Claude 4 tested: Good for algo-trading strategies? — Transcript

Full transcript

1. 0:00Hey guys, this is S. The Anthropic team
2. 0:01just released Cloud 4. Now, we already
3. 0:04know that so far Claude Sonnet 3.7 was
4. 0:07the top coding model, but what we care
5. 0:09about as algo traders is the performance
6. 0:11of the model for algo trading. That
7. 0:13means things such as writing strategies,
8. 0:15improving the existing ones, coming up
9. 0:17with trading ideas, converting pin
10. 0:19script strategies into Python, all the
11. 0:21good stuff like that. So, in this video,
12. 0:23I'm going to review the model and test
13. 0:25it with actual use cases so we can see
14. 0:27how good it is. So if you guys are as
15. 0:29excited as I am, let's get right into
16. 0:31it. In order to get the model to perform
17. 0:33best for algo trading, we're not going
18. 0:35to use the cloud platform itself. We
19. 0:37need a platform which allows us to add
20. 0:39our own custom instructions. I already
21. 0:41open sourced the instructions that I
22. 0:43made for Jess GPT which is built for
23. 0:44algo trading and you can find the
24. 0:46repository here. If you don't know what
25. 0:48instructions are and how you can do
26. 0:50that, you can watch my previous video
27. 0:51here where I explain how to self-host
28. 0:54this thing by yourself. But I'm not
29. 0:56going to do that. Instead, I will use
30. 0:58Jesse GPT, which is basically the cloud
31. 1:00version of this, and it's just ready out
32. 1:02of the box. You can get started for free
33. 1:04or even upgrade to premium. But if you
34. 1:07don't want to pay for it, or if you have
35. 1:08privacy concerns, then just self-hosted
36. 1:11with that video. So, let's get started
37. 1:12with the first question. So, let's paste
38. 1:14it here. Write a golden cross strategy
39. 1:16that uses the 50 and 100 moving
40. 1:19averages. Close the position when the
41. 1:20entry rule is no longer valid. So, this
42. 1:23is basically like the hello world of
43. 1:24algo trading. So let's press
44. 1:27enter. And it defined two moving
45. 1:30averages. The shoe long says the moving
46. 1:32average of 50 should be above 100. It's
47. 1:36not shorting which we didn't ask for.
48. 1:38And it's using half the capital for
49. 1:39opening the position. And it is closing
50. 1:42it when the entry rule is the opposite.
51. 1:44So this is actually perfect. And by the
52. 1:46way, make sure you are using close for
53. 1:48by clicking here and choosing this.
54. 1:50Anyways, let's copy this. go to Jesse
55. 1:52and create a new strategy which you can
56. 1:54do by clicking here or here. Now I'm
57. 1:57going to name it cloud for
58. 2:00test. I could just paste it here or I
59. 2:04could open my VS code editor and look it
60. 2:06up
61. 2:07here. So let's paste it here. All right.
62. 2:10So let's go to the back testing page and
63. 2:12look up cloud for test. The 4our time
64. 2:15frame is fine. The BTC USDT is also
65. 2:18fine. And we have one month of back
66. 2:21test. The fast mode. Let's enable it.
67. 2:25Right. So, this is good. Let's start
68. 2:27it. It went just fine. It took two
69. 2:30trades, made 5%. So, this is perfect.
70. 2:32And if I go to the view charts, we can
71. 2:34also look at the trades that it took.
72. 2:36So, it opened the long position. It
73. 2:37closed it here, opened and closed here.
74. 2:40So, this looks perfect. So, we're going
75. 2:42to give it a pass. Next, we're going to
76. 2:44ask it to convert this pin script
77. 2:45strategy into Python code. Now to find a
78. 2:48strategy on trading view, I'm going to
79. 2:49look up super trend strategy. Now I
80. 2:53already have added it to my chart. So
81. 2:56let's see how it looks
82. 2:58like. All right. So it opens a position
83. 3:01and closes it here. These are the
84. 3:03results. And to look at the source code,
85. 3:05let's click here, copy the whole thing,
86. 3:09go back
87. 3:10here, and do this.
