YT YouTube2Text

YouTube2Text → Transcripts → Grok 4 tested: Is it good for algo-trading strategies?

Grok 4 tested: Is it good for algo-trading strategies? — Transcript

by Algo-trading with Saleh · 5,341 words · 751 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00Hey guys, it's so XAI or the team behind
0:02the Gro model or as they are usually
0:04called Elon Musk just released Grock 4
0:07model. As always with everything Elon
0:09Musk, there's a lot of hype behind this.
0:12So in this video, what I want to
0:14particularly test is how good this model
0:16is for algo trade. So that means tasks
0:19such as writing strategies or debugging
0:21them or coming up with new trading
0:24ideas. So if that all sounds interesting
0:25to you, let's get right into it. Now
0:27this is the official benchmark released
0:29by their team. And if you take a look at
0:31the LCB which is stands for live coding
0:33benchmark which is the one we care for
0:36this task shows that Gro 4 without any
0:39tools of course because if it is using
0:41tools we don't know which type of tools
0:43it is using and we're definitely not
0:45using it here. So this is the one we
0:48care about and also this is the one that
0:50all the other models have such as 03 and
0:52Gemini 2.5. So it is showing significant
0:56improvements compared to at least Gemini
0:582.5 Pro. Now notice that they are not
1:00comparing it to Cloud Sonnet 4 and I
1:04think that's intentional because at
1:06least to my opinion cloth set is still
1:09the king of LLM models at least for
1:11Cody. But anyway, so it sounds like it's
1:13really good but I found it to be a bit
1:15more expensive than even Cloud Sonnet.
1:18So what I really want to see is if it
1:20actually blows me out of the water. So
1:22anyways, I updated our Jesse GPT to have
1:25the Grock for model and as you can see
1:28it is indeed more expensive than all the
1:30other models. Now if you don't want to
1:32use Jess GPT for either privacy concerns
1:35or if you don't want to pay us anything
1:37that's perfectly fine. You can watch
1:39this video of mine where I show you how
1:42you can self-host this thing by yourself
1:44because remember we want to test it for
1:46Algo trading and the instructions that
1:48I'm feeding the model are specifically
1:50for this use case. So either use JGPT as
1:54I am in this video or go with the
1:56self-hosted solution. In the previous
1:58video where I tested the cloud 4 model,
2:00we use a pass or fail ranking model. But
2:03in your comments, you guys suggested for
2:05me to use a number based model. So in
2:08this video, we're going to rank every
2:10task that the model does from 0 to 10.
2:13And in the end, we're going to end up
2:14with a total score. All right. So let's
2:16begin with the hello world question of
2:17algo trading. Write a golden cross
2:19strategy that uses the 50 and 100 moving
2:22averages. Close the position when the
2:24inter rule is no longer valid. It took a
2:27long time for the model to finish this
2:29question. I'm guessing this is because
2:31since they just launched the model,
2:33maybe their API is under heavy traffic.
2:36But anyways, so this seems perfectly
2:40fine. It defined a moving average with a
2:42period of 50 and 100 passes current
2:46candles is perfectly correct. It defined
2:49a trend property. It returns one for
2:52uptrend and minus one for a downtrend.
2:55Then in the shoot long it's asking
2:57whether or not the current trend equals
2:58one or an uptrend. In other words, it is
3:01not shorting which is fine because I
3:02didn't ask for it. It also defined this
3:05one which it didn't really have to. And
3:08when the current turn equals minus1, it
3:10is liquidating the current position,
3:12which is exactly what I asked it to do.
3:14And it is also very briefly explaining
3:16what it did. So I like this answer.
3:19Let's copy it. Go to Jesse and create a
3:22new strategy. Now I'm going to call it
3:24test rock 4.
3:28Let's paste it here. Go to back test.
3:32Let's close these.
3:36And I'm going to define
3:39the duration of it since the beginning
3:41of this year.
3:43Pick BTC UCT and choose the hourly time
3:49frame because I don't really care about
3:50the metrics of this strategy right now.
3:52So I'm not going to test all the time
3:54frames. I just want to see if it runs
3:55well. And let's pick test Gro 4. Now the
4:00fast mode should be on. Let's run it.
