I built a 24/7 Claude Code setup for trading — Transcript

Full transcript

1. 0:00Today, I'm going to show you a Cloud
2. 0:01Code setup that I just did on my Mac
3. 0:04mini, which is a remote machine that I
4. 0:06have access to. In my case, that is on
5. 0:07my home's network, but it could be
6. 0:09anywhere in the world. And this allows
7. 0:11me to run algo trading research for
8. 0:1424/7. So, not for minutes or hours, this
9. 0:17could literally be for days. And that
10. 0:19allows me to ask for very high sharp
11. 0:21ratio strategies. And I can even set
12. 0:23some high criteria thanks to Monte Carlo
13. 0:26to ensure the strategies that the agent
14. 0:28finds for me aren't over fit. So, this
15. 0:30is like a dream come true because until
16. 0:32recently we did not have access to
17. 0:34models that could do this. And thanks to
18. 0:36Cloud Code subscriptions, [music]
19. 0:37it's even going to be affordable for me.
20. 0:39If you have watched my previous videos,
21. 0:41I always say that in order to trade
22. 0:43successfully, you're going to need more
23. 0:45than just one good strategy. In fact,
24. 0:47the only holy grail of trading that I
25. 0:49have found is if you had a portfolio of
26. 0:5120, 30, or maybe even 50 strategies. And
27. 0:54this setup helps me to reach that goal
28. 0:56as fast as possible. Now, before I
29. 0:58continue, I got to say I am not advisor
30. 1:01and this video is for educational only.
31. 1:03So, with that out of the way, let's get
32. 1:05right to the video.
33. 1:07As always, I'm going to be using the
34. 1:08Jesse framework for the algo trading
35. 1:10side of things. But, I'm not going to go
36. 1:12through the installation steps in this
37. 1:14video. You can check out the
38. 1:15documentation, which is really easy to
39. 1:17follow, or you can watch my previous
40. 1:19tutorials about how to set up Jesse.
41. 1:21But, in this video, the goal is to do it
42. 1:23inside my remote Mac mini. So, in front
43. 1:25of me, I have the screen sharing app of
44. 1:28macOS and I've already added my Mac mini
45. 1:31machine. So, if I click on it, I'm going
46. 1:32to be able to log in to it. So, I'm
47. 1:35going to enter my password and there we
48. 1:36go. So, in it I have the absolute bare
49. 1:39minimum installed. And that is
50. 1:41intentional because I just need to run
51. 1:43Cloud Code and Python and Jesse in it.
52. 1:46That's it. But, just in case I have some
53. 1:48browser, some VPN app, and that's it.
54. 1:49So, nothing that concerns you. And I
55. 1:51could enter a terminal app inside here
56. 1:54and continue. But, I don't want to do
57. 1:57that. Instead, what I prefer to do is to
58. 1:59use this via my own terminal app inside
59. 2:02my own machine. So, I've already created
60. 2:05an alias for this. So, when I say SSH
61. 2:08Mac mini, it's going to enter it. In
62. 2:09case you're interested to see what's
63. 2:11behind this command, this is it. So,
64. 2:13nothing fancy. SSH, then my username,
65. 2:16and the local IP address of that machine
66. 2:18on my home's network. So, let's enter
67. 2:21it. And now I am logged in to that new
68. 2:23machine. Next, I'm going to CD into my
69. 2:25Jesse project on this machine. Now, for
70. 2:27me, that is going to be CD Jesse, and
71. 2:30then I'm going to go back, and then find
72. 2:33the gym project. There we go. Now, I
73. 2:35have also automatically activated the
74. 2:37correct conda environment. Now, in your
75. 2:39case, that's going to be conda activate,
76. 2:41and then the name of that environment,
77. 2:43which for me is Jesse 3.12. And I know
78. 2:46it's already activated because I'm
79. 2:47seeing it here. So, now if I, for
80. 2:49example, say pip list, I'm going to find
81. 2:52Jesse being installed here, right there.
