GLM 5.3 + MCP: Mean Reversion Crude Oil strategy with a 2.93 Sharpe — Transcript

Full transcript

1. 0:00ZAI just released GLM 5.3 and now we
2. 0:04have a new frontier coding model which
3. 0:06is open source. Now if you remember GLM
4. 0:085.2, even though it wasn't beating all
5. 0:11the other models out there, the result
6. 0:13of it was great. So because of that, I
7. 0:15actually do trust the benchmark results
8. 0:18that they are publishing here. And look
9. 0:19at this jump. In one of them it went
10. 0:21from 4.6 up to 28.3
11. 0:25and it is even beating Kimmy K3 which is
12. 0:28a great model. Now in some of them it's
13. 0:30not still Fable 5 or GPT 5.6 so, but it
14. 0:34is doing that in some others such as
15. 0:36this one and this one and in this one.
16. 0:38So again, this is really great guys. And
17. 0:41reading the numbers here, we can see in
18. 0:43most of them it is beating them which is
19. 0:45crazy that an open source model is doing
20. 0:48this. But you know how it is on this
21. 0:49channel. I don't just care about the
22. 0:51benchmark numbers. I actually want the
23. 0:52model to write me trading strategies in
24. 0:55Python and then run backtests on them,
25. 0:57rule testing, optimization and even
26. 0:59Monte Carlo so we can ensure the results
27. 1:01that we're getting aren't overfit. Now
28. 1:03we're going to need two setups. For the
29. 1:04Python side of things and also for it to
30. 1:08expose an MCP which we can use to hook
31. 1:10the agent into it, we're going to use
32. 1:12the Jesse framework which allows us to
33. 1:15write the code in Python and do
34. 1:16research. It also exposes an MCP which
35. 1:18we're going to use here. And next you're
36. 1:20going to need a GLM coding plan in order
37. 1:23to run the model and get the tokens. And
38. 1:26I subscribe to the light plan which
39. 1:29costs $18 per month and it's more than
40. 1:32enough for this experiment that I'm
41. 1:33doing. Now for the harness side of
42. 1:35things, I'm using open code which is a
43. 1:37great harness. It has support for both
44. 1:40terminal and they also have desktop app
45. 1:43at this point. So this is a great
46. 1:45harness and I highly recommend it. Now
47. 1:47in order to run Jesse itself, now that I
48. 1:49am inside my Jesse project, all I need
49. 1:51to do is to run the Jesse run command
50. 1:54and my conda environment is already
51. 1:57activated and this is the one that
52. 1:59actually has Jesse installed in it. So,
53. 2:01let's hit enter and once it is running,
54. 2:03we're going to get the address for the
55. 2:04MCP server. So, here it is. So, let's
56. 2:07copy it. And next, in order to run the
57. 2:09GLM model, I want to use the open code
58. 2:13harness. Now, the choices are actually
59. 2:15endless. We could have been using codex,
60. 2:18cloud code, we could have used it
61. 2:20directly inside as an agent like this or
62. 2:23many other choices that are out there
63. 2:25these days. However, I have found open
64. 2:27code to be one of the best ones,
65. 2:28especially with local models. In this
66. 2:31case, we're not using a local model, but
67. 2:32still I think it's going to be a really
68. 2:34good choice. So, let's run open code.
69. 2:37Now, if I hit enter, it's going to run
70. 2:39open code, but first I need to add the
71. 2:41Jesse MCP to it. So, I'm going to say
72. 2:43open code MCP add. Let's hit enter.
73. 2:47Next, it is asking whether we want to
74. 2:49add it just for the current project or
75. 2:51globally. And in this case, because this
76. 2:53is my Jesse project, I actually want it
77. 2:55to be here. So, let's hit enter. Let's
78. 2:57give it a name. So, we're going to call
79. 2:59it Jesse and then select MCP server
80. 3:01type. We're going to choose remote and
81. 3:04then I'm going to paste in the address
82. 3:06that we just copied earlier. There we
83. 3:08go. Does the server require or auth? No,
84. 3:10it does not. And that's it. So, now if I
85. 3:13run the open code command, I should have
86. 3:15access to that MCP server now. All
87. 3:17right. So, you can see that I've already
88. 3:19selected GLM 5.3
89. 3:21where the ZAI provider. Now, adding it
90. 3:24is actually surprisingly easy. So, you
91. 3:26just write this /models and then in
92. 3:30order to connect the provider, you press
93. 3:32control and A and then you pick ZAI. It
94. 3:35is super easy. So, let's just skip this.
