better before it starts!

you did not need to do this, but hey, it is your money

Maybe create a second price with (half of) the entry fees? Just an idea...

Thinking about my guesses at the moment

[Foxpup]

Code:

30	5	100000
29	4,9	79432,82347
28	4,8	63095,73445
27	4,7	50118,72336
26	4,6	39810,71706
25	4,5	31622,7766
24	4,4	25118,86432
23	4,3	19952,62315
22	4,2	15848,93192
21	4,1	12589,25412
20	4	10000
19	3,9	7943,282347
18	3,8	6309,573445
17	3,7	5011,872336
16	3,6	3981,071706
15	3,5	3162,27766
14	3,4	2511,886432
13	3,3	1995,262315
12	3,2	1584,893192
11	3,1	1258,925412
10	3	1000
9	2,9	794,3282347
8	2,8	630,9573445
7	2,7	501,1872336
6	2,6	398,1071706
5	2,5	316,227766
4	2,4	251,1886432
3	2,3	199,5262315
2	2,2	158,4893192
1	2,1	less

It is a bit rough, but these are the classes I could agree on (serial number, LOG(price_min), price_min). There are 30 of them but most of them are not really happening in 1 month timescale. Based on last year, I estimate that going over 100,000 USD/BTC is so unlikely that it can be represented by a single class. In the lower part, cutting the range short at 10^2.2 seems a good choice.

Just to be clear, our predictions are now expected to be in this format?

Code:

Price >      Probability
100000.0000    0.0001%
 79432.8235    0.0001%
 63095.7344    0.0001%
 50118.7234    0.0001%
 39810.7171    0.0001%
 31622.7766    0.0001%
 25118.8643    0.0001%
 19952.6231    0.0001%
 15848.9319    0.0001%
 12589.2541    0.0001%
 10000.0000    0.0001%
  7943.2823    0.0001%
  6309.5734    0.0001%
  5011.8723    0.0885%
  3981.0717    0.2000%
  3162.2777    0.2000%
  2511.8864    0.5000%
  1995.2623    1.0000%
  1584.8932    1.0000%
  1258.9254    2.0000%
  1000.0000    4.0000%
   794.3282    5.0000%
   630.9573   10.0000%
   501.1872   20.0000%
   398.1072   50.0000%
   316.2278    5.0000%
   251.1886    1.0000%
   199.5262    0.0100%
   158.4893    0.0001%
     0.0000    0.0001%
      Total  100.0000%

(Example only; not my actual prediction)

[wachtwoord]

But it is getting so mathematical that I should perhaps incentivize people by offering 1,000mBTC to the prize pot + all the 50mBTC participation fees. So in any event I will do all the work + pay 1,000mBTC OR $1,000 whichever is higher. (I knew this would happen.. )

FYI: This is sufficient incentive for me to join (even though I would be mainly guessing).

[Chalkbot]

I totally want to play, but I feel like I don't have enough time to learn how.

Edit:
Perhaps there should be a reward for someone to explain this shit like I'm 5.

[Foxpup]

I totally want to play, but I feel like I don't have enough time to learn how.

Edit:
Perhaps there should be a reward for someone to explain this shit like I'm 5.

For each price range, assign a probability to it. Probabilities for all price ranges must add up to 100%, and each individual price range must have a probability of at least 0.0001%*. On the last day of the month, we look at what range the price is actually in, and look at what probabilities everybody assigned for that price range. Those who predicted a higher probability than average are awarded points, and those who predicted a lower probability than average lose points. That's the average of everybody else's predictions for that price range, so you can still gain points from a totally wrong prediction as long as you're less wrong than everybody else.

*If anybody places a probability of 0% on a price range, and then the price actually does fall into that range, it becomes impossible to calculate the scores. That said, placing an extremely low probability on the winning price range will result in you losing an obscene amount of points, unless everyone else thought that price range was equally unlikely.

[snarlpill]

I'm thinking I'm into this, just need some more time (probably tomorrow) to read through it again and fully grasp it.

[rpietila]

I need all this feedback for making the final rules better before locking them in, (so please continue the discussion!).

When they are ready, I update the OP and declare the days of the starting of the competition. The first month will be shortened.

The forecasts will probably have to be in a standard form, and I will provide an Excel file that can be used for convenient calculation. Can we assume that anyone who participates will anyway have a spreadsheet?

Each month you not only forecast the price for the last day of that month, but also the last day of every month of 2014. The Excel file allows for making all these forecasts at one go. In total you submit 8 times and forecast 36 months (8+7+6+5+4+3+2+1).

[Foxpup]

Each month you not only forecast the price for the last day of that month, but also the last day of every month of 2014. The Excel file allows for making all these forecasts at one go. In total you submit 8 times and forecast 36 months (8+7+6+5+4+3+2+1).

Wait, how does the scoring work in that case? Since December will have 8 forecasts, which one is used for scoring? Are the scores for all of them added, or averaged, or what?

[rpietila]

Just to be clear, our predictions are now expected to be in this format?