82. 2:55All right. So, very quickly, before I
83. 2:57continue, I want to open another
84. 2:58terminal window. So, let's first see
85. 3:00which directory is this one. All right.
86. 3:02So, I'm going to copy this one. I'm
87. 3:04going to SSH into Mac mini one more
88. 3:06time, and then I'm going to say CD, and
89. 3:08then here. There we go. Now, the reason
90. 3:10I did this is because now I want to
91. 3:12firstly see if I can do Jesse run
92. 3:15without any issues, and I can. Now, I am
93. 3:17getting one error related to Python
94. 3:19language server, but I don't really need
95. 3:21this one, so I'm going to skip this. The
96. 3:23important part is that I am getting this
97. 3:25MCP command, and everything else seems
98. 3:28to be working just fine. Now, before we
99. 3:29continue, I'm going to stop this one
100. 3:31more time, and I'm going to run this
101. 3:34inside a tmux window. And the reason I
102. 3:37want to do that is because I want to be
103. 3:38able to close my local machine's
104. 3:40terminal without this SSH window getting
105. 3:43closed. So, for that, I'm going to say
106. 3:45tmux new, and then I'm going to give it
107. 3:48a name. There we go. And because I'm
108. 3:50seeing this green line here, I know that
109. 3:52everything is just correct. By the way,
110. 3:55forget this line here. I'm going to have
111. 3:57to fix this later, but for now
112. 3:58everything seems fine. So, now I'm going
113. 4:00to say Jesse run one more time and it's
114. 4:03telling me that it doesn't know Jesse.
115. 4:05That's because the correct conda
116. 4:06environment isn't activated. So, now
117. 4:08that that is done, it should work. And
118. 4:11there we go. So, one more time inside
119. 4:12this terminal window, I'm going to run
120. 4:15the tmux window one more time. So,
121. 4:17that's going to be tmux and then new and
122. 4:22then dash s and I'm going to give it
123. 4:24another name, but this time because I
124. 4:26want to run the cloud code in it, I'm
125. 4:28going to name it cloud. And there we go.
126. 4:31Now, I could run the conda environment,
127. 4:33but I don't think I need it, but it
128. 4:34doesn't matter. Just in case I'm going
129. 4:36to do that. And before running the cloud
130. 4:38code, I want to ensure to add this MCP
131. 4:41of Jesse in it. For that, I'm going to
132. 4:43go to Jesse's documentation on the
133. 4:46connecting cloud code and I'm going to
134. 4:48copy this command here. Going to go back
135. 4:51to my terminal and paste it here. But
136. 4:53there is a catch here, guys. So, you see
137. 4:55in this command, the MCP address has the
138. 4:58port 9002, but in here I'm seeing it is
139. 5:009003. So, I'm going to edit this into
140. 5:049003 and hit enter. That's it. You see,
141. 5:07it's telling me that this new MCP has
142. 5:09been added. So, now next time I run the
143. 5:11cloud in it, it's going to have access
144. 5:13to that MCP. So, let's say cloud and
145. 5:16then I'm going to create new session.
146. 5:19So, I'm going to remove this continue
147. 5:21flag here, but I'm going to leave
148. 5:23permission mode to bypass permissions
149. 5:25because I don't want it to keep asking
150. 5:26me for annoying messages, but many
151. 5:29people say this is dangerous, so you
152. 5:31need to be careful. Feel free not to
153. 5:32pass it, but again you're going to have
154. 5:34to deal with so much, but I am going to
155. 5:36do it especially because this is
156. 5:37supposed to be an isolated environment
157. 5:40meant for doing exactly this. So,
158. 5:42there's not much that cloud could even
159. 5:45mess with even if it wanted to, but in
160. 5:47all these months that I've been using
161. 5:49Cloud Code, I've never faced any issues
162. 5:51with it. So, I'm just going to go with
163. 5:52this. There we go. And you see, as soon
164. 5:54as I ran that, in this terminal, I can
165. 5:56see some requests being made into the
166. 5:58MCP, and everything seems to be fine.