95. 3:38All right. So, let's check to ensure we
96. 3:40have Jesse MCP. So, let's see. All
97. 3:43right. So, among my MCPs, I can see
98. 3:45Jesse is already connected. So, that's
99. 3:47great. Let's also ask it if it has
100. 3:49access to it or not. Hey there, do you
101. 3:51have access to GSM CP? All right, so it
102. 3:53did not write this part correctly. All
103. 3:55right, let's see what it says. By the
104. 3:57way, the UI of Open Code is absolutely
105. 4:00gorgeous. All right, so it says, "Yes, I
106. 4:02do. The GSM CP tools are all connected."
107. 4:05So, this is awesome and now we can give
108. 4:07it the actual prompt. Hey there,
109. 4:09research and find a mean reversion
110. 4:11strategy for trading crude oil on
111. 4:14Binance Futures, which has a symbol of
112. 4:16CL-USDT.
113. 4:18I need the results to be good from the
114. 4:20beginning of May 2026 until the end of
115. 4:23July. By good, I mean a Sharpe ratio of
116. 4:26at least 1.5. Risk 3% of the account on
117. 4:29each trade. Every time you develop a
118. 4:31strategy, validate the results using a
119. 4:33statistical significance test before
120. 4:35writing the full strategy. Only proceed
121. 4:37if the strategy's metrics demonstrate
122. 4:39genuine statistical significance. Use
123. 4:41optimization to improve the results. At
124. 4:43the end, apply Monte Carlo simulations
125. 4:45to ensure the results are not overfit.
126. 4:47Continue until you find a strategies
127. 4:49that fully meet the pre-specified
128. 4:51criteria. Use the 15-minute or 30-minute
129. 4:53time frame for trading side and hourly
130. 4:56as the anchor time frame. Do not prompt
131. 4:58me in the meanwhile. Good luck. Now,
132. 5:00while that's going, I want to quickly
133. 5:01remind you guys about our Telegram. It's
134. 5:03the fastest way to get notified about my
135. 5:05future work, whether it's a new tutorial
136. 5:08or a tool that I create. Also, don't
137. 5:09forget to check out our free Discord
138. 5:11where more than 5,000 members like you
139. 5:13and I are hanging out there and helping
140. 5:15out each other with algo trading so we
141. 5:16can all succeed together. The links for
142. 5:18both are down in the description. All
143. 5:20right, so it started working and the
144. 5:23speed is also pretty fine. It defined a
145. 5:25plan all by itself so it would know how
146. 5:28to proceed. And it also it started
147. 5:30writing the first version of the
148. 5:32strategy. Now, one thing I really like
149. 5:33about this release of GLM is that it is
150. 5:36actually pretty fast. I find it to be
151. 5:38significantly faster than the Chinchilla
152. 5:41model, which was released just a while
153. 5:43ago and it had fantastic performance,
154. 5:45but it wasn't exactly the fastest model.
