Claude 4.5 tested: Is it Good for algo-trading strategies? — Transcript

Full transcript

1. 0:00Hey guys, the Anthropic team just
2. 0:02released Claude 4.5 which is supposed to
3. 0:05be the king of the coding and in the
4. 0:07past they have proven to indeed be the
5. 0:09king of the coding but what I'm
6. 0:11interested as always is the performance
7. 0:13of these models for algo trading in
8. 0:16Python. So let's test it to see if the
9. 0:18model can write us some good strategies,
10. 0:20whether it can help us to debug our
11. 0:22existing strategies and also whether or
12. 0:25not the model has some actual trading
13. 0:27knowledge to improve the result of an
14. 0:29existing strategy. So if all that sounds
15. 0:31interesting to you, let's get right into
16. 0:33it. So here's a blog model that they
17. 0:35just released. And according to this,
18. 0:37the performance of this model is
19. 0:39supposed to be even better than Opus
20. 0:414.5, which was indeed a really, really
21. 0:44good model. but it costed like a kidney.
22. 0:48But this one costs exactly as much as
23. 0:50the Sonnet 4, which wasn't that
24. 0:53expensive. It was definitely affordable.
25. 0:55Not the cheapest model, but good enough
26. 0:57for most users. And here you can also
27. 0:59see a benchmark of it compared to the
28. 1:01previous models such as GPT5, which is
29. 1:03also a really good model. And Gemini 2.5
30. 1:05Pro, which at this point is considered
31. 1:07an old model, and I'm not even sure why
32. 1:10they are including it in this table. But
33. 1:12anyways, in agent encoding, you can see
34. 1:14that it definitely holds up even
35. 1:16compared to opus. But one that is really
36. 1:19important to us is this one. So, high
37. 1:21school math competition because it is
38. 1:23basically telling us how good the model
39. 1:25is in mathematics, which is also what we
40. 1:27use in algo trading. And you can see it
41. 1:29is beating all the other models by a
42. 1:31high margin except 5. So, that one is
43. 1:35also really good in math. So, that's
44. 1:36really interesting. So I guess the cloud
45. 1:40models in the past were really good at
46. 1:42coding but not as much good at
47. 1:44mathematics like the GPT5 was but in
48. 1:47this model they really made up for that.
49. 1:49So that's really brilliant. And here's
50. 1:51also another interesting metric and that
51. 1:53is financial analysis which is of course
52. 1:55what we care about and it is beating all
53. 1:58the other models including GPT5. So this
54. 2:00is also really nice to see. And here's
55. 2:02also another benchmark chart for
56. 2:04finance. And as you can see, Cloud 4.5
57. 2:06is beating all the other models, even
58. 2:09the Opus 4.1. And that is just
59. 2:12brilliant. Now, before I continue, I got
60. 2:13to say that since our point is to test
61. 2:16the model for algo trading. I don't want
62. 2:18to just run the vanilla model. I want
63. 2:21one that has some custom instructions
64. 2:23for algo trading. For that, you have two
65. 2:25options. So the first one is to use the
66. 2:27cloud version, which I call JessPT. I
67. 2:30made it myself. you can get started for
68. 2:32free or upgrade for more limits and that
69. 2:34one now has access to cloud sonet 4.5
70. 2:37and all the other models that are out
71. 2:39there. So feel free to use this one or
72. 2:41if you don't want to do that let's say
73. 2:43you have some privacy concerns or you
74. 2:45don't want to pay me anything that is
75. 2:47perfectly fine just go and watch my
76. 2:49previous video where I teach you how to
77. 2:51build your own AI for algo trading from
78. 2:53a scratch. So if you use that one you
79. 2:55don't have to pay me anything and you
80. 2:56will keep full privacy for yourself. But
81. 2:58if you want to get started as fast as
82. 3:00possible, Jess GPT is the best option
83. 3:03and that's the one that I'm going to use
84. 3:04in this video. Now, we're going to ask a
85. 3:06model 10 questions and see the results
86. 3:08and give it rating. And the first
87. 3:09question is really simple. Write a
88. 3:11golden cross strategy. So that's like
89. 3:13hello world question in the world of
90. 3:14algo trading, right? So let's give it a
91. 3:16hit. All right. So let's copy the code
92. 3:19and go back to Jess's dashboard. And
93. 3:22here I'm going to create a new strategy
94. 3:24and I will call it test cloud 4.5.