4:06and it runs just fine.
4:09They executed 25 trades.
4:12It didn't make a profit.
4:15The max thon is minus 36. The win rate
4:18is 28%. All right. So, this is fine. And
4:21I'm going to give it a perfect 10. Next,
4:25convert this pine script strategy into
4:27Python code. Now, I need a pine script
4:29code to feed the model. So I'm going to
4:32look up a strategy called super trend on
4:35trading view and it is this one super
4:38trend strategy
4:43and here are the supposedly results of
4:46its back test. Now to get the code I'm
4:49going to click here and this is the pine
4:53script that I'm looking for.
4:55So let's go back
4:57and do this.
5:02Now, unfortunately, no matter how many
5:04times I tried this question, it just
5:06doesn't work and it doesn't give me any
5:08answer. I'm not sure why that's the case
5:10because I tried it with Gro 3, the
5:13previous version of them and it works
5:14fine or with any other model on Jess
5:17GPT, but for this question, it's not.
5:20So, unfortunately, I have no choice but
5:23to give them zero for this question.
5:26Next, develop a mean diversion strategy
5:28using the Ballinger bands and the RSI
5:30with a seven period.
5:34All right. So, here it is. It didn't do
5:36syntax highlighting. That's because of
5:38this extra py that you see here. So,
5:40that's fine. Let's copy it. Go back and
5:44paste the code here. And also make sure
5:47to remove this extra py. Now, let's take
5:49a look. It defined the Ballinger bands.
5:52It named it BB, which is fine. define
5:53the RSI with the correct syntax and also
5:56the period of seven. In the shelong,
5:58which is the interior rules of the
5:59strategy, it's saying if the closing
6:01price is below the lower band and the
6:05current RSI value is also below 30.
6:08That's perfectly correct. It did the
6:11opposite for short positions. It's
6:13returning true for shoot cancel entry
6:16which is only helpful if we are using a
6:18limit order or a suborder to enter the
6:21positions which we are not because here
6:23you see we are using the current price
6:26which causes Jesse to submit a market
6:28order. Now here it's using size to
6:30quantity and passing the entire
6:32available margin. That's perfectly fine.
6:35And in the update position which only
6:37gets called after a position is already
6:40open. It's saying if it's a long
6:41position but the current closing price
6:44is above the middle band, I want to
6:46liquidate the position and for a short
6:48position do the other way around. That's
6:51perfectly correct. So let's go back and
6:54run a back test.
6:57And it runs perfectly fine. Actually,
7:00did I save it?
7:04No, sorry. I didn't save the code. So by
7:07pressing command S or this button we can
7:10save it. So let's go back and run it one
7:12more time
7:14and we're getting an error. Okay, now
7:16this is fine. It's because of that extra
7:18py. I forgot to remove it this time. All
7:22right. So one last time
7:26and it goes through. All right. So this
7:28is fine. It executed
7:31155 trades
7:33and yet it didn't lose that much.
7:35Actually, I think this strategy has some
7:38good potential in it. The win rate is
7:4063%. I like this. I might actually work
7:43on this later. So, we're going to give
7:45it a perfect 10. All right. Next, write
7:48a strategy that uses Fibonacci
7:50retracement levels for entry and exit
7:52points. So, this is a tough one. All
7:55right. After a lot of time, it finally
7:58printed out the result. So, let's copy
8:00it. Go to Jesse, edit the code, and
8:04paste it here and save it before I
8:06forget. Now, it named it fib
8:09retracement. It defined swing period as
8:1250. This is the look back period for
8:15swing high or low, which is indeed
8:17needed. The swing high is saying MP.max.
8:22to current candles. It's only selecting
8:24the swing period which again is the look
8:26back period. Now I think a better name
8:28for this would have been just look back
8:30period but anyways so it is selecting
8:35the index of three. Now in Jesse the
8:37indexes of candles goes zero for time
8:40stamp then one for open two for close
8:43three for high four for low and five for
8:46volume. So three for high is correct and
8:50four for low is also correct. Now the
8:52fib levels this is the important one.