155. 5:47The GLM model used to be so slow in the
156. 5:50past that after trying, I believe GLM 5,
157. 5:53I completely gave up and never gave it
158. 5:55another shot again. But now I'm happily
159. 5:58surprised. So, while this is going, I
160. 6:00want to explain something, guys. So, you
161. 6:02notice in the prompt that we gave the
162. 6:04model, I said that I want the result to
163. 6:08be good since the beginning of the May
164. 6:112026 until the end of July. So, that is
165. 6:14approximately 3 months. And 3 months
166. 6:17isn't actually enough for testing a
167. 6:20strategy. So, I never do this. But
168. 6:22there's a reason why I'm doing this. So,
169. 6:24let me show you the chart of crude oil
170. 6:26on Binance Futures. So, here it is. And
171. 6:29if you take a look, you can see that the
172. 6:32earliest date is for the beginning of
173. 6:34April. So, that means this is all the
174. 6:37data that we have. And if I even check
175. 6:39out other exchanges, you will see that
176. 6:41the same goes there. So, there's a real
177. 6:43problem here. And that is because all of
178. 6:45these real-world assets were added to
179. 6:48crypto exchanges approximately the same
180. 6:51date. And for some reason, all the
181. 6:52exchanges added at the same time. I
182. 6:54don't know why. But the point is
183. 6:56currently, we do not have access to more
184. 6:58data. At least not yet. Because you see,
185. 7:01if you check out the roadmap page of
186. 7:03Jesse's website, you can see that we
187. 7:05have this item here, which has a pretty
188. 7:08good amount of votes. And it is
189. 7:10specifically says adding historical
190. 7:12stock, forex, and commodity data
191. 7:14integrations. And when I say that, that
192. 7:16means we're going to add third-party
193. 7:18market data providers. So, Jesse can
194. 7:20access years of high-quality historical
195. 7:22data for stock, forex, and commodity.
196. 7:24So, once we do this, we're going to have
197. 7:26years of data to actually run useful
198. 7:29back tests. Now, you might be asking,
199. 7:31"Okay, so then what's the point of doing
200. 7:32this in the first place?" Well, the
201. 7:34thing is because of the fundamental news
202. 7:36that is going on out there, I find
203. 7:38myself wanting to trade on exchanges
204. 7:41such as lightyear, which offer contracts
205. 7:44for crude oil. And whenever I trade
206. 7:46manually, I always end up getting
207. 7:49burned. So, just so that I won't be able
208. 7:51to do this, I wanted to develop a
209. 7:53strategy just so at least I could trade
210. 7:55it with a bot because this way I know
211. 7:58that my position sizing is going to be
212. 8:00calculated, and that emotions are not
213. 8:02going to be a part of my trading. So,
214. 8:04that's why I decided to develop this
215. 8:05strategy, and I want it to be mean
216. 8:07reversion because I believe right now
217. 8:09we're going to be in a range for crude
218. 8:11oil because of the news that's going on.
219. 8:13Like, whenever the price goes up, Trump
220. 8:16says something just to bring down the
221. 8:18price. And whenever it comes down so
222. 8:20much, then like Iran does something, and
223. 8:23then the price goes up. So, the point is
224. 8:25I believe we're in a range. I don't
225. 8:26think the price is going to go up maybe
226. 8:30until at least there is some sort of big
227. 8:32news in the Middle East. And I don't
228. 8:34think it's going to come down that much,
229. 8:36either. So, I think this we're going to
230. 8:37be stuck in this situation for a few
231. 8:39months, at least. So, I wanted to take
232. 8:41advantage of this situation in my
233. 8:44trading results. The results are here,
234. 8:46and it did find what I was looking for.
235. 8:49So, we got a Sharpe ratio of 2.93 for
236. 8:51the backtest, which is significantly
237. 8:53better than what I asked for, which was
238. 8:55a Sharpe ratio of 1.5. But before I show
239. 8:58you the results, let me show you how
240. 8:59much token we spent. So, this is the
241. 9:01ending result. I got this 20 bucks
242. 9:04coding plan just for this video, and I
243. 9:07spent almost half my weekly tokens on
244. 9:10it. And it did a lot of research with
245. 9:12this. So, I want to say that this plan
246. 9:15is definitely worth it because the whole
247. 9:17step I got was indeed frontier, and the
248. 9:20fact that you can run this many tokens
249. 9:22with a $20 account is just crazy. I
250. 9:25mean, compare this to Cloud Code, and
251. 9:27you'll see what I mean. Because we got a
252. 9:29quality that is as good as Cloud Fable,
253. 9:31but we paid $20 for it, which is not
254. 9:34even available with the $20 Cloud
255. 9:36account. So, anyways, let's go through
256. 9:38the result. Now, here's what it printed
257. 9:40out for me and I couldn't click on them.
258. 9:43It was a bit annoying, so I just asked
259. 9:45it to print them one at a line. So,
260. 9:48let's click on this. So, this here's the
261. 9:49back test, two Monte Carlo and one
262. 9:51significance test. So, let's click on
263. 9:54them. Now, before I show you the result,
264. 9:56there was one weird thing going on and
265. 9:59that is it printed out two different
266. 10:01Monte Carlo results. So, I asked it
267. 10:02what's going on and which one is the one
268. 10:05that you had before and it said neither.