88. 3:17All right. So, it defined a super trend
89. 3:19like this, which seems fine. The trend
90. 3:23is either one if the price is bigger
91. 3:25than the super trends trend line. That's
92. 3:27perfectly correct. It's also checking
93. 3:30the previous trend. That seems fine. So,
94. 3:32it says open a long position if the
95. 3:34current trend is one and the previous
96. 3:36trend was minus one. So, it's trying to
97. 3:39catch it the moment that the crossover
98. 3:40happens. We don't really need to do
99. 3:42this, but if you do, you're not going to
100. 3:44open a long position while the trend has
101. 3:47been in uptrend for a while. So, it's
102. 3:49good practice, but I don't usually do it
103. 3:51myself. All right, so this seems fine.
104. 3:53Let's copy the whole thing. Go back and
105. 3:55paste it. Go to Jesse and rerun it, but
106. 4:00this time with the new
107. 4:01code. All right, so it took 17 trades,
108. 4:05but the time frame is hourly now. And
109. 4:08let's also take a look at the trades it
110. 4:10took. All right. So this is perfect
111. 4:12actually. So we bought here for a long
112. 4:15position and
113. 4:18we sold it here. So sell long. So this
114. 4:22seems perfect. And yes, here the trend
115. 4:24is changing. So yeah, this is really
116. 4:27really perfect. Now it's not using two
117. 4:29colors for the super trend like we do on
118. 4:32Trading View. I could have done that,
119. 4:33but we could just ask it because we
120. 4:36didn't. So this is also definitely a
121. 4:38pass. All right. Next, develop a mean
122. 4:40reversion strategy using the Ballinger
123. 4:42bands and the RSI with the seven
124. 4:45period. So here's a mean version
125. 4:48strategy. It defined the Ballinger bands
126. 4:50with a period of 20 and multiplier of
127. 4:53two. The RSI 7. It's like this. All
128. 4:55right. So we lost the formatting and
129. 4:58that's because it added an extra py
130. 5:00here. So let's copy the whole thing. Go
131. 5:02to the editor. Paste it here. But
132. 5:05remember to remove this. All right. So
133. 5:07let's take a quick look at the code. So
134. 5:09yes, the period is now 74 RSI. That's
135. 5:12correct. The entry rule is when the
136. 5:15price is below the lower band and the
137. 5:18RSI is below 30, which means it is
138. 5:20oversold. So we want to buy. So that's
139. 5:22perfectly correct for position
140. 5:25sizing. This also looks perfect. We have
141. 5:27a stop loss with the ATR multiplied by
142. 5:311.5. Submitted the buy order. So it also
143. 5:36submitted the stop loss and take profit
144. 5:37at the moment the position is opened. So
145. 5:40this looks perfect. All right. So here's
146. 5:42the thing though. Even though it
147. 5:43submitted the takerit and its price is
148. 5:46the middle band of the Ballinger bands,
149. 5:48it is also saying that if that is the
150. 5:51case, we want to liquidate the current
151. 5:53position when a closing candle occurs.
152. 5:56But we don't really need this because if
153. 5:58we have the takerit already submitted at
154. 6:00this price, then this condition will
155. 6:03never be needed to be executed. But I'm
156. 6:05just going to leave it be because it
157. 6:07should execute just fine. So let's rerun
158. 6:11it. All right. So we have an
159. 6:13error because this syntax is not
160. 6:17correct. So if I remove this, it should
161. 6:22work. And it does. Well, it does work,
162. 6:25guys, but it just got it wrong because
163. 6:28of that parameter. So, I'm going to give
164. 6:29it a
165. 6:31fail. Now, it's not a big fail, of
166. 6:33course, because it was just one tiny
167. 6:35mistake, but I guess that's the rules
168. 6:38for this video. Next, we're going to ask
169. 6:40it to write a strategy that uses the
170. 6:41Fibonacci retracement levels for entry
171. 6:44and exit points. So, this is a tough
172. 6:46one. Let's hit enter. Here's a Fibonacci
173. 6:49retracement strategy that identifies
174. 6:51swing highs and lows and uses Fibonacci
175. 6:54levels for entries and exits. So it
176. 6:57defined some default variables. It's
177. 7:00using the ATR and
178. 7:03ADX. I defined the swing
179. 7:05points. We have a look back
180. 7:08period which we define here. It is 20.
