¶1Today we're going to see what happens if I give JPT6 Astra and Fable $5.1,000 each to come up with a strategy that it's going to run on the 15minut BTC updown market on Cali. And my plan is when we have the winner, I'm going to run that for like couple of days maybe. So that's going to be in the future. And then we're going to see what strategy won between Astra and Fable. and we're going to see how that strategy performs over a longer time with actually real money that I'm going to put up.
¶2So, I think it's going to be a really interesting video and yeah, let's just do it. Uh, I'm going to show you a bit about the strategies each model picked and then we're just going to play through how they did and we're going to transfer into my real uh money experiment. So, the way I actually set up the strategies for each uh model was pretty simple. I just went into codeex on GPT6 Astra on code uh cloud code on fable 5.1 and I asked them just to come up with a strategy. Fable did even try to it kind of even created like an own HTML page here about the full strategy.
¶3I'm not going to go through that in depth here. But it did a really good interesting job actually to read everything it the work it did into coming up with a strategy. It ran some tests. It back tested and it it took this really serious but Astra was pretty much I was kind of underwhelmed by the strategy it picked. So you can see Astra basically at minute three into the 15minute window it's going to check is BTC at or above the target.
¶4So up or target is up below is down right and by that direction at minute 4. It also didn't do any kind of kelly sizing or anything like that. It just spends 20% of available cash that is kind of scaled to the balance keeping a reserve for later trades. That was basically it strategy and I thought I was bit h that was a bit lackluster. Uh fable like I said did a much better job.
¶5I think it it kind of went into some back testing. It kind of measures BTC's distance from the target volatility time left. I think it even put up like a Binance uh feed. I think looks for an edge from minute 2 otherwise targets a fallback entry at around minute 7. If it doesn't see any edge it has calculated at minute 2 sizes bet by edge up to 14% before fees stop new trades at 700 settle equity.
¶6Okay, so I said fable had the most interesting strategy. This was a bit straightforward for me. So, the way I ran this over the almost 24 hours was uh using today's sponsor hosting here. I'm going to tell you a bit more about them later, but I ran this on a KVM2 VPS here. And I think I set it to US just to be closer to Cali.
¶7Uh even though I don't know if we needed that, but it's a it's an advantage if you have a bit more lower latency. And with VPS's like from Hostinger, we can actually set that up. You can see I also kept this running as I will show you later on the real money experiment. But uh now I think we're just going to play back how it went over those 20 hours or something and you will see what performed the best here. So here is kind of the recording.
¶8You can see we ran it for 18 hours. That was kind of what we ended up with. So let's just play this now. Uh you can see I have an extension here. I don't know why it's that but uh you can see here uh it started out pretty good for Fable, but you can see I wanted just to pause this.
¶9You can see how volatile Astra is, right? You can see here it the sizing is so much bigger. So it gets this really volatile spike here. It's already up to 1,700. So it almost doubled.
¶10Uh but then it goes on a massive draw down here. You can see here down to 800 again. But you can see Fable is much more stable. It kind of controls its sizing and it doesn't always enter. Right?
¶11You can see Fable here really stable. it kind of doesn't do anything while Astra is still trying it's going down to 600 while stable fable is 1300 and when we comes to the extension you can see Astra falls off even more so you can see now there's just like 3 hours left and Astra is at 400 while Fable is now still a bit down but if you look at the final 3 hours you can see this is where we ended up so we ended up with Astra at 390. That's like lost 6 and over 600 bucks. While Fable ended up almost $200 up. Uh even though it had like a massive draw down here, if we just stopped it here, it would have been like 1,500 up, but uh yeah, it did recover a bit at the end here.
¶12So, let's take a look at what actually happened here. Draw down from peak. Also, this kind of reflects everything. You can see Astra here has massive draw downs. Well, Fable kind of still kind of doesn't have big drawdowns until we get down here.
¶13Now, you can see Fable goes on this massive draw down after almost 15 hours. It kind of runs into its first draw down here and it kind of all the way down to 600. So, next up, we can kind of see every trade, every outcome. If we kind of want to go through, this is more of like the win rate, right? You can see Astra had 43 wins, 21 losses.
¶14So for win rate, you can see Astra had more wins than Fable, but the sizing is all over the place. It doesn't make sense. So it goes on these big losses uh even though uh it had a higher win rate. So win rate isn't always what counts here, right? So if we look at kind of the gap between the two models, you can see sizing is kind of the biggest gap, right?
¶15So fill price and we have sizing. This is the biggest difference. So Fable was much more uh controlled on the sizing part. So when we put everything together here, you can see uh that the fee in uh trading fees is like 237 for Astra while 72 for Fable. So that's a big difference, right?
¶16So if we just look at the final score card, uh we kind of knew this, right? We ended up with a minus 60% return here and a nice almost 20% return for Fable. So that means that the strategy we're going to take over to the live and I'm going to run and I ran for a couple of days was the Fable 5.1 strategy. So I think we're just going to do that. But uh I want to show you how I actually set this up on Hostinger and how you can too to run these uh trading bots 24/7 on a VPS.
¶17Okay. So what you want to do is just follow the link you can find in the description. This will take you to uh Hostinger, right? So you can read about everything here, but I recommend at least what I have been choosing is the most popular KVM2 plan. But uh you can also find other uh options depending on what you need.
