Are they really going to destroy them? I don't see why they wouldn't use them for funding or something...

Predictions right now are showing a jump from mid 400s to just over 500 around the afternoon (EST) on 4/8. It's been consistently predicting this jump up for a while so I think we'll definitely at least see some kind of activity even if it's not right. I think we'll see a rise in price.

TwinWinNerD- Really? I've found that (with the exception of when the China ban thing happened) it almost never predicts the wrong direction of movement. It might predict the wrong direction if it, for example, says the price will go up by $2 but it ends up going down by $2 instead, which would be less than 1% error. For pretty much all of the major price changes I've watched it predict against, it's gotten the direction of movement correct but often overestimates the magnitude. There has been one or two instances where it predicted for the price to move up or down and it pretty much stayed the same, but I have not seen it predict a price change of more than like 2% where the actual price ended up moving in the opposite direction (except right after the China ban).

Why would anyone take a bitcoin over an ounce of gold right now? Which you could sell and immediately buy more than one bitcoin with lol...

It definitely seems like it's following that 6 month pattern. I think it will be a little less this time though - I would expect to see a spike sooner rather than later.

Be careful with the 20 day prediction, because it does have a large average error (about 8-8.2%). That one is not particularly reliable compared to the others, but I think it may be indicative of long-term trends. It has been much less than 20 days since I've had it running so I actually don't really know yet. We'll see I guess. The 24 hour prediction is pretty accurate most of the time though, and the 5 day one isn't quite as good but it still does reasonably well.

I implemented a 3-layer neural network that uses backpropogation. It trains on the entire history of transactions on bitstamp, so about 3 years worth of data.

Dr Bloggood - 20 days is the furthest I've been able to get the neural network to reasonably predict into the future, and unfortunately that one isn't actually especially accurate. The further forward you look, the more difficult it is to make predictions because real-world events that software can't anticipate will cause price changes.

toxic1978 - Yup! That's pretty much exactly what I'm trying to say.

catfish - That seems very interesting! So if I understand correctly, your idea was to have a neural network that could predict the behavior of individual traders in response to market changes, and then based on those predictions you could extrapolate to see what would happen to prices?

Also did they have multi-layered neural networks back when you were thinking of doing this? Because I know 3+ layers is relatively new and this would definitely be impossible with anything less than 3 layers. I guess computer speed back then might have made it difficult as well.

Haha I was waiting for someone to bring up quantum mechanics. I feel like that's the only legitimate argument you can make against what I said. I don't understand enough about quantum mechanics to really discuss this, but from what I've heard from people who actually study it and know a decent amount about it (nobody knows "a lot" about quantum mechanics) all of the things about randomness are still pretty theoretical or just not well understood.

It's a very interesting subject, but regardless it doesn't conflict with my point that bitcoin prices are NOT random.

The outcome of a coin flip is determined by the force with which you hit the coin, how far from the center you hit it, and tons of other physical properties that are obviously extremely difficult to measure. This is why it effectively is random for us, but it's not truly random, as you claim bitcoin prices to be. Theoretically if you could observe every physical factor that would affect the outcome of a coin flip, you could know with 100% certainty what the outcome would be. This means that it is not random.

The coin flip is a terrible analogy to bitcoin prices. Coinflips have a nearly even distribution of heads vs tails (it's not actually even). Bitcoin price movement is not binary, and it is not an even distribution. A coin flip would be a valid analogy to a market where the value of the commodity is ALWAYS either $1 or $2, and NEVER anything else. And also where it is $1 50% of the time, and $2 50% of the time. I hope this shows you how flipping a coin obviously doesn't even come close to making sense as an analogy to bitcoin prices.

And as I explained to you earlier but maybe you forgot or didn't read - it readjusted its prediction one hour after I made this post, and it was pretty close after readjusting. It accurately predicted the rise in price, just not to the correct degree. I've also repeatedly said it's not ALWAYS right. Saying that one prediction is wrong (which it wasn't) and therefore none are right is like saying seeing one person win money at a casino and saying that casino NEVER earns money.