167. 6:00Now, here, I'm also going to switch the
168. 6:02model into Cloud Opus 4.8. At the time
169. 6:06that you're watching this, maybe there's
170. 6:08a new model out there. Who knows? But,
171. 6:10let's go with this one just for now. All
172. 6:12right. Now, before I move on, let's
173. 6:14first test to see if Claude can access
174. 6:16Jesse's MCP. So, I'm just going to ask
175. 6:19it, "Hey Claude, how's it going? Can you
176. 6:22access Jesse's MCP in order to create
177. 6:24strategies, run backtests, or things
178. 6:27like that?" There we go. So, you see,
179. 6:28first of all, it's happy that I said hi
180. 6:31to it, something really important that
181. 6:33you should always do. And next, we can
182. 6:35see that it has access to tools such as
183. 6:38creating or editing strategies,
184. 6:40backtesting, research and validation,
185. 6:42optimization, Monte Carlo, significance
186. 6:44testing, and candle data. So, this seems
187. 6:47brilliant. Now, while that's going, I
188. 6:49want to quickly remind you guys about
189. 6:50our Telegram. It's the fastest way to
190. 6:52get notified about my future work,
191. 6:54whether it's a new tutorial or a tool
192. 6:56that I create. Also, don't forget to
193. 6:58check out our free Discord, where more
194. 6:59than 5,000 members like you and I are
195. 7:01hanging out there and helping out each
196. 7:03other with algo trading, so we can all
197. 7:05succeed together. The links for both are
198. 7:07down in the description. All right. So,
199. 7:08I cleared everything, and now we are
200. 7:11ready to give the agent its prompt, its
201. 7:14task to perform. And this is going to be
202. 7:16a huge prompt, because I want it to do a
203. 7:18lot for me, and I don't want it to just
204. 7:21perform a quick task and just come back
205. 7:23and give me the results. So, I want this
206. 7:24to be fully automated, and I want it to
207. 7:27work for hours or even days. So, I've
208. 7:29prepared this prompt for you, so you
209. 7:31guys can also copy and paste it, and
210. 7:33feel free to edit it however you like.
211. 7:35All right. So, I'm going to read it very
212. 7:36quickly. I need it to act as a 24/7
213. 7:38quant agent who can perform research. At
214. 7:41the end, I am going to need 10 different
215. 7:43strategies. Each of them must have a
216. 7:45Sharpe ratio above 1.5 over the past 4
217. 7:48years. So, this is a very difficult
218. 7:50criteria for the agent, and in many of
219. 7:53my previous tests, the model wasn't able
220. 7:55to perform this well. So, let's see how
221. 7:57it does now. And this is just too
222. 7:59strict, guys. So, feel free to lower
223. 8:01this, especially because if it can reach
224. 8:04that number on one single strategy,
225. 8:06that's going to be very, very difficult.
226. 8:08All right, continuing. It doesn't matter
227. 8:10if you try hundreds of different
228. 8:12strategies or different versions of it.
229. 8:13So, this is because I don't want it to
230. 8:14just go and write like 50 strategies and
231. 8:17come back and complain that I couldn't
232. 8:19find it. So, I wanted to keep trying.
233. 8:22Next, when you run a Monte Carlo
234. 8:23simulation, I want the Sharpe ratio of
235. 8:25the original backtest to be positioned
236. 8:27in the middle between the median and the
237. 8:29best 5%. So, this is to ensure that the
238. 8:31result isn't overfit. So, this is my
239. 8:33criteria for how I pick it. Again, this
240. 8:35is too strict, guys. Like, usually, even
241. 8:37if it's below 5%, that means it's good
242. 8:40enough, but this time I'm saying it must
243. 8:42be in the middle between median and the
244. 8:45best 5%. By the way, guys, first when I
245. 8:47started recording this video, I asked
246. 8:49the agent to run multiple agents itself
247. 8:51to take advantage of this feature of
248. 8:54Cloud called workflows. But when I did
249. 8:56that, the model often times would just
250. 8:58run some stuff in the background, tell
251. 9:00me that it's waiting for the result, and
252. 9:02then it just silently died. And that was
253. 9:04just horrible. So, that's why I switched
254. 9:06into this version where I just ask one
255. 9:09agent to do everything. So, let's remove
256. 9:12this from before. Next, you do not have
257. 9:14to pick only one symbol for this. It is
258. 9:16okay to try different trading symbols.
