YT YouTube2Text

YouTube2Text → Transcripts → Claude Tested 100,000 Price Action Strategies (Here’s What Works)

Claude Tested 100,000 Price Action Strategies (Here’s What Works) — Transcript

by AI Pathways · 3,737 words · 526 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00In this video, I use Claude to test over
0:02100,000 price action concepts to find
0:04which actually work. Lately, there's
0:06been a lot of trading videos focusing on
0:08price action and its core mechanics
0:10versus fancier indicators. So, in this
0:13video, I tested those price action
0:15concepts the same way they're taught,
0:16the way people typically use them, as
0:19well as a 40,000 sample to capture
0:21everything in between. This way, you're
0:22not just believing anecdotal evidence of
0:25one single type of trader. You're
0:26looking at all the data and numbers
0:28behind it. This whole process covered
0:30almost 5 years of 1-minute data on NQ
0:33and ES with over 300 million trades. So,
0:37by the end, I'll show you the exact
0:38price action setups that work and the
0:40time frames you should be running it on,
0:42as well as all the price action concepts
0:44that don't work. Now, if it's your first
0:45time on the channel, my name is Brendan.
0:47I studied math and econ at UCLA, worked
0:50in investment banking at Raymond James
0:51for over 3 years, and have spent the
0:53past 2 years building out trading
0:55systems for my clients. And as always,
0:57this is not financial advice, and I'm
0:58not guaranteeing any profits. I'm just
1:00showing you all the data and methodology
1:02behind these tests and how you can
1:04leverage them as well. Now, before
1:05diving into the results of the test, let
1:07me first go over what price action
1:09really is. So, the idea is that price
1:11action is just trading off the chart
1:12itself. So, number one here is just a
1:14level. So, on this chart here, you can
1:16see that this is support where price
1:18keeps bouncing off of it, and above here
1:20is resistance, and this is where price
1:21gets pushed back down. So, the idea with
1:23price action is that price is going to
1:25hold around these levels much more
1:27frequently than just random lines. And
1:29this idea of levels is supposed to work
1:31from 1-minute, 5-minute, all the way up
1:33to daily charts. Now, the second biggest
1:35component of price action is something
1:36called trend. So, an uptrend is just
1:38higher highs and higher lows as well,
1:41just as this chart section here shows.
1:43And more specifically, when price closes
1:45above the last high, this is called a
1:47break of structure. The idea with this
1:49is that it's supposed to confirm that
1:50the trend is still continuing. Now, next
1:52up, what we have is something called a
1:53change of character. This is the first
1:55time that the price breaks the other
1:57way. So, here on this chart, you can see
1:59that it actually closes below the last
2:01higher low. And just as a spoiler alert,
2:03this change of character concept and
2:05price action is actually the one that
2:06held up the best out of all of our
2:08findings. Now, that doesn't mean that
2:10all price action concepts work
2:12perfectly. So, for example, the next one
2:14is sweep wouldn't be a signal by itself.
2:16The idea here with the sweep is that
2:17when price goes just past a higher or a
2:19low where a lot of traders keep their
2:21stop losses and then it comes right
2:23back. But, this is an example of a price
2:25action concept that through all the data
2:27actually showed that it only reversed
2:29about half the time. So, this could be
2:30helpful as a signal, but not as a core
2:33concept that you should be running. Now,
2:34the next concept of price action is
2:36something called a confirmation. So, say
2:38that your signals trigger, instead of
2:39buying right at the level, you would
2:41wait for a candle like this one to close
2:43above it before you get in. And then the
2:44last part of understanding how to use
2:46price action concepts is actually what
2:48time frame to trade on. So, you can
2:50trade from the one-minute chart all the
2:51way to the daily chart. And later in the
2:53video, I'll show you the best ways to
2:54think about these time frames depending
2:56on which specific price action concepts
2:58work best. So, as a whole, we can pretty
3:01much say that almost all price action
3:03concepts at a quick glance do work.
