Jesse 2.0 Is Here! Algo-trading reimagined — Transcript

Full transcript

1. 0:00Hey guys, Sal. Today, I couldn't be
2. 0:01happier to announce the release of Jesse
3. 0:032.0. This is one of the biggest releases
4. 0:06we've ever had, so I couldn't be more
5. 0:07excited. And in this video, I'm going to
6. 0:09show you the highlights of it. So, let's
7. 0:11get right into it.
8. 0:12Now, right out of the bag, you're going
9. 0:14to see that the entire dashboard has
10. 0:16been redesigned, and we no longer have a
11. 0:18navbar here. Instead, we have a sidebar,
12. 0:20which you can also minimize. Now, I
13. 0:22actually like the minimized version
14. 0:24better because it saves me some space so
15. 0:26that I can see charts and and things
16. 0:28like that better. Next, you can see that
17. 0:30the home page have been redesigned, and
18. 0:32depending on whether you are a new user
19. 0:35or someone who has already run some
20. 0:37sessions, you're going to see different
21. 0:38things. But for me, as you can see, I
22. 0:40can see some of the recent backtest
23. 0:42sessions or any other sessions that I've
24. 0:45run, and I'm getting some useful stats.
25. 0:47But this part is specifically is the one
26. 0:49that I I would like to show you, which
27. 0:51is it shows us how much of the CPU and
28. 0:54RAM we are using. Now, this may not look
29. 0:56like something that you would actually
30. 0:57need. You would probably think, "Okay,
31. 0:58we already had apps with Well, yes, you
32. 1:00did, but if you were running Jesse on a
33. 1:03production server, then keeping an eye
34. 1:06on the resources of the your server was
35. 1:10going to be really important, right? So,
36. 1:11you want to make sure that you never run
37. 1:13out of resources, and this is going to
38. 1:16make it easy for you to keep an eye on
39. 1:18it. Next, we have new backtest metrics.
40. 1:22So, we have the max underwater period,
41. 1:25we have separate win rates for shorts
42. 1:27and long trades, but as you can see, we
43. 1:29also have new charts. So, before, we
44. 1:31only had the equity curve chart here,
45. 1:34but now, not only we have that, we also
46. 1:36have it in the log scale version. We
47. 1:38also have the max drawdown chart, which
48. 1:41also shows us the worst five periods,
49. 1:43and it will also show us like how much
50. 1:45each of them were losing, and we can
51. 1:47also click on the chart to see it in
52. 1:49bigger sizes. And as you can see, like
53. 1:51my worst max drawdown period for this
54. 1:53strategy was minus 17%, which was this
55. 1:56period. And if you also read this max
56. 2:00underwater period, you can also see that
57. 2:02it was 50 days. So, this period right
58. 2:05here was 50 days, which basically tells
59. 2:07me that if I were to run this strategy
60. 2:10for 50 days, I was going to have a
61. 2:12really hard time until I was able to
62. 2:14basically make back whatever I lost in
63. 2:17this period. Next, we have the
64. 2:18underwater plot, and then the monthly
65. 2:21returns heat map, which is also super
66. 2:23helpful, and the monthly returns
67. 2:25distribution chart, and trade panel
68. 2:27distribution chart. So, some of these
69. 2:29are completely new, and some of them
70. 2:31you've probably seen in other libraries
71. 2:32such as the QuantStats, which at some
72. 2:35point we were actually using inside
73. 2:37Jesse itself, but then we had to
74. 2:38discontinue that. And now, I'm happy to
75. 2:41say that we have access to these charts
76. 2:44again, but this time they are all built
77. 2:46in in the house. Also, I've made some
78. 2:48changes to the layout, so now the
79. 2:49performance metrics is in here in the
80. 2:52sidebar and not here, which actually
81. 2:54makes it really easier to read. Next,
82. 2:57and probably the biggest change in this
83. 2:59release is the rule significance test,
84. 3:01which is this statistical test that
85. 3:04which will tell you whether or not the
86. 3:06entry rule of the your strategy has any
87. 3:09predictive power or not. Now, as you can
88. 3:12see here, we have the distribution chart
89. 3:14and this line here, and the conclusion
90. 3:17of it is that this my strategy actually
91. 3:20has a very significant entry rule.
