YT YouTube2Text

YouTube2Text → Transcripts → GPT: Trading Strategy in Python makes 805% (+ Monte Carlo simulation results)

GPT: Trading Strategy in Python makes 805% (+ Monte Carlo simulation results) — Transcript

by Algo-trading with Saleh · 4,775 words · 637 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hey guys, it's so today I'm going to
0:01show you a swing trading strategy that
0:04not only produces good results in a back
0:06test, but it also has one of the best
0:08Monte Carlo simulations results that
0:10I've ever seen. As always, first I will
0:12show you the entry and exit rules of the
0:14strategy on Trading View. And then we
0:16will feed those rules into GPT to
0:19convert into Python code so we can run
0:21some actual back test on them. And
0:23before I continue, I got to say I am not
0:25a financial adviser. This is a tutorial
0:27and you should always do your own
0:29research before trading. So with that
0:30out of the way, let's get right into the
0:32video. Now firstly, we want to ensure
0:35this is a trending market. For that, the
0:37best indicator that I always use is the
0:39ADX. So let's add it to our chart.
0:42Now we need the threshold of this to be
0:4443. So let's add it here. So you see, I
0:48only want to take trades if the value of
0:50the ADX is above the threshold, like in
0:52here. Now, by the way, the time frame of
0:53the strategy is also the 4 hours. So,
0:56make sure you set your chart to this.
0:58Next, we want to make sure it is an
1:00uptrend or a downtrend and take a trade
1:02according to that. For that, we're going
1:04to simply use three moving average
1:06lines. So, let's look up EMA and add it
1:09three times to my chart. And then, we're
1:12going to change the periods. First,
1:14instead of nine, let's use 30. Next, I'm
1:18going to edit this to 46. And next I'm
1:23going to increase it to 80. Now also
1:27let's change the style of this to be
1:29more clear. So let's make this a little
1:31bit thicker. All right. So you see I
1:33want to take a trade when the order of
1:35lines is like this. I want the blue line
1:37which is the fastest email line to be
1:40above the one that is slow. And both of
1:42them must be above the slowest one which
1:44is also the thickest one here. Now again
1:46the periods of 30, 46 and 80. And we
1:50also want the ADX to be evolved with
1:51threshold. So this would have been a
1:53valid trade for example. So you could
1:55say this is a slow taking strategies
1:57kind of similar to the turtle strategy
2:00if you've heard about it. So we're not
2:02going to be early in the trend. But
2:04because this is a swing trading
2:05strategy, we don't really care about
2:07that because as long as we can catch a
2:09trend, that's good enough for us. So
2:11this would have also been a valid trade.
2:14So would have been this. So you see in
2:16this entire trend we would have taken
2:17two trades which is really good and
2:20that's it really that's all the interior
2:21rules that is to it and now let's talk
2:23about how to exit the trade. So first
2:25let's remove this. Now for that I'm
2:27going to use the ATR indicator which is
2:30one of my most favorites. So basically
2:32this will give us an average number of
2:35how much the price is changing on each
2:38candle. So for example, if we had opened
2:40the trade let's say here or here or
2:42whatever this value here which is by the
2:44way an absolute value. So in this point
2:47it would have been 939.1
2:50right? So let's zoom in a little bit. So
2:53we want to set the stop loss below let's
2:56say this value but not just by one let's
2:58say multiply it by a value such as I
3:00don't know like three or two. So that
3:02would have been for example in here and
3:04then take profit would have been let's
3:06say here right? So this is how we would
3:09have set it. So the ATR you want to
3:11multiply it by a certain number and then
3:14add or subtract it from our entry price.
3:16Now once you write the code it will
3:18become more clear for you if it's not
3:20already. Now because this strategy is
3:22not taking that many trades I want to
3:24actually risk a little bit more. So I'm
3:26okay with risking 5% for each trade and
3:29that is 5% of my capital. Right? So it
3:32doesn't matter how many percentage it is
3:34from our entry to our stop loss. Okay.