3:05Levels are real. Obviously, support and
3:07resistance lines do show up in charts
3:09and they're backed by data as well. But,
3:11what we want to show in this video is
3:13the best possible combination of
3:15concepts, when to use them, and when not
3:17to use them. Some work best only as a
3:19signal, while others should be a major
3:22core component of how you trade. And
3:24when you put all those different price
3:25action concepts together, you get 10
3:28major setups, which are the ones here.
3:29We're going to be testing these and
3:31derivatives of all of these, but these
3:33just cover the main ways that people
3:35trade price action. And really quickly,
3:37if you are interested in learning how to
3:38build your own testing and trading
3:40systems with Claude just like this, make
3:42sure to click the link in the
3:43description for my community. I
3:45currently run the largest AI focused
3:47trading community on school, which
3:49covers everything from stocks, options,
3:51crypto, and futures. And if you do want
3:53to work with me one-on-one to develop
3:55your own custom trading system, make
3:56sure to click the link in the
3:57description for that as well. Now, let's
3:59go into the methodology of this testing
4:01system. The problem with testing price
4:03action traditionally is that it's all
4:05drawn by eye. So, the idea is that if
4:07you ask five different price action
4:08traders what this level is on a chart,
4:10they'd have five different answers for
4:12you. So, this is an example of trader A
4:14and trader B that kind of see the chart
4:16differently. Now, obviously people
4:17trading like this never really check
4:19your own work. It's pretty much based on
4:21feel. So, our goal here is to take every
4:23single one of these judgments and turn
4:26it into something that we can actually
4:27test. So, on the left-hand side here,
4:29what we have are called dials. And these
4:31are all the things that we can test that
4:33relate to price action. So, how many
4:35touches a level needs, whether it's a
4:37line or a zone, which candle actually
4:39counts as confirmation. We also have
4:41things like where the stop goes, where
4:42the target goes, which chart you use,
4:44and which filters you add on top. Now,
4:46obviously in a trader's brain, these are
4:48all things they would judge just
4:49visually by looking at a chart. But, our
4:51goal here is to take all of this and be
4:53able to quantitatively test it. And this
4:55is exactly why we use Claude to build
4:57out this in-depth testing system to be
4:59able to turn these dials into something
5:01that's tangible. So, with Claude, we can
5:03take the way we trade, force it into
5:05explicit rules just like these here, and
5:06then test those rules across years of
5:09data in a few sessions. And just to show
5:11you the scale of this test, if we
5:13multiply out all of these different
5:14types of dials, we get to this number
5:16here, which is 20.3 trillion versions of
5:19looking at price action. Now, this
5:21pretty much explains why two price
5:23action traders would never typically
5:25trade the same way just because there's
5:27so many different versions of looking at
5:28a chart. So, in the test we ran, we took
5:30a sample of 40,000, spread it across
5:32every dial, so we have around 4,000 for
5:35each of the 10 setups that I just showed
5:36you. Now, the second most important
5:38thing about price action is that they
5:39don't actually spread out evenly across
5:41all these different types of versions.
5:43So, a lot of people cluster around the
5:45same settings like a clean level with
5:47two or more touches, a stop is just past
5:50the level or the last swing, things like
5:52that. So, on top of these 40,000
5:54versions of price action, we added
5:56additional 61,000 of common versions
5:58that were bucketed to get to over
6:00100,000 different ways that we can
6:02actually test price action. And now, to
6:04make one thing super clear that I always
6:06make in all of my videos, we're not just
6:08asking Claude whether price action
6:09works, right, in a chatbot. An AI's
6:12opinion is not a real test. What we're
6:14doing instead is having Claude build an
6:16entire testing system that can take data
6:18at scale. So, what we have here is
6:20almost 5 years of 1-minute futures data,
6:22so about 3.6 million bars on NQ and ES.
6:26Then after that, we built the full
6:27system test with Claude in layers. It
6:30covered over 300 million trades with
6:32real cost and slippage on top. We were
6:34fairly conservative in the assumptions,
6:36so that like if a candle touched the
6:38stop and the target in the same minute,
6:40we would count it as a loss. We also
6:41factored in luck by comparing against
6:43random trades, and then finally at the
6:45end, we did all of our testing out of
6:47sample on ES, so that the data was never
6:50really seen by the testing engine. So,
6:52here we can see the 2 years locked away
6:53meant that anything that we tested for
6:55all these different versions of price
6:57action had never seen the data before.