181. 7:11So that's correct because for Fibonacci
182. 7:13we always need a look back
183. 7:16period. For the highs, it's getting the
184. 7:18last 50 candles and their high prices.
185. 7:22And this is exactly correct. We also do
186. 7:25the same for
187. 7:28lows. All right. And then we have the
188. 7:30fib levels which it defined as a
189. 7:33dictionary like this. It multiplied it
190. 7:36by Fibonacci numbers. So this looks
191. 7:40good. So let's copy the whole thing. Go
192. 7:43back to the editor. Paste it here. go to
193. 7:46Jesse and run
194. 7:48it. It ran. Let's also take a look at
195. 7:53the trade it took. So, these
196. 7:57lines, they do seem fine. Actually,
197. 8:00let's rerun this with the 4 hours
198. 8:03because I don't want to take that many
199. 8:07trades. All right. So, this looks better
200. 8:10even in the P&L number.
201. 8:14So yes, it went long
202. 8:18here and it sold it here. So I'm
203. 8:21guessing yeah, because the closing price
204. 8:23was here. So it opened it at this line
205. 8:26at the line of uh
206. 8:2961.8 and closed it at the Fibonacci line
207. 8:32of
208. 8:3323.6. So this seems perfectly correct.
209. 8:37Now, of course, there are different ways
210. 8:38for trading Fibonacci lines, but I think
211. 8:41this is a good basic version of it. So,
212. 8:43this is definitely a pass. Next, update
213. 8:45my code to display the values of
214. 8:47important indicators of my strategy in
215. 8:49back test for debugging purposes. Now,
216. 8:51for that, we're going to have to feed it
217. 8:53a strategy. And I'm going to use this
218. 8:55one from our website called alligator
219. 8:57AI, which is free. So you just need to
220. 8:59be logged in and click on the get
221. 9:01strategy. Copy the whole thing and come
222. 9:04here and paste it
223. 9:08here. Let's see what it does. And notice
224. 9:12that the strategy itself, it doesn't
225. 9:14have the values for displaying the
226. 9:17indicators on the chart, which is really
227. 9:18helpful for debugging. And there we go.
228. 9:21So it added an after method and it is
229. 9:25using the add line to candle chart which
230. 9:26is the correct syntax for the jaw teeth
231. 9:29and lips of of the alligator. It's also
232. 9:32doing the same for the long-term EMA for
233. 9:35the CMO
234. 9:36oscalator. This looks really perfect. We
235. 9:38could also copy the whole strategy
236. 9:40because sometimes it changes the names
237. 9:42of some parameters. So it's better idea
238. 9:45to just copy the whole thing. Paste it
239. 9:48here.
240. 9:49go back to Jesse and rerun this. Now,
241. 9:53because it added just too many charts, I
242. 9:56have to zoom out my page a little bit. I
243. 9:58apologize for it, but at least we can
244. 10:00see it now. So, here's the chart, the
245. 10:03candlestick charts, and you can see the
246. 10:05moving averages and the lines of the
247. 10:07alligator indicator. And this one is the
248. 10:11CMO. This one is the stoastic RSI. And
249. 10:14you can also see that it defined the
250. 10:16thresholds that we use in the strategy.
251. 10:18the ADX again with the threshold of 30
252. 10:23which you can find here and also the
253. 10:25trend property which we defined which is
254. 10:27either zero or one. You can also see it
255. 10:30here. So this looks perfect. So this
256. 10:32again is the pass. Next write a trend
257. 10:35following strategy that uses 15 minutes
258. 10:37time frame for entry and 4 hours time
259. 10:39frame for confirmation. Right? So it
260. 10:41defined the for candles like this using
261. 10:44the get candles method and it hardcoded
262. 10:47the time frame which is correct and then
263. 10:49it's using that for moving averages to
264. 10:52give us the long-term trend. All right,
265. 10:54so this seems perfect. Let's go back to
266. 10:56the code editor, paste the whole thing
267. 10:59and rerun
268. 11:00it. And there we
269. 11:03go. It's also profitable with some good
270. 11:06numbers. So this is also a pass. Next,
271. 11:09here's the quote for my trend following
272. 11:10strategy in the 15 minutes time frame.