3:37So what we care about is 5% of our
3:39capital. And for example, if I were to
3:41open a long position here and I wanted
3:43to risk 5% of my capital because you see
3:46the distance between here until here is
3:49actually less than 5%. That means that
3:52we would have had to use leverage in
3:55order to accomplish that. So unlike what
3:56people think about leverage, which is
3:58you know you are always risking more and
4:00it's always risky and dangerous, that's
4:03not really the case. like sometimes it
4:04could actually be used in a limited way
4:08and that is really helpful. All right.
4:09So now that we have both the entry and
4:11exit rules of the strategy, we can feed
4:14those into GPT to give us some Python
4:17code so we can run some back test on it.
4:19Now to do that you have two options. The
4:20first one is to use the cloud version
4:21that I made myself. It is basically the
4:23same GPT with custom models such as
4:25cloth on it, GPD5 or whatever, but with
4:29some custom instructions that I fed it
4:31specifically for Algo trading in Python.
4:33Or if you want to self host this thing
4:34yourself to use your own API keys if you
4:37have some privacy concerns or whatever,
4:39you can check out my channel and watch
4:41this video where I show you how to do
4:43exactly that. So again, you don't have
4:46to use the cloud version that I'm going
4:47to do, but it's going to be easy for me.
4:50So that's the one that I'm going to
4:51stick with. Now, as for the model, I
4:53have picked GPT5 mini, which is also the
4:55cheapest one, and you can get started
4:57with it absolutely for free. So, that's
4:59the one I'm going to use. Let's paste in
5:02the prompt. All right. So, write a trend
5:03strategy with these rules. Go long when
5:06you have an uptrend and the ADX is above
5:0843. For the trend, use three EMAs. A 30
5:11period EMA is 46 period and 80. The
5:15trend is up when the 30 EMA is above the
5:1746, which is above 80. And the current
5:20price is above all of them. Enter at
5:22price market risk 5% of account per
5:24trade. Set the stop loss at the
5:26enterprise minus 1.8 times the ATR value
5:29for take profit. Use the enterprise plus
5:313.3 times the ATR value. So you see when
5:34I win, I'm going to win bigger than when
5:36I loot. So even if the win rate of the
5:39strategy is going to be like 40% or 45%,
5:42I'm still going to sit at a win. So I
5:45don't need to win all the time. So,
5:46that's one of the good things about a
5:48strategy where your wins are going to be
5:51bigger than your losses. And for short
5:53positions, do the opposite. That's it.
5:55Very simple. Let's hit enter. All right.
5:57So, it gave us the code. Let's copy the
5:59name of it. Triple EMA trend and go to
6:03Jess's dashboard and generate a new
6:05strategy. And I'm going to paste the
6:07name here. Now, here's the code. Now,
6:10you could edit the code right from
6:11within the editor of Jess's dashboard.
6:14But what would have been better is if I
6:16open it inside a better editor such as
6:19cursor or VS code which is completely
6:21free and open source. So let's open it
6:23there. And now we can read the code. All
6:26right. So you see this is the base
6:29strategy code that Jesse generated for
6:31us. So let's go back to Jess GPT and
6:34copy this whole thing and paste it here.
6:38Now let's give it a read. So it defined
6:40a new property for EMA30 and in it it is
6:43simply returning TA. EMA which is the
6:45correct syntax within Jesse framework
6:47for defining an indicator. It is passing
6:50the current candles by saying
6:51self.candles and the period of it is 30.
6:54Define another one for EMA 46 EMA 80 and
6:59then it defined the ATR simply saying
7:01TATR and then passing the current
7:03candle. So you see it is very easy to
7:05define indicators inside Jesse and then
7:08it defined a new property for the ADX
7:10and next we get to the entry rules of
7:13the strategy using the should long
7:14method which is basically asking should
7:16we go long right now. So in it we just
7:19define some inter rules. So it's saying
7:21if the EMA 30 is above 46 EMA 46 is
7:24above 80 the price is bigger than all of
7:26them. So I find this to be a little bit
7:29hideous and unnecessary, but I'm going
7:32to leave it as it is because it works
7:34and you know unless if the model makes a
7:36mistake, we don't really need to change
7:38it. And if the current ADX is also above
7:4143. So this is perfectly correct. And
7:44then for the shoot short, which is for
7:45taking the short positions, it's doing
7:48the exact opposite. So this is also
7:50correct except remember that the ADX
7:52must still be above 43. So this is
7:55perfectly correct. And then for the go
7:58long method which is where we define the
8:00position sizing right. So assuming that
8:03the inter rules are correct and now we
8:05want to go long what should be our
8:07position sizing what should be the entry
8:09price or the stop-loss price or things
8:11like that. So we're going to say the
8:14entry is the current price which is a
8:17market order in other words and Jesse is
8:19smart enough to detect this behind the
8:22scenes. All right. And then for this
8:23stop loss, we're seeing the entry minus
8:25the current ATR multiplied by 1.8. And
8:28then the quantity, we want to risk 5%.