6:59So, out of the 100,000, only 88,000 had
7:02enough trades to even be graded on. Then
7:05after that, 24,000 made money after
7:07costs, and then only 8,310
7:10cleared a decent risk-adjusted bar with
7:12a sharp of over .5. About 8% of the
7:15versions of price action did actually
7:17make money after costs and are viable,
7:19but we obviously care about the
7:21specifics around this and how to
7:23actually think about it. Now, the first
7:24claim around using price action for
7:27trading is that it has a super high win
7:28rate. Some people say like 80%. This
7:31chart here shows all of the original
7:3340,000 strategies by win rate and
7:35profit. So, you can see that win rate
7:37never really clears that 80%. Obviously,
7:39it does happen in some versions, but all
7:42in all, it's pretty rare to have this
7:43high of a win rate. And as you'll see
7:45later on in the video, win rate doesn't
7:47really track profit at all. The
7:49correlation is pretty low, it's only
7:51.27. Now, next on a big-picture scale is
7:54figuring out which time frame works best
7:56for price action as well. Some people
7:58claim that it works on every single time
8:00frame, but obviously some work better
8:02than the others. Now, as expected, if
8:04you trade on the 1-minute time frame,
8:05you can see that costs take around 40%
8:08of the gross profit. That's just the
8:10nature of trading more often every
8:12single day. So, if you are just starting
8:14out, it's much much easier to be
8:15profitable if you're just taking slower
8:17trades. So, trading on like the 1-hour,
8:204-hour candles, rather than trading
8:221-minute candles, just because you're
8:23taking a lot less trades. And with
8:25factoring costs, your edge doesn't have
8:27to be significantly higher per trade.
8:29Now, the next concept is levels, and
8:31this was actually a very positive
8:32finding. So, we took every level, drew a
8:35line next to it at random, and then
8:36scored both the same way. So, obviously
8:38support and resistance would be the
8:40levels. And as you can see, levels here
8:42held 51.1% of the time, meaning that
8:45prices do react to levels much better
8:48than any other lines on the chart. But,
8:50more importantly than levels is the
8:51concept of how often a level gets
8:53touched. This is a pretty big concept of
8:55price action, where the more times you
8:57touch support or touch resistance, that
8:59means the stronger the level is. But,
9:01what we found here is that a level
9:02touched once beat its random line by
9:04like 9 percentage points, and then a
9:06level touched four times beat it by 6.1.
9:09This means that the extra touches, so
9:11continually touching support or
9:13resistance, didn't make it any stronger.
9:15In fact, it was actually the opposite,
9:17where levels touched three or more times
9:19actually bounced a bit less often, and
9:22then broke through a bit more often as
9:23well. Now, obviously you can find
9:25examples of this not being true, but
9:28across basically 300 million trades,
9:30this is pretty much how it's shaken out.
9:33So, the main takeaway is that levels
9:34obviously do matter, like support and
9:36resistance levels, but the amount of
9:38times it's been touched isn't as
9:40relevant for the level. But now, the
9:41most important finding for this whole
9:43entire video is the idea of flip zones.
9:46So, this is what I covered all the way
9:47in the beginning. So, when old
9:49resistance turns into support after
9:51price breaks through it. So, on the
9:53first retest, these held 11.5 percentage
9:56points more often than random lines,
9:58which is this dot all the way out here,
10:00and 10.9 points more on ES. So, this
10:03means that flip zones are probably the
10:05strongest, most important in all of
10:07price action, apart from just levels of
10:09support and resistance. Next is trend
10:11rule, so the break of structure as well
10:13as a change of character. So, trend
10:15concepts on their own don't predict
10:17where price goes, but as a trade, if you
10:19go deeper down, it was the best out of
10:21all 10 setups. And typically, the way
10:23you would use this is to wait for a
10:24sweep first, but when we test this on
10:27the locked away 2 years that the test
10:29never saw, it actually lost money. The
10:31best way to run this setup was without a
10:33sweep. Now, next we have confirmation,
10:35which is just to make sure in typical
10:37price action concepts to not catch a
10:39falling knife. So, here we took the
10:40exact same trades two ways, buying right
10:43at the level or waiting for a candle to
10:45confirm it. So, as you can see here,
10:46waiting does give a slightly worse
10:48price, but has an increased edge of
10:50around $719.
