YT YouTube2Text

YouTube2Text → Transcripts → GPT: Mean Reversion strategy in Python makes 813%

GPT: Mean Reversion strategy in Python makes 813% — Transcript

by Algo-trading with Saleh · 5,239 words · 721 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hey guys, it's Salah. In this video, I
0:01will show you a mean reversion strategy
0:03in Python which not only has a high win
0:06rate, but it also works on multiple
0:07assets. So, at least until this point,
0:10it is the best mean version strategy
0:12that I have ever written. As always,
0:13first I will show you the entry and exit
0:15rules of the strategy on trading view
0:17and then we will use GPT to convert
0:20those rules into Python code so we can
0:22run some back test on it. So, you can
0:24see that even if you are not a great
0:25programmer, you can still do research
0:28with Python for algo trading. Before we
0:30continue, I got to say I am not a
0:32financial adviser and everything that
0:33you see in this video is simply a
0:35trading tutorial. Trading always has
0:37risk. So no matter what the win rate
0:39that you see for the result of this
0:41strategy, you should always be careful
0:43and do your own research before trading
0:45this strategy or anything like it. So
0:47with that out of the way, let's get
0:48started.
0:50Now, this is a mean reversion strategy,
0:52which means unlike trend following
0:54strategies, we're not going to begin
0:56with a leading trend indicator because
0:58we don't want the price to continue to
1:00its direction. Instead, what we want is
1:02for it to go back to its mean, which
1:04means we're going to take a position the
1:05opposite side of the current direction
1:07of the market. Now, a great indicator
1:09for that is Ballinger band. So, let's
1:11look it up.
1:14By the way, the time frame of the
1:16strategy is 1 hour. All right. So simply
1:18put, we want to buy at the lower band of
1:20the Ballinger bands, which would be this
1:22green line here. And if it's a short
1:24position, we want to do opposite, which
1:25means we're going to open your short
1:27position at this price. Now, in case
1:29you're not already familiar with the
1:30Ballinger bands, this blue line here is
1:33a moving average. And then these two
1:35other lines are simply two standard
1:37deviations away from that mean. because
1:39statistically speaking whenever the
1:41price goes two standard deviation away
1:43from the mean it is more likely to go
1:46back to its mean. Now I don't remember
1:48the exact numbers for this but it's a
1:50very interesting theory. However, what I
1:52just described isn't actually an interle
1:54of the strategy. It's basically where we
1:57want to open the position. Now the next
1:59indicator which is the main indicator of
2:01the strategy is simply the RSI. So let's
2:04look it up on trading view.
2:08Now, as you already know, the RSI is
2:11simply an oscillator. Usually, we want
2:14to short when it goes above the
2:16threshold, which would be here, and we
2:19want to go long when it's vice versa.
2:22But the way I want to use it in this
2:23strategy again, because it's mini
2:25reversion, is actually quite different
2:27because I actually want to buy at a
2:28pullback. So, I want to be a direction.
2:32I want it to be a trend. But instead of
2:34buying instantly, we want to bet that
2:36the price is going to have a pullback.
2:39So I'm actually looking forward for
2:41scenarios like this. So let's say there
2:43is a strong trend, but I want to buy at
2:45pullback like in here for example. So
2:47that when it goes up again, I'm going to
2:50make a profit. But here's the thing. If
2:52I only buy when the price is for example
2:55like here above the threshold, the
2:57window of opportunity is going to be
2:58very small. So instead, what I'll be
3:01doing is I will still use the RSI but on
3:04a bigger time frame. So let's do that on
3:07trading view
3:10and change the time frame to
3:134 hours.
3:16Click okay.
3:18All right. So now from here all the way
3:22until here would be a valid opportunity
3:24for us to open the position. All right.
3:26Next, we're going to have to think about
3:28situations where maybe the RSI is like
3:31this, but there isn't an actual trend or
3:33volatility in the market like in this
3:35case. So, even if there isn't, for
3:37example, in here, the RSI could still be
3:40overbought. For example, in here, right?