92. 3:23And we also have some And we also have
93. 3:26some raw data, which can help you
94. 3:28further, such as the annual return of
95. 3:30the entry rule, the observed mean, and
96. 3:32the number of observations and the
97. 3:34simulations that we run for this. Now,
98. 3:36to show you an example to understand
99. 3:37what this does, well, let me show you
100. 3:39some actual code, and for it, we're
101. 3:41going to run the rule significance test
102. 3:44inside the terminal because I also added
103. 3:47to the research module of Jesse. And in
104. 3:49fact, just so you know, the one that I
105. 3:50added to the research module is
106. 3:52completely free for everybody, but the
107. 3:54one that's is accessible via the
108. 3:56dashboard is for premium users only. But
109. 3:59in the usage, they all do the same. So,
110. 4:01let's just check it out. Now, here, we
111. 4:03have a strategy that if I were to
112. 4:05actually run it, I would have made
113. 4:06money. So, you can see the entry rule of
114. 4:08the strategy is 7.4%.
115. 4:12Now, by the way, this is just for the
116. 4:13entry rule of the strategy, okay? So,
117. 4:15it's not the entire strategy. Now, if
118. 4:17you don't fully understand what I'm
119. 4:18talking about, it doesn't matter because
120. 4:19I am going to record a full video just
121. 4:22about this topic because it definitely
122. 4:24deserves its own separate video because
123. 4:27I spent months on this topic, and I had
124. 4:30to read the whole book just to
125. 4:31understand it. But for now, in this
126. 4:32video, I just want to show you the
127. 4:33highlight of it. So, here, we have a
128. 4:35strategy that is making money. But the
129. 4:37reason that it is making money is
130. 4:39because it was trading purely a huge
131. 4:41uptrend. And in fact, if I show you the
132. 4:43code for the strategy that I am running
133. 4:46here, all that it is doing
134. 4:49is that it's trying to open a long
135. 4:51position with 80% certainty. And in
136. 4:54other times, it either doesn't open a
137. 4:56position or it opens a short position.
138. 4:58However, because I was running this test
139. 5:02during a huge uptrend in the Bitcoin
140. 5:04market, then it is simply making money.
141. 5:07But as you can guess, such a strategy
142. 5:09does not have any predictive power,
143. 5:12right? We're just being lucky here. If I
144. 5:14were to execute this strategy during a
145. 5:17bear market, we were going to get a
146. 5:18slaughtered without any doubt. And as a
147. 5:20result, if I run this strategy, the
148. 5:23result of the test is that it says two
149. 5:25random signals have no edge, and that my
150. 5:28entry rule is not significant because
151. 5:30the P value is above, well, uh this
152. 5:33value. Although, in the final results,
153. 5:35I've actually said that I want a P value
154. 5:37below 10%, okay?
