I Tested the Strategy From a $1,000 Trading Book — Transcript

Full transcript

1. 0:00This is Toby Crabel. He wrote what might
2. 0:02be the most famous day trading strategy
3. 0:05book ever written and then he refused to
4. 0:07print it ever again. Now today a single
5. 0:09copy can almost run you a thousand
6. 0:11dollars. Now if you're anything like me,
7. 0:12I'm wondering what's actually in it and
8. 0:14does the strategy still work? Well I got
9. 0:16a hold of it and I coded the main system
10. 0:18idea from scratch and tested it and I've
11. 0:21got to say what I found really surprised
12. 0:23me. So in this video I'm going to take
13. 0:25his opening range breakout strategy,
14. 0:27backtest it on years of minute by minute
15. 0:29data on QQQ and show you exactly what
16. 0:32happened. Now the strategy is actually
17. 0:33simpler than you would actually think
18. 0:35because it doesn't try and predict what
19. 0:37way the market's going to go. It waits
20. 0:39for the market to go dead quiet and then
21. 0:41places two orders and lets the market
22. 0:43pick the winner. Well, at least in
23. 0:44theory. So who is Toby Crabel? Well he
24. 0:47actually started out as a professional
25. 0:49tennis player. Then he became a trader
26. 0:50and then he worked on the Victor I don't
27. 0:53think I'm going to be able to say last
28. 0:54name but
29. 0:55Victor Niederhoffer and in 1920 22 he
30. 0:58went out on his own. And this was the
31. 1:00first number that actually stopped me in
32. 1:01when I was reading about him. From 1991
33. 1:04to 2002
34. 1:06his trading went through a 12 year
35. 1:08streak without a single losing years. 12
36. 1:11years is an incredible amount of time.
37. 1:13And then Financial Times actually once
38. 1:14called him the most well known trader on
39. 1:16the counter trend side of things. At
40. 1:18peak he was managing about 7.5 billion
41. 1:20and he only ever wrote one book that was
42. 1:22in 1990 and has never been reprinted
43. 1:25since. Today his firm runs billions but
44. 1:27it did peak around seven and a half
45. 1:29billion. In this book though he laid out
46. 1:31his research and sold it and then
47. 1:33refused to ever let it get reprinted
48. 1:34again which is why now a copy can almost
49. 1:36cost you a thousand dollars on Amazon.
50. 1:38So I bought the strategy and coded it
51. 1:40and ran it on 15 years of data. And I'll
52. 1:42be honest, when I first saw the result I
53. 1:45thought it would be something incredible
54. 1:46but keep on watching to see how it turns
55. 1:48out.
56. 1:50Now speaking of the strategy, let's go
57. 1:51over what it is. Well it's an opening
58. 1:54range breakout strategy. Now many of you
59. 1:55are are probably aware of something
60. 1:57similar to this idea. Ironically, he
61. 1:59actually coined most of this phrase
62. 2:00around opening range breakout from this
63. 2:02book. Now the whole observation from
64. 2:03this book was market breath. Overall,
65. 2:05the market can go through quiet periods
66. 2:07where the range gets a lot tighter and
67. 2:09tighter, and then it explodes out in one
68. 2:11direction. So overall, the idea of
69. 2:13contraction then expansion. So the first
70. 2:15main thing we have to do for this
71. 2:16strategy is wait for a quiet day where
72. 2:18the market has cooled up tighter than
73. 2:20usual, and then the next morning, we're
74. 2:22not going to guess the direction.
75. 2:23Instead, what we're going to place an
76. 2:25order above the open and an order below
77. 2:26the open. Whichever way gets hit first,
78. 2:29that's the trade we're going to take.
79. 2:30Overall, you're letting the market tell
80. 2:31you which way it's breaking towards. But
81. 2:33then the second question comes in, well,
82. 2:34is how far do we place this order above
83. 2:36and below the open? Is it just right at
84. 2:38the open kind of price? Or are we
85. 2:39placing it with a decent amount of
86. 2:40margin? And this is where he uses
87. 2:42something called the stretch. The
88. 2:43stretch is a 10-day average of the
89. 2:46smaller of high minus open and open
90. 2:48minus low. I.e., on a normal day, how
91. 2:50far does price usually poke past the
92. 2:53open before coming back? And then we're
93. 2:54going to place an order exactly one
94. 2:56stretch above and below, as kind of
95. 2:58demonstrated here.
