I Turned My RSI Divergence Strategy Into an Indicator — Transcript

Full transcript

1. 0:00I've been trading the RSI divergence
2. 0:01manually for years, and I've given you
3. 0:03my secret sauce. And in this video, I'm
4. 0:05going to give you the sauce, and I'm
5. 0:07going to explain an experiment that I
6. 0:09ran. The RSI divergence is probably the
7. 0:11trading setup I trust more than anything
8. 0:14else I use. And the problem is, I've
9. 0:16always recognized it visually. I look at
10. 0:19the chart, and then I look at the RSI,
11. 0:21and I just know that's the divergence
12. 0:23that I'd take, or that one's garbage.
13. 0:25So, I wanted to see if I could take
14. 0:26something that exists in my head and
15. 0:29turn it into an actual indicator. Not
16. 0:31some generic RSI divergence indicator. I
17. 0:34mean my version of it. The exact thing
18. 0:36that I'm looking for when I'm trading
19. 0:38manually. And it turns out to be way
20. 0:40harder than I expected. Because
21. 0:42explaining my trading setups, like word
22. 0:44for word in English, but in a way that a
23. 0:47computer can understand, cuz they can't
24. 0:50see like my hand movements. They're two
25. 0:52completely different things. So, here's
26. 0:54the setup. I first had to explain to AI
27. 0:56what I'm looking for. And this is the
28. 0:58secret sauce. For regular bullish
29. 1:00divergence, my rule is very simple.
30. 1:03Price makes a low, and then price makes
31. 1:05another low at roughly the same level or
32. 1:07lower, but the RSI does something
33. 1:10different. The first RSI low has to be
34. 1:12outside of the normal 70 to 30 range
35. 1:15within the RSI indicator. So, for a
36. 1:17bullish setup, I want the first RSI to
37. 1:20be below 30. Then price comes back down
38. 1:23again, making a lower low or a double
39. 1:26bottom. But this time, the RSI makes a
40. 1:29higher low. And the second RSI low is
41. 1:32back within the 70-30 range. That's the
42. 1:35important part. That's the secret sauce.
43. 1:37I don't take every RSI divergence. That
44. 1:40transition is like my secret recipe.
45. 1:42First RSI low, below 30. Second RSI low,
46. 1:46higher than the first and back above 30.
47. 1:48Then once I see the RSI make that tiny
48. 1:50little zigzag, basically, I can see that
49. 1:53the second it forms on the chart, I
50. 1:55enter a buy. Bearish divergence is
51. 1:57exactly the opposite. Price makes a
52. 1:59high, then later makes an equal high or
53. 2:01a higher high. The first RSI peak is
54. 2:04above the 70 range, and the second RSI
55. 2:07peak has to be lower than the first, but
56. 2:09now back within the 70-30 range. When
57. 2:12the RSI makes that little turn
58. 2:14downwards, I sell. That's what I've been
59. 2:16trading manually. Then there's hidden
60. 2:18divergences, and I also wanted the
61. 2:19indicator to detect hidden divergences.