95. 3:30Actually let's put it like this because
96. 3:31this is supposed to be the name of a
97. 3:33Python class. So let's create it and
98. 3:36let's paste the whole thing here. And
99. 3:38let's also give it a read. So let's zoom
100. 3:40in a little bit. All right. So it named
101. 3:43it golden cross. It defined an EMA 50 by
102. 3:47simple syntax. return TA. EMA which is
103. 3:49the correct syntax in just for defining
104. 3:51an indicator. Then it is passing the
105. 3:53current candles and then it is defining
106. 3:56the period. So perfect. It did another
107. 3:58one for EMA 100 and then it is defining
108. 4:00the entry rule which is should long like
109. 4:02whether or not we should open a long
110. 4:04position in it. It is simply saying if
111. 4:06the current EMA 50 is bigger than EMA
112. 4:08100. So it is detecting the moment that
113. 4:11the crossing is happening. This is
114. 4:13perfectly correct. And then in the
115. 4:15golang function which is where we define
116. 4:17the position sizing it is defining the
117. 4:19quantity using the size to quantity
118. 4:21utility function of Jesse. It is passing
119. 4:24the entire available margin that we have
120. 4:26and then the current price as the entry
121. 4:28price and the fee of the current
122. 4:29exchange. And to submit the buy order it
123. 4:32simply says the quantity and then the
124. 4:34price which is the current price. So we
125. 4:37want to use a market order to open the
126. 4:39position. So this is perfect. And of
127. 4:41course for short positions it's doing
128. 4:43the opposite. So instead of self buy it
129. 4:45is using selfell and then you define the
130. 4:47update position which is where we
131. 4:50basically say if it's long position but
132. 4:52the entry rule is now the opposite
133. 4:54liquid the current position or vice
134. 4:57versa for a short position. Again this
135. 4:59is perfectly correct. It also defined
136. 5:01the should cancel entry function which
137. 5:03isn't really necessary in this strategy
138. 5:06because we are using market order to
139. 5:08open the position. If we were using a
140. 5:10limit order or even a stop order, we
141. 5:12would have had the question of like what
142. 5:14should happen if the order didn't get
143. 5:16filled and now another candle has
144. 5:18closed. So, should we cancel it to maybe
145. 5:21submit a new one or should we just leave
146. 5:22it as it is? So, that's why we use this
147. 5:25function in Jesse. But in this case,
148. 5:27because we're using a market order, we
149. 5:29don't need that. So, we could even
150. 5:30remove the sync and it should still
151. 5:32work. But I'm going to leave it be. So,
152. 5:34let's go to the back testing page. Now,
153. 5:36the exchange is Binance Piocial Futures.
154. 5:38The route let's choose BTCUCT
155. 5:41and for time frame let's use hourly and
156. 5:45change this into test cloud 4.5 and for
157. 5:50period could be anything really. Let's
158. 5:52just pick since the beginning of 2025
159. 5:55and ensure the fast mode is on and is
160. 5:59the benchmarking. All right, so let's
161. 6:01run the first back test. It's a perfect
162. 6:04go. But actually, you know what? Because
163. 6:06this was a really hard period to trade,
164. 6:08just to make sure it is fun, let's
165. 6:11change this into 2024. So, we could see
166. 6:15some profitable results, right? I mean,
167. 6:17it doesn't really matter because the
168. 6:19point of this testing isn't that, but
169. 6:21still, I'd like to see some positive
170. 6:23results if possible. All right. So, this
171. 6:25looks a lot better now. Now, it executed
172. 6:2879 trades. It made 42% in profit. The
173. 6:32maxon is minus almost 30%. So this is
174. 6:34fairly okay. The venote is 34%. All
175. 6:38right. So this is definitely a pass.
176. 6:40Next question. Convert this pine script
177. 6:42into Python code. Now for this next
178. 6:44part, we need an actual pine script of
179. 6:46trading view. Right. So let's go to it
180. 6:48and look up the word super trend
181. 6:52strategy. Yeah. So this one here. I
182. 6:55should be able to get the pine script.
183. 6:57All right. So here we go. Let's copy it.