96. 3:01Now that's the rough idea, but let me
97. 3:02give you the actual full rules. First,
98. 3:04we actually have to define what a quiet
99. 3:06day is. To backtest or quantify any
100. 3:08strategy, we need to get this into
101. 3:10criteria we can code or we can at least
102. 3:12put data through. And he actually gives
103. 3:13up gives us three setups for this, which
104. 3:15is amazing. So first, he does NR4, which
105. 3:18is today's range is the narrowest of the
106. 3:20last four days, which is very nice and
107. 3:22simple logic. Then he has NR7, which is
108. 3:24basically the same idea, but it's the
109. 3:26last seven days. And then lastly, he has
110. 3:27inside day. And this is where today's
111. 3:29whole range fits inside yesterday's. So
112. 3:32lower and high and higher and low, for
113. 3:33example. And if any of those three
114. 3:35qualify, then we are armed for tomorrow
115. 3:37to then place our orders. We're going to
116. 3:38be placing a buy stop one stretch above
117. 3:40the open and a sell stop one stretch
118. 3:42below. The first one to trigger wins and
119. 3:44the other side cancels out. Our stop
120. 3:46loss is incredibly simple as well. It's
121. 3:47simply the opposite band. So if we go
122. 3:49long at the top, the price falls all the
123. 3:50way back to the bottom band and we're
124. 3:52out. And if we're still in a trade at
125. 3:54close, we'll exit it at the closing
126. 3:55bell. Every trade is done by the end of
127. 3:57the day, nothing is held overnight at
128. 3:59all. For the risk per trade, this is
129. 4:00going to be two times the stretch, so
130. 4:02the top band to the bottom band, and
131. 4:04every position exits by 4:00 p.m. close.
132. 4:06Now, one more rule is if neither order
133. 4:08triggers early in the session, we cancel
134. 4:10them out. Korbel actually found that the
135. 4:12edge is strongest right after the open,
136. 4:14so we give it until around 10:30 in the
137. 4:16morning, and if nothing is triggered by
138. 4:17that time, we're going to be walking
139. 4:18away from the day and no trades are
140. 4:20going to be placed. To give you an
141. 4:21example idea of how a trade looks, this
142. 4:23is one of the trades that I backtested
143. 4:25using code, and this was actually from
144. 4:272025. As you can see, this was an 8R
145. 4:30winner, which was pretty insane for the
146. 4:31portfolio, and it entered over here and
147. 4:34exited over here.
148. 4:36Now, speaking of, how did I actually
149. 4:38test this strategy so that it can be
150. 4:40automated or we can just first backtest
151. 4:42it over years of data without me having
152. 4:44to manually do this myself. Now, of
153. 4:46course, I actually did this via code,
154. 4:47and if you want a general snippet of how
155. 4:49that code looks like, I've left some
156. 4:51over here. First part of the code is all
157. 4:52around that kind of tightness and
158. 4:54contraction where we're trying to look
159. 4:55for NR4 and NR7 and inside days so that
160. 4:58we can arm those particular days and
161. 5:00then look for the next day if we're
162. 5:02going to be placing those stretch
163. 5:03orders. Now, this is also another really
164. 5:04important thing to note, is we have no
165. 5:06look-ahead bias. Every one of those
166. 5:08signals is shifted by one day. The
167. 5:09stretch, the arming, everything is
168. 5:11calculated from days that have already
169. 5:12closed before the day's even opens, and
170. 5:15this is to avoid us having a look-ahead
171. 5:17bias in our backtest and getting false
172. 5:18results. For example, if you compute the
173. 5:20stretch using today's high and low, then
174. 5:23trades today, you've used information
175. 5:25you would have never known at that point
176. 5:26of time. Third is how the actual order
177. 5:28fills. A buy stop fills at the worst of
178. 5:31the band or at the bar's open. So, if
179. 5:32the price gaps straight through your
180. 5:34level, you fill at the gap and not the
181. 5:36level. And that's what normally really
182. 5:37happens in actual live markets, and
183. 5:39that's why it's kind of coded in. Now,
184. 5:40I've also tested two version. He
185. 5:42actually mentioned both of these in the
186. 5:44book, which is the original ORB
187. 5:45strategy, so this is in both directions
188. 5:47where we're placing both orders every
189. 5:48single day that is armed, long and
190. 5:51short. And then there is one where it's
191. 5:52a trend filtered version. So, add a
192. 5:5450-day filter and above it we're going
193. 5:56only long and below it we're only
194. 5:57looking for shorts. Now, I ran this
195. 5:59primarily on QQQ, but I also did a quick
196. 6:02backtest also on spy later in this
197. 6:04video. All of it was using 1-minute
198. 6:06data. And just before I show you the
199. 6:08results, let me just give you a quick
200. 6:09word from our sponsor because it ties
201. 6:11directly into how I backtested this
202. 6:13strategy. Now, to be clear, this is a
203. 6:15paid sponsorship, but I am free to share
204. 6:17my own thoughts and opinions on the
205. 6:18actual software. So, everything I showed
206. 6:20you today, the code, the shift, the fill
207. 6:21logic, it took me quite a while to
208. 6:23write. And I've been coding since I was
209. 6:25around 12, so more than a decade now.