62. 2:22This is where the Fibonacci trading
63. 2:24strategy comes in. Let's say price makes
64. 2:26a strong bullish impulse move up and
65. 2:29then starts retracing. If I draw a
66. 2:30Fibonacci tool from the bottom of that
67. 2:32impulse to the top and then price pulls
68. 2:35back. If price reaches the 0.618
69. 2:37retracement and the RSI is showing a
70. 2:40hidden bullish divergence, that's the
71. 2:42setup I'm interested in. So, visually
72. 2:44price is still making a higher low, but
73. 2:46RSI is actually making a lower low
74. 2:48compared to the RSI at the original
75. 2:50bottom of the impulse move. Price has
76. 2:53held its bullish structure, but momentum
77. 2:55has reset much deeper. For me, if price
78. 2:58reaches that 618 area, I can see the
79. 3:00hidden divergence. That's enough
80. 3:02confirmation for me. I don't need the
81. 3:03same little RSI zigzag that I would wait
82. 3:06for for a regular divergence. The
83. 3:08Fibonacci itself is part of the
84. 3:09confirmation. And obviously the bearish
85. 3:12version is just flipped. Now, here's
86. 3:13where it gets interesting because when I
87. 3:15explain this to another trader, they
88. 3:17probably understand what I mean
89. 3:19immediately. They look at the chart,
90. 3:20they can see the two lows, they can see
91. 3:22the two RSI points, they can see the
92. 3:24divergence, but a computer doesn't see
93. 3:27any of that. It sees numbers. So, every
94. 3:29part of my explanation suddenly has to
95. 3:31become a mathematical rule. What is a
96. 3:34low mean? What is roughly the same low
97. 3:36mean? How many candles exist between the
98. 3:38two points? How far does RSI need to
99. 3:40move before I can officially call it a
100. 3:42second peak or a zigzag? What counts as
101. 3:46a meaningful Fibonacci impulse? How
102. 3:48close does price need to get to the
103. 3:500.618 retracement? Does it have to touch
104. 3:54it exactly? Can it wick through it? Can
105. 3:56it miss it by a few pips? All of these
106. 3:59things feel obvious when I'm looking at
107. 4:00a chart, but suddenly you need to
108. 4:02quantify them and there needs to be
109. 4:04numbers attached. And this is the part I
110. 4:06think most traders underestimate when
111. 4:07they say, "Oh, I'm just going to have AI
112. 4:10code my strategy." Because AI can write
113. 4:12the code, that's not necessarily the
114. 4:14hard part. The hard part is figuring out
115. 4:16exactly if you explained your strategy
116. 4:19well enough for a computer to understand
117. 4:21it and code it. So, we built the first
118. 4:24version of the indicator. For a regular
119. 4:26bullish divergence, AI interpreted my
120. 4:28rules as find the first RSI zigzag below
121. 4:3230, store the RSI value and then
122. 4:35corresponding price low. Then wait for
123. 4:37another RSI zigzag. If price makes an
124. 4:40equal or lower low, but RSI makes a
125. 4:43higher low, that's the second low and
126. 4:46the second RSI low is back within that
127. 4:4870-30 range, then wait for the RSI to
128. 4:51make that small upward turn. Once that
129. 4:53happens, print a buy signal. And for
130. 4:55bearish divergence, it's the exact
131. 4:57opposite. And for hidden divergences,
132. 4:59this one was an absolute show that
133. 5:02took me days. Cuz I had to do something
134. 5:04completely different. It had to identify
135. 5:07an impulse move, calculate the Fibonacci
136. 5:09retracement automatically, watch the
137. 5:120.618 area, compare the current RSI to
138. 5:15the RSI at the original swing, and
139. 5:17determine whether the price structure
140. 5:19and RSI structure were diverging. That's
141. 5:21already a hell of a lot more complicated
142. 5:23than just saying, "Find an RSI
143. 5:25divergence." And then I put it up on the
144. 5:28chart. And this is the part that I was
145. 5:29genuinely worried about cuz it was
146. 5:31either going to look like dog or be
147. 5:33relatively okay. Because it's easy to
148. 5:36write something that compiles with no
149. 5:38errors. But it's much harder to write
150. 5:40something that looks like what I would
151. 5:42actually see. So, I put it up on the
152. 5:44charts and it was surprisingly close.
153. 5:47I'd say around 80% of the way there. It
154. 5:50was identifying most of the regular
155. 5:51divergences I would take. It was finding
156. 5:54hidden divergences. It was drawing them
157. 5:56in places that made sense to me. And
158. 5:58then something started happening. It was
159. 6:00also identifying more divergences than I
160. 6:03personally would have noticed. And then
161. 6:05I looked at them. A lot of them were
162. 6:07actually valid. They followed the rules
163. 6:09I'd given it. I just wouldn't have
164. 6:10necessarily spotted them manually. And
165. 6:12that was interesting to me because now
166. 6:14the computer wasn't just copying what I
167. 6:16saw, it was applying my own rules more
168. 6:19consistently on the charts than I was.
169. 6:22And at the same time, it would
170. 6:23occasionally miss something, especially
171. 6:25some of the hidden divergences that I
172. 6:28would have caught manually. Maybe it was
173. 6:29the fact that they were further apart.
174. 6:31And that shows exactly where the problem
175. 6:33is. The computer doesn't structure
176. 6:34things the way that you do. When I look
177. 6:36at a chart, I don't subconsciously
178. 6:38calculate every part of the structure.
