Bitcoin quote simulator.

[sinuari]

Good afternoon.

I've just finished my degree in computer engineering and I've come up with the idea of programming a Bitcoin quote simulator (I already programmed one a couple of years ago but it was a bit poor, the new one would be much more ambitious).

I have been working on AI for several months and the idea is to take advantage of my new knowledge to improve what I already had. Using several parameters as input to get the trend and maybe some other interesting data as output, it would be through regression.

The price is not going to be a direct output because it is statistically intractable, but it can be inferred once I have the trend, mean and variance for certain periods.

I am open to suggestions for choosing the input data, both raw data (interest rates, stock market indices, gold prices) and elaborated data (if there is not a library that includes them I can always program them).

The development will be done in Python using TensorFlow.

The results of the simulator can be seen here (as soon as I have the first models).

There is plenty of quotation data, the rest is better if they are publicly available or more or less simple.

Thanks in advance to the charitable souls who can contribute with their suggestions.

[Kryptowerk]

Just read your proposal. Sound like a cool fun program (app?). - Which platform do you plan to release this for? Will it be open-source? Do you plan to sell it or give away for free?

Not sure if I get your idea right - do you want a program to write fictional quotes (speech) with reference to Bitcoin or do you want a program that gives you the current rate (quote) for Bitcoin? Why would you need AI / huge data for that? Do you want to make projections into the future?
Slightly confused, English is not my native language, haha.

[jackg]

There's quite a few back testing solutions already available if you're after something that'll test a strategy.

You can quickly search online for historic prices and get daily, weekly, monthly or yearly data downloads (you can normally only download a years data).

Indicator values are available to download too from many places (I think tradingview allow it but I'm not certain and they might be easier to generate yourself).

The difficulty here is how you're actually making these predictions, markets are irrational and hard to predict, day to day activity is a lot easier but can still be skewed.

[Husires]

I am open to suggestions for choosing the input data, both raw data (interest rates, stock market indices, gold prices) and elaborated data (if there is not a library that includes them I can always program them).

There are already a lot of similar ideas and most of them tried to be based on artificial intelligence libraries of old historical data and try to improve the results accordingly, which is the basic learning tool due to the presence of data on the price from the year 2013.
Adding some variables such as exchange rates, political events, Covid-19 and some extreme variables will help improve the quality of the results.

What are your basic ideas to help you?

[so98nn]

Just read your proposal. Sound like a cool fun program (app?). - Which platform do you plan to release this for? Will it be open-source? Do you plan to sell it or give away for free?

Not sure if I get your idea right - do you want a program to write fictional quotes (speech) with reference to Bitcoin or do you want a program that gives you the current rate (quote) for Bitcoin? Why would you need AI / huge data for that? Do you want to make projections into the future?
Slightly confused, English is not my native language, haha.

Yeah, I am stuck here as well. While reading the OP's idea I was thinking about the pricing initially. For example, their program would give us updated price data in the current timeline as well as some future predictions or something. Since they have added the term "raw data" addition and previous data it supposed to be related to price quotes only?

@Op can always fetch the data from price sites like CMC, which has already done great work by adding the bitcoin price data since the beginning. I am pretty sure as you start coding you will have another level of idea's by fetching the data from multiple markets and thus pin poininting towards highly accurate price quote.

Strategy is to collect and verify data from hundreds of sources together if its an AI.

[sinuari]

A few years ago, I developed a simulator based on approximations to normal using the Kolmogorov-Smirnov test. It performed relatively well in bullish markets but struggled to detect changes in trends.

The idea is to improve it using AI. I want to be able to predict, at least, the trend and, if possible, the approximate price for the next day or week.

Basically, I will try to use price, trading volumes, moving averages, market sentiment indicators, inflation rates, and interest rates.

Maybe in future iterations, I can add a news processing system to improve it, but that is not the primary focus for the initial development.

First, I want to develop it, and if it proves useful, I could sell it, use it to make money, or make the source freely available.

I will work with Python, and if I decide to make it freely available, I'm still unsure whether I will make it open source or use a server.

[nioctib_100]

A few years ago, I developed a simulator based on approximations to normal using the Kolmogorov-Smirnov test. It performed relatively well in bullish markets but struggled to detect changes in trends.