269. 10:07So, in fact, this is the one that we
270. 10:09usually had. So, that is the one week
271. 10:12block. So, let's also click on this. All
272. 10:15right. So, let's begin with the rule
273. 10:17significance test. As you can see, the
274. 10:19return of the strategy is beating the
275. 10:21simulations by huge margin and that is a
276. 10:24good sign. And the P value is 0.01,
277. 10:28which means the entry rules of the
278. 10:29strategy have statistical significance.
279. 10:32So, that's good. The annualized return
280. 10:34is 131.
281. 10:36So, this definitely has potential. So,
282. 10:39this test is passing. Let's take a look
283. 10:41at the back test. Here's the equity
284. 10:42curve. So, we can clearly see it's going
285. 10:44up while the price was actually going
286. 10:47down for most of the time. Let's look at
287. 10:49the drawdown chart. So, this is how it
288. 10:52looks like. Basically, when the price
289. 10:54was going down, the strategy did really
290. 10:56great. When it was going up again, at
291. 10:59first it lost and then it started
292. 11:01earning again. So, you should kind of
293. 11:03say that when there is a pivot point in
294. 11:05the price, the strategy doesn't do well
295. 11:08and here is the monthly return. So, we
296. 11:10had two positive months following by one
297. 11:12negative month, but the negative wasn't
298. 11:14big. So, looks great so far. The profit
299. 11:17is 16%. The max drawdown is minus 5.59%,
300. 11:22which is really low, which means we can
301. 11:24easily add to the size of the position
302. 11:26to increase this P&L number here. The
303. 11:28average win to loss ratio is 0.73.
304. 11:32The win rate is 70%. So, this is pretty
305. 11:35typical for a mean reversion strategy.
306. 11:37The sharp ratio is 2.93, which is not
307. 11:40It's awesome. The average trades per
308. 11:43month is 11. Per week, it's two. And per
309. 11:47day, it is 0.37. All right. So, overall,
310. 11:49I'm seeing things that I really like.
311. 11:51But, take this with a grain of salt
312. 11:53because we're just running the back test
313. 11:55on 3 months of data. And that just isn't
314. 11:59a good amount for a back test to have
315. 12:01real significance. But, moving on with
316. 12:04Monte Carlo, these are the results that
317. 12:06we can see. And you see, it starts to
318. 12:08get better. So, this is the one that we
319. 12:11usually run, which is with the one-week
320. 12:13block. And looking at the last one, this
321. 12:16is the one with three-week block. And
322. 12:18with this one, it is definitely passing.
323. 12:20Like, this equity curve looks really
324. 12:22great. It's in the middle of
325. 12:24simulations. And the sharp ratio of the
326. 12:26original back test is 2.72,
327. 12:28while the best 5% is 4.15. The median is
328. 12:320.75.
329. 12:34So, this value is almost in the middle,
330. 12:36which is really great and what I like to
331. 12:37see. And even the median itself is
332. 12:39positive. So, these are all good signs.