273. 11:13The strategy is losing money because it
274. 11:15is taking too many trades and there are
275. 11:16too many false breakouts during the
276. 11:18ranging markets. Please improve the
277. 11:20strate actually. Let's feed it the
278. 11:21strategy that it just generated for us.
279. 11:24But first, let's make sure that we are
280. 11:26paying for trading fees because it was
281. 11:28set to zero. So, let's run
282. 11:33it. All right. So, the numbers are
283. 11:35slightly different now. But also, let's
284. 11:38lower the time frame to be 15
285. 11:43minutes. And now we're not profitable
286. 11:46anymore. We're taking 39 trades, which
287. 11:49is indeed too many. But we're paying
288. 11:51this much in trading fees. So, if we
289. 11:53weren't paying fees, we would have been
290. 11:55profitable. So, that's really the
291. 11:57problem with the strategy right now. So,
292. 11:58let's go here and just make sure there
293. 12:01is no ADX. All right.
294. 12:04So this is the only entry rule. We are
295. 12:10using the ADX here and I don't want to
296. 12:12do that because usually when I see a
297. 12:14strategy that is taking too many trades,
298. 12:16the first thing that I do is using the
299. 12:18ADX indicator. But there are also some
300. 12:20other methods and what I want to see is
301. 12:23whether or not the model is capable of
302. 12:25reasoning to know what a good strategy
303. 12:28is to know what to do. So not just
304. 12:30writing the ADX for me. I'm not going to
305. 12:32tell it to do that. It needs to know
306. 12:34that is a good idea for this case. So
307. 12:36let's copy this. Remove
308. 12:39it. And yes. So also remove it from
309. 12:43here. Make sure it's not anywhere else.
310. 12:46Let's go back and run it one more time.
311. 12:49Now it's actually making a profit now.
312. 12:52So that's a bit unexpected. So this is a
313. 12:57exception case I guess because it
314. 12:59increased the number of trades. But now
315. 13:01we are even more profitable even though
316. 13:04we are paying for fees. So here's the
317. 13:06thing. Let's change the duration of this
318. 13:09to two
319. 13:12months. And now we are losing and we are
320. 13:16also taking way more trades. So let's
321. 13:18copy the
322. 13:20strategy. Go to GSGPT and just paste the
323. 13:23whole thing. All right. So let's copy
324. 13:26the whole
325. 13:27thing. Go back. Paste it here. And again
326. 13:30it added an extra py here. So make sure
327. 13:33that is removed. So let's check out the
328. 13:36entry rule now. So it added a couple of
329. 13:40things. So it's checking for market
330. 13:42structure which is using the ADX
331. 13:44indicator which is absolutely correct.
332. 13:46It also added a volatility filters which
333. 13:49is using the Ballinger bands width which
334. 13:52is a perfect method actually one very
335. 13:54subtle but effective which I also use
336. 13:57sometimes it's checking for the momentum
337. 13:59strength which is using the RSI and the
338. 14:03MACD. All right. So it basically wants
339. 14:07both the RSI and MACD to have certain
340. 14:09conditions. It's wrote it in a a bit
341. 14:11complex way. I don't love it but it
342. 14:14should work. It's also checking for the
343. 14:17price position which is using moving
344. 14:21averages. It also added a time spacing.