184. 6:59And now I should be able to paste the
185. 7:01whole thing. Hit enter. All right. So
186. 7:04let's copy the whole thing. Go back and
187. 7:07paste it here. All right. So let's take
188. 7:09a look at the code. So it define some
189. 7:11hyperparameters. Now I don't like it
190. 7:13when the models do this. Maybe I need to
191. 7:15fine-tune the instructions to
192. 7:17specifically tell them to avoid doing
193. 7:19this because I prefer them to actually
194. 7:21write the period of the indicators
195. 7:22instead of trying to optimize like since
196. 7:24the beginning. But anyways, it defined
197. 7:26the super trend like this and this is
198. 7:30the correct syntax dispassing the
199. 7:32candles. The period is this which again
200. 7:34is the syntax it is using for uh
201. 7:36parameter optimization in Jesse. Then it
202. 7:38defined the inter rule. So basically
203. 7:40when the trend is one we want to go
204. 7:43long. Now here's the thing though this
205. 7:45isn't right because in Jesse's
206. 7:46indicators with the super trend at least
207. 7:48the trend value actually gives us a
208. 7:51single value. So it's a scalar value
209. 7:54like you know in here it will give us
210. 7:58this line here and then we're going to
211. 8:01know that if the current price is bigger
212. 8:02than this it's an uptrend. If the
213. 8:05current price is below it then it's a
214. 8:06downtrend right? So it will give us
215. 8:09that. It won't just give us one or minus
216. 8:11one. So this syntax just won't work. So,
217. 8:15actually, let's just give it a run and
218. 8:18see if it runs or not because I know it
219. 8:20probably won't. Okay, I think I forgot
220. 8:23to save it.
221. 8:24Yep. Let's run it one more time. And you
222. 8:27see it didn't execute anything because
223. 8:29just those rules were true. So,
224. 8:31unfortunately, this is a fail. Next,
225. 8:33develop a mean reversion strategy using
226. 8:35the Ballinger bands and the RSI with the
227. 8:37seven period. Let's copy this. Go back
228. 8:41and change the code.
229. 8:43Save it this time. So this is really
230. 8:45important and let's run it. All right.
231. 8:47So we got an error saying that the mult
232. 8:50argument is incorrect. So let's go and
233. 8:52see what's going on. So apparently when
234. 8:54it defined the Ballinger bands, it also
235. 8:56defined this which is incorrect. So
236. 8:58let's remove this. By the way, let's
237. 9:01also take a look at other code. So it
238. 9:03defined the RSI with this syntax which
239. 9:05is correct. The ATR indicator with this
240. 9:08which is also correct. The shoot long is
241. 9:10saying if the current price is below the
242. 9:12lower band of the current Ballinger
243. 9:14bands and the RSI is below 30 which is
244. 9:18considering it an oversold situation and
245. 9:21for shorting positions it's doing the
246. 9:23opposite. It is indeed risking 2% perish
247. 9:26trade using the risk to quantity utility
248. 9:28function of Jesse. So this is also
249. 9:30perfectly correct. And once it opens the
250. 9:33position, it is submitting the stop loss
251. 9:35and the track take profit using this
252. 9:37syntax which passes the quantity of the
253. 9:39current position and the price using the
254. 9:42middle band. So this is perfect really.
255. 9:45So let's just give it another run. Now
256. 9:48while that's going, I want to quickly
257. 9:49remind you guys about our Telegram. It's
258. 9:51the fastest way to get notified about my
259. 9:53future work, whether it's a new tutorial
260. 9:55or a tool that I create. Also, don't
261. 9:57forget to check out our free Discord
262. 9:58where more than 5,000 members like you
263. 10:00and I are hanging out there and helping
264. 10:02out each other with algo trading so we
265. 10:04can all succeed together. The links for
266. 10:06both are down in the description. And
267. 10:08there we go. So, it executed
268. 10:09successfully. Now, the problem is it is
269. 10:12executing too many trades. So, that's
270. 10:14why it is losing a lot because we are
271. 10:16paying a lot of money in trading fees.
272. 10:18But otherwise, it just went fine and I'm
273. 10:21going to give it a pass. But because it
274. 10:22made one mistake, I'm going to give it a
275. 10:249 out of 10. Next, write a strategy that
276. 10:27uses Fibonacci retracement levels for
277. 10:29entry and exit points. All right, let's
278. 10:32copy the code and give it a read. All
279. 10:34right, so it defined the swing high as
280. 10:36this return MP.max. So the maximum value
281. 10:40of the last 50 candles and then it is
282. 10:43selecting the high value of every single
283. 10:46one of them. So this is perfect. And
284. 10:47then for swing low, it did the opposite.