8:30Right? Now, to do this, we're going to
8:32use the risk to quantity utility
8:34function of Jesse. And what it needs
8:36from us to calculate things is our
8:38current capital or available margin. So
8:41we pass self available margin and then
8:43the number five for the number of how
8:47much we want to risk the percentage and
8:49then the entry which we defined here the
8:52current price and then the stop which we
8:54defined here and then we also need to
8:56set the fees. It's not necessary but
8:58it's better to do it. And for that we're
9:01saying fee rate equals self.fe rate
9:03which is the fees we're going to pay in
9:05the current exchange. And then to submit
9:07the actual buy order, all we need to do
9:09is to say self.by equals quantity and
9:12then the entry. That's it. Very very
9:15simple. Now for a short position, we're
9:16doing the exact opposite. So basically
9:18for the subus, instead of subtracting
9:20the ATR, we are adding it. And also
9:22instead of self buy, we are using
9:25self.ell to submit a sell order for a
9:28short position, which of course makes
9:29sense. Now assuming that everything went
9:31fine and we are now sitting on a
9:33position now is the time to submit both
9:35the takerit and the stop loss. So we're
9:38saying if it's a long position and that
9:40is very simple in JC simply say if self
9:43is long then we want to calculate the
9:45stops and take profit prices. Now we
9:47didn't have to do it like this. We could
9:48have simply write it here but this model
9:50chose this and we're going to go with
9:52that. So the stop loss equals the
9:54current positions entry price which is
9:56the current price also minus the current
9:58ATR minus 1.8. Now by the way the
10:01current ATR is what we defined here. All
10:03right. Now the take profit is going to
10:06be similar except we are adding it by
10:093.3 times of the ATR. And now is the
10:12time to actually submit the orders. Now
10:14this is the syntax in Jesse. So for a
10:16stop loss we say self to the stop loss
10:17and then we pass the quantity which we
10:19are passing the quantity of the current
10:21position and then the stop loss. Now for
10:22take profit still we're doing the same
10:25thing but except instead of saying self
10:27stop loss we say self dot takeprofit and
10:30if it's a short position again very easy
10:32in Jesse simply say is short we do the
10:36opposite right so instead of subtracting
10:38and adding we are adding and subtracting
10:40that's it now it is also doing something
10:42extra that I didn't ask so it's saying
10:44it's using the update position method of
10:45Jesse which we use when let's say for
10:49some conditions we want to liquidate the
10:51current position we don't want to wait
10:53until the stop loss or the take profit
10:55price is hit but that is not what I
10:57wanted to do in this strategy so this is
10:59extra and I'm going to remove it is also
11:01defined the should cancel entry method
11:03and it's returning true this method is
11:06used when we are using a limit or a
11:08stop-loss order in Jesse which is really
11:10necessary right because let's say you
11:12submit your limit order but the next
11:14candle closes and still your price
11:16hasn't been hit so now you ask yourself
11:18should I update the previous order or
11:20should I just leave it as it is but
11:21Because in this strategy we are using a
11:23market order. We don't need this and I
11:26can just remove it. All right. So now
11:28that we have everything we are ready to
11:30run some back test. Now by the way the
11:32LLM model added an extra py here. So
11:35make sure to remove that if it happened
11:37for you because that will cause an
11:39error. So let's go to JC go to the back
11:41testing page. Change the symbol into BTC
11:43USDT. the time frame into four hours and
11:47the strategy name must be triple EMA
11:51trend. And for the duration, let's begin
11:54since 2021 up until 2022
11:58and the first month.