8:54It's saying that diff equals swing high
8:56minus swing low and it define the three
8:58levels as swing low plus the difference
9:03multiplied by this Fibonacci number and
9:07then this one and this one. Okay, this
9:09is not only true but I kind of like how
9:11it wrote it. It's very simple and
9:13practical. Now then we defined an EMA
9:17fast with a period of 50. Another one
9:19with a period of 200 for slow. The
9:22uptrend is when the EMA fast is above
9:24EMA slow. The downtrend is the opposite
9:27of that. And for shoot it's saying f me
9:30the Fibonacci level of 61.8
9:34the previous close and it's saying if
9:37it's an uptrend and the previous close
9:39is below this Fibonacci while the
9:42current close is above it. Right? It's
9:44trying to detect whether or not we have
9:46just crossed it. Then we want to go
9:49long, right? So let's draw the Fibonacci
9:52line over a chart. Now here's the 0.5
9:57line. Here's 0382
10:02and this is 0.618.
10:04Now it's saying that whenever the price
10:06crosses this line like in this case, we
10:09want to go long. So it's some kind of a
10:13breakout strategy. It's not a mean
10:15reversion. All right? So we're not
10:17trying to open a long position when the
10:19price is here, for example. Instead, we
10:21wanted to cross this line. So it is
10:23technically correct, I guess, but it's
10:25not how I'm used to Fibonacci. It's
10:28perfectly fine, though. All right. So
10:29let's see how it's opening the position.
10:31It's using market order.
10:33And here you can see the entry is the
10:35current price. The stop loss is set at
10:38the Fential line of 38 which is this
10:42line. For quantity, it is risking 3% of
10:45the entire account.
10:47And for a short position, it's doing the
10:49opposite. Now, when we open the
10:51position, we are submitting the take
10:53profit and we're using the TP level of
10:56this. So if it's a long position, it's
10:59saying I want the swing high plus the
11:03difference between swing high and the
11:05swing low multiplied by this. So again,
11:09this is really interesting. It's not
11:11using the typical Fibonacci levels for
11:14exiting the trade. It's basically
11:17expecting the price to continue to go up
11:19again because it's supposed to be a
11:21breakout strategy according to the
11:23model. But even for that, it is using
11:26the Fibonacci lines because it believes
11:28these are magic numbers. So this is
11:31certainly interesting. But again, this
11:33isn't how I'm used to Fibonacci. So I
11:36don't know. Let's run and see if it
11:38gives us maybe good results. But I know
11:41one thing. It is not submitting the
11:43stops here. If it had submitted the
11:47surplus here, it would have been fine.
11:49But it's not. It is defining the sub
11:50price but it didn't do something like
11:53this. I mean if it did it like this
11:57surplus it would have been fine. So
12:00because in Jesse you don't have to
12:02define the surplus inside the unopen
12:04position. You could also do it here but
12:07that only works if it's a futures
12:09account. If it's a spot you cannot do
12:11that but it's not doing that. instead
12:14inside the update position is saying
12:16that
12:18if it's a long position and we're not in
12:19upturn anymore liquid the current
12:21position so basically we are not
12:23submitting a stop loss I don't love that
12:26to be honest especially because it did
12:28calculate the stop loss here it should
12:30have submitted it here so that's at
12:31least one mistake
12:34so let's go to the back test and run it
12:36and see if it goes through all right so
12:38here's the result
12:41it runs fine there is also aren't great.
12:45Let's also try it with another time
12:47frame at least because I'm curious.
12:51All right, so with the 4 hours, it's
12:53actually doing really well. So the draw
12:55down is good. This is a good P&L. The
12:58win rate is really high and the sharp is
13:012.3. So I really like this. So you know
13:04what? Let's also back test it for one
13:06year earlier
13:10because I'm really curious about this
13:11one.
13:14All right. So,
13:17it is losing here, but starting here, it
13:21just went up and up.
