Topic: Raystonn's BTC/USD Probability Forecast (Read 3731 times)

Are you kidding me??? Have you been reading the comments? Nobody here has been grieving but you! People are making reasonable and intelligent suggestions as to how you should improve on what you are doing here and you are dismissing or ignoring everything. Did you expect to post these predictions and just have everyone blindly love you for it? People are providing you with solid and constructive criticism and you are ignoring it and then calling it "grief". In what world does that make any sense?

Stop being so sulky and actually take people's advice - then you might find a bit more success and you'll have a much better overall model of prediction. You are the only one grieving here.

+1

Just don't get discouraged. There is no "grief" as you say, it's just the internet forum phenomenon that escalating is the thing everybody is looking for. I, too, believe that you found a system that works for you, but there is simply no chance to prove it or make it a major useful thing for the community as long as you can't predict the intervals of your predictions.

Seriously, I hope you find a way to make money with it.  Don't listen to the naysayers.  If you find a system that works for you, then by all means do it.

If you want the numbers for the past predictions I'll tally them up.  But I don't think I'll bother with future predictions as I hear nothing but grief.

There really doesn't seem to be much interest here.  So at this point I'm going to discontinue the public predictions.  Current price is 584.88.

And I still want him to compare against buy and hold.  He hasn't done that for a while.

C'mon Raystonn, show us the numbers!

That is a very interesting question, and of course it would depend on the actual percentages. There are some good AI techniques out there that could find the optimal manner of making buy/sell decisions given this limited information. The first thing that comes to mind for me is a technique called Q-Learning, which involves learning how to make optimal decisions in an unknown environment. I'm sure there are dozens of other techniques that could be applicable too.

I have no reason to doubt that Raystonn has an algorithm that he is using for his predictions, and he wishes to test it to see how well it works. But Raystonn, you need to be more rigorous in your overall approach. Neither your predictions, nor your assessment of your algorithm, are useful in their present form.

Naaaah, I think he is indeed doing some research, but as he admitted, he can't predict when parameters change so that there are no fixed intervals. Therefore the analysis might not be pulled out of his behind, it might even work for him, but it is difficult to prove to others and has only restricted value for others. Interesting, though.

I think he is pulling numbers out of his ass and wasting a forum thread and everyone's time for reading it.

About an hour after that prediction, the price rose above $590. Did you make your next prediction then? No, because your above prediction would have been "wrong." So you waited another hour and a half for it to swing down, so you could declare the above prediction "right:"


I don't actually know for sure that that's what you're doing, but it sure as heck looks like it.

Incidentally ... as I type, the price is $580.

Thanks, that was pretty much exactly what I was looking for! I figured someone must have done the theory on this already.

@K128kevin2: I wasn't necessarily thinking about the numbers in this thread in particular. I just wondered how to act on it IF you had perfect predictions. As the wiki article oda.krell linked points out, the best way to deal with imperfect information is simply to bet less.

At least the "how much to invest" part can be answered with certainty by the Kelly criterion, if you do assume that your highest ranked goal is maximizing your (profit) growth rate.

I don't think you could profit at all using the information on this thread given you have no idea whatsoever of the time intervals between predictions. Furthermore, the percentages are not accurate. For almost every single one of his posts, the price has gone in the direction of the higher probability, which is extremely unlikely given the probabilities and number of posts he has made. And even furthermore, you don't know how much the price is going to move. I am almost positive that there is no way you could actually make money by following this unless you just got very lucky.

This thread got me thinking, what's the best trading strategy based on this kind of information?

More specifically, say you get a string of perfectly accurate up/down probability predictions. (Ie. if it says 63% up, that means ~630 out of 1000 times it WILL actually go up.)

How do you choose when to buy/sell based on such information? And how much, or how often? You could do the same placements as percentages (ie. 63% BTC / 37% USD based on the number above), or you could go all-in up or down on every guess, or some hybrid strategy. What strategy would give the most profit over time?

I compare the bid and ask orders sitting on the book at Bitstamp.  It's a proprietary technique.

Current price: $582.00
Forecast until my next post:
    50.8% probability of going up.
    49.2% probability of going down.

how are these probabilities determined?

Current price: $587.00
Forecast until my next post:
    44.4% probability of going up.
    55.6% probability of going down.

Your last two predictions were 60% chance of up (most recent) and 49.1% chance of up (the one before that). By your definitions, do numbers like that qualify as being in a strong trend, or range bound?

That's a range bound market.  The price went both up and down.  In a strong trend this is a very low probability scenario.  In any case, my calls are not tied to price action.  They are based on specific analysis of bid and ask order quantities on the book.

In a market that is not range bound, what I suggest (and what I suspect you are in fact doing) would not be guaranteed of success 100% of the time, but it would work a lot of the time. I gave a very clear example of a situation where it would work very nicely.

What you suggest would only work if I was picking tops and bottoms in a range-bound market. The minute we hit a trend my position would be destroyed, and I'd have to sit with it and hope it came back.  This is not what I'm doing.  I have clearly reported a couple losses when conditions changed and my system changed its forecast.