333. 12:42There is a catch here. And that is the
334. 12:44model did something weird. Now, before I
335. 12:46show it to you and explain it, I want to
336. 12:48say that I'm not disappointed in the
337. 12:50model because the bar that I set for the
338. 12:52model was pretty high. So, I asked for a
339. 12:55mean reversion strategy, which is really
340. 12:57difficult to pull off, especially when
341. 12:59the amount of data that we have is
342. 13:00limited. But, the catch is this. In
343. 13:02order to meet the criteria that I set
344. 13:05for it for Monte Carlo test, it tried to
345. 13:08change the configurations of it. And by
346. 13:10the way, let me show it to you what I
347. 13:12mean by that. So, here's a form for a
348. 13:14Monte Carlo on Jesse's dashboard. And
349. 13:16right down here, you can see we have
350. 13:19this configuration for the pipeline
351. 13:21type, and it is set to moving block
352. 13:23bootstrap. And the batch size is 17 days
353. 13:26by default. Now, this is actually not
354. 13:29the default value. It's what the model
355. 13:31changed. So, the default value is 7
356. 13:33days, and that's my point. That in order
357. 13:36for the result to meet the criteria that
358. 13:38I set for the model, it changed the
359. 13:40parameter of the Monte Carlo test
360. 13:43itself. So, this is kind of like running
361. 13:46too much optimization for a strategy
362. 13:48until you get good results. And the
363. 13:50point of running Monte Carlo is to
364. 13:53prevent that because we don't want the
365. 13:56strategy to be overfit, right? In order
366. 13:58to do that, you have to pick a set of
367. 13:59parameters and just stick with it. So,
368. 14:01if you were to just keep changing
369. 14:03everything, including the Monte Carlo
370. 14:05itself, until we get the results that we
371. 14:07want to see, it's not going to play out
372. 14:09right. So, that's what I did, and when I
373. 14:11did that, we got this result that looks
374. 14:13good. But before doing it, this was the
375. 14:16actual result that we were getting. Now,
376. 14:18it's not horrible, okay? So, it's still
377. 14:20the Sharpe ratio of the original
378. 14:21backtest is below the best 5%, but what
379. 14:25is alarming to me is that not only this
380. 14:28number has a lot of difference until the
381. 14:30median number, but also the fact that
382. 14:32the median itself is negative. So, it's
383. 14:34not even positive. So, if this was the
384. 14:37test that we were judging the strategy
385. 14:39based on, which it should be, then I
386. 14:41would say that it is not passing the
387. 14:43Monte Carlo. So, the model kind of did
388. 14:45something tricky, and I'm not sure if
389. 14:47it's cool, but on the other hand, I did
390. 14:49not ask it inside the prompt to not do
391. 14:51this. I could have said, "Hey, stick to
392. 14:53the inputs of the the default Monte
393. 14:56Carlo values." And also, again, the bars
394. 14:58we set for the model were pretty high,
395. 15:00like higher than what I gave to other
396. 15:02models. So, I'm not disappointed in it.
397. 15:04Now, to be fair, all I was expecting
398. 15:06from this strategy was a replacement for
399. 15:09my manual trading, right? I said it
400. 15:11before starting this whole process that
401. 15:143 months isn't enough, and that we're
402. 15:16going to come back to this once we have
403. 15:18more historical data, which is exactly
404. 15:20what this item of Jesse's road map is
405. 15:23promising to do. But, before we move on,
406. 15:26let's actually take a look at the code
407. 15:28that the agent wrote, and also the
408. 15:30trades that it took. So, let me ask it
409. 15:32to add charts to the strategy and run
410. 15:37the backtest again. So, once it does
411. 15:40that, I will be able to show you the
412. 15:42exact trades that the strategy took.
413. 15:44Now, we can do it right now, but you see
414. 15:47this is a naked chart, so we don't have
415. 15:50any indicators on it, and it's not super
416. 15:52helpful because we won't be able to know
417. 15:54why exactly the model to decide, for
418. 15:57example, to go long here. So, we would
419. 16:00have no idea here. So, that's one
420. 16:02problem, and let's see if we can fix it.
421. 16:06Now, while this is going, let's go back
422. 16:08and take a look at the strategies code.
423. 16:10So, here, if I click on the strategy
424. 16:12snapshot, we can see the exact code that
425. 16:14Jesse loaded in order to execute the
426. 16:17backtest, which by the way, once I
427. 16:18upload the code of the strategy on
428. 16:20Jesse's strategy page, which is where
429. 16:23you can see the results of the backtest
430. 16:25for other trading periods, symbols, and
431. 16:28time frames. So, once I do that, you
432. 16:31will be able to download the code of the
433. 16:32strategy, and this should be done by the
434. 16:34time you're watching this video. So,
435. 16:35when you do, remember that the prompt
436. 16:39that I wrote for this is going to be
437. 16:42included in the comment section. So,
438. 16:44remember that. Anyways, moving on. So,
439. 16:46it defined the hyper parameters here
440. 16:48right at the beginning, and we can see
441. 16:50defined the BB period, which stands for
442. 16:52Bollinger Bands period, and then the
443. 16:54Bollinger Band deviation, the ATR
444. 16:56multiplier. So, these are the values
445. 16:58that it is using. So, it did not
446. 17:00differentiate between the upper band
447. 17:02deviation and the lower band deviation,
448. 17:04and we just have one ATR multiplier and
449. 17:07the period. Now, I generally say the
450. 17:09less number of hyper parameters for the
451. 17:12strategy, that is actually better
452. 17:14because the strategy is going to be less
453. 17:16likely to be overfit. So, that's a good
454. 17:19sign. Here's how I define the Bollinger
455. 17:21Bands. So, simply return
456. 17:23ta.bollingerbands
457. 17:25and then it's passing the current
458. 17:26candles. The period is defined like this
459. 17:29in order to pass the period number. And
460. 17:31then for both the upper deviation and
461. 17:34lower deviation, it is using this value.