285. 10:50So it is using MP.min and then it is
286. 10:54selecting the low value of the candles
287. 10:57and then it define the fib levels with
288. 10:59this syntax. So the diff equals swing
289. 11:01high minus swing low and then it's
290. 11:03returning a dictionary with the level
291. 11:06value. So this is perfect. It's also
292. 11:09very simple to read. I like it. And then
293. 11:11for the trend, it's using the EMA is
294. 11:14saying whether or not the current price
295. 11:16is above or below the EMA to return one
296. 11:19for an uptrend and minus one for a
297. 11:21downtrend. So this is also great for
298. 11:23shoot long. It is saying if it's not an
299. 11:26uptrend return. Okay, so that's cool.
300. 11:28Now assuming it is, it's saying if the
301. 11:31price is near 61 or 78 Fibonacci levels,
302. 11:36we want to go into a long position. So I
303. 11:38love what it did here. This is really
304. 11:41good. And then for a short position,
305. 11:43it's doing the opposite. It defined the
306. 11:45position sizing and it is submitting the
307. 11:48stop loss and take profit again using
308. 11:50fib levels. So this is also correct. All
309. 11:54right. So I think it nailed this
310. 11:55question overall. So let's go and run
311. 11:58it. And it does. It's executed a fairly
312. 12:01good number of trades actually. So this
313. 12:03was really really good. Like one of the
314. 12:05best I've ever seen. The next question
315. 12:08is update my code to display the values
316. 12:11of important indicators of my strategy
317. 12:13in back test for debugging purposes. So
318. 12:16basically we wanted to update an already
319. 12:18existing strategy and we needed to show
320. 12:21us some charts for it. Now there is a
321. 12:23specific syntax for this in Jesse. So we
322. 12:25just need the model to understand that
323. 12:28syntax and use it correctly. Now to feed
324. 12:30it a strategy I'm going to get it from
325. 12:31our website the strategies page. So this
326. 12:34one is called temo trend following. So
327. 12:37let's click here and here to copy the
328. 12:39code of it and paste it here. Now I'm
329. 12:42going to remove these because this is
330. 12:44exactly what we want the model to give
331. 12:46us. So let's just remove it to make sure
332. 12:48it won't cheat and hit enter and see
333. 12:51what it gives us. All right, let's copy
334. 12:54the whole thing, paste it here, save it.
335. 12:57All right, so this is exactly what I was
336. 12:58hoping it would give me. It defined the
337. 13:00after function and it is using the add
338. 13:02line to candle chart syntax to add the
339. 13:06values of the tema indicator to the
340. 13:08current candle chart which I will show
341. 13:10you in a second and then it defined the
342. 13:13another chart using the add extra line
343. 13:15chart. So you define a new chart because
344. 13:18the values for the ADX which is an
345. 13:20oscillator. So it's going to be between
346. 13:231 to 100 it's going to differ from the
347. 13:26values of TMA indicator right. So for
348. 13:28BTC this could be in like 100,000 but
349. 13:31this one is going to be between 0 and
350. 13:33100. So we needed another chart and it
351. 13:36did the same is for CMO. So this seems
352. 13:39perfectly fine. Let's just go and run
353. 13:41it. Now this error we're getting is
354. 13:43because the strategy is using the 4our
355. 13:45time frame but we didn't define the data
356. 13:48route for it in our back test. But in
357. 13:50the when you run the back test with JC
358. 13:52it is smart enough to take care of it
359. 13:53behind the scene but it will still let
360. 13:56you know about it because when you're
361. 13:58live trading the same strategy you need
362. 13:59to ensure those routes exist otherwise
363. 14:01you will face an issue. All right so it
364. 14:04went just fine but I forgot to enable
365. 14:07the interactive chart so we can actually
366. 14:09see those charts. So let's run it one
367. 14:11more time. And now if I click here,
368. 14:14here's the chart.
369. 14:17And you see not only we have the
370. 14:19candles, we also have the demo values
371. 14:21which are these blue, green and red
372. 14:24line. So we can see for example when it
373. 14:26took this trade were the values
374. 14:28correctly in line or not. And then we
375. 14:31have the values of CMO and the ADX. It
376. 14:34also define the thresholds. The same
377. 14:37goes for here. So we have the long level
378. 14:39and the short level for CMO and we can
379. 14:41see whether or not the value of it is
380. 14:42above these threshold. So this is
381. 14:45perfect. So this is another 10. Next
382. 14:47write a trend following strategy that
383. 14:49uses 50 minutes time frame for entry and
384. 14:514hour time frame for confirmation. It
385. 14:53named it trend following 50 minutes.