259. 9:18So, I've given it a list of trading
260. 9:20symbols that I have already imported
261. 9:22candles for, but even if I hadn't, the
262. 9:24Jesse MCP is able to import these by
263. 9:27itself, so that's also pretty cool. It's
264. 9:29okay for you to find multiple strategies
265. 9:31on the same symbol. Do not feel
266. 9:32pressured to absolutely use all the
267. 9:34symbols. So, this is also something
268. 9:36else. Like, I want model two, for
269. 9:37example, to write multiple strategies
270. 9:39for BTC, let's say. So, it doesn't have
271. 9:42to feel like it needs to find all the
272. 9:4410,
273. 9:45with each of them being on different
274. 9:47symbols. Next, make sure to find me all
275. 9:50sorts of trading strategies, and not
276. 9:51just trend following. So, mean
277. 9:53reversion, or maybe even greater
278. 9:54strategies where we enter and exit at
279. 9:56multiple points. Those are all fine.
280. 9:58Notice that some of these symbols are
281. 10:00low-cap altcoins. Some of them are so
282. 10:03volatile or pouring quality that their
283. 10:05prices are more likely to go down. This
284. 10:07could be an opportunity to develop
285. 10:09short-only strategies. So, I don't want
286. 10:11the model to always think that the
287. 10:13strategy must go both long and short at
288. 10:16the same time. So, it's fine for some
289. 10:18altcoins, for example, to only short
290. 10:20them. So, that's also a valid scenario.
291. 10:23You can also come up with whichever time
292. 10:25frame that works best. However, I would
293. 10:26prefer it if it was either 15 minutes,
294. 10:2830 minutes, or hourly. Feel free to also
295. 10:31use an anchor time frame, such as the 4
296. 10:33hours, in your analysis, but not for
297. 10:35trading time frame. The most important
298. 10:37part is not to stop until you have found
299. 10:39all 10 strategies. In the meanwhile,
300. 10:41feel free to keep updating different
301. 10:44report files in markdown format, so that
302. 10:46I can easily monitor your progress. Keep
303. 10:48the reports in a new folder called
304. 10:50reports inside your project directory.
305. 10:52This may take hours or even days, and
306. 10:54that is perfectly fine with me. Do not
307. 10:56interrupt to ask for permission. I trust
308. 10:58your judgment. Only ping me if something
309. 11:00is wrong, and you absolutely need my
310. 11:02input. So, sorry, guys, if this was too
311. 11:03long. It's just that I needed to be as
312. 11:06specific as possible in the prompt, and
313. 11:09I think this is going to perform better.
314. 11:10So, let's go back to Cloud Code, and
315. 11:13paste this whole thing in, and hit
316. 11:16enter. So, let's go and come back a bit
317. 11:18later. All right. It's been like 24
318. 11:20hours, but it wasn't like what I
319. 11:22expected. It wasn't exactly an
320. 11:25uninterrupted 24 hours. The results are
321. 11:28here, and they are acceptable, but
322. 11:30again, they're not as good as I expected
323. 11:33them to be. So, let's go through it. So,
324. 11:35the first issue that I had was
325. 11:37apparently Cloud Code these days has
326. 11:39some sort of memory. And because as I
327. 11:41mentioned just earlier in my previous
328. 11:43prompt, which actually didn't work, I
329. 11:45was using sub-agents and that wasn't
330. 11:47working fine. So, when I asked the model
331. 11:50to do this this time without the
332. 11:52sub-agent, but you know, I didn't
333. 11:53specifically mention that, it started
334. 11:56using sub-agents again. So, I had to
335. 11:58actually mention it that "Hey, I do not
336. 12:01want you to use sub-agents anymore." So,
337. 12:03that's the first thing that I had to do.