462. 17:36And the default values are the ones that
463. 17:38are actually loaded. So, here we have
464. 17:41the period set to 40, which if I'm not
465. 17:43mistaken by default it's is 20 for the
466. 17:46Bollinger Bands. And the deviation is
467. 17:482.45, which by default I think is two.
468. 17:51And the ATR multiplier. Now, let's see
469. 17:54how it used this. All right, it's been
470. 17:56used here for setting the stop price.
471. 17:58So, we're saying the stop price is going
472. 18:00to be the current price minus the anchor
473. 18:03ATR, so the ATR basically on the anchor
474. 18:07time frame candle, which in this case
475. 18:08was the hourly, multiplied by a certain
476. 18:11ATR multiplier. So, pretty textbook.
477. 18:15Then we used the risk to quality utility
478. 18:18function of JC to define that 3% risk
479. 18:21per each trade. So, we're passing the
480. 18:23current available margin, the number
481. 18:25three for 3%, the current price, which
482. 18:28is going to be the entry price of the
483. 18:29strategy, the stop price, and then we
484. 18:32set the fees to what it is on that
485. 18:34exchange. So, that said, then it is
486. 18:35submitting the buy order, the stop loss,
487. 18:38and take profit. And the take profit, by
488. 18:39the way, the price of it is going to be
489. 18:41the middle band. Now, I'm going to show
490. 18:42you this in a second once we can take a
491. 18:45look at the interactive charts. But
492. 18:47anyways, moving on, the entry rule of
493. 18:49the strategy is simply whenever the
494. 18:51current closing price, or the current
495. 18:53price, it's basically the same thing,
496. 18:55whenever it goes below the lower band of
497. 18:57the current Bollinger Band indicator, we
498. 19:00want to go long. That's it. And whenever
499. 19:02it goes above the upper band, we want to
500. 19:04go short. And this is typical in a mean
501. 19:06reversion strategy, right? Whenever it
502. 19:09goes too far from the mean, we want to
503. 19:11bet that the price is going to go back
504. 19:14to the mean. So, that's the whole point
505. 19:15of a mean reversion strategy. If this
506. 19:18was trend following, we would have
507. 19:19wanted to do the opposite. Now, for a
508. 19:21short position, it did exactly the
509. 19:23opposite of a long position, and that's
510. 19:27basically it. We're also saying that
511. 19:28whenever a candle closes, which in Jesse
512. 19:30we use the update position built-in
513. 19:33method, we're saying that if it is a
514. 19:36long position and the current closing
515. 19:38price is above the middle band, you want
516. 19:40to liquidate the position. Now, this
517. 19:41wasn't really needed because we already
518. 19:44submitted the take profit here. So, once
519. 19:46the price reaches this, it's going to
520. 19:48get closed automatically. So, this block
521. 19:50could have been removed. We don't really
522. 19:51need it. And so is the case for this
523. 19:54one. So, the should cancel entry method
524. 19:56in Jesse basically says, do we need to
525. 19:59cancel the current entry orders and try
526. 20:01again? But because inside the go long
527. 20:04and go short, we're using the current
528. 20:06price for opening the position,
529. 20:08basically we're using a market order
530. 20:10here, then this is unnecessary, and this
531. 20:12also could have been removed. Anyway, so
532. 20:14this strategy is overall pretty simple,
533. 20:16which is actually a good thing because
534. 20:19the simpler the strategy is, usually it
535. 20:21plays out even better. So, anyways,
536. 20:23let's move on. Here's the new link for
537. 20:26the strategy. So, let's open the trade
538. 20:28chart, and as you can see, now we have
539. 20:31actual indicator values. Now, this one
540. 20:33is the ATR on the anchor time frame,
541. 20:35which is the hourly. We don't really
542. 20:37need it right now because it's just used
543. 20:39for setting the stop price. So, I'm
544. 20:40going to toggle this. Now, let's take a
545. 20:42look at some of the trades. Starting
546. 20:44with the losing one, actually. So, let's
547. 20:46scroll down. So, let's zoom in a little
548. 20:48bit. All right, so here it is. Here's
549. 20:51where it's went long, and this is where
550. 20:54it closed it. So, this was a buy order
551. 20:57and a sell order. And it should have
552. 20:59been a buy order, so it is correct
553. 21:01because the price closed below the lower
554. 21:04band of the Bollinger band, right? So,
555. 21:06this was supposed to be a buy, but the
556. 21:08price didn't go back to the mean, it
557. 21:10went down. So, that's why it's a loss.