I looked up who that is - that book looks kind of awesome! I might check it out.

Have you been reading what I wrote at all? Do you still not understand that there is no randomness? This isn't an opinion, it's a fact. NOTHING IS RANDOM. EVER. Not even the random number generators in computers - it's based off of complicated strings and your cpu clock.

Do you realize that if you can predict whether the price of something is going to move up or down correctly more than 50% of the time, you are successfully predicting future prices? Nobody said they can predict prices perfectly 100% of the time.

I don't understand how you don't understand this -_- It just plain is NOT random. All I'm claiming is that I can predict the price within a reasonable margin of error MOST of the time. And it's not like you can argue against this, because it's objectively true. My software does it on the ENTIRE HISTORY OF BITCOIN DATA every hour and predicts within an average of 1.3% error. You're arguing against something that is just clearly objectively true.

Edit: Read the name of the post!!! I said it's a prediction, that that it WILL be $500! ugh please try to understand...

By adjusting the predictions every hour, accuracy is greatly increased. It allows itself to change its prediction if something happens that changes the course of prices. I know it can seem like it's changing its mind, but it rarely significantly changes the prediction.

I'm not sure what chart you've been looking at, but the 24 hour chart has been right much more often than it has been wrong over the past 3 weeks. The 5 day hasn't done quite as well but it has done decently. The 20 day one is not reliable, as I've explained before. I've started collecting data to prove this, and you'll be able to see charts with predicted prices compared against actual prices. This will be up within a week hopefully.


That is a potential problem, though I don't think it is one yet. I don't think my site is popular enough yet to have a significant influence over the market. If/when it does become popular enough, I expect that changes in price might be more dramatic than it anticipates, and they might occur sooner as well. It definitely shouldn't predict the wrong direction of movement or anything though.

Once again, saying that it is random is literally the same as saying it is magic. And I do mean literally.

I'll give you a brief explanation of how it works, but I doubt this will help:

I use a 3-layer neural network that trains on 3 years worth of historic data from bitstamp (about 4 million transactions I think). It looks at 60 inputs (average hourly prices in a row) and spits out 24 outputs (next 24 hourly averages). It trains about 30-35 times by making predictions at a few tens of thousands of points throughout the historic data and adjusting the weights of the edges between nodes according to the errors of its predictions. After about 30-35 times training on that set, it can predict 24 prices with an average error of about 1.3%.

And you have to realize, I don't know a lot about finance - I know a lot about computer science and data. I don't own any stocks and I don't have enough money to buy any stocks. I own 0.6 bitcoins and I can't afford to buy any more, so I'm just holding them. You need to have money to spare in order to try to make money on the bitcoin or stock market. I don't have that. It's not easy to make money in these markets unless you have plenty of extra money to spare, and even then I doubt it's easy.

Thats not what I said at all.  I said markets are random so its impossible to predict them. They are by nature unpredictable.   If you think you can predict future prices youd be a billionaire by now. Just put all capital into options and you can double it every day

I was just curious if your software was a good trader.   Thats why I offered the challenge.   Maybe the chaellenge should be human and computer starting w same amount of money and see who makes more after a certain period.

Its a cop out if you can't explain what you are doing to predict these prices.  Telling me to google neural networks doesn't make you know future prices more than me.   I admit don't know future prices.   
[/quote]

To say that something is, by nature, unpredictable is to say that it is random, which is literally the same as believing in magic. Nothing in the world is random - everything has an observable cause. If something has a cause, then it is not random. If it doesn't, then it is magic. That's what magic means. Maybe we don't have the means or technology to observe something, but it is still theoretically observable. This is besides the point though.

People VASTLY overestimate the power that comes with the ability to accurately predict bitcoin prices MOST of the time. This does not mean you will instantly become rich, or even become rich ever. It means that there is a higher percent chance for you to make some money on the bitcoin market. To think that someone could become a billionaire as an undergrad in college speculating on bitcoin prices is absolutely ridiculous. This isn't a magical get-rich-quick tool and I've expressed this repeatedly to everyone.