210. 6:27But nowadays, we have AI and you could
211. 6:28do a lot with it. Horizon is a web app
212. 6:31where you can build and test a strategy
213. 6:32like this without writing any code at
214. 6:34all. You can describe the idea and then
215. 6:36it can instantly backtest it for you
216. 6:38using actual data. Now, if you try and
217. 6:39do this through a lot of traditional AI
218. 6:41platforms, they won't actually have
219. 6:42solid data and solid backtesting
220. 6:44techniques. Whereas Horizon actually
221. 6:46does. It also has a whole execution part
222. 6:48where you soon you'll be able to connect
223. 6:49to your brokerage and it can
224. 6:51automatically automate those strategies
225. 6:52that you've tested. Let me actually just
226. 6:54quickly show you a quick example. I
227. 6:55built a really simple orb strategy using
228. 6:58Horizon. My prompt was actually, "Make
229. 7:00me a simple orb strategy on the
230. 7:0130-minute timeframe on QQQ." It was able
231. 7:03to build everything for me, give me all
232. 7:05the results, and then I was able to ask
233. 7:07a second question to compare it on how
234. 7:10the stop loss would affect the results.
235. 7:12As it then showed me for a 2% stop loss
236. 7:15I got this net profit over here. Whereas
237. 7:17when I changed it to a 4%, I was
238. 7:18thinking maybe it would improve
239. 7:19performance. It did not at all. It was a
240. 7:21really nice way to see everything in one
241. 7:23panel. I can also see all the trades
242. 7:25taking place here. I can also click into
243. 7:27any of these trades. So, if I just want
244. 7:28to look at this one for example, I can
245. 7:30zoom in and see exactly how that trade
246. 7:31is placed. It also has more information
247. 7:33on performance broken down by drawdown
248. 7:35and rolling sharp. Ultimately, the idea
249. 7:36is really simple. All you have to do is
250. 7:38go to horizon.trade, type into a prompt
251. 7:40that you want to get tested, and then it
252. 7:42can automatically backtest it for you.
253. 7:44If we actually have a look, I can click
254. 7:45my strategies, I can see everything I've
255. 7:47tested before in a really nice
256. 7:48interface. Soon you'll also be able to
257. 7:50connect brokers, which you'll be able to
258. 7:51automate those strategies really easily
259. 7:53for you. Now, just as a reminder,
260. 7:54Horizon is a software tool for strategy
261. 7:57testing and execution. It does not
262. 7:58provide any financial advice or
263. 8:00investment advice. Trading always
264. 8:01involves substantial risk of loss, and
265. 8:03backtesting is simply a tool for
266. 8:05evaluating historical performance and
267. 8:06never guarantees future results. Now, if
268. 8:08you want to try it out, feel free to
269. 8:10click the link in the description.
270. 8:12This is being tested on 15 years of
271. 8:14minute data on QQQ. Every single arm
272. 8:17day, both orders are placed and exits at
273. 8:19the close. Here, we can see the equity
274. 8:21curve. On first glance, this looks
275. 8:23incredibly impressive. The green line
276. 8:24over here is the actual strategy, and
277. 8:26yellow is just a simple buy and hold.
278. 8:27And as we can see, it's really not even
279. 8:29close. There's a wide margin where this
280. 8:31strategy is really outperforming.
281. 8:32Overall, it did around 1,300% and it had
282. 8:35a 19% CAGR. It also, most importantly,
283. 8:37had a very low correlation to spy, which
284. 8:40means it can be really useful when
285. 8:41adding it into a portfolio of
286. 8:43algorithmic strategies. It did have
287. 8:44quite a steep drawdown of 30%, but we
288. 8:47are getting quite an above average CAGR
289. 8:49of 19%. We can also see how those annual
290. 8:51returns look like over all those years.
291. 8:53We can see in 2016, it was up around
292. 8:5567%, and in 2018, it had one of its best
293. 8:58years, being up 97% in a single year.