13:24So, this is actually good. Like even in
13:26here, it didn't do that bad. So, I think
13:30this strategy has some good potential in
13:32it. So, I'm going to give it a nine out
13:35of 10. I will subtract one because based
13:38on how it calculated the stops itself,
13:41it should have also submitted the order
13:43but it didn't. All right. Next, I'm
13:45going to ask it to update my code to
13:47display the values of important
13:48indicators of my strategy in back tests
13:51for debugging purposes. So, this is a
13:54typical task that I use AI every day
13:56because I'm a bit lazy and I don't want
13:58to do it manually. Now, for this one, we
14:01need a code of a strategy. And I'm going
14:02to use this strategy which I submitted
14:05just a few days ago. If you haven't
14:07watched the video for it, go and check
14:09it out. So, here it is. Let's do this
14:13and actually we already had the
14:16indicator values. So, let's remove it
14:19and
14:21hit enter.
14:23All right. So, let's copy the whole
14:25thing.
14:30Go back, update the code, paste it here,
14:32and save it. And this is how it did it.
14:35Add line to candle chart for tema 10 and
14:3980. It created a new extra line chart
14:42for the long-term values, which used
14:464our time frame.
14:48It added a new one for ADX, a new one
14:51for CMO, and
14:54it also added one just for the trend
14:56lines. So, it's a bit too much, but it
14:59should run just fine. So, let's go back,
15:02bring this back to the beginning of
15:052025, and run it again.
15:10And here it is. So, it executed nine
15:13trades.
15:16All right. So, let's go back and make
15:18sure the back test duration is since the
15:20beginning of turn 25. And also, let's
15:23make sure the interactive charts option
15:25is enabled. And now let's run it.
15:29Here it is. And if I open this option.
15:33All right. Now we have so many charts
15:35that it doesn't allow us to see our
15:37candles chart. Now that's not great,
15:40but there is a fix for it. So let's
15:43first take a look at these numbers. So
15:46for the long-term tema, it looks
15:49perfectly fine.
15:51The ADX is also fine. It's an isolator
15:56between zero and 100. The seam all also
15:59looks perfectly fine. It also defined
16:02the thresholds for upper and lower at 40
16:05and minus 40. It also did a fine job for
16:08the trend. But here's the thing. I don't
16:10think it should have defined the values
16:12of tea 70 and 20 in another chart. It
16:16should have simply added it to the main
16:18candle chart which we cannot see here.
16:20So let's close this.
16:24go back and edit it.
16:27Now, you see here, it shouldn't have
16:29done this, but I'm just going to disable
16:31it because
16:33I want to lower the number of charts.
16:35So, let's disable these two. Go back and
16:38run it again.
16:41Now, we can actually see the candles
16:44chart and also where we bought and where
16:47we sold. Of course, we can also check
16:49out the trades. For example, we can see
16:53the orders of this trade which caused us
16:57to lose. So, we shorted but the price
17:00went up. And here we can see that the
17:03ADX was indeed above the threshold and
17:06everything was according to the strategy
17:08but still it lost money. And we can also
17:11see that the tema 10 and 80 are added to
17:14this chart which makes it really easy
17:16for us to see. So, this is good, but it
17:19didn't do a perfect job. Like, I didn't
17:22like that it added one new chart just
17:25for a value that it could have easily
17:28added here. But I don't want to be too
17:30harsh on it. So, I'm going to give it
17:31nine. Next, write a trend following
17:33strategy that uses 50 minutes time frame
17:36for entry and 4 hours time frame for
17:39confirmation.
17:40All right. So, let's copy the whole
17:42thing.
17:44Go back to Jesse and edit the code.
17:47Paste it here. Now, let's take a look.
17:51It's supposed to be a trend following
17:53strategy that uses the 15 minutes time
17:55frame. Now, this one depends on our pick
17:57of the time frame when running the back
17:59test. But as for the 4 hours, it's
18:02defining the candles 4 hours. It's using
18:04the get candles and it's specifying the
18:08time frame as 4 hours. That's correct.
18:11It defined the EMA of 50 using that for
18:14hours again like this. Now I like this
18:17syntax that it first defined a property
18:20which makes it easy to access here
18:22instead of selecting the candles one
18:24more time in here. It did it again here
18:27for the trend. It's saying if the last
18:29value of this now we usually don't do
18:31this in Jessie because we can simply
18:33write it as this. Right? We don't need
18:35this. But because it passed the
18:37sequential as true it did have to do
18:39this. Now I don't approve of this. It's
18:42not as readable as I like it to be. But
18:44it's correct and it should go just fine.