3:42And yes, this is a good candle, but it
3:44doesn't mean that it's a trend, right?
3:46Now, yes, in this case, if we did open a
3:49long position, it would have made us
3:50profit, but there will be many false
3:52breakouts, and we need to be careful
3:54with those. So, the next indicator that
3:56I'll be using is simply the ADX. So,
3:58let's add it to our chart. And because I
4:00don't want to lose so many trades
4:02because this is also some kind of
4:04squeeze strategy, you could call it. So,
4:06we don't want to always wait until the
4:08trend is really, really strong, right?
4:10We want to be early with it. For that
4:12reason, unlike my previous strategies,
4:14instead of setting the ADX threshold to
4:16a number such as 40, in this one, I'm
4:18going to set it to 20. So, let's do that
4:21on Trading View. So, wherever this line
4:25is above this threshold, we're okay to
4:27open the position. But this could also
4:29have some false breakout. So, what I'm
4:32going to do is to do the same thing, but
4:34also on the bigger time frame. So, this
4:39will be again 4 hours. We'll do this.
4:42But for the 4 hours, I'm going to set
4:44the ADX threshold to 40, which would be
4:47this luck. Okay? So, by doing this, I
4:49won't be able to buy here. I can only
4:52buy from here all the way until here.
4:55All right. So, we have all the entry
4:57rules except one more thing. We're going
4:59to need one last filter for the trend.
5:01Now, again, I know this is not supposed
5:03to be a trend following strategy, but
5:05still we need the trend indicator except
5:07that we didn't begin with that. I'll
5:09show you what I mean. So, usually in a
5:11trend following strategy, you want to
5:13use a trend indicator with the current
5:15time frame. And that could be anything.
5:17It could be the moving average, the
5:18super trend, the comma indicator,
5:20whatever. In this one, I'm going to use
5:22the super trend. So, let's add it. And
5:25also remove the RSI and the ADX so we
5:30can see this better. Now, let's also
5:33disable Ballinger bands so that this
5:35will be more clear to see. All right.
5:36So, you see, for example, here we have
5:39an uptrend because the super trend line
5:42is below the current price and it's also
5:44displaying the color green. And here we
5:47have a downtrend because the super trend
5:49value is above the current price. Also,
5:52it color is red. Now, this is important
5:54because that's how we're going to write
5:55the Python code for it. Now, here's the
5:57thing. If we used super trend as it is
5:59right now, for example, in here, this
6:02would have been considered a downtrend.
6:04But as I explained, I actually want the
6:06price to have a pullback, right? So, no
6:08matter where it's going before, I want
6:11it to go actually against that price for
6:14me to be able to open the position. So
6:16if it changes the direction of the trend
6:17very quickly, it won't be good enough
6:19for me. So what I'm going to do is to
6:21simply change a time frame one more time
6:24from whatever my charts is, which in
6:27this case would have been the hourly
6:29into 4 hours.
6:32Click okay. And now you see this period
6:35here is no longer a downtrend, but it is
6:38an uptrend because it is displaying the
6:40green color. And just like that also
6:42this one here isn't considered an
6:44uptrend because on the 4hour time frame
6:47it is still a downtrend. Now this is
6:49something that we would have never done
6:50in a trend following strategy but in a
6:52mini version that's perfectly fine. All
6:54right. So we talked about the entry
6:55rules of the strategy but now let's talk
6:57about the exit rules for both the stop
6:59loss and take profit. So for that I'm
7:01going to disable super trend and enable
7:04the Ballinger bands again. Zoom it in.