My personal thesis is that most abrupt trend changes come from more social factors (i.e. not because the numerical trendline crossed the projected SMA line or whatever), and as a result, it is very difficult to predict future trends and price data from a purely quantitative viewpoint. I think the formula that will truly make it will be the one that can incorporate social "data" into the mix, which is very difficult to do, but I'd argue not totally impossible. Probably would be rather elementary (in relative terms) using today's capabilities, but you have to build for the future - what is going to exist in 10 years is different from today. Whoever makes a formula or calculation that can somehow quantify and build upon social data, which really should be somewhat possible using todays technology, is on the right path to creating something that will change the world of price prediction calculations.

[bettercrypto]

Congratulations on your completed course, and if there is any new idea you have learned over the past few months please share it with us on this forum platform somehow.

        You mentioned Tensorflow from python, it's good as far as I can see, which is an open-source machine learning library, which I think fits what you're planning to do in terms of development that you're talking about.

This is a link that I think can help increase your knowledge https://github.com/topics/bitcoin-price-prediction, Good day to you.

[joniboini]

if I decide to make it freely available, I'm still unsure whether I will make it open source or use a server.

If you're planning to monetize it then you'll probably need to host it on your own anyway, so I guess you'll have to use a server. You can still make the base code open-source I guess, and add new features that are only available on your customized service. They are not mutually exclusive. There are dozens of open-source software out there that keep being updated and people support the dev through Patreon for example. Not sure how it would work on your project though, since the market is quite niche.

[Flexystar]

if I decide to make it freely available, I'm still unsure whether I will make it open source or use a server.

If you're planning to monetize it then you'll probably need to host it on your own anyway, so I guess you'll have to use a server. You can still make the base code open-source I guess, and add new features that are only available on your customized service. They are not mutually exclusive. There are dozens of open-source software out there that keep being updated and people support the dev through Patreon for example. Not sure how it would work on your project though, since the market is quite niche.

May be OP still can go for AWS or other cloud services and fetch the data for open source purposes. They have various options available and plus its always costly when you build up your own server. Hardware server specifically are very rough to manage. You need to have back up servers in a different location than the current location and thus it adds up more cost to the project. Anyways, all of this is only manageable if OP really wants to do this long term and not just for trial purpose as freshman. They will surely need good amount of money to pour into this before perfectly running server can get established.

Also, heartiest congratulation to OP. Do get involved with GitHub community and have all your answers easily found. 

[SamReomo]

Thanks in advance to the charitable souls who can contribute with their suggestions.

I'm not really sure about such system, but if someone gets successful in building something like that which you mentioned then things could be really good for investors. And, if the accuracy of such simulator is very well then I think it can be a sellable piece of software, and the whales might pay you for such simulator without rethinking.

If you're planning to add the function of trending news, then I think it should be far better to train your AI in a way that if could create better simulative environment and more accurate quotes with the help of news trends. If the AI is able to learn the news and is able to simulate the data and can come up with a high quality quote that's mostly accurate then I'm sure that everyone would love something like this.

[joniboini]

May be OP still can go for AWS or other cloud services and fetch the data for open source purposes. They have various options available and plus its always costly when you build up your own server. Hardware server specifically are very rough to manage.

Oh yeah, that was what I'm referring to. I thought that was what OP meant in the first place, not building his own network and maintaining it on his own. If he wants to do that, I guess it really depends on his budget.

And, if the accuracy of such simulator is very well then I think it can be a sellable piece of software, and the whales might pay you for such simulator without rethinking.

I doubt they would base their trading decision on a prediction from a software that has limited backtesting/data though. I believe OP is also aware of that, hence why he plans on adding new features in the future. He can definitely ask for a monthly/yearly plan as long as the features are in demand, but I doubt he'd be able to sell the code to other companies that easily. CMIIW.

[ImThour]

I love Python, it really helps save time and write scripts like these which are fun to program and can also be shared with others and they can run it online or on Google Colaboratory.
https://colab.research.google.com/

OP if you need any help after you are done with your program, do share the GitHub link. I would love to contribute to this project of yours.
Also, after this program if you get any interesting ideas to improve the user experience on this forum, go ahead. That will be a banger project.

[sinuari]

My intention today is to write the code and test it, and then I'll see what I can do.
I'm not thinking about monetising it. I have a job and other projects to make money.
If it's useful and a good opportunity, who knows.