338. 12:05By the way, I just changed the theme and
339. 12:07now everything seems better on the dark
340. 12:09mode. So, the first thing I needed to do
341. 12:12was to specifically ask it, "Hey, do not
342. 12:14use sub-agents for running things." So,
343. 12:15that's the first thing I had to do or,
344. 12:18you know, interruption that I faced and
345. 12:20I had to fix. Next, it kept working and
346. 12:23working, but it couldn't really find a
347. 12:26strategy that would have given it the
348. 12:28Sharpe ratio that I was asking for. So,
349. 12:30that was a real for the model, which to
350. 12:33be fair was expected. Like, I did not
351. 12:36expect this to be able to find something
352. 12:37really great in just a few minutes. So,
353. 12:40that's why I said, "Hey, keep working
354. 12:41for hours or days." But unfortunately,
355. 12:44the model kept complaining. So,
356. 12:46beginning with here, it said, "To be a
357. 12:48straight with you, based on the evidence
358. 12:49so far, tennis strategies each robustly
359. 12:52with a Sharpe ratio more than 1.5 looks
360. 12:54beyond what this single symbol can do."
361. 12:57So, it asked if it could just stop and
362. 13:00give me a lesser number. And this isn't
363. 13:03what I wanted. So, I said, "Keep
364. 13:04grinding for the 1.5." So, that's the
365. 13:06first time I said this. And then, just
366. 13:09like a while later, again it complained.
367. 13:11So, it said, "I want to be straight with
368. 13:13you. Across 12 optimizations and 45
369. 13:16backtests, the evidence suggests that a
370. 13:19robust Monte Carlo near median 1.5
371. 13:22Sharpe is unreachable on single symbol
372. 13:24in universe." So, I said that, "You know
373. 13:26what? The numbers you are telling me
374. 13:28like 12 optimizations and 45 back tests
375. 13:31are nothing for this type of work. You
376. 13:33should be thinking about hundreds of
377. 13:34back tests. Keep going." So, this again
378. 13:36is something I didn't expect Opus 4.8 to
379. 13:39do because they said this model has been
380. 13:41optimized to perform tasks for
381. 13:44long-running hours, and this was
382. 13:46literally nothing. So, this was really
383. 13:48weird for me because if you're not
384. 13:49reaching my token limits, why don't you
385. 13:52just keep working? Like, this is
386. 13:53supposed to be loop. Why is the model
387. 13:55stopping early? And it's not just the
388. 13:58only problem I've had with Opus lately.
389. 14:00I feel like it has become really dumb
390. 14:02ever since the Cloud Fable came out, and
391. 14:05of course they shut it down later, but
392. 14:07ever since I feel like Cloud could was
393. 14:09not giving me the same quality as
394. 14:10before. So, I just looked it up, and on
395. 14:13Reddit I found this, and apparently many
396. 14:15people are experiencing the exact same
397. 14:17thing. So, maybe this is the reason. So,
398. 14:19maybe in the near future they're going
399. 14:21to fix this. All right, so this is
400. 14:23something to remember, but even with all
401. 14:25this, I was still able to get what I
402. 14:28wanted out of this experiment. So, let's
403. 14:30continue. Again, it continued working
404. 14:32after I asked this, and finally at this
405. 14:35point it had a breakthrough. So, first
406. 14:36strategy over the gate. So, this one
407. 14:39using the Soul UCT on the dual anchor,
408. 14:44so 4 hours and daily, was able to
409. 14:46achieve a sharp ratio of 1.5 while still
410. 14:50it was passing my Monte Carlo filter in
411. 14:53order to ensure the result isn't
412. 14:54overfit. Then it kept working,
413. 14:57and at this point it was able to find me
414. 15:00five separate strategies that were all
415. 15:03seeing the criteria that I had set for
416. 15:06it. Now, this point I asked it to print
417. 15:08out more details because I wanted to see
418. 15:10them in a markdown file cuz for example,
419. 15:12it wasn't even telling me on this table
420. 15:15which symbols they were for or which
421. 15:17time frame, and that was a bit annoying.