7:08All right. So assuming we open the
7:10position at the lower band. So I could
7:12be here for example. I want the takerit
7:15to be at the upper band which would be
7:17here at this red line. And of course do
7:19the opposite for a short position. Now
7:21what about the stop loss? Well for that
7:23just like my previous strategies I'm
7:25going to use the ATR indicator. So
7:27basically it will tell us on average how
7:30much every candle has changed in the
7:32price. So for instance, let's say from
7:33here to here is the actual average price
7:36that every candle is changing because in
7:38one of them is changing less but in
7:40another like here for example is
7:42changing a lot. Right? So let's say this
7:44is the average and then I want six times
7:46of that value to be below my entry. So
7:49if my entry for example is here I want
7:51it to be six times of this below this
7:54which would end up something like here
7:56for example. Now if you want the exact
7:58value of it, of course you could look up
7:59the ATR indicator on trading view and
8:02here you would get that value. So in
8:04here it would have been 725.
8:08So you need to calculate six times of
8:12that and subtract it from here. Now the
8:14ATR indicator isn't really helpful for
8:16manual traders at least based on what
8:18I've seen. But in algo trading it is
8:20actually very handy. Now, I should also
8:22emphasize this that usually in a trend
8:24following strategy, we want to have a
8:26high risk-to-reward ratio, which is why
8:29we want to make sure that when we win,
8:30we win bigger than the amount when we
8:32lose, which is why even when we do use
8:35the ATR, we usually use values such as
8:37two or three. But because this one is a
8:38mini version, we usually do the
8:40opposite, which means the win rate of
8:42the strategy is going to be higher.
8:44Something like 70 80% is like the
8:47minimum for such a strategy. But when we
8:50win, we win less. And when we lose,
8:52we're going to lose in bigger sizes,
8:54which is why our stop loss is being this
8:56much below the current entry. If this
8:58sounds a bit too complicated, just
8:59forget about it once I show you some
9:02back test results. You'll see what I'm
9:03talking about. All right. So, now that
9:04we have both the entry and exit rules of
9:07the strategy, we're ready to convert it
9:09into Python code. So, we can run some
9:10back test on it. Now, for this part, I
9:12want to simply feed the entry and exit
9:15rules to an LLM model and have it write
9:17the Python code for us. And as always,
9:19we have two options. One is to use Jesse
9:21GPT, which is the cloud version of it
9:24that I made myself with some
9:25instructions specifically for aggregate
9:27trading in Python. Or if you don't want
9:29to use my service because you have some
9:30privacy concerns or if you don't want to
9:33pay me anything, that's perfectly fine.
9:35You can just come to my channel and
9:37watch this previous video that I made
9:39which will show you how to selfhost this
9:42by yourself. But again, I'm going to use
9:43JGPT. So, let's get started. All right.
9:45So first we need to ensure we have
9:48picked the right model. Now this is
9:50right now set on quen 3 coder which I
9:52reviewed a while ago but for this I'm
9:55going to go with claw 3.7 sonnet which
9:58I'm always happy with. So I kind of see
10:00no reason to change this. So let's paste
10:03in the in rules. Now I simply ask the
10:05model please write a mini version of
10:07strategy with below rules. Use the RSI
10:09on a bigger time frame specifically the
10:114 hours time frame. For long positions
10:13the RSI must be above 70. For short
10:15positions, do the opposite. As for the
10:17trend, use the super trend indicator.
10:19Once it is displaying an uptrend, you
10:21want to go long. This should also use
10:23the bigger time frame. For the ADX, it
10:25needs to be above 40 on the bigger time
10:27frame. It needs to be above 20 on the
10:29current trading time frame. For short
10:31positions, do the opposite. For entries
10:33in long positions, the entry price must
10:35be the lower band of the current
10:36Ballinger bands indicator. Set the stop
10:39loss as the current entry price minus
10:41six times of the current ATR. Risk 2% of
10:44the accounts per each trade. For take
10:46profit in long positions, close at the
10:49upper band of the current Ballinger
10:50bands. For short positions, take profit
10:53at the lower band of the current
10:54Ballinger bands. So basically whatever
10:56just showed you on the chart, we are
10:58simply asking the model to write it for
11:01us. Let's hit enter and wait for the
11:03response. All right, so it wrote it for
11:05us. It is called mean version. So let's
11:07copy this name. All right. Let's go to
11:09Jess's dashboard. Create a new strategy
11:11and paste this name here. But I will
11:13also add the name RSI at the end of it.