386. 14:56That is okay. Define the 4our time frame
387. 14:59using the get candles method of Jesse.
388. 15:02When we use this, we need to pass the
389. 15:04name of the current exchange, the name
390. 15:06of the current or whichever symbol we
391. 15:07want and then the time frame for that it
392. 15:10hardcoded four hours. So this is
393. 15:12perfectly correct because when it
394. 15:13defines it as a property like this, we
395. 15:16are able to use it in multiple points in
396. 15:18our strategy. So I really like this
397. 15:20syntax. Then it defined the EMA fast,
398. 15:23EMA slow, the big trend. So for big
399. 15:26trend it is using the bigger time frame
400. 15:29candle. So this is perfectly correct. If
401. 15:32it's an uptrend, it's just returning
402. 15:33one. If it's a downtrend, it returns a
403. 15:36minus one. And then it define the ADX,
404. 15:38the ATR, and the shoot long. So it's
405. 15:42checking for the trend, the ADX, and
406. 15:45whether or not the fast is bigger than
407. 15:47the slow. Once the position is open, it
408. 15:49is submitting the stop loss and take
409. 15:51profit using the ATR indicator. So this
410. 15:54might be actually a good profitable
411. 15:57strategy. So let's give it a try. But
412. 16:00you know what? I'm going to disable this
413. 16:02interactive chart now because the back
414. 16:04test will go a little bit faster when it
415. 16:06is off. All right. So, this is
416. 16:08profitable. It's not that bad. Of
417. 16:10course, it's not beating the market
418. 16:11because we didn't really work on the
419. 16:12strategy, but the win rate is also
420. 16:14acceptable. The sharp is also
421. 16:16acceptable. So, yeah, this is perfectly
422. 16:18a pass. Now, next, we're saying here's
423. 16:20the code for my current trend following
424. 16:22strategy in the 15 minutes time frame.
425. 16:24the strategy is losing money because it
426. 16:26is taking too many trades and there are
427. 16:28too many false breakouts during the
428. 16:30ranging markets. Please improve the
429. 16:31strategy. So I'm talking about the
430. 16:33strategy that we just back tested but
431. 16:36you know in the period that we selected
432. 16:38it isn't actually losing money but I'm
433. 16:40guessing that is because in the code it
434. 16:43was smart enough in the first place to
435. 16:45use the ADX indicator right. So if that
436. 16:47wasn't the case I believe it would have
437. 16:49lost money. So, you know what? Let's
438. 16:51just go back and remove that single line
439. 16:53cuz I want to see if it is able to
440. 16:56improve the result or not. So, let's
441. 16:58just remove this one and this one. All
442. 17:00right. So, no more checking for the ADX
443. 17:03because it is actually the best defense
444. 17:06against a ranging market. So, this is
445. 17:08good. Now, let's go back and run it one
446. 17:10more time. There we go. So, you see now
447. 17:12it's almost losing money because it is
448. 17:14taking too many trades. Now again it's
449. 17:16still is losing my almost losing money
450. 17:19because the result are still good. So
451. 17:21anyways let's just go back and copy the
452. 17:24whole code and here I'm going to paste
453. 17:26it in and we're going to see what the
454. 17:29obvious version is going to be. All
455. 17:30right let's copy it. Actually let's read
456. 17:33what it did. So it added the ADX filter
457. 17:35so it only trades when the ADX is above
458. 17:3725 which is what it did in the first
459. 17:39place. It also added the choppiness
460. 17:41index which is also another great
461. 17:43indicator against a ranging market. So
462. 17:46it avoids choppy or sideway markets when
463. 17:48it's below 50. Yes, that is perfectly
464. 17:50correct. It added the Ballinger bands
465. 17:53with again a really good one because it
466. 17:55measures like what's the strength of the
467. 17:58current wave or or price action that we
468. 18:02are in. And we want it to be above
469. 18:03three. So basically when this is really
470. 18:05low like below three, it's like the
471. 18:07market doesn't even know where it wants
472. 18:08to go. Like the changes are really
473. 18:10small. It is also waiting a little bit
474. 18:12between the trades to not take too many
475. 18:15trades and it is exiting on a weak
476. 18:18trend. So it liquidate the position when
477. 18:20the ADX drops below 20. So these are
478. 18:22really great. Let's go and update our
479. 18:26code and re-execute it. So yeah, it
480. 18:30definitely improved. So it is now taking
481. 18:34124 instead of like 400 or something. It
482. 18:37is also making more. The max rodon is
483. 18:39lower. The win rate is also better and
484. 18:43the sharp is also acceptable. But it
485. 18:45isn't as good as like the first version
486. 18:47that it wrote. But it doesn't matter
487. 18:49because still the version that I fed the
488. 18:51model I specifically asked that like
489. 18:54avoid ranging markets and it added
490. 18:56multiple filters to do just that. So it
491. 18:59was really good. Next, develop a
492. 19:02strategy that combines the Williams
493. 19:03percentage R, MACD, and Ballinger bands.
