This is like if a sharpshooter claims that they are pretty good and you ask them "If you are pretty good, then why can't you hit a mosquito's left eyeball from 2,000 meters away with a blindfold on?"

It's not perfect, as I've said many, many times. It is sometimes wrong, which is why it readjusts its predictions every hour. If you read the older posts you would have seen that it readjusted its prediction one hour after I posted this, saying that it would not reach $500. The neural network successfully predicted a pretty significant rise in price.

It doesn't have to be 100% perfect. I was arguing against the ridiculous claim that it is literally impossible to predict bitcoin prices. I'll give you that it's impossible to predict bitcoin prices 100% correctly 100% of the time, but it's very possible to predict them correctly within a reasonable margin of error the majority of the time. This is better than most (probably any, actually) humans could do.

For you to argue that it is impossible to anticipate price changes using software is like arguing that it is impossible to go to the moon. You honestly could make a more convincing argument that it is impossible to go to the moon. You see here right in front of you, on my website, that I - as an undergrad studying computer science, far from a financial expert - was able to do it myself. This means that obviously people with much more experience than me can do it. And they do it on a regular basis. Use Google... you'll see that these types of strategies are commonly employed and proven to be successful.

Also honestly I find it hard to believe that anybody can believe that THIS is impossible. After all that humanity has done, all of our complex and great technological achievements, you think that predicting stock prices with reasonable accuracy is impossible? It's very possible and I have done it. So have many others.

It's not impossible and people do it all the time. People make ridiculous amounts of money by predicting stock price movement seconds ahead of time and making super short term trades, seconds apart. This is just trying to do it on a longer term scale. And it seems to be working pretty well. Definitely better than any human could do.

Oh that's interesting! I would have guessed that more people would be in the 21-30 range than 31-40. Seems like it's changed since 2011

Edit: Just realized there are only 4 votes so far lol... so maybe not actually. We'll see.

Really? That's awesome! I'm going to set it up today to start saving predictions and match them up against historic data on a chart. What method(s) have you been using to analyze the market?

Please respond to Easy2Mine's post.

It's been only 10 hours, not 24. Plus it predicted this dip along the way, so actually the prediction so far has been very accurate and extremely successful

It has lowered the predicted peak price, but we'll see what happens.

Thanks for sharing it! I am planning on starting to collect/generate/display data for charts that will show predictions against actual prices. I am going to start that either tonight or tomorrow some time. A lot of people have been asking for something like that, and I think it would be interesting to see.


Yes, it trains on the historic bitstamp transaction data, which consists of all transactions for the past (about) 3 years. In order to determine what future prices will be, it looks at a sample of data leading up to the current price. So for example, the 24 hour prediction neural network looks at 60 average hourly prices in a row and tries to predict the next 24.

I experimented a lot with different distances into the past that it would look to make predictions. I tried having it just look at one day's data to predict the next, and I tried having it look at like 30 days worth of data to predict 1 day in the future, and I found that around 60 hours tended to work the best for the 24 hour one. The 5 day one looks at about 20 days, and the 20 day one looks at the past 150 days.

Looks like it changed its mind about hitting $500 =/ might just get to upper $400s. We'll find out soon.

I tried to do a dow jones prediction using a neural network, but the problem was that I was having trouble finding the most up to date data. I might revisit stock price predictions in the future though.

Some of you have seen my website which tries to predict bitcoin prices with software using neural networks:

www.btcpredictions.com

It successfully predicted the drop in price that happened yesterday morning (EST) and it was very accurate throughout the day. It also predicted this rise that we are seeing now, and it's saying that prices will go into the upper $400s, possibly hitting $500! Here is an image of the predictions as of 11am EST:



The software is not always this accurate, but it has been performing exceptionally well recently! I think that we will see prices over $500 within the next 24 hours.