18:46For an uptrend is it is returning one.
18:48For a downtrend it's returning minus one
18:51and zero for neutral. Then need to find
18:53the MACD like this. And in should long
18:56it's saying that if the last value of
18:58MACD is bigger than the last value of
19:00MACD signal and if the trend is one, it
19:04should go long. It's doing the opposite
19:06for short trend. All right. So it seems
19:08fine. And for the stop loss, it's saying
19:10the entry minus two times of the ATR.
19:12This is also perfectly correct. So
19:14everything seems fine. But again, I
19:16don't love it that it did here because
19:18unless you specifically need the
19:20previous values, never use the
19:23sequential value. It just makes your
19:24code a little bit more complex and
19:27harder to read. All right, let's run it.
19:31And here it is. All right, so it runs
19:33just fine. We executed 21 trades. Oh, we
19:37forgot to set the time frame to 15
19:39minutes. So, if I run it again and this
19:42time we should get an error saying that
19:44we are missing the 4 hours time frame
19:45because I didn't specify it. But since
19:48this is back, Jess is smart enough to
19:51take care of it for us. But if you are
19:53trading a strategy like this in a live
19:55trading mode, you have to also define
19:58the data route for it, which is like
20:00this. So the same symbol except that the
20:03time frame is now 4 hours instead of 15
20:05minutes. Now if I run this one, I
20:08shouldn't get any errors.
20:11All right. So it's fine that it's losing
20:13money because I found scalping
20:14strategies to lose money unless if you
20:17have so many filters because it's taking
20:19271
20:21trades and of course it's paying a lot
20:23of money just for the fees. So it makes
20:26sense
20:28and the win rate isn't really great. But
20:30anyways, so it ran just fine, but I'm
20:33going to give it a nine again because I
20:35didn't like how it defined the indicator
20:38values and it was using the sequential
20:41property which wasn't really needed. Now
20:43the next question is actually really
20:45interesting. So I'm going to say here's
20:47the code for my trend following strategy
20:49in the 15 minutes time frame. The
20:51strategy is losing money because it is
20:52taking so many trades and there are
20:55false breakouts during the ranging
20:57markets. Please improve the strategy. So
20:59I'm going to paste the strategy that
21:01they just gave me. And remember it was
21:03losing money, right? So I want to see if
21:06the model is now smart enough if it has
21:09some actual trading knowledge to know
21:12what it should do to improve the results
21:14of such a case. So this isn't really
21:17related to the syntax, it's related to
21:19the knowledge and this model is supposed
21:21to be really good at that. So let's see.
21:24All right. So here's the code. Let's
21:26copy it. go back and paste it here. Now
21:31let's see what is changed.
21:34So in here other than checking for the
21:36MACD values and the trend it also added
21:40both the ADX and the chop indicator. Now
21:42these are really good. In fact I always
21:44use them myself. The ADX gives us
21:46basically whether or not the price is
21:48volatile enough or not. And the
21:51choppiness is actually ensuring that
21:53there is less randomness in the price
21:56because if there are too many random
21:57stuff happening in the market, we want
21:59to stay out of it, right? Because we
22:01cannot really predict things if
22:03randomness is all over the place. So I
22:05actually like this and it shows me that
22:07the model has some good trading
22:10knowledge. Now for the chop, it's using
22:12the 50 threshold and for ADX 25. Now
22:15both of them are fine. Let's go back and
22:17run it one more time. Now remember this
22:20is the result that we had. So we ended
22:23with 5,696.
22:27Now we are doing significantly better.
22:30Now we're still losing money in the end,
22:32but it's only 2.6%.
22:36Right? So I'm going to give it a perfect
22:3810 because it did a good job. Next
22:40question. Develop a strategy that
22:42combines the Williams percentage R MACD
22:45and Ballinger bands. So I expect this
22:47one to be really easy first. Okay. So
22:50this seems fine. Let's go back and edit
22:54the code. Paste it here. Save it. Now
22:57let's take a look. The volume R is
23:00defined like this. Now to be honest, I'm
23:02not sure about the syntax of this in
23:04Jessie. So we're going to find out when
23:06we run it. The MACD is fine. The
23:09Ballinger bands is also fine. For the
23:11entry, it's saying if the Williams R is
23:14below minus 80, which is correct, and
23:17the MAC is above its signal and if the
23:20closing price is below the lower band.