345. 14:24Okay. All right. So, this one is really
346. 14:26good because it says a minimum of eight
347. 14:29candles or 2 hours in this case because
348. 14:32we're trading 15 minutes time frame must
349. 14:34have been passed between trades to avoid
350. 14:37overtrading. So this is also really
351. 14:39effective and the way it does it is by
352. 14:41saying if the current index minus the
353. 14:43last trade index is bigger than 8. Now
354. 14:46the current index is always updated
355. 14:48behind the scenes by Jesse. So we just
356. 14:51need to make sure the last trade index
357. 14:53is also updated which it is doing that
358. 14:58inside the unopen position. So when it
359. 15:00opens a position it stores the value of
360. 15:02the current index inside this variable
361. 15:04so that we can later use it. So this is
362. 15:07really beautiful code. So let's go and
363. 15:10make sure it also
364. 15:13runs and it
365. 15:15doesn't. So this is really a shame
366. 15:18because the syntax of it was really
367. 15:21great so far. So let's see where it made
368. 15:24the mistake. So it's basically here. All
369. 15:27right. So it's passing the Ballinger
370. 15:29band values to SMA getting the average
371. 15:32of them. It looks fine but it's not
372. 15:33running. And unfortunately, I'm going to
373. 15:36have to give it a fail even though the
374. 15:38code was really beautiful. All right,
375. 15:40next. Develop a strategy that combines
376. 15:42the Williams percentage R MACD and
377. 15:45Ballinger bands. So, this is an easy one
378. 15:48actually. All right, so let's copy the
379. 15:50whole thing. Go back to my cursor
380. 15:53editor, paste it, and remove this.
381. 15:58So it defined the Williams percentage R
382. 16:00with the syntax which is correct the
383. 16:02MACD the Ballinger bands and these are
384. 16:06the entry conditions. Now I like what it
385. 16:08did with the comments and how it is
386. 16:10defining the entry conditions because
387. 16:13this is really easy for example if we
388. 16:15need to remove one of them or add it
389. 16:17again. So and the comments are also
390. 16:20really helpful for the
391. 16:21user. All right so this looks perfectly
392. 16:24fine.
393. 16:26Everything else is also correct. I like
394. 16:29the definitions here. All right. So,
395. 16:31let's go and make sure it
396. 16:37runs. And there we
397. 16:39go. Now, if we run it on the 4
398. 16:45hours, it is now profitable and it took
399. 16:49six trades. So, this is definitely a
400. 16:51pass. All right. Next, write the entry
401. 16:53functions of the strategy for me in a
402. 16:55way that it should use two limit orders
403. 16:57for opening the position and it should
404. 16:59only risk 2% of the capital per trade.
405. 17:02So, here's the thing. So far, with every
406. 17:04strategy that we wrote, it was using
407. 17:06market orders for entering and also in
408. 17:08just one point. And what I want to see
409. 17:11if the model gets how we can do it with
410. 17:13multiple points using limit orders. So,
411. 17:16let's hit enter. And it just wrote that
412. 17:19function. So it is also listening to
413. 17:21what I said. It didn't write the whole
414. 17:23strategy and the example it is using it
415. 17:28is using the ATR which makes perfect
416. 17:30sense. It also did the same for a stop
417. 17:32loss. For the quantity it is risking 2%
418. 17:36using the risk to quantity utility
419. 17:38function of Jesse. So that's also
420. 17:40perfectly correct. It is entering with
421. 17:4260% of the capital for the first order
422. 17:46and the second one for with 40%. And it
423. 17:49is using the correct syntax which is
424. 17:51this. So in Jessie if we want to submit
425. 17:54more than one order we don't just enter
426. 17:56it with quantity first and then the
427. 17:58price. We need to define a list of tulip
428. 18:02values. And that's how it did it which
429. 18:04is perfectly correct. So this is also
430. 18:06definitely a pass. And last and
431. 18:08definitely not least write a parading
432. 18:11strategy for ETH and BTC. So this is a
433. 18:14tough one because it needs to write two
434. 18:16strategies and also guide us on how to
435. 18:18run it because we want to long one
436. 18:20strategy while we short the other one.