294. 9:02And Korbel was definitely right about
295. 9:03the edge being strongest right after the
296. 9:05open. The data from most recent years,
297. 9:07as we can see from my backtest, was that
298. 9:08a lot of returns were coming from the
299. 9:10first 15 minutes or from trades being
300. 9:12executed at 10:15 to 10:30. It's quite
301. 9:14impressive how his observation has held
302. 9:15up even from 1990. So, at this point,
303. 9:18I'm thinking, well, I've got a
304. 9:1935-year-old strategy from a
305. 9:20thousand-dollar book and it's crushing
306. 9:22the market with almost zero correlation.
307. 9:24Case closed, let's just run this live
308. 9:26now, and I never have to worry about
309. 9:27money. But, unfortunately, there's one
310. 9:29thing I haven't added yet, and that is
311. 9:31commissions and slippage. In the real
312. 9:33world, when you trade, you might get
313. 9:34away with some brokers not charging you
314. 9:35commission, but you will always suffer
315. 9:37some kind of slippage. And slippage can
316. 9:39be a really sneaky one. When you send a
317. 9:41stop order to buy, you don't get filled
318. 9:43at your exact price. By the time it
319. 9:44executes, the market has usually moved a
320. 9:46little bit against you or in your favor.
321. 9:48For example, maybe you wanted 418, but
322. 9:50you actually got 418.10. That gap is
323. 9:53slippage. And on any order that you're
324. 9:54placing like a market order, for
325. 9:56example, you're going to get some of
326. 9:57that slippage. Now, you could have
327. 9:59course use limit orders, but then you
328. 10:01have a whole liquidity problem of if you
329. 10:03even get filled at that particular
330. 10:04price. So, here's how I modeled it. I
331. 10:06added commissions and slippage. For
332. 10:08commissions, I just added one basis
333. 10:09point, and for slippage, I added two
334. 10:12basis points on every single fill. Now,
335. 10:14I did have a percentage of these that
336. 10:16were adverse fills, and some of them
337. 10:17were not adverse fills to at least be
338. 10:19fair. Now, this might be lethal for this
339. 10:21particular strategy. The reason being is
340. 10:23one, it trades a lot. It's got about a
341. 10:25thousand trades over that time, and
342. 10:27that's going to be a lot of slippage and
343. 10:28commissions added on top. Second, the
344. 10:30edge isn't actually that big. On
345. 10:32average, per trade, we're only making
346. 10:34about 0.13 R. When that edge is so
347. 10:36small, when you add commissions and
348. 10:38slippage, it can get easily eaten away
349. 10:40as demonstrated here. Now, modeling
350. 10:41costs is actually one of the first
351. 10:43things I try and teach in the backtest
352. 10:44bootcamp because it's the single most
353. 10:46common thing a winning backtest can fool
354. 10:48you on into thinking that this strategy
355. 10:50is going to be really profitable when
356. 10:51you take it live. Now, if you want to
357. 10:53properly backtest your strategies, feel
358. 10:54free to click the link in the
359. 10:55description to check it out.
360. 10:58So, let's actually see the real results.
361. 11:00When I add commissions and slippage,
362. 11:02sadly, while the equity curve isn't
363. 11:04terrible, it definitely doesn't beat spy
364. 11:06anymore over that time period. The sharp
365. 11:07has also halved, and the drawdown gets a
366. 11:09lot deeper to about 45%. You can also
367. 11:12see how it affects the holes entries at
368. 11:14different times where in some segments
369. 11:15actually just become straight
370. 11:16unprofitable. Additionally, it does kind
371. 11:18of get worse. I tested this on spy, and
372. 11:20on spy, this strategy just completely
373. 11:22sucks. Overall, the lesson from this is
374. 11:24to test things properly. Always have
375. 11:26costs and slippage, and make sure to do
376. 11:28multiple different checks like I did
377. 11:30here of correlation, and you could do
378. 11:32further things like in sample and out of
379. 11:33sample and walk forward optimization.
380. 11:35Additionally, the idea doesn't seem that
381. 11:37bad. While it may not be performing very
382. 11:39well on QQQ and spy, this could be
383. 11:41something to test in crypto or other
384. 11:43more inefficient markets that may still
385. 11:44actually have that edge. And if this is
386. 11:46the kind of thing you want to get good
387. 11:47at building systematic strategies like
388. 11:49this one and testing them honestly and
389. 11:51then automating them on a machine so
390. 11:53that it runs completely without you,
391. 11:55that's exactly what we do inside the
392. 11:56crypto momentum group. It's a community
393. 11:58of people doing this properly with the
394. 12:00code, the templates and the guidance to
395. 12:02go from an idea to running something
396. 12:03live. Link in the description if that
397. 12:05sounds like your thing. Well, I hope
398. 12:06this video was helpful and there'll be
399. 12:08many more to come.