494. 19:06Okay, so it defined the Williams R
495. 19:08percentage with this syntax, which is
496. 19:10perfectly correct. Defined the MACD with
497. 19:12TA MACD is also correct. The Ballinger
498. 19:16bands, it named it BB, which I really
499. 19:19like. Then it defined the ATR. uh for in
500. 19:22rules of the strategy is saying open a
501. 19:24long position if the volume's percentage
502. 19:27r is below minus 80 and the maxis h
503. 19:31value is bigger than zero and the
504. 19:33current price is below the ballinger
505. 19:35bands but it is multiplying it by 1% so
506. 19:40why is it doing that all right so it
507. 19:42wants to make sure we are near the lower
508. 19:45band not like right next to it so this
509. 19:49is interesting I don't how I feel about
510. 19:51it. I It's not exactly what I asked it
511. 19:54to do, but it is interesting
512. 19:56nonetheless. Let's see if there's any
513. 19:58explanation about this. Well, no, there
514. 20:00isn't. All right, let's go and run it.
515. 20:04There we go. It runs just fine. It
516. 20:08didn't make any profit. The max is minus
517. 20:1015. So, this is good enough and it's
518. 20:13definitely a pass.
519. 20:15Next, write the entry functions of the
520. 20:17strategy for me in a way that it should
521. 20:19use two limit orders for opening the
522. 20:21position and it should only risk 2% of
523. 20:24capital per trade. So, in this test,
524. 20:27what I want the model to do is to just
525. 20:29give me two functions. I don't want it
526. 20:31to write two whole strategies because
527. 20:33some models, they're good, but they
528. 20:35overdo stuff like you ask them to write
529. 20:38a function and they give you a hold of
530. 20:39strategy and I just don't like that. So,
531. 20:42let's see how this one does. There we
532. 20:44go. So it actually gave me just a
533. 20:46function the entry function with the
534. 20:48correct syntax in Jesse. It defined two
535. 20:51entry prices using ATR
536. 20:55and then it defined the surplus. It does
537. 20:57know the correct syntax in Jesse for
538. 21:00submitting multiple entry orders which
539. 21:02is you know you define a list instead of
540. 21:04just giving the quantity and a price. So
541. 21:07this is perfect. It gave us the short
542. 21:09version of it. All right. So this is
543. 21:11definitely a pass.
544. 21:14And last but not least, write a pair
545. 21:16trading strategy for ETH and BTC. So for
546. 21:20this one, it's going to be a bit more
547. 21:22complicated because it needs to write
548. 21:24two strategy classes. And the way P
549. 21:27trading strategy works is that one of
550. 21:29them goes long while the other one goes
551. 21:31short. Now, if this is confusing to you,
552. 21:34watch one of my previous videos called
553. 21:35Praing Strategy from Scratch in Python
554. 21:38where I explain everything including the
555. 21:40math behind it from scratch. All right,
556. 21:43so here's the leading strategy. So,
557. 21:45let's copy the name of it and go to
558. 21:48Jesse and paste it in. I will name it
559. 21:53cloud 45 and then whatever name it gave
560. 21:56me. Let's copy the whole thing. All
561. 21:58right. So you see it defined the BTC
562. 22:01returns. So it's using the prices of the
563. 22:03BTC prices but using the prices to
564. 22:06returns utility function of Jesse. It is
565. 22:08converting them into returns. It's doing
566. 22:10the same thing for ETH. So yes in Jessie
567. 22:12you have access to prices of other
568. 22:15trading routes from one single class. So
569. 22:18it's really cool. It define the Z score.