179. 6:40My brain sees the impulse, it sees where
180. 6:42I would naturally draw Fibonacci, it
181. 6:44sees the retracement, it sees the RSI
182. 6:46relationship, and I recognize the setup
183. 6:48almost instantly. But the code has to
184. 6:51decide which exact swing low starts this
185. 6:54exact impulse move, which exact swing
186. 6:57high ends it. And that's much harder
187. 6:59than it sounds. Sometimes the chart has
188. 7:02several possible swing points. To me,
189. 7:04one of them is obviously the correct
190. 7:06one. To the computer, it's all just
191. 7:07highs, lows, opens, and closes. Unless I
192. 7:10give it a way to rank them. That's why
193. 7:12hidden divergences has to be the hardest
194. 7:14part of this entire project to
195. 7:16translate. The regular divergences are
196. 7:18much cleaner. They're easier to reduce
197. 7:20to basic information like first extreme,
198. 7:23second extreme, price relationship, RSI
199. 7:26relationship, confirmation. But
200. 7:28Fibonacci is so subjective. And I think
201. 7:31this is a very important lesson if
202. 7:33you're trying to build your own
203. 7:34indicator using AI. Your first problem
204. 7:36is not coding. Your first problem is,
205. 7:39can you explain what you do in a way
206. 7:41that removes ambiguity. Because most
207. 7:44discretionary traders have rules they
208. 7:45don't even realize they're using. You
209. 7:47might say, I buy support. Okay, what is
210. 7:50support? How many touches? How far
211. 7:52apart? Do wicks count? Do candle bodies
212. 7:55count? How close does price actually
213. 7:57have to be to that support? What
214. 7:58invalidates it? And you might say things
215. 8:00like, I wait for momentum to turn. What
216. 8:03exactly is momentum? RSI? Is it MACD? Is
217. 8:06it candle size? Is it the rate of
218. 8:08change? What does the word in your brain
219. 8:10turn mean mathematically? You might say
220. 8:12things like, oh it doesn't look like a
221. 8:14real breakout. That sentence makes sense
222. 8:16to a trader. You see the hesitation, but
223. 8:19it means absolutely nothing to a
224. 8:20computer. And that's the real hurdle.
225. 8:23You have to take all of that
226. 8:24subconscious pattern recognition that
227. 8:26you've developed over the years of
228. 8:28practicing trading and actually trading
229. 8:30in the live markets and slowly convert
230. 8:32it into like explicit rules. And again,
231. 8:34this is where AI comes in as the most
232. 8:36useful tool in human history. Because I
233. 8:39don't need to know how to Pine Script
234. 8:41well enough to build all of this from
235. 8:42scratch. I can describe what I see. I
236. 8:45can attempt to convert it into the code,
237. 8:47then I can put it on the chart and
238. 8:48compare it. Did I catch this one? Did I
239. 8:50miss that one? Why did it take that one?
240. 8:53I didn't even see this thing. Then we
241. 8:54adjust. So then the workflow isn't
242. 8:57describe a strategy once and AI builds a
243. 8:59perfect indicator. It's more describe
244. 9:02what you see in the best way possible
245. 9:04and be as precise as possible. AI makes
246. 9:08an interpretation, then you put it up on
247. 9:10the chart, compare it to what your eyes
248. 9:13actually see, then refine the
249. 9:15interpretation. And the best way to do
250. 9:17that is to weed out the differences.
251. 9:21Look at the trades you would have taken,
252. 9:23take screenshots of them, and say,
253. 9:25"These are the good basket. These, I
254. 9:27don't know what the you're talking
255. 9:29about." And give that to the AI. And it
256. 9:31will make the changes in accordance to
257. 9:33what you're showing it. Your indicator
258. 9:36doesn't perfectly have to match in
259. 9:38version one. Mine definitely isn't, but
260. 9:41if I can get 80% of the way there, then
261. 9:44you've got something that you can work
262. 9:45with. Because now the computer can scan
263. 9:48thousands of candles, find every setup
264. 9:50that matches those specific rules, and
265. 9:53eventually we can start answering the
266. 9:54question I actually care about. Why do
267. 9:57some of these divergences work, and why
268. 9:59do others fail? That's when this will
269. 10:01get interesting. If you guys enjoyed
270. 10:03this video, please share it with your
271. 10:04friends. Thank you so much for watching,
272. 10:06and we will see you in the next one.