11:15So it will be mean version RSI. Let's
11:18click on create. And now let's go back,
11:21copy the whole code and paste it here.
11:24Now I could review the code on the
11:26editor inside J's dashboard or we could
11:29open the VS code which is a
11:31significantly better IDE. So that's the
11:34one that I'm going to do. So mean
11:35reversion RSI. So let's look it up. All
11:38right. So first it defined the higher
11:40time frame candles with the syntax get
11:43candles. Then it's passing the current
11:45exchange symbol and specifically the
11:47time frame of 4 hours. So this is
11:49perfect. The higher TF RSI is returning
11:53TA. RSI for the RSI indicator. But
11:56instead of passing the current candles,
11:58it is passing this higher TF candles. So
12:00I really like what it did here. So first
12:02it defined a property and later we can
12:06pass it very easily whenever we need to.
12:09Next it defined the higher time frame
12:10super trend like this TA. Trend again
12:13passing the higher time frame candles.
12:15The higher time frame ADX is again
12:18TA.ADX with the higher TF candles. And
12:21the current TFADX is simply passing TA.
12:26ADX but instead of passing the higher
12:28time frame candles it is passing the
12:30current candles which in Jesse you can
12:33always get it by simply using
12:35selfcandle. So you don't need to define
12:37this anywhere. Now for the Ballinger
12:39bands again it is using this syntax and
12:42passing the current candles because we
12:44don't want the Ballinger bands on the
12:45bigger time frame. We want it on the
12:48trading time frame. All right. Next, it
12:50defined the ATR by simply returning
12:52TA.ATR and then passing the current
12:55candles. And then it defined the integer
12:57rules of the strategy. So, it's also
13:00displaying it here in a beautiful
13:01comment. So, the RSI needs to be above
13:0470 on 4 hours. The super trend is
13:06uptrend on 4 hours. The ADX is above 40.
13:09Now, let's take a look here. So, the
13:11higher TF RSI must be above 70. The
13:15higher TF super trend must be one.
13:18This is incorrect.
13:20And then the higher time frame ADX must
13:22be above 40. And the current time frame
13:25ADX must be above 20. So this is really
13:27good. But this part is not correct. So
13:31let's first see how many times it is
13:33defining this.
13:35Okay. So just once. So instead of this,
13:39I'm going to first of all rename this
13:41into simply trend.
13:44and I will get the super trend value
13:47like this. Then I will say this if the
13:51current
13:53price is above the value of S. trend and
13:59by the way this S dot trend is this
14:02value here. So if the current price is
14:04above it that means we are in uptrend
14:07and if it is below it like in here then
14:12we are in a downtrend right? So I'm
14:14going to say if the current price is
14:16above S. trend return one which means we
14:20are in uptrend otherwise we're going to
14:23return zero. So let's go back
14:28and now this is correct except that we
14:31should remove this
14:33and for short positions we should also
14:36do this. All right. So everything else
14:38looks good. Now let's go to the position
14:40sizing which is right here. Now the
14:43entry price is going to be the lower
14:45band of the current Ballinger bands. So
14:47I know this is correct. The stop loss is
14:50going to be the current entry price
14:51which would be this minus 6 times of the
14:54current ATR and we are risking 3% of per
14:58capita. How? Well by using this utility
15:00function risk to quantity. First it gets
15:03the current available margin which is
15:06how much money your account has for
15:07trading. Then it is passing the number
15:09for risk. So you see the second
15:12parameter is risk per capital. It is
15:14passing three. Next we have the entry
15:16price the surplus price and then we
15:19going to have to set the fee rate and we
15:22are passing the current fee of the
15:24current exchange that we're trading. Now
15:25this one you don't really know. Just
15:28pass it like this and you're good to go.