570. 22:21Again if this is confusing to you just
571. 22:24watch my previous video about pair
572. 22:26trading if that's what you're interested
573. 22:27about. But what I'm curious about is if
574. 22:30it knows how to communicate between
575. 22:33multiple trading strategies of Jesse.
576. 22:35So, and it does because it's using the
577. 22:37shared wires property which is exactly
578. 22:39how you're supposed to do it. This seems
579. 22:41perfectly fine. So, let's go back and
580. 22:43see the following strategy which is
581. 22:45supposed to be for ETH. So, for this
582. 22:47one, it's define the entry and exit
583. 22:50functions and in them it is simply
584. 22:53reading from shared VS. So for example,
585. 22:55if position is supposed to be one, we're
586. 22:57going to go long on this. If it's minus
587. 23:00one, we're going to go short on it. And
588. 23:02then in here, if it's supposed to be
589. 23:03closed and we do have a open position,
590. 23:05it will just liquidate it. So this is
591. 23:07perfectly great. Let's just copy the
592. 23:09whole thing and create another one. So
593. 23:12I'm going to name it Claude 45
594. 23:15follower and paste it here. So let's go
595. 23:18here and choose claude 44 BTC ETH pair.
596. 23:24Let's add another one for ETH.
597. 23:29Change the time frame to hourly to be
598. 23:31the same as this. And then claude 4.5
599. 23:36follower. Now we need to disable the
600. 23:38fast mode or otherwise we'll get an
601. 23:40error. Now let's just run it and see how
602. 23:42it goes. There we go. So you see these
603. 23:45are the prices of BTC and ETH and you
604. 23:48can see they are correlated and
605. 23:51configurated together which means like
606. 23:54the cases where the price of ETH is
607. 23:56higher than BTC. So for example if we
608. 23:58were to short ETH here and long BTC like
609. 24:02at this point it would have been vice
610. 24:03versa. So we would have been sitting on
611. 24:04a profit or in here we could have done
612. 24:07the other way around. So we could have
613. 24:09longed ETH and went short on BTC and
614. 24:13then at this point it would have been
615. 24:15the other way around. So we would have
616. 24:17been sitting on a profit. So the logic
617. 24:19of the strategy makes sense. But you got
618. 24:21to be careful not to take too many
619. 24:23trades and that's exactly what it's
620. 24:24doing here. So it took way too many
621. 24:26trades. It's paying way too much for
622. 24:29trading fees. So in fact if we added
623. 24:31this to this number here we would have
624. 24:33been on a profit already. Right? But
625. 24:36since trading fees are a fact of
626. 24:38trading, so we just need to keep working
627. 24:41on the strategy. But nonetheless, it is
628. 24:44working and it is definitely a pass. So
629. 24:48there you go guys, here's the final
630. 24:50score. 89 out of 100. So in this
631. 24:53question, it got a nine. In this one, it
632. 24:56got a zero. So that's just it. And the
633. 24:59result that we seeing here is beating
634. 25:02Cloud 4. So it definitely had some
635. 25:04improvements. But if we compare it to
636. 25:06GPD5 mini that's I did a while ago. It's
637. 25:10not beating that one because that one
638. 25:12got 100 out of 100. So it was perfect
639. 25:15score with every question. But this one
640. 25:17isn't beating that. But I wouldn't say
641. 25:20that GPT5 mini for example is a better
642. 25:22model than this because it depends on
643. 25:24what you're doing. So in oneshot
644. 25:26questions like this yet GPD5 is beating
645. 25:29this but when it comes to actual coding
646. 25:31or debugging my experience during the
647. 25:35past months has been that the cloud is
648. 25:37still a better model but when it comes
649. 25:39to debugging something really complex
650. 25:42GP5 beats clot. So that was my
651. 25:44experience. Now, this one, here's the
652. 25:47result of it oneshotting the questions.
653. 25:49But when it comes to like iterating over
654. 25:52them, we don't know yet until we give it
655. 25:55some try. But overall, it seems like a
656. 25:58good model, and I'm happy that they gave
657. 26:00us this upgrade. Now, I hope you enjoyed
658. 26:02this video. If you did, please make sure
659. 26:04to give it a like and subscribe to the
660. 26:06channel for more videos like this. Thank
661. 26:08you so much for watching. I'll see you
662. 26:10in the next one. Now what does CL