To say that I can't explain it in a single post is not a cop out. If you really want to understand it you have to do some research. There are a lot of things in life that can't be explained in a paragraph, and this is one of them. Also it's not my job to explain this to you. I have tons of other things I need to be working on. Asking me to explain to you how neural networks work would be like asking a chemistry professor how chemistry works. Sure, I could give you some brief, vague, high level answer but you still wouldn't fully understand how predictions are made.

I know 2 women who own bitcoins, but I don't know any who spend them actively

I feel pretty confident that it will go down before up. I would set my buy at around $415-$420 if I were going to buy soon.

www.btcpredictions.com - this isn't always right but it's very good at predicting direction of movement. I will be very surprised if we see $480 before $430.

I think that when we hit $480, it's not going to be from a slow gradual climb - it will be part of a spike that takes us above $500. That's my prediction anyway.

That's the attitude xD


Glad you like it! And yeah I think the fact that it gets more and more accurate as time goes on is one of the coolest things about a project like this. There really is a relatively small amount of data available since bitcoin is so young, but as more data becomes available the accuracy should improve more and more!

I'm sorry but I'm not going to bother trying to get my software to predict S&P right now because I have a ton of other work to do. If you want to test accuracy, you should just make your own bitcoin price predictions and test their accuracy compared to my charts on this website.

If you want to know how it works look up artificial neural networks on Google. This isn't some simple process that I can explain to you in one post. It's a very complex computational model that people have been developing and researching since the 1940s. You could teach multiple college courses just on neural networks alone. To be honest, it seems like you are refuting 70 years worth of research based on your gut feeling that humans would be better at this - and if we're going to be realistic, computers are better than humans at nearly all tasks involving any type of computation. They're surpassing us quickly in other realms as well. They have been able to beat us in chess for a long time now, they can beat us in almost any other game, and they can even drive cars better than us now. Is it so hard to believe that they can predict stock prices better than us?

Thanks, I'm glad you can actually appreciate it


hmmm, you obviously haven't been following them then. The only time they were consistently off during the past month was for about a 24-36 ish hours immediately after the China ban. They've been wrong before at other times but definitely not consistently.

You should search Google for studies using neural networks for stock price prediction. You will find plenty of examples of it being done successful.

I will reiterate my main point that I expressed before though: Neural networks can find patterns that humans can't. Just because we can't see patterns or understand them doesn't mean that they aren't there. Neural networks can find complex and extensive patterns that no human could possibly ever understand. I use 60 inputs to predict the 24 prices for the 24 hour prediction. The hidden layer has 200 inputs. This basically is like representing a function with 12,000 variables. Obviously this is extremely complicated, and obviously there is no way a human could understand a function like this.

No matter how often you trade and how much time you spend trying to understand patterns in price fluctuations, you will never come even remotely close to being able to understand these patterns as thoroughly as a properly implemented neural network. It's like trying to beat a calculator at math - you can't do it.

The thing that you have to understand is that just because we, as humans, can't see predictable patterns in a set of data doesn't mean that they aren't there. Neural networks are REALLY good at finding patterns that humans cannot see. The first time I ever implemented a neural network was to recognize hand-written digits. It was able to correctly classify some hand written digits that were written so poorly that I, a human, could not tell for sure if they were one digit or another.

Human behavior is not totally unpredictable. Especially given large numbers of humans. This is besides the point though. The actual cause of price fluctuations is completely irrelevant actually. My computational model has no idea what is causing prices to change because it doesn't matter at all. All that matters is that there are repetitive patterns in the data (which there are). The neural network learns these patterns, and is able to use them to accurately predict bitcoin prices.

You can say you don't believe that bitcoin prices can be mathematically modeled, but it is still true - it's exactly what I'm doing (relatively successfully) here.