23:22So, this seems perfect to me. It's also
23:25risking only 3% of the available margin
23:28per each trade, which is a good practice
23:31in my case. It's also risking 3% per
23:34each trade, which is good practice.
23:38And for update position, it's saying if
23:41it's a long position, but the Williams R
23:44is now below minus 20 or we're not in
23:49uptrend anymore or if the price is above
23:52the upper band of the Ballinger bands,
23:54then I want to liquidate it. And also
23:56the opposite for a short trade. So this
23:58seems really good. Let's run it.
24:03All right. So this is good especially
24:06that we are trading it with the 15
24:08minutes. If I go back and let's say we
24:12run this with hourly.
24:16Okay, same numbers. Let's also try 4
24:19hours.
24:23Now we ended with a profit but it only
24:26took two trades.
24:29So let's also try 30 minutes cuz I'm
24:32really curious. So, while that's going,
24:34I want to quickly remind you guys about
24:35Apex Omni, which is my favorite decks
24:38made by the team behind Bybit. It has
24:40low trading fees, instant order
24:42execution, and of course, no KYC
24:44whatsoever. If you're going to check
24:46them out, please consider using our
24:47link, which not only will give you
24:49trading fee discounts, but it will also
24:51support me in order to make more videos
24:53like this. Okay, it did better. It took
24:5618 trades. The max draw is minus one. We
25:00made 6.7%.
25:02The win rate is 55%. The sharp is 2%.
25:05>> So, while that's going, I want to
25:06quickly remind you guys about Apex Omni,
25:08which is my favorite deck made by the
25:11team behind. It has low trading fees,
25:13instant order execution, and of course,
25:15no KYC whatsoever. If you're going to
25:18check them out, please consider using
25:19our link, which not only will give you
25:21trading fee discounts, but it will also
25:23support me in order to make more videos.
25:26>> The sharp is 1.55.
25:29I really like this. Like if I go back
25:31here and just increase the size of the
25:35position with a huge number. Let's say
25:37five
25:42and save it.
25:45So this is pretty interesting. I think I
25:46found a good strategy kind of by
25:48accident.
25:50So now when the maxon is minus 12%,
25:54we're making 84% over 1.5 years. So this
25:59is good and it definitely has some real
26:01potential. All right, so I'm going to
26:03give it a perfect 10, but I wish I could
26:05have given it a bigger number. Next,
26:07write the entry functions of the
26:09strategy for me in a way that it should
26:11use two limit orders for opening the
26:13position and it should only risk 2% of
26:15capital per trade. Again, this is an
26:18easy one, but I want to see if it's
26:21smart enough to just give me one
26:23function, not the entire strategy,
26:26because sometimes that's all you need.
26:27Right now, of course, I could have also
26:29told the model about this, but I didn't
26:31do it. I want to see if it's smart
26:33enough to figure it out. All right, so
26:35it didn't. That's not great.
26:40Now, for entry, it is indeed using two
26:44market orders. So it got that right. So
26:48it's using half the quantity for the
26:51first order and for the price it's
26:53saying the current price minus the
26:56current ATR multiplied by half and also
27:00another one exactly one ATR below the
27:03current price. So this is perfect and it
27:06is also risking 2% per each trade. So
27:10it's correct but it's not giving it in
27:12one single function as I wanted it. It's
27:15just giving me the entire strategy. So,
27:18it's good, but I'm going to give it
27:20nine. Last but not least, write a pair
27:23trading strategy for ETH and BTC. All
27:26right. So, for this one, it's supposed
27:28to give us two strategies and it did do
27:32that. So, you see we have one master
27:34strategy which is guiding all the other
27:38routes and then we have follower
27:40strategies. The first strategy is
27:42supposed to make all the decisions and
27:44it is indeed doing that and using the
27:46shared VS dictionary in Jessie it's
27:50storing these variables in a way that
27:51the second strategy is able to access
27:54it. So in case you didn't know this is
27:56the way for strategies in Jesse to
27:58communicate with each other. Now in this
28:00case we have only two strategies or two
28:02symbols but if you were trading let's
28:04say 10 or 20 or 50 symbols still this
28:07would have been the way to trade them
28:09all. Now, let's copy this one.