¶18Here we will get 8 GB of RAM, two CPU cores and we get 100 GB. That could be good and a really good bandwidth here. So I'm just going to click on choose this plan. Right? And then you will get to your cart here.
¶19Right? So you can see I don't need this for 24 months. Let's just do 1 year to get good value. And down here have a coupon code. Click on that.
¶20Fill out all about AI to get your 10%. Yeah, we got it. Perfect. And yeah, just click continue and sign up. And after you have completed everything in the cart and you have checked out, you will get to your age panel here in hosting.
¶21All right, but I'm just going to show you how I did this. I went to the API for dev tools and just click on API and then I kind of get to this page here. So you can see we have a few options here. We could actually connect this to the hostinger MCP. That's an option if you want to try that.
¶22But the way I did this, I you can see here I created my Kali 1000 API token and then I just follow kind of I went back to codeex again. You can launch the VPS from codeex and just I just find that very efficient but of course you have the option to do that from the dashboard too if you want to. So these are the three steps I use to launch my VPS with my trading bot. So I added my Hostinger API key to a local ENV file in my repository. Uh Codex used the key to authenticate with Hostinger.
¶23So it did everything for me. And from there, Codex could be pretty much helped me with everything. It created like an SSH key, uploaded a Python bot and it strategy files. So, for me, this was like a super efficient way to actually launch a VPS on Hostinger by just signing up, getting the API key, and let Codeex help you launch everything. And just one last tips, if you are struggling with anything, you can just point Codeex to docs.hostinger.com, and you should be able to get all the information you need to actually set up Hostinger um yeah uh on the VPS and launch your trading bot.
¶24So, a big thanks for Hostinger for sponsoring this video. Go follow the link in the description if you want to use them to set up your uh VPS where you can run your trading bots. Okay, so here you can see how our live experiment has been doing. We started with $100. Uh I know it says virtual bankroll here, but uh you can see on Cali like these are kind of the 15minut BTC up and down and you can kind of see the sizing here is really small.
¶25We got in at 69 cents. Nice. And yeah uh also we have a 69 win rate. So not bad. So yeah let's see here now.
¶26So you can see we have this nice graph here all over all the trades. We have been running this for how long now? At least 48 hours maybe a bit almost 72 hours I think. So let's take a look here. Now you can see we have done 36 wins, 138 settled, right?
¶27Uh 444 positions. The max draw down as you will see soon was $17. We have an open position that is yeah we talked about that and you can see down here now. So let's to take a look at the graph here. So you can see we started out really good right in the beginning.
¶28We almost had like a straight uh up to the right curve up to like almost $22 and then we had this 17 draw down and then I decided let's look into it maybe try to change up something. So I created like version 1.1 and when we restarted ah this looks good right we went straight up again and we kept just climbing but then we had this draw down again. So it's been up and down but uh lat lately like lately from yesterday I guess Bitcoin has been doing really good and we did actually capitalize on that. You can see we have some beta here like we are capturing like the we have some draw downs but mostly this is probably just beta right because the market has been going really up and we have followed it all the way up again to like our record curve peak that is plus $27. So if you kind of look at the sizing here we are doing like this Kelly half kelly almost like fractional max order we can do is 14.
¶29I think we have had a few times we put up some uh contracts at 14 and we have an active source that's like a separate price from Binance. So basically we did just copy the the strategy we took from uh from fable and we just put it into live and yeah so far has been doing really good right we can say 70 7% return so far um I know I recorded maybe some part of the video yesterday I couldn't really remember if I did or not what I if I said anything about the return but you can see so far we have been doing pretty good but of course like I said this is a part of the beta. I think yesterday we were all the way down here and yeah, we've been kind of riding the the upwards curve of uh of Bitcoin. So, I would like to say that this is not like a good thing if you kind of want to make steady money. I know it kind of looks like this now, but basically this is a really good project if you kind of want to get into learning about setting up your AI trading bots.
¶30This is a really good project because you don't have to start with like a hundred. You can start with like 20 because we had the draw down at 20 so far. Let's say you had just maybe you can do like a quarter Kelly you can do even smaller positions. So what you learn about is kind of how to set up your agent or AI trading bot uh like we did in Hostinger. Uh that's a really good use case.
¶31Kind of learn about the VPS. You can see we lost here now, right? We lost the position and you can see we kind of uh yeah we lost the 69 cents. So, it goes up and down. That's just the way it is.
¶32But, uh, yeah, like I said, this is a really good project if you want to learn to set up a bot. It's it's perfect for that actually because you don't really you could just re reduce this and just learn about the mechanics and how to set everything up. So, so yeah, that was basically what I wanted to talk about today. And I got to say going from these um starting out with two different models doing like a simulated run with thousand and picking the model that performed best bringing it over to live. It was a really fun experience and it's definitely going to be something I'm going to try again when we get like new models.
¶33Maybe we get like a Opus model. We can try to or like a new Fable or Opus model. We can do it again maybe versus Astra and see if Astra can perform a bit better because this strategy was maybe lackluster. I didn't like this strategy. It was very boring and very Yeah, maybe I should have prompted it better.
¶34I don't know. But I gave the Fable the same prompt. So yeah, really strange. Uh because I I usually really like Astra. I think it's a pretty good model actually.
¶35But uh yeah. So yeah, thank you for tuning in and don't forget to check out today's sponsor, Hostinger, if you want to set up your VPS and all the other things they offer. And if you want to support our channel, follow the links in the description.