This is exactly what I was saying but he didn't have any response -_-

Conditions are always changing. If the price has moved against your prediction, do you report that? Or do you wait for the fluctuation to swing in your favor?

Let me illustrate: Suppose that at 1 pm, the price is at $570, and Alice and Bob disagree on the next price movement. Alice says: 99% chance up before my next forecast, 1% chance down. Bob says: 99% chance down before my next forecast, 1% chance up. Reality: at 2 pm, the price is at $575, at which time Alice declares that her prediction was correct and she gives her next forecast. Then at 6 pm the price hits $565, at which time Bob declares that his prediction was correct and he gives his next forecast.

I don't.  The market does.  When conditions change, I report it.

How are you setting the time interval between forecasts?

Current price: $566.10
Forecast until my next post:
    60.0% probability of going up.
    40.0% probability of going down.

Current price: $570.92
Forecast until my next post:
    49.1% probability of going up.
    50.9% probability of going down.

I'll update that later today.

Current price: $575.00
Forecast until my next post:
    35.8% probability of going up.
    64.2% probability of going down.

Any update on your dollar amounts?  That's the part of the thread I find most interesting.

Current price: $525.55
Forecast until my next post:
    66.7% probability of going up.
    33.3% probability of going down.

Actually this is false and that is not how it's done in the real world. Furthermore, as I've stated, I'm not asking for the details of how you make predictions - I'm asking for the general method. My website uses neural networks to make predictions. What do you use to make predictions?

Furthermore, if you don't make posts in set time intervals that are always 100% the same, then your predictions are meaningless and wrong. You can post anything and be correct if your next post comes at the right time.

And even furthermore, literally every single one of the predictions you have made has involved the higher probability price movement occurring, which is actually extremely improbable, meaning that the percent chances that you are giving are actually inaccurate. It's like if you say there is a 60% chance of x being true when you randomize its value to true or false, and you do this 100 times and it is true every time, you were almost definitely wrong in initially saying there was a 60% chance of x being true.

I have one loss in there.  But it's small.  I post the current price at time of post.  So it's all as verifiable as Bitcoin's ledger.

Current price: $529.54
Forecast until my next post:
    54.0% probability of going up.
    46.0% probability of going down.

Thanks for the account tracking.  It makes this thread more interesting.

And it's very impressive that you are (so far) 100% correct.  I assume you would show us if you were wrong and aren't hiding those, right?

Current price: $522.00
Forecast until my next post:
    51.5% probability of going up.
    48.5% probability of going down.

No, you don't.  Trading system information is considered proprietary in the markets.  If I gave out the basis of my systems to everyone they would quickly stop working as everyone used them with tweaks to front-run all the others.  Working systems tend not to last forever.  Working systems whose mechanics are released into the wild tend not to work past the end of the week.  You can choose to ignore the systems, or you can look at track record to assess.  This is how it's done in the real world.  Welcome to Wall Street, Bitcoin.

You really are fundamentally misunderstanding what I'm saying. I never said that you wait until the price goes your way, and I've made it very clear that I am not saying that. The obvious point that I made was that we don't know what you do to come up with these numbers and because we have no idea what you do, your predictions are meaningless. And as I said before, you don't have to tell us. Just know that nobody will care about your predictions or take them seriously unless you say where these numbers are coming from. Not many people are going to come here and trust you blindly.

Current price: $494.52
Forecast until my next post:
    59.5% probability of going up.
    40.5% probability of going down.

Current Account Values as of Last Prediction:
System 1: Account at $1.10.  [Held BTC long forever.  Account value is: $494.52/448.69 ($1.1021417905458111390938064142281)]
System 2: Account at $1.12.  [Long from 498.53 to 494.52.  Loss from trade is $4.01 per BTC, or 4.01/448.69 ($0.00893712808397780204595600525976)]
System 3: Account at $1.11.  [Long from 498.53 to 494.52.  Loss from trade is $4.01 per BTC, or 4.01/448.69 ($0.00893712808397780204595600525976)]
System 4: Account at $1.09.  [Flat]
System 5: Account at $1.09.  [Flat]

Systems:
I will track profitability by showing profits for 4 systems using these signals, plus 1 that just holds long.  The accounts will start with USD$1.00:
1) Holding 100% of the account long forever.
2) Trading 100% of the account long or short depending on which probability is greater than 50%.
3) Same as #2 but shorting is not allowed.  Will go flat instead of short.
4) Trading 100% of the account long (up > 66.6%), short (down > 66.6%), or flat (neither is > 66.6%).
5) Same as #4 but shorting is not allowed.  Will go flat instead of short.

Results:
Here are my first 8 forecasts:

Price at forecast: $448.69.  Forecast: 81.9% prob of up, 18.1% prob of down
Price at forecast: $453.00.  Forecast: 60.5% prob of up, 39.5% prob of down
Price at forecast: $455.40.  Forecast: 76.3% prob of up, 23.7% prob of down
Price at forecast: $491.00.  Forecast: 36.2% prob of up, 63.8% prob of down
Price at forecast: $486.70.  Forecast: 56.1% prob of up, 43.9% prob of down
Price at forecast: $490.99.  Forecast: 55.8% prob of up, 44.2% prob of down
Price at forecast: $492.20.  Forecast: 61.6% prob of up, 38.4% prob of down
Price at forecast: $498.53.  Forecast: 55.3% prob of up, 44.7% prob of down

All systems initially bought in with 1/448.69 BTC.