Can I quote you when we never hit $80 dollars? You can quote me when we reach $500 within the next 24 hours

EDIT: Prediction is already starting to come true

Okay you could have made $2 extra if you set it to $418 instead of $420 =P

I think the lowest it will go in the short term is around $420... I would buy at that price if I were buying.

www.btcpredictions.com

I think the price will be a good amount higher in 2 months, so hopefully you won't have to shut everything off. I will be very surprised if it is lower.

I experimented a bit with having it only train on more recent portions of the data in hopes that this might make it better at predicted future data. I might try this again when I work on a 4-layer network, which I'll be starting soon. Like you said, it's hard to know if it might do better on future data by only training on recent data.

The reason that I say the data is large enough that the neural network would have to be able to generalize to attain a reasonable error is because a function that fit to the historic bitcoin price data without generalizing would have to be more complex than three-layer neural networks are capable of. Three layer neural networks are limited to being capable of representing continuous functions, and a continuous function just wouldn't be able to fit to all that data in a way that it wouldn't for new data. This is part of why I'm going to work on a 4-layer neural network (since they are capable of representing any function).

I can see your point about how testing it on the set it is trained on does not show how well it generalizes to new samples. However, regardless of how a neural network is trained, the most accurate test of its performance will be testing it on the sample that is closest to what it will be used for. This is especially true since the data is large enough that the neural network would not be able to attain a reasonable error without being able to generalize.

If I were to train it on 60% of the data and test it on the other 40%, it would definitely perform worse. The reason, however, would just be because it hasn't had anywhere near as much data to train on. A better way of testing the neural network's ability to generalize to new data might be to collect bitcoin price data for about a month and then test the neural network now on that next month's data. This might perform worse than on the test/training sample, but I doubt it would be significantly different.

I think the potential issue you are pointing out would be more relevant on a neural network that was training on a much smaller amount of data. I could see a neural network learning a specific pattern in a set of data that may not necessarily be indicative of new data. However, with something like 3 years worth of bitcoin transactions, I doubt something like that would happen.

I disagree. What could possibly a better test of how accurate the neural network would be at predicting bitcoin prices than actual bitcoin prices (all of them over the past three years)? Testing its accuracy on any other data set would be misleading because the test data would be less indicative of what will happen in the future than the training data.

What you are saying about measuring accuracy would be true for an application like facial recognition, handwriting recognition, or sound analysis, but it is not for bitcoin. The reason is that bitcoin is unique, and sounds/faces/handwriting are not. What I mean by this is that you can train a neural network to recognize John's handwriting, and it will be able to read Joe's handwriting. However, I doubt the equivalent for bitcoin would work. If you trained a neural network to recognize, say, stock price patterns, it would probably perform worse at recognizing bitcoin price patterns than it would if it had been trained on bitcoin price data.

Haha no problem! I think that if enough people followed my charts, it could possibly cause trends to happen sooner than predicted. Also, prices might rise higher and lower than predicted.

I put it online for people to see for a few reasons:
1. Just for the sake of sharing it
2. I'm looking for jobs, and this looks good on a resume
3. I don't really like to buy/sell bitcoins frequently, I'm more of a long term investor
4. If I did decide to use this to try to make money, I don't think having other people see it as well would affect my ability to do that

---

Creo que si mucha gente seguiria a mis predicciones, possiblemente los precios predicidos vendrian mas prontos. Tambien creo que los precios se alzarian mas altos y bajarian mas bajos. Es una pregunta interesante.

Muestro los predicciones gratis por el sitio por algunas razones:
1. Porque me gusta compartirlo
2. Estoy buscando trabajo, y esto mejora mi hija de vida
3. Generalmente yo no compro y vendo btc frequentamente. Yo soy inversor a largo plazo (long term investor? No se si lo dije correctamente)
4. Si en el futuro quiero usar los preddiciones para ganar dinero, creo que podria hacerlo aunque otra gente se pueden verlos

I factor in a lot of different potential indicators of price when making these predictions, but the Magic 8 Ball definitely plays the most crucial role