155. 5:40So, even above 5%, it might have some
156. 5:42significance, but in here, you can see
157. 5:45it's actually even slightly above 10%,
158. 5:48so it definitely has no edge. And if I
159. 5:50show you the distribution chart, you can
160. 5:52see the returns of the strategy is not
161. 5:56by far more than the random ones, so
162. 5:58we're going to conclude that it does not
163. 6:00have any actual edge. Here, I have
164. 6:01another strategy, which actually does
165. 6:04have some edge. So, it has some actual
166. 6:06entry rule that is checking the data and
167. 6:08making decision based on that, so it
168. 6:10does have some predictive power. And if
169. 6:12I run the same test on the same period
170. 6:14for this one, I'm going to get this
171. 6:15result, which basically is below 10%, so
172. 6:18we're going to conclude that it does
173. 6:19have some significance. Now, by the way,
174. 6:21I I wrote this part before, so
175. 6:24I need to update it, so please ignore
176. 6:26this part that it says it's not
177. 6:27significant. So, the important thing is
178. 6:29that the P value is low. Also, one other
179. 6:31reason that I'm getting this result and
180. 6:33and not something like, "Yeah, your
181. 6:34entry rule is super significant." That's
182. 6:37because the period of the backtest is
183. 6:39actually very low, so sometimes you need
184. 6:41to increase that to get a more accurate
185. 6:43results. And if we take a look at the
186. 6:45distribution and if you take a look at
187. 6:46the distribution chart, we can see that
188. 6:49the returns of the uh actual strategy is
189. 6:52actually better than most of the
190. 6:53simulated ones. Also, once again, the
191. 6:55one that I have here is actually a
192. 6:57better example, so this one is running
193. 6:59on a full year and on a lower time
194. 7:01frame, so it means that we are running
195. 7:03more simulations. And the more
196. 7:05simulations you run with any kind of
197. 7:07randomization test, the more accurate
198. 7:09results you're going to get. And as you
199. 7:10can see, it's telling me that the entry
200. 7:12rule of the strategy is indeed very
201. 7:15significant, and the P value is actually
202. 7:17even below 1%. So, there you go. All
203. 7:19this test does is is going to tell you
204. 7:22if the entry rule of the strategy has
205. 7:25some predictive power or not. It's not
206. 7:26going to tell me if my strategy is ready
207. 7:28for production or whether or not it is
208. 7:30overfit. It is nothing like that, but
209. 7:33the way this is helpful is because
210. 7:35imagine you have an idea for an entry
211. 7:37rule to write a strategy based on it.
212. 7:40Now, before you spend time, like hours
213. 7:42on that strategy to get the position
214. 7:45size right right, to get the exit rules
215. 7:48right. So, before doing any of those
216. 7:50things, you just simply test your entry
217. 7:52rule. And if it says that it is not
218. 7:55signifi- significant, then you're just
219. 7:57going to drop it right there. You're not
220. 7:59going to waste time on it. But if it
221. 8:00says that the entry rule is significant,
222. 8:03then you can spend time on it to develop
223. 8:05an actual strategy, and hopefully, you
224. 8:07will get some good results. But again,
225. 8:10so it doesn't tell you that the the
226. 8:11strategy is definitely awesome, but it
227. 8:13can tell you if it definitely isn't.
228. 8:16Does that make sense? Now, anyways, you
229. 8:18definitely need to do more research on
230. 8:20this topic because it is actually very
231. 8:22helpful and important, and I am making a
232. 8:23video about it, so uh stay tuned for
233. 8:25that. But in the meanwhile, you can also
234. 8:27read the documentation that I wrote for
235. 8:29it, and we have a page fully describing
236. 8:32the steps and why it matters, and like
237. 8:35the underneath method that I am using as
238. 8:38a statistical test for this mode, and
239. 8:41how to exactly read the results that you
240. 8:43are getting. And as I mentioned, you can
241. 8:46also use this inside the research
242. 8:48module. So, if you need to run it via a
243. 8:50program, or you can have the AI to write
244. 8:53some code for you, if that's what you
245. 8:55want to do, you can use the research
246. 8:56module of Jesse, which you can find the
247. 8:58documentation for it right here. And the
248. 9:00next thing that I did is that I added
249. 9:02the optimization mode as a function to
250. 9:05the research module. So, before, we
251. 9:06didn't have this. You could only run
252. 9:08optimization sessions via the dashboard