28:13Go back to Jesse. But I want to create a
28:16new strategy. I'm gonna name it test for
28:20hair one
28:23and save it. Next, I will create another
28:27one. I will call it test rock for pair
28:32two
28:33and copy this one
28:37and paste it here.
28:40Notice that second strategy doesn't have
28:42trading logics by itself. It's only
28:44using the values of shared VS to see
28:48what the decision has been made for it
28:50by the other strategy. Now, if what I'm
28:53doing here is a bit too complex for you
28:55for this specific strategy, make sure to
28:58watch my previous video called pair
29:01strategy from a scratch in Python. With
29:03that one, I explain everything in depth
29:06about how to go about it. So, let's go
29:08to the back testing page. And now I'm
29:11going to define two trading routes. The
29:14first one for BTC and the second for
29:16ETHUCT.
29:18The time frame is 15 minutes. test Grock
29:22pair one and test Grock pair two. And
29:27I'm going to disable the fast mode
29:29because it doesn't work for strategies
29:31that use multiple trading routes like
29:34this one. And let's start it.
29:39And it's saying we're missing candles.
29:41Okay. So that's because I don't have
29:42enough candles until this date. So let's
29:46lower the ending date of the back test.
29:50Okay, so it's saying that this strategy
29:53name isn't defined. So let's go back and
29:57edit the code here. Okay, so this is
30:00because we only pasted the strategies
30:03code but not the initial imports. So
30:05that's fine because you know it does
30:07have it here. It's assuming that we also
30:10had it when it define this one. So I
30:14mean it's okay. It should have given it
30:16to us, but I'm going to pretend that
30:19didn't happen. Let's go back and run it
30:21one more time.
30:24And now it does go through. Now we're
30:26losing money. That's fine because it's
30:28taking a lot of trades.
30:31And we're paying a lot of money in
30:33trading fees. In fact, if I go and lower
30:36the trading fees, let's say to zero, as
30:39if we own the exchange we were trading
30:41this on,
30:43and run it again. Now, we are ending up
30:47with a little bit of money. Okay, so
30:49this was really good, but because it
30:51forgot the imports, I'm going to give it
30:53a nine. All right, there we go. The
30:55ending number for this LLM model by the
30:58Grock team is 85 out of 100. However,
31:02this is mainly because we gave it a zero
31:04for this question. But I'm guessing that
31:07might have to do with the fact that they
31:09have just released the model and maybe
31:11their API just was failing in this case
31:14and maybe even tomorrow or the next day
31:16if you try it might work and if that's
31:18the case then this is actually a really
31:21really good model. In every question we
31:23basically gave it either 10 or nine. So
31:26it's really good. So overall, I think I
31:28really like this model and maybe at
31:30least whenever Claude Sonnet fails to
31:33get the job done, I tried it with this
31:35one. Or maybe vice versa. Maybe I begin
31:37this one and switch back to CLA when
31:39this isn't able to get it done. But
31:41let's not forget that this model is also
31:43a bit more expensive and I am not sure
31:46if I can say it is worth the price. Now,
31:49before I leave, we're going to have a
31:50giveaway for this video as well. The
31:52random person who likes this video, post
31:54a comment, and subscribes to the channel
31:56is going to win 1 million B token. Now,
31:58let's pick the winner for the previous
32:00video.
32:02[Music]
32:07And the winner is T5G. I hope I'm
32:10pronouncing your name correct. It's
32:12Chinese if I'm not mistaken. Thank you
32:14so much for your comment. Please reach
32:15out to me so I can send you your bunk
32:17tokens. Thanks for watching and I'll see
32:19you in the next one.
32:23[Music]

About this transcript

This page contains the full transcript of Grok 4 tested: Is it good for algo-trading strategies? by Algo-trading with Saleh, generated from the public captions YouTube serves with the video. The transcript has 5,341 words across 751 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.