12:01Make sure the fast mode is on and so is
12:02the benchmark. All right, so let's run
12:04this. And while that is going, I'm going
12:05to open another tab and it starts since
12:082022 up until 2023.
12:12Run it. And then another one with 2023
12:16up until 2024.
12:19And then 2024 up until 2025.
12:24And lastly,
12:26starting 2025
12:28until this month, which is September. So
12:31let's run it and let's take a look. So
12:34in 2021, this is how our equity curve
12:37looked like, right? So we took 14
12:39trades. We ended with 78% profit. The
12:43max was -3%.
12:46The win rate was 71%. So that is really
12:48great for a strategy where the average
12:52win to loss ratio is 1.44. So this is
12:55really great. The average holding time
12:57is 265 hours. So that is definitely
13:00long. Now if you are a swing trader,
13:03this might be fine for you. But if you
13:04are a scalper, you're going to have a
13:06hard time holding a position for that
13:09long. So just consider this and the
13:12sharp is close to two. So I love this
13:14results at least for this year. The
13:16calmer ratio is also 5.63.
13:20That is really good.
13:22All right. So let's go to the next year
13:25for 2022 which was a bare market. And
13:28here's our results. So we ended with 67%
13:32and a max of minus3. And this is for
13:362023.
13:38This is this is for 2024 and this is for
13:412025. Now if I go to the benchmarking
13:43page of Jesse, we can see all the
13:45results next to each other which makes
13:47it really easy for us to compare the
13:50results. So here is the maxon. You see
13:52all the numbers here and then we can see
13:55the sharp and compare all of them
13:56together. Now by the way the maxon of
13:59the strategy is a bit lower than my risk
14:01tolerance. So we could even go and
14:04increase the quantity. So let's go to
14:06the go along method for example and we
14:09can multiply this quantity number by a
14:11certain num multiplier such as two or
14:141.5 or whatever you're comfortable with.
14:17So, in case you were wondering why
14:20aren't we beating the market? Actually,
14:21we are beating the market here a little
14:24bit and also here. But in here, we're
14:26not beating the ending P&L number that
14:29the market made for us, right? Like if
14:31we just held BTC, we would have made
14:33more. But with the strategy, we didn't
14:36make as much. But because the max number
14:38is actually really low, if we had
14:41multiplied the quantity or the size of
14:43our position by three, we would have
14:45definitely beaten the market, right? So
14:47in case you are wondering for that,
14:49well, this is the solution to that. But
14:51when you do that, you also increase your
14:53risk. So that means if the strategy
14:55wasn't doing fine, you would have lost
14:57more. So you need to be very careful
14:59with that. But anyway, so let's go back
15:00to the benchmarking page and we can see
15:04the start and ending dates to know which
15:07result this is. We can very quickly
15:10rerun one of the results if that's what
15:11you wanted to do. We can also rerun the
15:13entire back test if that's what you
15:16needed. Let's say I change a certain
15:19part of the code. So let's multiply this
15:21by 1.5 for example
15:24and also for the shorting part. And
15:27let's say I wanted to see how the
15:29results look now, right? So we just
15:31click on this. And now if I go to the
15:34result, this is how it looks like.
15:38So you see now we are almost beating the
15:40market. Now if I had multiplied by two,
15:43we would have definitely done that. And
15:45here we're also making more. And also in
15:48here. Now let's also run another back
15:50test. It's starting 2021 up until ending
15:54date.
15:55Actually, I think I have a little bit
15:58more data because we're not in the
16:00beginning of September anymore. So,
16:02let's run it like this. And by the way,
16:04if you didn't have data like me,
16:06actually in this case, you can just go
16:08to the import page and choose BTC.