10:52So, the key here is that waiting for
10:53confirmation, again, is a good price
10:56action concept that you should do,
10:57especially on Nasdaq futures. Then, we
10:59have stops. So, a common rule is to only
11:01stop out when a candle closes past your
11:04stop instead of as soon as the price
11:06touches it. And when this happens, it
11:08does win more often, 40% against 29%
11:11here, but the losses get bigger, so
11:13overall it does cost money. So, all in
11:16all, a plain hard stop does perform
11:18better. Next, we wanted to make sure
11:19that we had all the context as well. If
11:21we traded only during the New York
11:23session, so 8:00 a.m. to 4:00 p.m.
11:25Eastern time, this actually improved our
11:27edge per trade significantly. And like I
11:29talked about in the beginning of the
11:30video, we did an overall test to make
11:32sure which of these concepts held up,
11:34and then at the end we kept 2 years of
11:36data locked away that the test never saw
11:38to get a true final result. And before
11:40that final test, we just did a
11:41consistency check here to find which
11:43setups actually passed the first period
11:45on both markets. And out of every single
11:47setup on the chart, only the 4-hour
11:50change of character technically passed
11:52on NQ and ES, and it did that with both
11:54sets of versions. So, here we have the
11:56final test, which is just the 2 years of
11:57data that stayed locked away. So, no
11:59strategy saw this data, and nothing was
12:01tuned on it. So, most importantly, we
12:03first wanted to look at the overall time
12:05frame. This shows how many of the common
12:07versions
12:08on those 2 years by chart. So, on the
12:101-minute chart, around 10% or one out of
12:1310 strategies made money. And then once
12:15we get up into the 1-hour, 4-hour,
12:16daily, we can see that the percentage
12:18increases significantly. So, if you were
12:20to look at this quickly, if you were
12:21just trading the 4-hour chart instead of
12:23the 1-minute chart, you would have 45%
12:26of those price action concepts making
12:27money versus 9% on the 1-minute chart.
12:30So, now I'm going into the best version
12:32of all the different price action
12:33concepts, we have what is tested here,
12:35which is the change of character on the
12:374-hour time frame. So, this is just the
12:39plain set up, like I said, with no
12:41sweep. You enter at the candle close.
12:43The stop goes past the last swing, and
12:45the target is the same size as the risk.
12:47On the two locked years, which is
12:49everything after this dotted line, which
12:51again the system never sees, it made
12:52around $62,000 on a single NQ contract
12:56and around $39,000 on ES. Now, this is a
12:58good sign just because a strategy that
13:00was meant to fit on old data usually
13:02stops working on new data, but this one
13:04made money on both markets. And it's not
13:06just one version, so 39 of the 48
13:08versions here on its usual settings made
13:11money on the locked years, and so did
13:13two-thirds of the common versions as
13:15well. Now that we saw the best overall
13:16results from all this testing, what's
13:18more important about this video is which
13:20price action concepts make sense. Since
13:22you don't just want to test that one
13:23very specific rule, you want to make
13:25sure that you're using all of these.
13:26Sometimes these can be put directly in
13:28strategies, other times these can be
13:30used more so as filters or things to
13:32know before you trade. So, first is that
13:34the change of character on the 4-hour
13:36chart performed the best. It was the one
13:38set up that held up even in the locked
13:40away years. And now secondly and more
13:42importantly, and something that I almost
13:44always stress on the channel as well,
13:45especially if you're automating, is to
13:47trade on higher charts. Since the higher
13:49the chart, the more versions made money.