Current Account Values as of Last Prediction:
System 1: Account at $1.11.  [Held BTC long forever.  Account value is: $498.53/448.69 ($1.1110789186297889411397624194878)]
System 2: Account at $1.13.  [Long from 448.69 to 491.00, short from 491.00 to 486.70, long from 486.70 to 498.53.  Profit is $58.44 per BTC, or 58.44/448.69 ($0.13024582674006552408121420134168)]
System 3: Account at $1.12.  [Long from 448.69 to 491.00, long from 486.70 to 498.53.  Profit is $54.14 per BTC, or 54.14/448.69 ($0.12066237268492723261048831041476)]
System 4: Account at $1.09.  [Long from 448.69 to 453.00, long from 455.40 to 491.00.  Profit is $39.91 per BTC, or 39.91/448.69 ($0.08894782589315563083643495509149)]
System 5: Account at $1.09.  [Long from 448.69 to 453.00, long from 455.40 to 491.00.  Profit is $39.91 per BTC, or 39.91/448.69 ($0.08894782589315563083643495509149)]

Current price: $498.53
Forecast until my next post:
    55.3% probability of going up.
    44.7% probability of going down.

Current price: $492.20
Forecast until my next post:
    61.6% probability of going up.
    38.4% probability of going down.

So you're saying I just wait until the price goes my way?  My last call is profitable already, and has been for quite a while.  If you like you can look at a chart and the time of my posts for the other calls.  I think you'll find no huge negative excursions.

But also if the time interval is between posts, and his posts aren't made in set intervals, he can just say any prediction and be right by not posting again until the prediction is reflected by btc price

Whatever his method is, it seems to be working!

Since you are tracking $1, can you give us the current total along with these?  It would be nice to see how much money you have made against buy and hold.

Current price: $490.99
Forecast until my next post:
    55.8% probability of going up.
    44.2% probability of going down.

Raystonn, do you mind me asking exactly what you do to get the # and %?
Your original posts say from stamps order books.  Do you put the numbers on a spreadsheet etc?

Current price: $486.70
Forecast until my next post:
    56.1% probability of going up.
    43.9% probability of going down.

Current price: $491.00

Forecast until my next post:
    36.2% probability of going up.
    63.8% probability of going down.

Like I said, you don't have to share the details. With my predictions I use artificial neural networks. The specific details of the implementation are not public, but I let people know that the method I am using to make predictions is artificial neural networks. If people have no idea where your numbers are coming from, they won't care about the predictions or take them seriously. For all we know, they are made up randomly.

I haven't yet decided if I will be sharing implementation details of my method.  I will come back to this question later.

lol dude this is not an answer to my question at all. If you don't want to say the method you are using then you can just say that, even though that would be ridiculous. You saying that you make predictions based on the order book is like me saying that I make predictions based on past prices. If nobody has any idea how you are coming up with these predictions, nobody is going to take them remotely seriously. What technique are you using to make predictions? You get data from the order book, but what do you do with it?

As stated in the first post, the method is "based on all orders in the Bitstamp order book".  So it's based on the sitting orders, and not specifically on the price action of trade executions.

I mean what is the general method that you are using? You don't have to go into the details of the specific implementation.

It's definitely mathematical.  I could code it up, as it's purely mechanical.  But right now it's so simple I haven't bothered.

How are you making predictions? Those seem like very specific figures, which leads me to believe you are using some sort of mathematical calculations or software to generate this. I do something similar on my website (www.btcpredictions.com) using artificial neural networks to make predictions. I'm very curious about what you are doing though

I need to clarify what happens with a net short position:  Should the probability of going down exceed the probability of going up, the virtual account will be net short.  This will be done at leverage appropriate such that profit on a $1 drop when short will equal profit on a $1 rise when long.

For purposes of tracking profitability of the calls, I will assume we started with $1 and invested it completely based on the probability.  If I show a 75% probability of going up and 25% going down, then the full amount of the virtual account is invested 75% long and 25% short.  This effectively means the account is flat if the probabilities are 50% up and 50% down.

Current price: $455.40

Forecast until my next post:
    76.3% probability of going up.
    23.7% probability of going down.

Current price: $453.00

Forecast until my next post:
    60.5% probability of going up.
    39.5% probability of going down.

These forecasts will be based on all orders in the Bitstamp order book.

I will track profitability by showing profits for 5 systems using these signals, starting at USD$1.00:
1) Holding 100% of the account long forever.
2) Trading 100% of the account long or short depending on which probability is greater than 50%.
3) Same as #2 but shorting is not allowed.  Will go flat instead of short.
4) Trading 100% of the account long (up > 66.6%), short (down > 66.6%), or flat (neither is > 66.6%).
5) Same as #4 but shorting is not allowed.  Will go flat instead of short.

-----------------

First forecast:

Current price: $448.69

Forecast until my next post:
    81.9% probability of going up.
    18.1% probability of going down.