16:14So, let's do that and change this into
16:18Binance Pial futures. Click on import
16:22and you can easily do that. And you see
16:24it will skip the candles that already
16:26exist and it will only try to fetch the
16:28new ones. So let's give it a moment to
16:31finish this. I want to quickly remind
16:33you guys about Apex which is my favorite
16:35deck allowing you to trade perpetuals
16:37without providing any KYC while keeping
16:39the custody of your funds. If you sign
16:41up via my link, you will get 25% trading
16:44fee discounts. And you will also be
16:45supporting my work. Now that we have the
16:47candles, if I go back to the back
16:49testing page and rerun this, hopefully
16:52it should go just fine this time. All
16:55right, so this is the result of the
16:58equity curve. If we started trading this
17:00since the beginning of 2021 up until
17:04just a few days ago, this would have
17:06been how much we made. So 1,425%
17:10in profit with a max of minus 27%. So
17:13this is really great. The win rate is
17:1654%. For a strategy with a profit factor
17:19of 1.4, this is really good. And then
17:22the average holding time is 118 hours.
17:25Again, be careful with this. You're
17:26going to need to be patient with it. And
17:29the sharp is 1.61, which is really
17:32amazing. All right, so the results are
17:34really good, but the main reason I
17:37wanted to show you this strategy is the
17:39result of its Monte Carlo simulation.
17:41Now, if you have been following me, you
17:43know that recently I released a tool to
17:46run Monteol simulations with Jesse,
17:48which is really helpful. So, let's run
17:50that. Let's go and copy the name of the
17:53strategy and then go to that script that
17:56I gave you guys. Again, watch the video
17:58if you haven't or check out our
18:00documentation. And then I'm going to
18:01paste the name of it here. So, the name
18:03of the strategy, the time frame is 4
18:06hours. The symbol is BTC and the
18:08exchange is binance per features and we
18:11don't have any data out so I don't need
18:13to enable this part. And for the number
18:15of simulations I have picked 200 and as
18:18for the beginning of it let's set it to
18:22turn 21
18:24up until
18:26so basically the same thing that we did
18:28for our back test and the trading fees
18:33and everything else is just fine. All
18:35right. So, let's go to the terminal and
18:36I will simply say Python test Monte
18:39Carlo, which is the name that I gave
18:41that the script for you. It could be
18:42anything. Let's run it. Now, this is
18:45going to have to run 200 back tests very
18:49fast, right? But I'm also using 12 CPU
18:52cores, so it should take approximately a
18:55few minutes. All right. So, by the way,
18:57I'm getting an interesting error. The
19:00multiprocessing that I'm using is
19:02actually running out of space because I
19:04don't have enough empty space on my SSD
19:08or in my RAM. So, let's stop this cuz
19:11this isn't going as fast as I need it to
19:13be. So, we have multiple options. We can
19:17lower the amount of candles that we're
19:19going to use, right? So, let's bring
19:22this down to 2022 2023 actually because
19:27I wanted to include one bull market and
19:30one bare market. And yeah, so this
19:33should be fine. Let's give it another
19:36try.
19:37By the way, you might also get some
19:39other errors like this. And that is fine
19:41because you see since in this simulation
19:45we are generating simulated data.
19:48Sometimes that data doesn't play nice.
19:51So for example, the prices could be
19:52really low or things like that and that
19:55makes sense, but we don't need all the
19:57back tests to actually go through for us
20:00to have a result that we can read,
20:02right? So let's just wait and see how
20:04this turns out. All right, so the result
20:07is two tables and two chart pictures
20:10that we're going to take a look at. Now
20:12let's begin with this one, which gives
20:14us the Monte Carlo of the trade. So what
20:17if we took the same number of trades but
20:19the order of them were different? So
20:21instead of like winning, winning and
20:22losing, what what if it was losing and
20:25then winning and winning, right? So how
20:26would have our equity curve changed
20:28then? So let's take a look at the chart.
20:31So you see this is how actually in many
20:34of the simulations we would have done
20:35better. But in some of them, but very
20:39few actually things would have been
20:40worse, right? But also notice that the
20:44ending number for all of the equity
20:46curves, the original and the simulations
20:48are the same because these are the same
20:51trades. We're just changing the order
20:52which will change the draw down which
20:54will also change some of other metrics
20:56such as the sharp ratio which is
20:57basically how much you're making
20:58compared to how much risk you're taking.