422. 15:19I also asked it, "Please try finding
423. 15:21other stuff for other symbols. These are
424. 15:23all for sole. And then at this point it
425. 15:25was able to find 10 different strategies
426. 15:28with the criteria that I had set for it.
427. 15:31The sharp ratio of all of them as you
428. 15:32can see are at least 1.5. The max
429. 15:34drawdown of all of them is acceptable.
430. 15:37And even the number of trades that they
431. 15:39were taking was pretty good. So this
432. 15:41seems really good. And except this one
433. 15:42which is on the borderline, all of them
434. 15:44are also passing the Monte Carlo
435. 15:47criteria which means that they are
436. 15:49hopefully not overfit. Or you could say
437. 15:51they are less likely to be overfit than
438. 15:54any other thing that it could have found
439. 15:56until that point. Now, I got these
440. 15:58results today. So after my last prompt
441. 16:01when I told the model, "Hey, just keep
442. 16:03going. Try at least hundreds of back
443. 16:05tests and do not stop." Apparently that
444. 16:07one worked. So you could say that's a
445. 16:09really good prompt that I'm going to use
446. 16:11from now on. So just tell the model,
447. 16:14"Hey, I need you to perform at least
448. 16:15hundreds of back tests before stopping."
449. 16:18You could even increase this to like
450. 16:20thousands because if you do have a an
451. 16:22Anthropic coding subscription and if
452. 16:24your limits allow it, why not just let
453. 16:27the model do this? And the fact that I
454. 16:29was doing this on my Mac mini was also
455. 16:31pretty helpful because if I had to do
456. 16:33this on my own main machine, I would
457. 16:35have needed my resources. So I couldn't
458. 16:37just let it do keep doing research for
459. 16:39this long. All right. Now that
460. 16:41everything is ready, I can go to the
461. 16:44reports file that it generated for me.
462. 16:46So notice that I have opened that
463. 16:49servers files inside my local Z editor.
464. 16:53So it has this option called open
465. 16:55remote. And then you simply pass the SSH
466. 16:59address like the username and the IP for
467. 17:01it. Now in case you don't remember from
468. 17:03before, I was saying SSH Mac mini and I
469. 17:06said that if I were to show you the
470. 17:08command below it, it's going to be this.
471. 17:10So SSH soleh at sign the local IP of
472. 17:14that machine on my network. So notice
473. 17:16this local address here, 192.168.1.103.
474. 17:23This is important because Well, first of
475. 17:25all, you're going to need to copy this
476. 17:27and paste it to your Z editor in case
477. 17:30you're also using this, or whichever
478. 17:32editor really, such as Cursor. They all
479. 17:34have this option which allow you to SSH
480. 17:36into that remote machine and edit its
481. 17:38files just very easily. So, I've already
482. 17:41done that. I have access to all the
483. 17:42files on the machine. And specifically,
484. 17:44this is the project that I care for. So,
485. 17:46I named it Jim. And we have this reports
486. 17:49directory, right? So, this one, accepted
487. 17:51strategies, is the one that has all the
488. 17:53strategies that were meeting those
489. 17:55criteria. Now, this is in markdown
490. 17:57format, so by simply pressing the
491. 17:59command, shift, and V, I'm going to be
492. 18:00able to preview it in proper formatting.
493. 18:04But this one is a bit zoomed out, and it
494. 18:07makes it a little bit difficult to read.
495. 18:09Also, for some reason, I cannot click on
496. 18:11the links, probably because it doesn't
497. 18:12have the proper styling. But anyways,
498. 18:15so, my point is, let's actually go back.