13:51It's much, much harder to find a
13:53successful strategy trading on the
13:551-minute chart versus the 4-hour or
13:571-day chart. Now, third is trading only
13:59the New York session time on futures.
14:01Fourth is skipping trades that go the
14:02same way as a 3-day run, since both kept
14:05adding money everywhere we checked.
14:07Fifth is waiting for that confirmation
14:08candle at least on the Nasdaq. Sixth is
14:11using a plain hard stop, and then
14:13seventh is not waiting for a sweep
14:15before the change of character. Since
14:17the closed base stop and the sweep,
14:19again, both of them cost money every
14:20single time we checked. So, basically
14:22the version of price action that
14:24survives all this testing is pretty
14:25simple. It's right here on this chart.
14:27It's really just one setup on the right
14:28chart, a simple entry and a hard stop.
14:31And this is an actual trade from the
14:32best version, too. So, here you can see
14:34the trend. Here's the first close the
14:36other way, which is the change of
14:37character, again, on the 4-hour chart.
14:40So, after this change of character on
14:41the 4-hour chart, you enter at this
14:43close, and then this is where you set
14:45the stop, and then this is where you set
14:47the target with no sweep or trend filter
14:49on top. Now, if this sounds familiar,
14:51it's because it's literally the same
14:53thing we keep finding in all these
14:54videos I make, which you can see from my
14:569,000 strategy test or my ICT test. In
15:00that 9,000 test, I said that mean
15:01reversion was the only strategy that
15:03really held up out of thousands of
15:05different variations. In the ICT test,
15:07the version that held up was fading the
15:09sweep, which was, again, a mean
15:10reversion style strategy. And what we
15:13have here is a reversal setup, which is,
15:15again, a mean reversion type of
15:17strategy. Only it's that it's slightly
15:18different. It's called a change of
15:20character. It's on the 4-hour chart. But
15:21again, the core concepts of this
15:23movement on the chart are all exactly
15:25the same. When price stretches too high
15:27or too low, it's always going to revert.
15:29It's literally the core concept of
15:30trading, especially if you're trading
15:32quicker time frames. And just like the
15:349,000 strategy video test, this is the
15:37floor, not the ceiling, right? This is
15:39where you would start. These are setups
15:41that are average across every single
15:42version that we tested at one contract
15:45per trade. So, obviously, you can make a
15:46ton more, or you can just completely
15:48botch it and make a ton less, too. This
15:50isn't a finished strategy. It's a
15:52foundation that you build your own risk
15:54management and sizing on top of. Now, if
15:56you want to build this testing system
15:57yourself, it's the same format as
15:59always. The system is built in layers,
16:01and each layer is a separate prompt. So,
16:03again, if you're using ChatGPT or
16:05Claude, feel free to screenshot all of
16:07these prompts layer by layer, and it'll
16:09build out the same exact testing stack
16:11for you. So, the first layer just builds
16:13the data and then turns every price
16:14action rule into dials. The second here
16:17builds the trade engine with real costs
16:19and tests what price actually does at
16:21levels. The third runs every version
16:23through the funnel. The fourth runs the
16:25final test on the locked away data. And
16:27then the last one builds the dashboard
16:29that I've been basically presenting this
16:31whole entire time. And the bigger point
16:33here that I also stress on all my videos
16:35is that this isn't just about price
16:37action. Whatever style you trade, if it
16:39currently lives in your head as a feel,
16:41you can do the same exact thing here and
16:43automate it or just turn it into rules
16:45and find out if it holds up at scale
16:47with real numbers and data behind it.
16:49Now, if you enjoyed the video, make sure
16:50to leave a like, comment, and subscribe
16:52as it greatly helps out the channel. And
16:54if you do want to go deeper, like I
16:55mentioned, my community is in the
16:57description. And if you want to work
16:58one-on-one, there's also a link to book
17:00a call.

About this transcript

This page contains the full transcript of Claude Tested 100,000 Price Action Strategies (Here’s What Works) by AI Pathways, generated from the public captions YouTube serves with the video. The transcript has 3,737 words across 526 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.