15:30And then to submit the actual buy order
15:32we're using self buy equals quantity and
15:35then enterprise. So notice that we are
15:37not subitting the surplus but we did the
15:39calculated so that we can use it here to
15:43estimate how much we should be buying in
15:45order to risk only 3% of capital per
15:48that position. Now for a short position
15:50we're doing the opposite. Now the should
15:52cancel entry is true. Now about this one
15:55we need to talk. So let's go back to the
15:57chart
16:01and bring back the Ballinger bands. So
16:03let's remove this. All right. So suppose
16:06that at this point for example we decide
16:10to submit the buy order and of course
16:13because we are setting a price lower
16:15than the current price it's going to be
16:16a limit buy order right so it won't be a
16:19market order and we said that the buy
16:22price is going to be at the lower band
16:24of the Ballinger bands. So in this point
16:26that would have been here right but as
16:29you can see on the next candle the price
16:31didn't reach here. Not only it didn't
16:33reach here but the value of the lower
16:35band is now here. So we need to update
16:38the order or submit another one and
16:41cancel the previous one. Right? And if
16:43we keep doing this at this point now
16:46this one actually isn't touching this
16:48but in this point it is right. So if we
16:52had kept our earlier entry price which
16:55was actually here then even at this
16:58point we would not have opened the
17:00position. So we want to wave to tell
17:03Jesse that hey keep cancelelling that
17:05order and submitting a new one if of
17:08course the in rules are still valid and
17:11to do that in Jesse we need to use the
17:13should cancel entry method and inside it
17:16simply return true if we were returning
17:18false it wouldn't have done this which
17:20means if your strategy is using a market
17:23order through open positions you're not
17:25going to even need to define this at all
17:27anyways after we have an open position
17:30so Assuming our order have been
17:32executed,
17:34we will use this method on open position
17:37which is where we submit both the stop
17:39loss and take profit. So we're saying if
17:41is a long position, this stop loss is
17:43going to have the quantity of the
17:45current position which we pass by
17:48self.position quantity and then the
17:50price of it is going to be the entry
17:52price of the current position minus 6
17:54times of the current ATR. For the
17:56takerit, the quantity will again be the
17:59quantity of the current position and the
18:01price is going to be the upper band of
18:03the current Ballinger bands. And of
18:05course, if it's a short position, we
18:06want to do the opposite. So, very
18:08simple. And then we also have the update
18:11position. Now, this one is keep updating
18:14the takerit price. So, assuming that the
18:17Ballinger bands is changing, which it
18:19is, it is also updating our takeprofit
18:22order. But actually, I don't want it to
18:25do this. So, it's doing something that I
18:27didn't ask for. So, I'm going to remove
18:29this. Now, you see the more specific you
18:32are with your request to the model, the
18:34more likely it is to follow it. Exactly.
18:36Right. And the more vague requests you
18:39give it, it might do some new stuff like
18:41this. So, I should have been more
18:43specific in my request, but it's
18:44perfectly fine. I'm okay with changing
18:47some things in the code, especially when
18:49I'm just removing stuff and all the rest
18:51is perfectly fine. All right. So now
18:52that we have everything working, let's
18:54go to Jesse so we can run our first back
18:57test. So let's remove these.
19:00Choose mean reversion RSI. Change the
19:04time frame to hourly and the
19:08symbol to BTC.
19:11And for the duration of the back test,
19:13I'm going to do since the beginning of
19:152024 up until 2025.
19:19And I will also choose
19:22this state. All right. So let's
19:25go back and change the fast mode to
19:29enabled. The benchmark is also on. So
19:33let's give this a try.
19:36By the way, the trading fees are set to
19:40almost the double number that it is
19:42actually on Binance. I do this on
19:44purpose because I want to leave some
19:46room for the slippage when the actual
19:48trading is happening or for the funding
19:50fee or any extra fee that isn't visible
19:53inside my back test. I also change the
19:55leverage number to 10, but it doesn't
19:57actually mean that I'm using 10 times
19:58leverage. Right? So, we'll get to that
20:00now. Let's take a look. All right. So,
20:03we took nine trades. We made 9.23%
20:07in profit. Now, the equity curve looks
20:10good, but we're not beating the market
20:13yet, right? So, but let's go down. And
20:15you see the max rodon is near zero.
