An AI-powered Twitter bot predicts Bitcoin’s price

[tenrox]

I’d like to share my story about how I made an AI-powered Twitter bot that predicts Bitcoin’s price.
There’s a summary at the bottom for those who don’t have time.

The bot’s Twitter username: @predictBTC

Introduction

First of all, I’d like to introduce myself. I’m a student who loves cryptocurrency, artificial intelligence, and coding. I’ve been coding since I was 15-years old. I made several AI projects like face mask detector, image captcha solver, etc.
Where did I get the idea?

Every time Bitcoin goes into a correction, I get crushed. I was so frustrated because every time I long, the price dumps. Every time I short, the price pumps. It was mainly because of my fear and greed emotions.

I tried to look up ways to control them, and I stumbled upon an article that mentioned that I should use trading bots. “How I didn’t think of this?” I said to myself, “Lemme try making one”.

My First Try

I spent the next week trying to make a trading bot. My goal was to make one that scores at least a 51% win rate. I decided to make it use MACD and RSI indicators to predict the color of the next candle. Around 600 lines of code later, it was the time for back-testing.

“Back-testing is the general method for seeing how well a strategy or model would have done ex-post. Back-testing assesses the viability of a trading strategy by discovering how it would play out using historical data. If back-testing works, traders and analysts may have the confidence to employ it going forward.” — Investopedia.

A Horrible Fail

How did the back-testing go? Let’s say it didn’t end up very well. After testing the bot on all of the top ten cryptocurrencies’ historical data, it scored 50.72%. It’s not bad, you might think. But wait until you see what I discovered.

I ran the back-testing simulation with a $1,000 demo balance. After running the bot through Bitcoin’s historical data, it ended up being around $100,000. I was so happy until I discovered that if the bot had bought and HODLed, it would have ended up being $300M. So technically, it’s worse than just buying and HODLing.

I tried changing the indicators the bot uses, changing the trading timeframes, I even tried to train the bot on chart patterns, but it just wouldn’t pass that $300M mark. I give up.

My second try

Around two months later, I learned how to code artificial intelligence programs. I decided to try again. This time, the bot will use artificial intelligence. I have to give it Bitcoin’s last candlesticks, and it should generate predictions for me.

I successfully trained the AI model with around 80 years of candlesticks data. It was time for back-testing. I still remember how I was sweating as I had just done a chole ting workout. I ran the simulation, and the $1000 demo balance ended up being $1M.

Hope at the end of the tunnel

I was going to punch my computer, but I noticed something weird with the bot’s trend accuracy. I ran a more detailed test and found that the bot’s predictions direction accuracy is around 60%. It means that the bot is not bad at predicting what direction will Bitcoin’s price go. But still, it’s not accurate at predicting the exact price of Bitcoin.

I knew that it was impossible to make such a bot that passes that $300M mark, so I decided to use my current one with 60% trend accuracy to help me do technical analysis. I also had the idea to make a Twitter bot out of it to share its forecasts with the globe.
Nothing great comes that easy.

I thought it would be easy to create an accurate predictions bot as those blogs said. It was a tough journey. It took me at least a month of late nights to make it. The lesson I learned is to never ever give up. I know you hear that a lot, but nobody takes it seriously. If I had given up after failing on my first try, I wouldn’t be writing this now to inspire others.

Summary

Before I learned about artificial intelligence, I made a price predictor bot for Bitcoin that burned weeks of my time in exchange for inaccurate predictions. It was a horrible fail.

Months later, after taking AI courses, I decided to try again. This time, it worked. But it wasn’t as accurate as I expected. I accepted that no such thing predicts a cryptocurrency’s price with 100% accuracy.

I left my dream of letting that bot trade for me, and I used it to create a Twitter bot that helps others expect the unexpected.

Takeaways

Never give up. I know you hear that a lot, but nobody takes it seriously. If I had given up after failing on my first try, I wouldn’t be writing this now.
Nothing great comes that easy. The best things take work.
It all starts with a tiny, stupid idea, then one thing leads to another, and suddenly, you find something amazing.

[mochi86_]

First of all, the countless nights spent and the trials and errors you've been through to achieve something like this is highly respected as a fellow CS majoring person (I'm still in uni, lol).

I am actually very new to investing and whatnot but have a decent grasp of crypto and blockchains.

I just wanted to point the minor mistake of mentioning that the twitter bot goes by the handle of @predictBTC but it shows up as "We've moved" and written with a different @ in the description redirecting to @predictCCbot. Unless that wasn't a mistake? idk, just wanted to point that out. 

[YUriy1991]

I just wanted to point the minor mistake of mentioning that the twitter bot goes by the handle of @predictBTC but it shows up as "We've moved" and written with a different @ in the description redirecting to @predictCCbot. Unless that wasn't a mistake? idk, just wanted to point that out. 

Yes. I see it too, as you said, it should be more practical and not make people doubt and confused.

Takeaways
Nothing great comes that easy. The best things take work.