Code:

Price >      P(31.5)     P(30.6.)     P(31.7.)     P(31.8.)     P(30.9.)     P(31.10.)     P(30.11.)     P(31.12.)
100000.0000    0.0001%     0.0001%    0.0002%    0.0005%    0.0015%    0.0100%    0.0500%     0.2500%
 79432.8235    0.0001%    ...
 63095.7344    0.0001%
 50118.7234    0.0001%
 39810.7171    0.0001%
 31622.7766    0.0001%
 25118.8643    0.0001%
 19952.6231    0.0001%
 15848.9319    0.0001%
 12589.2541    0.0001%
 10000.0000    0.0001%
  7943.2823    0.0001%
  6309.5734    0.0001%
  5011.8723    0.0885%
  3981.0717    0.2000%
  3162.2777    0.2000%
  2511.8864    0.5000%
  1995.2623    1.0000%
  1584.8932    1.0000%
  1258.9254    2.0000%
  1000.0000    4.0000%
   794.3282    5.0000%
   630.9573   10.0000%
   501.1872   20.0000%
   398.1072   50.0000%
   316.2278    5.0000%
   251.1886    1.0000%
   199.5262    0.0100%
   158.4893    0.0001%
     0.0000    0.0001%
      Total  100.0000%

(Example only; not my actual prediction)

That's what I was thinking. This way the format stays fixed every month and it is easy to collect and analyze the data.

The downside is that for the months that are near, the bulge in the probability distribution will likely be only a few classes wide, and not utilize the whole spectrum. But it will nevertheless be interesting how people decide to guess and outsmart each other. The scoring system should balance the guesses so that their aggregate (scaled geometric average) is pretty close to the actual probability distribution.

[rpietila]

Yes, we can divide the prize pot to several people. What do you think of the following:

50% goes to winner
25% goes to 2nd place
12.5% goes to 3rd place
6.25% goes to 4th place
...and so on until Nth place would receive less than 100mBTC. Instead, he receives nothing, and the balance is prorated to the others.

With only 4 competitors, there would be 1,200mBTC in the pot.

1st => 600mBTC
2nd => 300mBTC
3rd => 150mBTC
4th => 75mBTC too little! => 0mBTC

The remaining 150mBTC are given to others:

(4p final distribution):
1st => 686mBTC
2nd => 343mBTC
3rd => 171mBTC
4th => 0mBTC


With 50 competitors, similarly:
1st => 1806mBTC
2nd => 903mBTC
3rd => 452mBTC
4th => 226mBTC
5th => 113mBTC
6th-50th => 0mBTC.

 

[windjc]

This has got to be the worlds most complicated guessing game. Its not really fun unless its simple enough to easily follow and comment about. Instead, this looks like some statistics exam. I'm not playing unless I can understand what the heck is going on.

Otherwise way too much time and energy for a numbers game.

[rpietila]

Each month you not only forecast the price for the last day of that month, but also the last day of every month of 2014. The Excel file allows for making all these forecasts at one go. In total you submit 8 times and forecast 36 months (8+7+6+5+4+3+2+1).

Wait, how does the scoring work in that case? Since December will have 8 forecasts, which one is used for scoring? Are the scores for all of them added, or averaged, or what?

My thinking would be to treat all 36 forecasts as equal, so that each of them is evaluated and scored separately, and scores added.

[rpietila]

This has got to be the worlds most complicated guessing game. Its not really fun unless its simple enough to easily follow and comment about. Instead, this looks like some statistics exam. I'm not playing unless I can understand what the heck is going on.

Otherwise way too much time and energy for a numbers game.

By reading the thread you hopefully get some idea of the purpose of this. It is also +EV on average, because not only I take the time to organize it, I also pay the prize(s). If it is still not worth your time to attend, have a nice day!

[dnaleor]

Yes, we can divide the prize pot to several people. What do you think of the following:

50% goes to winner
25% goes to 2nd place
12.5% goes to 3rd place
6.25% goes to 4th place
...and so on until Nth place would receive less than 100mBTC. Instead, he receives nothing, and the balance is prorated to the others.

(...)
 

Or you can just keep the amounts below 100 mBTC

[dnaleor]

Each month you not only forecast the price for the last day of that month, but also the last day of every month of 2014. The Excel file allows for making all these forecasts at one go. In total you submit 8 times and forecast 36 months (8+7+6+5+4+3+2+1).

Wait, how does the scoring work in that case? Since December will have 8 forecasts, which one is used for scoring? Are the scores for all of them added, or averaged, or what?

My thinking would be to treat all 36 forecasts as equal, so that each of them is evaluated and scored separately, and scores added.

or, you could attach different "weights" to the guesses.


for the july forcast, you get the following:
1 * "first prediction" (end of april)
2 * "second prediction" (end of may
3 * "third prediction" (end of june)
=> the 3th prediction for july will have a weight of 3/6


For the december forecast, you get the following:
1 * "first prediction"
2 * "second prediction"
...
8 * "eight prediction"
=> the 8th prediction for december will have a weight of 8/36