499. 18:17All right, so, the point is I have all
500. 18:19the strategies that we found. It's
501. 18:22telling me which files they are in. So,
502. 18:23for example, I could just copy this one
503. 18:25and search for it. So, here I can see
504. 18:28the parameters that it has defined. And
505. 18:30I'm guessing these are the results of
506. 18:32the optimization that you ran. So,
507. 18:33instead of keeping the other parameters,
508. 18:35it tends to do this. And we can see how
509. 18:38it defined the ATR indicator, the EMA,
510. 18:41the Donchian channel. And here are the
511. 18:44entry rules of the strategy. So, it says
512. 18:46if the agree is one and closing price is
513. 18:49above the upper band of the Donchian
514. 18:50channel, go long. If it's the opposite,
515. 18:53go short. And here's the position
516. 18:55sizing. So, everything seems very neat
517. 18:58and perfectly fine. All right, so, let's
518. 19:01go back. Now, I can also see the result
519. 19:03of the Monte Carlo that it ran, the
520. 19:06backtesting. So, it says it's It had a
521. 19:08Sharpe of 1.5. So, So looks awesome. But
522. 19:10here's the thing. Suppose I want to take
523. 19:13a quick look at the details of the back
524. 19:16test or Monte Carlo. And you see it has
525. 19:18generated URLs to the dashboard of
526. 19:20Jesse. So this is really great and
527. 19:22helpful. But if I were to just copy
528. 19:24this, go to my Chrome browser and paste
529. 19:26it in, it's not going to work. Why?
530. 19:28Because this is using the local address,
531. 19:30which is 127.0.0.1,
532. 19:34which is for my local machine. But I
533. 19:37want the address for that remote Mac
534. 19:39Mini machine, right? And you remember it
535. 19:40from before that I said this is going to
536. 19:42be 192.168.1.103.
537. 19:49So if I replace it with that, now it's
538. 19:51going to work. So I'm going to enter my
539. 19:53password and there we go. Now I'm able
540. 19:55to actually see the result. So let's do
541. 19:57something. I'm going to copy this
542. 19:59address from here, go back to my code
543. 20:01editor,
544. 20:02open a new AI window, and I'm going to
545. 20:06attach this file, and I'm going to ask
546. 20:08it to replace that previous address with
547. 20:11this one. So replace 127.
548. 20:15Sorry, 001
549. 20:17with this in all this file. All right,
550. 20:20so it's going to do it for us. And while
551. 20:22this is going, I want you to also pay
552. 20:25attention to this. So suppose this is
553. 20:27something I always want to do. Now I
554. 20:30could simply edit the agents.md file of
555. 20:33this project to do this for me. So if I
556. 20:37do it here, then I would know that it
557. 20:40would always do this. So whenever it
558. 20:41wants to generate the report file for
559. 20:43me, it will use the address, so this one
560. 20:47for example, that would always work for
561. 20:49me no matter where I'm opening it from.