Must be perfected to completion OP Don't be half-hearted because if someone else comes and happens to also use your application, unfortunately later complain because of the inaccuracy even though we all know no one is able to read BTC price movements in the market perfectly, so in my opinion you should complete one job one by one perfectly,  Then on the other hand, at least you have done with good intentions and steps in terms of making products in the form of price prediction bots, whether you will be devoted to Bitcoin only or there are additions for others.

[Poker Player]

In other words, all the effort for this:

Months later, after taking AI courses, I decided to try again. This time, it worked. But it wasn’t as accurate as I expected. I accepted that no such thing predicts a cryptocurrency’s price with 100% accuracy.

I left my dream of letting that bot trade for me, and I used it to create a Twitter bot that helps others expect the unexpected.

I don't know what you mean by the last sentence, but I hope you're not trying to sell the bot as a good price prediction system. Your story, as a story of effort and overcoming is fine, but the flaw for me is that you are looking for a goal that cannot be reached. No matter how much AI it is it is not going to be able to foresee a panic suddenly breaking out in the markets. Surely if it is good it could make probabilistic predictions with a high percentage of success but never 100%.

[pooya87]

It is impossible to see the future! Writing some code to predict it, is definitely not going to work.
This is why your "bot" fails.

We aren't yet in the unpredictable territory either. I'd say at the time being because of the halving and the similar patterns in history we are in a more predictable phase where speculation about the price is easier in comparison to other times when certain events cause unpredictable volatility.

[BitMaxz]

I guess this Twitter bot predicts that Bitcoin will skyrocket which is obvious due to this coming event block halving.
As a technical analysis, we can only predict possible prices depending on the previous price action you can rely on the support and resistance as a prediction you don't need to predict the future because no one knows what will be the price of Bitcoin but there is a high possibility that Bitcoin price will rise more than the previous price once $73k broke.

[Reatim]

It is impossible to see the future! Writing some code to predict it, is definitely not going to work.

The thing is even though we could get an idea of how the market would move next through analyzing historical patterns and data, we need to remember that how the market is still driven by decisions of people.

Most would say that marker movements are all mathematical but I would argue there is also some psychological factors that affect the movement as well. And that is something that is hard to predict let alone an AI bot. Your efforts are noted and maybe there is some other way you could “predict” the movement as accurate as you can.

[Kakmakr]

I think the reason why so many of these Bitcoin trading bots fail, is because Bitcoin is like no other commodity. What other stock or share has the same volatility as Bitcoin?

People base their inputs on models and trading patterns that are more suited for traditional financial instruments and not something like Bitcoin that has crazy volatility.

I have seen several day traders that failed miserably in predicting the future Bitcoin price, with all their higher education and years of experience. 

[davis196]

1.Twitter doesn't exist anymore. The platform is called X.
2.What's the point of this forum thread, OP? Do you want to brag about your achievement? Do you want to sell this bot(maybe as software-as-a-service or something)? Can you prove that you have created this AI BTC price predicting bot? Sharing a Twitter/X username doesn't work as a valid proof.
Maybe you should move this forum thread in the Project Development forum, since this looks more like a project to me.
Anyway, good luck with your AI programming journey and I hope that you will make a better AI price predicting bot in the future.

[Die_empty]

Takeaways

Never give up. I know you hear that a lot, but nobody takes it seriously. If I had given up after failing on my first try, I wouldn’t be writing this now.
Nothing great comes that easy. The best things take work.
It all starts with a tiny, stupid idea, then one thing leads to another, and suddenly, you find something amazing.

Nothing good comes easy that is why we need to keep trying even after failing several times. Almost all the great inventions that we are benefiting from today took many pains and disappointments before they became stable and useful. I will not depend on any bots to predict the price of Bitcoin, however great job. People who believe in the power of artificial intelligence might find your invention useful.

It is impossible to see the future! Writing some code to predict it, is definitely not going to work.

The thing is even though we could get an idea of how the market would move next through analyzing historical patterns and data, we need to remember that how the market is still driven by decisions of people.

The price of Bitcoin is not just affected by the decisions of people but I institutions and the government making the price unpredictable. I am not sure there was a prediction by this time last year that the SEC would approve Spot Bitcoin ETF. Many people were sceptical that Gary Gensler who was anti-Bitcoin would give up on his history of disapproval.

[Outhue]

That bot isn't a time machine, stop trying so hard to make it look like one.

Even the best predictors in the crypto space are all based on past history and data, and there is room for something new to happen, which will wipe off every past history in the chart.

You can't always be right with Bitcoin, you can only me prepared for any possibility because it goes two sides either bull or bear, instead of trying to be a future reader it's wiser to be prepared for these two sides.

Every bots that are based on predicting the future will always fail, I am guessing that this is all about Bitcoin halving, but still nothing can predict the future, know this and know peace.

Good luck with your back-to-the-future bot.

[kentrolla]

We need to understand one thing that no bot can be accurate because Bitcoin is not static that we can consider only the price chart or past data because there are numerous factors which decides the market and it's impossible to have all the logics implemented into the bot for prediction hence I would say we shouldn't waste time in bot prediction unless we have coded it with certain logics based on our strategy.

