I unfortunately posted this in the wrong section first. Hope it fits better here

I found out that neural networks (NN) have been broadly used to predict Bitcoin (or other cryptocurrency) prices. The common approach is to use historic price data to train the NN. When I did this, I run into troubles in times of high volatility when there is a huge jump or a crash. And this was the problem everybody else run into using ONLY historic price data.

To solve this problem I used a second NN to quantify the news to get a sentiment which I also included for the price prediction. At the moment I am using headlines and started also including tweets from twitter.

I would like to know what you think about this approach. If you are interested in this method I would be more than happy if You visited my Blog about this topic

BR, O.

Hey, thank You for the feedback. I think you will say I am right when I tell you that you can make rough estimates of future prices based on price history. Somehow there are patterns in the prices which are somehow conected to future prices. You can imagine a neural network (NN) as a black-box where you input the price history and it is learning about the patterns to predict future prices. A very good and very basic introduction is given in this youtube-video.

But the price is not only driven through the history, the main influence comes from the news. If .g. the European Union would say "Trading Bitcoin will be forbidden", this would result in a large crash. This is an extrem example, but this is basically what I tried in my blog: Quantifying news. This headline would give a negative sentiment and therefore the neural network would predict a huge drop in Bitcoin price.

I hope it's a bit more clear now how this basically works. I linked a few pages in the blog so that you can get further information.


Thanks for this hint. Unfortunately I just know a bit about the prediction-part. But to blogging and posting I am completely new

If you are interested in predictions that far in the future, I have to tell you that I cannot predict it. And I am completely sure that nobodyis able to make predictions for a timerange that large. If you make such a prediction just based on historical data this wil not work because the price is influenced by much more that just the price history. If you just assume that Trump will twitter "Trading Bitcoin will result in jail", or if the European Union will forbid Bitcoin trading, it will crash and I think it will never come back. And if you go to even longer timeframes, you have also to think about other things. E.G quantum computing could raise huge problems, .
making Bitcoins unsecure. But yeah. If e.g. the European Union tells that Bitcoin will be officially allowed as currency, there will be an immens rise. These are extrem news, but I hope you get what I wanted to explain. I try to quantify the news to make reasonable trading decisions.

I try to get a solid idea about how the cryptocoin will perform in the near future, not more than one day. This basically means that if it should rise, i buy. And as soon as there is a solid reason based on news and historic prices it will drop, I sell.

If you just want to buy and hold cryptos, you dont know what will happen. Until now, Bitcoin recovered from every crash. And yeah...maybe it will never collapse. But if someone tells you that he is 100% sure it will never crash, he is definetely lying.

My goal is to make reasonable trading decisions

Thanks for the hint, but unfortunately I also included these technical indicators which gave just a minor improvement, way less than 1 percent. The NN should be able to detect patterns on its own. Calculate technical indicators based on historical data doesnt give you more information. If the network is trained well, there should at least be NO IMPROVEMENT becasue NO NEW DATA are fed in. I also would say that it should get even worse because assuming the same time for training the performance should get better the less redundant information you get. And addin gindicators should be redundant because its nothing new, its based on historical data.

I think thats what we all want to know

One approach is to make future predictions based on historic price data. But this wont obviously now allow you to predict crashes or huge rises in the price. Therefore you need to take news into account. If you have no idea how to do so, there are a lot of pages which want to sell you their opinions, called sentiment. One website about this is this one.

If you dont wont to spend money and trust on your own work, you should read this post which gives you a basic understanding on how to predict Bitcoin prices based on historic price data AND news!!

I found out that neural networks (NN) have been broadly used to predict Bitcoin (or other cryptocurrency) prices. The common approach is to use historic price data to train the NN. When I did this, I run into troubles in times of high volatility when there is a huge jump or a crash. And this was the problem everybody else run into using ONLY historic price data.

To solve this problem I used a second NN to quantify the news to get a sentiment which I also included for the price prediction. At the moment I am using headlines and started also including tweets from twitter.

I would like to know what you think about this approach. If you are interested in this method I would be more than happy if You visited my Blog about this topic

BR, O.