253. 9:10of Jesse, which is pretty beautiful and
254. 9:12helpful and and it's like awesome, but
255. 9:15there are some cases that you want to
256. 9:17run the optimization session via a
257. 9:19program. Again, maybe you want to have
258. 9:21your AI to write the code for you or
259. 9:23control it even. If that's what you want
260. 9:25to do, then the research module is the
261. 9:27one that you want, and there were
262. 9:28actually many requests for the optimize
263. 9:31mode function in the research module,
264. 9:33and that's what I did. So, now we have
265. 9:35it. And by the way, just so you know, we
266. 9:37are going to have Jesse's own MCP
267. 9:39version soon. So, that is also
268. 9:41definitely coming. Now, while we are in
269. 9:43the documentation, I should also give a
270. 9:44shout-out to the machine learning stuff
271. 9:46that we added. So, if you watched my
272. 9:48previous video about machine learning,
273. 9:50then you know this was added just
274. 9:52recently, but it wasn't in the previous
275. 9:54major release, so you need to know this
276. 9:56that this is also new, but at the moment
277. 9:58it is only supported inside the research
278. 10:01module, and I'm not even sure if you're
279. 10:02going to add it to the dashboard because
280. 10:04the way you're going to use machine
281. 10:06learning is very flexible and very
282. 10:08unique to your own strategy, and I'm not
283. 10:11sure how to add that to the UI, like
284. 10:13what sort of use case you're going to
285. 10:15have for it. But as a function, it is
286. 10:17super helpful and you can use it in
287. 10:19unlimited ways, and I made a video about
288. 10:21it. So, go watch it if you want to know
289. 10:23like how I use it. But definitely the
290. 10:26documentation for it is really helpful.
291. 10:28Like if you read this whole thing, you
292. 10:30may not have to read like a whole book
293. 10:32about machine learning just to be able
294. 10:34to use it like I had to. And just so you
295. 10:36know, our machine learning supports
296. 10:38regression models and the binary models
297. 10:41and multi-class classification. So, it
298. 10:43is pretty flexible. That I wouldn't say
299. 10:45like it is handling like 100% types, but
300. 10:48I think it can handle most use cases of
301. 10:51machine learning in trading. And by the
302. 10:53way, this is the video that I was
303. 10:54talking about. So, if you haven't
304. 10:55watched it, definitely make sure to give
305. 10:57it a watch. It is a very interesting
306. 10:59topic. Now, you might be thinking,
307. 11:01"Okay, so now that you're on version
308. 11:03two, does that mean there have been some
309. 11:05breaking changes?" Well, I made my best
310. 11:08to lower the breaking changes. And the
311. 11:12the only one that actually most of you
312. 11:13have to worry about is related to the
313. 11:16backtest function of the research
314. 11:18module. So, if you were using just this
315. 11:20backtest using the function, so not the
316. 11:22mode in the dashboard, then you're going
317. 11:24to check to ensure that you're using the
318. 11:27correct API for the charts and things
319. 11:30like that. So, that's the only change
320. 11:32that we had. Now, we also had some small
321. 11:34changes. So, for example, if you're in
322. 11:36the backtest settings and you want to
323. 11:39quickly look up the exchange that you
324. 11:41need to change the setting. Now, before
325. 11:43you had to just look for it, which was a
326. 11:45bit annoying, but now we have a search
327. 11:46box for it. So, these sort of small
328. 11:48adjustments have been done in the entire
329. 11:51dashboard, but I didn't document those.
330. 11:52But hopefully you're going to be happy
331. 11:54about it because these are based on your
332. 11:56direct feedback in the past few months
333. 11:59and years. So, that's it, guys. I've
334. 12:01been focused on development only for the
335. 12:04past like God knows how many of weeks,
336. 12:06and starting now I'm going to go back to
337. 12:08do some research for myself and making
338. 12:10more videos for you guys. So, definitely
339. 12:12stay tuned for that because new videos
340. 12:14are coming soon. And as always, if you
341. 12:16had further feedback about like any
342. 12:18changes you want in the dashboard or new
343. 12:20features that you want to be added or
344. 12:22improvements to the existing ones, just
345. 12:24ping me on our Discord server. so much
346. 12:26for being a part of the Jesse community
347. 12:28and thanks for watching this video. I'll
348. 12:29see you in the next one soon.