558. 21:12So, moving on with this one, we can see
559. 21:14again the price closed below the lower
560. 21:17band, so it went long. And here, the
561. 21:19price went back to the middle band, so
562. 21:21that's why it took a profit here.
563. 21:23Looking at another one, all right, let's
564. 21:25zoom in. So, again, it closed below the
565. 21:28lower band, and then it went back to the
566. 21:30mean. Now, usually in my strategies, I
567. 21:33would set some sort of filter to not
568. 21:36take these trades, because this is
569. 21:38basically a squeeze moment of the
570. 21:40Bollinger Band. So, the price is not in
571. 21:43any kind of range anymore, and I usually
572. 21:46want to avoid taking trades here. So,
573. 21:48this isn't really a good sign, because
574. 21:49even considering the trading fees, it
575. 21:51may not even be worth it. But anyways,
576. 21:53let's move on to one of the big ones,
577. 21:55actually. So, let's sort this based on
578. 21:58best first. So, let's take a look at
579. 22:00this. All right, we already did. Let's
580. 22:02take another one. All right, so here,
581. 22:03the price actually closed above the
582. 22:06upper band, so we went short, and then
583. 22:08it reached the mean. So, again, another
584. 22:11case of a mean reversion strategy. All
585. 22:13right, so this is why these indicator
586. 22:15values are pretty useful, because if it
587. 22:18wasn't behaving exactly the way I
588. 22:19wanted, I could have just identified and
589. 22:22fixed it. Or, in this case, for example,
590. 22:24that I said because this is a squeeze
591. 22:25moment, I don't want to take trades
592. 22:27during this. So, this is something I
593. 22:29want to avoid, and it can simply be done
594. 22:32by using um an indicator such as the
595. 22:35Bollinger Band width, which basically
596. 22:37tells us this exact thing, that you
597. 22:39know, the Bollinger Band is not too big
598. 22:41anymore, so it's pretty tight. So, we
599. 22:44could use that as a filter. And because
600. 22:47I am seeing this visually, it is very
601. 22:49easy for me to identify. But if I
602. 22:51didn't, if I just read the numbers for
603. 22:53the trades, I could not have done this.
604. 22:55Before I leave you guys, I want to
605. 22:56mention two things. First of all, if you
606. 22:58follow me on my profile on Jesse's
607. 23:00website, you will be notified whenever I
608. 23:03submit a strategies like this one. So,
609. 23:05that's pretty helpful. And second, if
610. 23:07you haven't already, make sure to open
611. 23:09the road map page of Jesse, which is
612. 23:11simply at jesse.trade/roadmap,
613. 23:13and vote on these features. So, let me
614. 23:15know what you guys want me to build
615. 23:17next. If you enjoyed the video, please
616. 23:19give us a like and let me know what you
617. 23:21think in the comment section. And don't
618. 23:22forget to subscribe to the channel if
619. 23:24you haven't already, because I publish
620. 23:26videos just like this one all the time.
621. 23:28Thank you so much for watching. I'll see
622. 23:29you in the next one.