Interesting short-term price prediction

gentacomp

full member

Activity: 211

Merit: 100

Quote from: K128kevin on April 11, 2014, 09:38:42 AM

Hey guys, I'm sure a lot of you have seen my project/website which uses artificial neural networks to make bitcoin price predictions:

www.btcpredictions.com

I've mentioned before that it has some trouble with predictions when events in the news are significantly affecting the price (like now) but I thought I would post an interesting prediction it is making right now anyway. It seems to think that the price is going to rise relatively quickly into the upper $400s and then fall back down to the $420s over the next 24 hours, peaking around midnight EST. I definitely have some doubts about its correctness myself but it will be interesting to see what happens.

Often times when the 24 hour chart predicts big fluctuations like this, what happens is that fluctuations occur but to a much smaller degree than predicted. I wouldn't be surprised if we see prices go up to around $450 or $460 and then fall back down to something like $430. At any rate though, I think there will be some volatility today and I look forward to seeing what happens!

Any other option ? I mean for LTC etc...
Your 5 days almost perfect, and I like it

K128kevin2

sr. member

Activity: 322

Merit: 250

Quote from: m3g4tr0n on April 14, 2014, 06:21:34 PM

Yes and it makes that 1.3% average error look deceptively good. It's typically much further off with every hour you look into the future.

Well actually no, 1.3% average error does not really make it look deceptively accurate - it's a solid representation of its average accuracy. It is sometimes less accurate than that, but it is also sometimes much more accurate than that. And yes, it does get less accurate the further you look into the future, but at the same time it gets more accurate the closer you look to the present. 1.3% is just the average.

m3g4tr0n

member

Activity: 74

Merit: 10

Quote from: jamesc760 on April 14, 2014, 09:02:21 AM

I think what confuses people is that the graph (and thus the prediction) is a dynamic one, it re-calculates every hour on the hour. It becomes more accurate as target time/date apporaches, you might say.

Yes and it makes that 1.3% average error look deceptively good. It's typically much further off with every hour you look into the future.

jamesc760

sr. member

Activity: 448

Merit: 250

I think what confuses people is that the graph (and thus the prediction) is a dynamic one, it re-calculates every hour on the hour. It becomes more accurate as target time/date apporaches, you might say.

K128kevin2

sr. member

Activity: 322

Merit: 250

Quote from: birr on April 13, 2014, 06:15:42 PM

How many inputs do you use, price and volume, or what? Didn't see any mention in this thread or on your website.

It does not specify this on the website. I decided not to write about the details of the structure of the neural network on the website because the vast majority of people would have no idea what I'm talking about, but that's why I put my email there and invited people to ask any questions if they have any. But I can answer this here.

The inputs do not include volume (though I'm thinking of including it in a new version of the neural network that I'm working on). They correspond with changes in price over a certain time period leading up to the present, and it is different for each neural network. They have between 60 and 120 inputs.

birr

hero member

Activity: 868

Merit: 584

How many inputs do you use, price and volume, or what? Didn't see any mention in this thread or on your website.

K128kevin2

sr. member

Activity: 322

Merit: 250

seleme - I'm still working on compiling that data. I should have a page added to the website with charts comparing past predictions to actual prices up some time next week hopefully, depending on how much work I have.

jamesc760 - Yeah I'm curious to see if that will happen too especially given that it is right after the exchanges are supposed to close in China. Remember that it is making this prediction having no idea about this news, so it's hard to tell whether or not it will be accurate.

aminorex - I will do that

glad to know people appreciate it!

Edit: This is K128kevin... I think my account was hacked and I have no idea why but whatever

aminorex

legendary

Activity: 1596

Merit: 1029

Sine secretum non libertas

Quote from: SportsBet on April 12, 2014, 06:18:39 AM

Stop spamming the forum with your project.

Please continue to inform the forum of your project.
Please disregard newbie detractors.

jamesc760

sr. member

Activity: 448

Merit: 250

A 5-day high of $485 at 7AM on April 16th?

Let's see if it stands.

seleme

legendary

Activity: 2772

Merit: 1028

Duelbits.com

What is the historical reliability of predictions on your site without news driven moves?

K128kevin

member

Activity: 84

Merit: 10

Hpotsirc - Yeah it seems to be much better at predicting the time of peaks/dips, which I guess is really what is most important. I'm still working on some new methods though that might help it become more accurate at predicting actual magnitudes of price changes.

jdun - Yeah I'm not sure what will happen, I wouldn't be surprised whether these predictions are right or completely wrong. As I've mentioned, the neural network does not know about real life events like the exchanges closing that day. In the past month ish, when news stories have occurred it has caused predictions to be less accurate for a short time. I am very curious to see what will happen in the couple days following the 15th though.

jdun

member

Activity: 84

Merit: 10

So you're predicting a large spike from the 15th to the 16th? I guess this is reasonable. I bet it won't spike on the 15th though. The traders will probably need a few days to regroup their thoughts after the exchanges officially close. For example, look at the stock market after it re-opened right after 9/11. It dropped the first day, stayed low for three days, then climbed back up. Three days is the key.

SportsBet

member

Activity: 75

Merit: 10

Stop spamming the forum with your project.

Hpotsirc

newbie

Activity: 43

Merit: 0

Well as you said the height of the top wasn't predicted correctly, however the time was correct. Really worth getting up that early Grin

Let's see how it works out through the day.

disclaimer201

legendary

Activity: 1526

Merit: 1001

You don't need a neural network to be able to tell you that a lot of volatility will happen now. Ask any one person with a working brain. Having said that, I approve this message.

K128kevin

member

Activity: 84

Merit: 10

Hey guys, I'm sure a lot of you have seen my project/website which uses artificial neural networks to make bitcoin price predictions:

www.btcpredictions.com

I've mentioned before that it has some trouble with predictions when events in the news are significantly affecting the price (like now) but I thought I would post an interesting prediction it is making right now anyway. It seems to think that the price is going to rise relatively quickly into the upper $400s and then fall back down to the $420s over the next 24 hours, peaking around midnight EST. I definitely have some doubts about its correctness myself but it will be interesting to see what happens.

Often times when the 24 hour chart predicts big fluctuations like this, what happens is that fluctuations occur but to a much smaller degree than predicted. I wouldn't be surprised if we see prices go up to around $450 or $460 and then fall back down to something like $430. At any rate though, I think there will be some volatility today and I look forward to seeing what happens!

Topic: Interesting short-term price prediction (Read 1944 times)