21:00So if I take a look at this, you see the
21:02max draw down of the original was minus
21:0420% for that period. But the median had
21:07a max draw down of 32%. The best 5% had
21:10a draw down of 17% and the worst had 84.
21:14Now you should never really read the
21:16worst because it's always like this. So
21:19I don't really think this strategy is
21:21going to take this draw down. And you
21:23see also the sharp is changing. Now
21:25let's take a look at this part which is
21:27the monte color for the candle. So
21:29basically not only we run the same back
21:31test on the original data, we also
21:33generate some new data based on the
21:35original candles. Now, this is a very
21:38complex but interesting topic. The
21:40method I'm using here is called moving
21:42block bootstrapping. So, we're basically
21:45using the real data. We're just changing
21:47the order of them. And you could also
21:49assume that this is a way to stress test
21:52the result of your strategy, right? So,
21:54let's take a look at this. So, this
21:57would have been the original back test
21:59and these would have been the
22:00simulations. So the ending number isn't
22:02really great, but if you take a look at
22:04here, many of the simulations were
22:06actually doing better. So it's
22:08definitely good enough. Now let's read
22:10the table and you see the max draw down
22:13was - 20% for the original. We already
22:15showed that earlier. The median had a
22:18max draw down of minus 28%. So this is
22:21what I really like because my risk
22:23tolerance allows me to take as much as
22:25minus 30% draw down. And if the median
22:29of the simulations is only minus 28%
22:33that means it's good enough. So even if
22:36I wasn't as lucky as the original back
22:38test the maxon would have still been
22:40acceptable for me. And in the best 5% it
22:42was actually minus 70%. So it was even
22:45better. And what I want to see is if the
22:47original back testing was worse than the
22:49best 5%. So I don't want it to be inside
22:52the best 5%. All right. Now for the
22:55sharp for the original it was 1.988
22:58but for the median it was 1.13 and you
23:01see the original is actually lower than
23:03the best 5%. Another way to read this is
23:06to forget about the original just read
23:08what you see for the median which is
23:101.13
23:11and the question is is that good enough
23:14for me to trade it? So what if I wasn't
23:16lucky forget the best 5% forget the
23:18original what if the median was
23:20happening? Not only that, even the worst
23:225% has a sharp of 0.04,
23:26which isn't great, but it's not
23:28negative. So, you see, even in the worst
23:305%, I wasn't ending with losing money.
23:33So, that means if I run this strategy
23:35long enough, it will hopefully generate
23:39me profit. Now, we never have 100% thing
23:43in trading. So, don't think that if
23:45you're going to run the strategy, you
23:46will definitely make money. There's no
23:48printing machine. There's no guarantee
23:50in trading, but I like the odds that I'm
23:52seeing here. And you can also read the
23:53other metrics if you like to. But
23:56overall, I'm happy with these results.
23:58Now, something interesting that you
23:59should know is that I actually released
24:01this strategy more than a year ago on my
24:04channel and on our website. But this is
24:07the premium version of it. It was called
24:10Trend Swing Trader V2. So if you want to
24:13see the result of it for other periods
24:16or time frames or symbols, you can do
24:18so. But one interesting fact is that
24:20back then when I release this, let's say
24:22you were wondering, does the result of
24:24it also work for fusion data? Well, now
24:27we have the result. If you were a
24:30premium user, you would have had access
24:32to the exact same thing that I just ran.
24:34And you would have seen that let's say
24:36in 2025
24:38the strategy well here it wasn't making
24:41much but actually I don't have the
24:43result of more of them. All right we do
24:45have it here. Let me open it. All right.
24:47So since the beginning of 2025 up until
24:50September still it was making money and
24:54it was also doing the same thing in
24:552024. So you see if you wanted to do
24:58some forward testing here it is. If you
25:01enjoyed the video, please consider
25:02giving a like and subscribe to the
25:04channel for more videos and strategies
25:06like this one. Thank you so much for
25:08watching. I'll see you in the next one.
25:10[Music]
25:13[Applause]
25:14[Music]

About this transcript

This page contains the full transcript of GPT: Trading Strategy in Python makes 805% (+ Monte Carlo simulation results) by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 4,775 words across 637 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.