[Jet Cash]

I think your mistake was to measure your success in Fiat. If you used your bot to build a Bitcoin holding via trading, and measured its success in Bitcoin, then you would have very different results. I did some scalping on Coinbase to build a few Satoshi, and I've kept those. I ended up with more Satoshi than just a simple buy and HODL. It's a bit harder now with new government regulations and taxes, so you will have to factor in those.

[Cryptomultiplier]

I guess this Twitter bot predicts that Bitcoin will skyrocket which is obvious due to this coming event block halving.
As a technical analysis, we can only predict possible prices depending on the previous price action you can rely on the support and resistance as a prediction you don't need to predict the future because no one knows what will be the price of Bitcoin but there is a high possibility that Bitcoin price will rise more than the previous price once $73k broke.

This AI- powdered bot idea sounds not so different from what crypto signals do. They just predict and no matter how accurate any instrument predicts the price of BTC, BTC will always be affected by demand factor based.
Its price will always be hard to predict given that other variables like government policies and agencies and the halving event already happened and one is waiting for the right time to invest.

I did try to visit the site on X and there's no content there as it has been moved.

[Richbased]

I guess this Twitter bot predicts that Bitcoin will skyrocket which is obvious due to this coming event block halving.
As a technical analysis, we can only predict possible prices depending on the previous price action you can rely on the support and resistance as a prediction you don't need to predict the future because no one knows what will be the price of Bitcoin but there is a high possibility that Bitcoin price will rise more than the previous price once $73k broke.

This AI- powdered bot idea sounds not so different from what crypto signals do. They just predict and no matter how accurate any instrument predicts the price of BTC, BTC will always be affected by demand factor based.
Its price will always be hard to predict given that other variables like government policies and agencies and the halving event already happened and one is waiting for the right time to invest.

I did try to visit the site on X and there's no content there as it has been moved.

You ain't far from the truth because this bot were programmed by humans as well so there is every possibility that it's predictions may near the actual price but not exactly what it would be. Just like you said the demand is a key factor that determines the price movement of Bitcoin and the higher the demand, the higher the supply thereby this process is actually what will have effect on the price of BTC so basically, a bot can't really predict the in-flow that would come to the market after halving considering the fact that it is the determining factor.

However, knowledge will surely increase but uncertainty can occur in the future and create an error that were not discovered now but regardless these AI bots can stand as a prerequisite that will pave way for future speculations and predictions to meet reality.

[pinggoki]

I think your mistake was to measure your success in Fiat. If you used your bot to build a Bitcoin holding via trading, and measured its success in Bitcoin, then you would have very different results. I did some scalping on Coinbase to build a few Satoshi, and I've kept those. I ended up with more Satoshi than just a simple buy and HODL. It's a bit harder now with new government regulations and taxes, so you will have to factor in those.

It's probably because it's easy to understand it and it would be a bit difficult if it's changed in a way that you'd have to be questioning how exactly it works. Some people can't really understand how it would do though and with the number that it might show in bitcoin, some might not unconsciously consider this bot to be this powerful. I've heard of scalping before too, I haven't tried it yet but I'm definitely not the kind to be not trying anything new if I know that there's a possibility that I can get some more wins but for now I just do DCA and that's it, not that crazy but also not guaranteed a big profit in short-term

[Aanuoluwatofunmi]

I’d like to share my story about how I made an AI-powered Twitter bot that predicts Bitcoin’s price.
There’s a summary at the bottom for those who don’t have time.

I will have to commend you for giving this a try and having it to this particular stage, but if we are to also look into how acceptable is the use of twitter bots and other bitcoin price predictions, some may want to track down their bitcoin price through the speculations made from the use of the market price performance chart then relying on bot, except, but if the bot is only about a particular or current tie range, then people will like to follow up with the use and not by receiving predictions on the future price which they may probably doubt how true it could be when they know they can speculate by themselves. 

[btc78]

That bot isn't a time machine, stop trying so hard to make it look like one.

Even the best predictors in the crypto space are all based on past history and data, and there is room for something new to happen, which will wipe off every past history in the chart.

Even though there is usually a saying that goes “history is bound to repeat itself” it does not mean that everything from the past will accurately repeat itself. At some point, there would be a change that is caused by numerous factors that were not present before.

For example lots of people speculate the impact of bitcoin etf on its price pre-halving. No one probably thought of etf becoming a possibility during bitcoin’s early times and thus there was no way you could predict other factors that may or may not happen in the future that could affect bitcoin’s price.

[goldkingcoiner]

The price cannot be predicted exactly, even with AI. That is because there are too many unknown variables which effect the price. And you don't need AI to see the most probable direction that the Bitcoin price will go if you have a basic understanding of Technical Analysis and Fundamental Analysis.

[AHOYBRAUSE]

The price cannot be predicted exactly, even with AI. That is because there are too many unknown variables which effect the price. And you don't need AI to see the most probable direction that the Bitcoin price will go if you have a basic understanding of Technical Analysis and Fundamental Analysis.

Exactly. That’s just how it is.
It’s almost the same like an octopus predicting the fifa World Cup results some years back. Nobody can know what happens, anything is possible.
It’s always funny when people actually believe in this stuff and make their decisions according to AI or whatever "predictions".