562. 20:51Even from the same machine with this
563. 20:53address is going to work. So this is
564. 20:54also a good tip to remember. All right,
565. 20:56so we already have this address for back
566. 20:58test. So let's do it one more time. All
567. 21:00right, so this is for Monte Carlo, and
568. 21:02we can see that the result of the
569. 21:04original back test is in the middle of
570. 21:06the simulations. This is a good sign and
571. 21:08the sharp ratio of the original backtest
572. 21:10is 1.30, which is even a little bit less
573. 21:14than the median number. So, this is
574. 21:15really good for whenever you want to see
575. 21:18if the strategy is overfit or not. So, I
576. 21:20like this result and going back to its
577. 21:23backtesting results. So, especially this
578. 21:25one, which is shows me the log scale
579. 21:27equity curve, we can see that it is very
580. 21:29nice and steady beating the benchmark,
581. 21:32which is solid UCTT chart and here we
582. 21:34can see the max drawdowns of the
583. 21:36strategy. So, we can see the five worst
584. 21:40periods. So, all of them seem
585. 21:41acceptable. Next, we can see the monthly
586. 21:44returns heat map chart. We can see it's
587. 21:46profitable almost every single year
588. 21:48except on 2024, where in five months in
589. 21:52a row it was losing. So, this is also
590. 21:55important to remember that the strategy
591. 21:57is not going to be profitable every
592. 21:58single month. And here we can see the
593. 22:01monthly return distribution of the
594. 22:02strategy. All in all, everything seems
595. 22:04good. Next, we can see the average
596. 22:07holding time of the strategy is around
597. 22:0822 hours. The minute is 34%. So, this
598. 22:12isn't a high win rate strategy, so
599. 22:14that's something that is very important
600. 22:16because if you are going to trade this,
601. 22:18approximately you're going to lose seven
602. 22:20out of every 10 trades. So, mentally
603. 22:22this is going to be very tough. So,
604. 22:24that's very important to remember and
605. 22:27the average trade per month is around
606. 22:2912. So, definitely not bad. Next, we
607. 22:31have the Skeltner Breakout strategy
608. 22:34here. So, let's copy it and search for
609. 22:36it. All right, so here's the code. Here
610. 22:38are the hyper parameters of the strategy
611. 22:41and here we can see the other metas that
612. 22:44it defined, the entry rules for long
613. 22:46positions, for short positions, the
614. 22:49position sizing for long positions. This
615. 22:51is how it submits the buy order, how it
616. 22:54defines the size of the position. So, it
617. 22:57sees risking a certain percentage for
618. 23:00each position and that certain
619. 23:02percentage is go here. So, it is 2%. So,
620. 23:06again pretty standard for adjusted
621. 23:09strategy. So, let's go and check the
622. 23:12back test results and also the Monte
623. 23:14Carlo results. All right, so again an
624. 23:16acceptable equity curve, which is
625. 23:18growing nice and steady. The max
626. 23:20drawdown of the strategy is very low,
627. 23:22which means we can easily add to the
628. 23:24size of the position. The minute is 42%,
629. 23:27so I actually like this one better. The
630. 23:29Sharpe ratio is 1.61, so this is
631. 23:32definitely better than the previous
632. 23:33strategy. Here's the worst five drawdown
633. 23:37periods and here's the monthly return,
634. 23:40so it is profitable every single year.
635. 23:42Even on 2024, this one isn't like losing
636. 23:45five months in a row, so this is really
637. 23:47good to see. And so, all in all, this
638. 23:50one seems like a better candidate for
639. 23:53me. So, let's check out the Monte Carlo.
640. 23:56So, the Sharpe ratio of it is 1.55.
641. 24:00The median is 1.33.
642. 24:02The best 5% is two. So, this one is
643. 24:04almost in the middle. So, again a pretty
644. 24:08solid strategy, I would say. I actually
645. 24:10very much like this one and I preferred
646. 24:13to the previous one. I think this one
647. 24:15could actually be good for me. So, I
648. 24:17could go through the other strategies as
649. 24:19well, but you get the idea because the
650. 24:21point was for you to learn how to set up
651. 24:25your own AI Quant to do research for
652. 24:28you. Now, I am going to submit at least
653. 24:31one of these strategies on our
654. 24:33strategies page on our website. So, if
655. 24:35you want to check it out or see the
656. 24:37results of the strategies on other
657. 24:39period symbols or time frames, feel free
658. 24:42to do so. So, there you go, guys. You
659. 24:43just saw both the strong points and the
660. 24:46weaknesses of this setup and how to fix
661. 24:48the weaknesses. The results weren't
662. 24:50exactly what I asked for initially, but
663. 24:52it was more than acceptable and I now
664. 24:55know exactly how to proceed in order to
665. 24:58reach my initial goal. Now, if you like
666. 25:00the video, please support me by liking
667. 25:02the video and posting a comment to let
668. 25:04me know what you think and subscribing
669. 25:06to the channel because I make videos
670. 25:08just like this one regularly and you
671. 25:10don't want to miss them. My name is
672. 25:11Saleh, guys. Thank you so much for
673. 25:13watching. I'll see you in the next one.