Topic: rpietila Wall Observer - the Quality TA Thread ;) - page 224. (Read 907229 times)

Game on! 

4466-100000;0%
2818-4465;0%
1778-2817;0%
1413-1777;0 %
1122-1412; 0%
1000-1121; 0%
891-999; 0%
794-890; 0%
708-793; 0%
631-707; 0%
562-630; 15%
501-561; 75%
447-500; 10%
398-446; 0%
355-397; 0%
316-354; 0%
251-315; 0%
0-250; 0%

4466-100000;0%
2818-4465;0%
1778-2817;0%
1413-1777;0 %
1122-1412; 0%
1000-1121; 0%
891-999; 2%
794-890; 3%
708-793; 6%
631-707; 16%
562-630; 26%
501-561; 24%
447-500; 14%
398-446; 4%
355-397; 3%
316-354; 2%
251-315; 0%
0-250; 0%

oh boy, I hope it doesnt go to $1000 - $100,000. Ill get a terrible score

Read again. Even under my current model, you get infinitely many negative points if your prediction for a probability is zero, and it does happen.

So a 76% chance price will be higher than today in the next 30 days. Sounds good.

Sure that helps but I would still employ the same tactic as getting it completely right gives so much points and getting it wrong only means you get a zero score. I would suggest a negative component for being very far away. How about this technique:

* The goal is to get as low a score as possible because the score is an indication for how of you are
* You provide a list of n predictions (i is the index of these predictions) as you did before, where we call each probability P_i
* All the probabilities must add to 1 (or 100% as you put it)
* For each prediction we take the middle of the range in our formula M_i
* The actual price at the end of the term is X

Then we sum for each of the predictions the following:



And of course we add all the score for each round and the lowest score wins.

**The only do we calculate and square all mistakes and then add them weighted on the users estimated probability. This gives a good indication of how wrong each of the participants is.

I am willing to predict where the price is 2014-5-17 (30 days from now) volume weighted average Bitstamp price:

Price in USD range; probability

900-100000; 1.0%
800-899; 2.0%
751-799; 3.0%
709-750; 4.0%
688-708; 5.0%
651-687; 8.0%
621-650; 13.0%
589-620; 18.0%
501-588; 12.0%
431-500; 10.0%
401-430; 9.0%
381-400; 8.0%
331-380; 6.0%
0-330; 1.0%

(sideways until june, with slight ticks upwards on the way there)

LOL, did you make any changes?


The competition would be the sum of many rounds, so under this scoring system you would be very lucky make a high score as missed predictions would penalize you so heavily, compared to the right ones.

so here´s my bet

4466-100000; 0.1%
2818-4465; 0.4%
1778-2817; 2.0%
1413-1777; 2.0%
1122-1412; 3.0%
1000-1121; 2.5%
891-999; 3.5%
794-890; 5.5%
708-793; 8.0%
631-707; 14.0%
562-630; 18.0%
501-561; 17.0%
447-500; 9.0%
398-446; 6.0%
355-397; 4.0%
316-354; 2.0%
251-315; 2.0%
0-250; 1.0%

What I meant was that with these competitions your goal is usually not to have a nice average score. The goal is to become number 1 (or sometimes 2nd or 3rd) to get a high prize. There is not so much difference between slightly lower places or merely ending up last. Therefore my strategy would be to make very concentrated predictions (bets ) probably to the market price at the time of making the prediction (because I don't believe I have predictive capabilities). If I make the ranges small and the percentages large I will either score very highly or extremely low. You polarize the score you'll get and therefore maximize your winning odds.

If you're going to award prizes I really shouldn't have brought this up as this is the only edge I would bring to the table. I've just given it up completely

How would you do it? What if your only competitor was me, and my published numbers? How would you game the system to the maximum vs. me?  

EDIT: After a few trial rounds this will become a serious competition with cash prizes, so better get used to it 

Very interesting approach. I do however have the feeling this system is gameable with a very concentrated strategy.

Now let's start the Accountable Prediction(TM) competition. (I may move this to another thread if it catches on).

The first task is to predict, what is the price in 2014-5-17 (daily average Bitstamp).

Give your answer as a probability distribution where each USD value between 0-100,000 is assigned a % probability and these values add to 100%. All answers given in 24 hours from the timestamp of this post take part in the competition.

And example and my official prediction is the following:

Let's suppose that price ends up being $600. I assigned 18.0% to this band which consists of 69 dollars. Therefore my probability density is 18.0% / 69$ = 0.2609 %/$.

If the price were $400 instead, 6.0% / 49$ = 0.1224 %/$.

An outlier like a doubling the price to $1,000 would have a quite small probability density: 2.5% / 122$ = 0.0205 %/$.

Once the price is determined (daily volume weighted average rounded to the nearest full dollar) and the probability density of each prediction at that price is calculated, a geometric mean of the predictions is calculated. In this example, if the figures above were the contestants' probability densities at the "winning number", their geometric mean is: (0.2609*0.1224*0.0205)^(1/3) % = 0.0868 %.

Then we award plus points to the ones who did better than the mean, and minus points to the ones who did worse:

0.2609 / 0.0868 - 1 = 2.004
0.1224 / 0.0868 - 1 = 0.410
1 - 0.0868 / 0.0205 = -3.237

The point system compares the relative accuracy of different people's forecasts, and the average of the points over several prediction rounds gives a better idea whether this person has the hunch over future prices or not. To predict effectively you should know the system which penalizes severely if you have grossly underestimated the probability of something, yet it happens. It ruins your whole score. Also you should know the distribution of historical results, which for 30 days period is listed here.

Wow well nhl players dont really get the attention they deserve.. in the states ncaa will get played over stanley cup finals.. a shame because there is no sport that can compare to the challenge of winning the stanley cup.. Here in Canada im sure he wouldnt walk around like that.

Supposing that support holds at 480 through the lower volume weekend, the declining resistance trend line from the November 2013 peak becomes close enough to the support that there could be a breakout next week - resuming the rally I am biased to expect, because many share the belief that the bottom of the previous bubble collapse is behind us.

Here is the 12 hour chart extended in to the next couple of weeks.

That is a comfort. Thanks.

Tell him you'll donate $100,000 (1 bitcoin), but ONLY IF he takes it in bitcoin.

One of the top NHL players used to live near me.  He would go to events with my neighbor 2 doors down, who lives in a 1500 sq ft house.  I saw this player at a nearby mall leaving with his wife and kids.  I said, "Hi" and asked him if anyone recognized him.  He said I was the first one all day.

It is well-known that he made $8 million that year and had no particular security and wanders around in public anonymously.  I think you may be overestimating the problem.

If you are in the US, just make sure you make less than $73,800 joint or whatever the equivalent is for single ($36,900).  If you are over, max out your 401k, IRA, charitable giving, buy an electric car, etc.  If you need to quit your job so your income is less than that, then do it.

There is a case here but not including earlier data or giving it a lower weight does run the risk also of skewing the results. One thing we must keep in mind here is that there are some long exponential trend lines going back to April 2011 that have held, so getting a good grip on what happened before the 2011 peak is very important.

U are surely talking about MtGox?