20:18That's ridiculously low, which is
20:20awesome. And the win rate is 100%. Now,
20:23of course, this isn't realistic. You
20:25should never expect a win rate of 100%.
20:29And the sharp ratio is 2.48, which is
20:32really good. And all right, so
20:34everything else looks really good. Now
20:36the first thing I want to do is to go
20:38back to the code and increase the size
20:41of my position. So I'm going to add the
20:45quantity by five and again for short
20:47positions. Now one very important thing
20:49when you do this you need to remember
20:51that you are multiplying the entire size
20:54of the order by a specific number. So
20:57you're not risking 3% per each capital
21:00anymore. So this is kind of invalid now.
21:03But I'm doing it because I just want to
21:05show you the actual equity curve that
21:07we're getting now because I want to show
21:08you how you can beat the market returns
21:11at the end. But in an actual production
21:13ready, you want to use another formula.
21:15You want to increase maybe this number
21:17or even use some other formula for
21:19position sizing such as the Kelly
21:21criteria. But that's a topic for another
21:23video. So let's go back and rerun this
21:26again.
21:28Now, by the way, we're getting this
21:30error because even though the strategy
21:32is using the bigger time frame, which is
21:334 hours, I didn't define it here. You
21:37see, I should have added a data route
21:39and chosen the same
21:42route. Sorry, symbol and then it should
21:45have worked. But I didn't do that. And
21:47in back test, that's fine. Jess, it will
21:49give you an error, but it will take care
21:51of it behind the scenes and the back
21:52test will go through. But if you don't
21:55do this in a live trading session,
21:57you're going to face an issue and your
21:58session will get terminated. So be
22:00careful with that. All right. So let's
22:02remove this. And now the equity curve
22:04looks significantly better. Right? We
22:06are closer to beating the market. But
22:08we're still not fully beating it. Of
22:10course, I could simply add to the size
22:11of my position more, especially because
22:13it has still a very low max draw down,
22:16but this isn't a realistic number in all
22:19the periods, right? So I'm not going to
22:21do that. All right. So first I want to
22:23show you the result of the back test on
22:25other periods. So let's change this to
22:28beginning of 2022 up until 2023
22:32and the first day of that year. Let's
22:36run this one more time. I'm going to do
22:392023 up until 2024 and one more time on
22:432024 till 2025
22:46and 2025 until
22:51almost the 8th month which is just a few
22:53days ago. So let's run the whole thing.
22:56All right. So in 2022 you see we are
22:59ending with losing a little bit by 9%.
23:03The max draw is minus 22% and the win
23:07rate is 75%. So even though it's not
23:10horrible, we're ending with a loss. So
23:13that's the first lesson that even though
23:15it has a win rate of 100% in 2025 and
23:192024, it doesn't mean it's always
23:22profitable. In 2023, which you can see
23:24the date here, by the way, we're ending
23:27with 97% profit. The max thon is minus
23:3012% and the win rate is still 100%. So
23:34it looks really great for these three
23:37years but not so great here but it's not
23:39bad either. So let's do one interesting
23:41thing. I want to run the back test
23:43starting this date until this ending
23:46date. Right? So we can see the whole
23:48equity curve.
23:51So let's run it.
24:00All right, there we go. So now if we
24:02were trading this strategy in the past
24:043.5 years, we would have beaten the
24:08market, right? Not only with the ending
24:10P&L, but the max draw down is minus 22%.