437. 18:22Now if you don't know what type of a
438. 18:23strategy I'm talking about, watch my
439. 18:25previous video here where I described
440. 18:27how Edward Torb used this strategy to
441. 18:30make his 800 million fortune. All right,
442. 18:32so here's a pair trading strategy for
443. 18:34ETH and BTC. You'll need two strategy
444. 18:36files and two routes configured. So
445. 18:38that's correct. So the first one is the
446. 18:41leading strategy for BTC route. So let's
447. 18:44copy it. Go to Jesse and define one.
448. 18:47Let's call it
449. 18:50claude pairs
450. 18:52one. And another one
451. 18:56for
452. 18:59claude pairs
453. 19:02two. Now for the first one, let's paste
454. 19:05it here. All right. So this is the first
455. 19:07one.
456. 19:09This is the second one which is the
457. 19:10follower strategy. So let's also copy
458. 19:12this one but also let's read what it
459. 19:14says. So route configuration configure
460. 19:17two routes in J. Route one is for BTC
461. 19:20which is the leading and route two is
462. 19:23EUCT which is the following how it
463. 19:25works. Calculate Z score of BTC and ETH
464. 19:28price spread. Enters positions when
465. 19:30Zcore exceeds plus or minus 1.5
466. 19:33threshold. Exis when Zore returns to
467. 19:35neutral. checks configuration daily to
468. 19:37ensure pairs relations exist. Uses
469. 19:40volatility based position sizing for
470. 19:42optimal risk distribution. BTC route
471. 19:45leads all decisions and communicates via
472. 19:47shared VS. Now in case you're not
473. 19:49familiar in Jesse, if you have more than
474. 19:51one strategy and if those strategy files
475. 19:54need to communicate with each other, we
476. 19:56need to use the shared VS variable which
477. 19:58can be accessed from any file. And this
478. 20:01is brilliant. So let's go back to Jesse
479. 20:04and go to the second pair and paste it
480. 20:08here. Now if I go to the back test page
481. 20:11and let's add another trading route and
482. 20:15this one is cloud pairs one cloud press
483. 20:192. This one must be
484. 20:22ETC. So let's run it. Now it's telling
485. 20:26me that I cannot run it because the fast
486. 20:28mode is on. So in case you don't know
487. 20:30the fast mode is really great but it
488. 20:32doesn't work when you are dealing with
489. 20:33multiple routes. So this should be fine
490. 20:36now. Let's run
491. 20:37it. All right. So we're getting this
492. 20:40these errors. Let's cancel it and go and
493. 20:44check the code. So the way it defined
494. 20:46the get candles. So instead of this it
495. 20:49should have
496. 20:50said self exchange so that it would have
497. 20:54worked anywhere. I think that's why we
498. 20:55were getting
499. 20:57that. Yeah. So now it actually crushed
500. 21:00it. So it made one tiny mistake.
501. 21:02Otherwise, it was great and it was even
502. 21:05profitable with some really good
503. 21:07numbers. Now, unfortunately, I'm going
504. 21:08to have to give it a fail, but it did a
505. 21:11pretty good job. It just made one tiny
506. 21:13mistake. So, here's the overall result.
507. 21:15And even though it failed in three
508. 21:17questions, it actually did make them. It
509. 21:20just made one really tiny mistake. So,
510. 21:22I'm not even sure if I had to give it a
511. 21:24fail. Now, this was a new type of video
512. 21:26that I just did. If you guys have any
513. 21:28feedback, especially for the questions,
514. 21:30maybe you have some other questions you
515. 21:32want me to ask the model or maybe we
516. 21:34need to change the rating system, just
517. 21:35let me know by posting a comment on the
518. 21:37video. And before I leave you, we're
519. 21:38going to have a giveaway. A random
520. 21:40person who likes the video, post a
521. 21:42comment, and subscribes to the channel
522. 21:44is going to win 1 million bunk token.
523. 21:46All right, let's pick the winner from
524. 21:47the previous video. And the winner is,
525. 21:50"Thanks for a new strategy to test out.
526. 21:52Thank you so much for your comment.
527. 21:53Please reach out to me so that I can
528. 21:54send you your bank tokens." Thanks for
529. 21:56watching and I'll see you in the next
530. 21:58one.
531. 21:59[Music]
532. 22:00[Applause]