24:14Which is significantly lower than the
24:16max draw down that BTC had in this
24:19period. We ended with 176% in P&L. The
24:22win note is 96% which is absolutely
24:25amazing. The sharp ratio is 1.47
24:29which is a realistic number and very
24:32good. And the average holding time would
24:34have been 33 hours. But actually I think
24:36we made a mistake because if we check
24:39out here it's only taking long
24:41positions. So no short positions. So
24:43let's go back to the code and see why
24:45that is the case. All right. So here for
24:47the trend value when it is an uptrend
24:51I'm returning one but when it's a
24:53downtrend I'm returning minus one. So
24:56that was my mistake. So let's go back
24:59and rerun this one more time. And I'm
25:01also going to rerun this one for 2022.
25:07There we go. Now we are beating the
25:09market and making a profit. So we're
25:13ending with 16% in profit. And in this
25:16one we are beating the market even
25:18further. So we ended with 253%
25:22and the max ardon is still minus 22%.
25:25Which means the sharp is now better. It
25:27is 1.57.
25:29But the win note is slightly lower. It
25:31is 94.59.
25:34All right. So this is actually great.
25:36But there's one problem with this
25:38strategy. You see, I don't like this
25:40total closed trade number because over
25:433.5 years, it only took 37 trades. And I
25:48think that's very low because the bigger
25:50the number of trades that the strategy
25:51is taking during a period, the higher
25:54the statistical significance of that
25:57strategy. In simple words, it means you
25:59can rely more on the result of the back
26:02test that is in front of you. And 37 for
26:053.5 years isn't good enough for me. And
26:08that brings me to the next good thing
26:10about the strategy that I found, which
26:12is it doesn't work just with BTC. It
26:15also works well with other coins. So,
26:18let's go back and I'm going to add some
26:21more trading routes. So, I'm going to
26:22add ETH
26:24again on the hourly
26:28and I'm going to add
26:31Doge again on the hourly. And by the
26:34way, let's also change this to be mean
26:37the version RSI. Now, this is enough for
26:39me for now, but you can find others by
26:42yourself. And now, if you run it again,
26:45we're going to get an error because the
26:47fast mode doesn't work when you are
26:49trading multiple trading routes, right?
26:51So, we need to disable this. And I'm
26:53also going to disable the benchmark so
26:55we can see the equity chart better. So,
26:57let's run this one more time.
27:00Voila, we have something with bigger
27:02number of trades. So it is 75 now. And
27:06look at this equity curve. It's
27:08beautiful.
27:10Now the P&L is 813%.
27:14The max RON is still minus 22%. So it
27:18didn't even get worse.
27:21And the win rate is 92%.
27:24And the average holding time is a little
27:27bit higher. Now by the way, you need to
27:28be careful with this cuz it is near to
27:3170 hours, right? So that's a few days.
27:34So if you trade this strategy at any
27:36point, you're going to have to hold the
27:38position for days maybe in a loss and
27:41psychologically that's going to be
27:42difficult. So that's something to be
27:44aware of. And then the sharp is now 1.95
27:48which is amazing. The winning streak is
27:5132 and the losing strike is two. Right.
27:54Overall, I think this looks great.
27:56Probably the best strategy I've ever
27:58made so far. And let's not forget that
28:01I'm still trading only three assets. We
28:04could continue this for some other
28:05assets and find some good ones and
28:07improve the results even further. Now,
28:09as always, I am going to submit both the
28:12code and the result of the back test of
28:14the strategy on other time frames or
28:16symbols on our strategies page which you
28:19can check it out. And if you already
28:21have a free account, you can also
28:22download the code of the strategy
28:24assuming it's a free strategy like this
28:26one that I just shared. Now guys, a lot
28:28of effort goes into research before
28:30making a video like this. So if you want
28:32me to continue making them, please
28:34support me in any way that you can.
28:36Liking the video, posting a comment with
28:38your opinion, and also subscribing to
28:40the channel helps me a lot. So thank you
28:42so much for your continued support. I'll
28:44see you in the next one.
28:48[Music]
28:57[Music]

About this transcript

This page contains the full transcript of GPT: Mean Reversion strategy in Python makes 813% by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 5,239 words across 721 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.