YT YouTube2Text

YouTube2Text → Transcripts → I Tried to Improve fxalexg’s Trading Strategy

I Tried to Improve fxalexg’s Trading Strategy — Transcript

by Revelio Trading · 3,368 words · 532 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00In the previous video, we tested Alex
0:02Gonzalez's trading strategy with the
0:03exact same rules he shares on his
0:05YouTube channel. The verdict was not
0:08nice. $10,000 turned into $21
0:12after 10 years of trading.
0:14But there was one thing that made me
0:16believe this strategy could still be
0:17saved. The win rate was always around
0:2020% with an average risk to reward ratio
0:23of 1 to 3 and 1/2. To break even at that
0:26risk to reward, we need 22.2%.
0:30We were close. Close enough that maybe
0:33if we found a few bad rules that were
0:34dragging the whole thing down, this
0:36could actually become profitable. But we
0:39already investigated in video one, time
0:42frames, pairs, patterns, sessions,
0:44months. Nothing worked. So today, I'm
0:47bringing the big guns. I'm taking the
0:49strategy apart rule by rule to answer
0:52one simple question. Can we improve this
0:55trading strategy?
1:00When I looked at Alex's rules, two
1:01things immediately felt weird.
1:04First, every area of interest had to be
1:06between 5 and 60 pips.
1:09Think about what 60 pips means. A daily
1:12zone is not the same thing as a weekly
1:14zone. And on Euro pound, for example, 60
1:17pips is huge. Price can stay inside it
1:19for weeks. While on pound New Zealand
1:22dollar, price usually crosses 60 pips in
1:24a few hours.
1:26The same fixed number applied to
1:28different time frames and to markets
1:30that move completely differently.
1:32It's like having only one t-shirt size
1:34for every human on Earth.
1:37So the first repair was obvious. What if
1:39the zone width was based on volatility
1:41instead?
1:43For that, I used the ATR. Most of you
1:46are probably familiar with this
1:47indicator. It simply measures the
1:49average movement of the market. So
1:51instead of applying a fixed range, we
1:53have a range that adapts to each time
1:55frame and each currency pair.
1:58I tested three levels: 0.25 ATR, 0.5
2:02ATR, and 1 ATR.
2:05The return was bad for all the versions,
2:07but I want to focus on the profit
2:09factor.
2:11The profit factor simply measures how
2:12many dollars you make for every dollar
2:14you lose.
2:160.835
2:17means that the strategy makes 0.835
2:20dollars back for every 1 dollar lost.
2:23It's easy to understand that if this
2:25figure is above one, the strategy is
2:27making money, while if it's below one,
2:29the strategy is losing money.
2:32Think of the profit factor as the
2:33quality figure, while the number of
2:35trades is the quantity figure, and the
2:37return is the consequence.
2:40So, looking at the profit factor, there
2:42isn't a single version that makes money.
2:44They are all returning less than 1
2:46dollar for every dollar lost.
2:49So, the ATR doesn't seem to add anything
2:51meaningful to the strategy. The ATR 1
2:54loses less money simply because it takes
2:56less trades, certainly not because it
2:58has some sort of edge.
3:00But there is a second rule I found quite
3:02weird. An area of interest on the daily
3:05timeframe expires after 2 years, while
3:08an area of interest on the weekly
3:09timeframe expires after 5 years.
3:13So, a level that the market has
3:14respected for 6 years is removed because
3:17the area of interest had a birthday.
3:20While a level that the market has
3:21smashed through many times is still
3:23valid because it's young.
3:26It doesn't sound logical to me.
3:28So, I built a different concept. An area
3:31of interest is removed when the number
3:33of breaks catches up with the number of
3:35rejections.
3:36It sounds more logical to me, but the
3:38question is not whether it sounds more
3:40logical. The question is whether it
3:42makes money.
3:44So, I tested it across the different ATR
3:46values and the original pip range.
3:49Results didn't improve. They got worse
3:52for every variation, four to zero.
3:56I have to admit, I thought we would have
3:57seen at least a small improvement here,
4:00but I was wrong.
4:02So, we admit defeat. We trust the data,
4:05and we stick with Alex's rules on this.
4:07But, this is not a bad start. Because
4:10even if we didn't find anything here,
4:13these tests tell us that the problem is
4:15somewhere else. And this brings us to
4:17the next chapter, where I spent most of
4:20my days building something that could
4:22tear the strategy completely apart and
4:25rebuild it piece by piece.
4:31In the previous video, I said I wanted
4:32to use machine learning to analyze this
4:34strategy and see what works and what
4:37doesn't. But, to use the right tool of
4:39machine learning, we first need to
4:41understand the problem. We have so many
4:43rules that are interconnected. The trend
4:46on the higher time frame must be aligned
4:47with the trend on the time frame we are
4:49trading. The price needs to be in an
4:51area of interest, possibly retest that
4:53level. There must be a candlestick
4:55pattern and more. And we need that exact
4:58configuration to take a trade. But, with
5:01that exact configuration, we already saw
5:03that we lost money. So, I decided to use
5:06a technique that is technically not
5:08machine learning, but it's used to
5:10validate the most advanced machine
5:12learning systems on Earth. Its name is
5:15ablation.
5:16Now, you may think it's something super
5:18complicated. The beautiful part is that
5:20once you strip the math from it and only
5:22look at the concept, it's dead simple.
5:25To prove it, I have here my friend, star
5:27of one of my favorite TV shows, Dr.
5:29House. Imagine a patient walks into the
5:32hospital with several symptoms. House's
5:35team decides to treat the symptoms with
5:37five different pills as standard
5:39protocol. The patient seems to get
5:41better for a while, but then crashes.
5:43The team is brainstorming what to do.
5:46After all, they are following the normal
5:48trusted treatment. But, Dr. House
5:50doesn't trust anything. He asks the only
5:53question that matters. Which of these
5:56pills is actually doing something? Maybe
5:59one pill is the cure and the others are
6:01dead weight. Maybe only a combination of
6:03two or three pills works. Maybe they are
6:06all doing nothing. Or maybe one of the
6:08pills is the very thing killing the
6:11patient. So, House does something that
6:13sounds brutal, but is pure logic. He
6:16tells his team to remove all the
6:18treatments and only give the patient
6:20pill one. He tests every combination.
6:23Pill one alone, pills one and three,
6:26pill two, no pills at all, and so on.
6:29Every mix measured the same way with the
6:31question, does the patient get better or
6:34worse?
6:35In machine learning, this technique is
6:37called ablation. Our patient is Alex's
6:40strategy. It arrived already on its
6:42cocktail of five rules taken every day
6:45for 10 years. And we have watched it
6:48bleed through our two videos. So now,
6:51every rule becomes a pill. We can give
6:53it or take it away. And we test every
6:56combination. Here are the five pills.
6:59Rule one is the higher time frame bias.
7:02Alex's rule is that we need to have the
7:04weekly and daily time frame agreeing on
7:07the direction or the daily and the
7:094-hour. Rather than giving or removing
7:11the pill, here we test five different
7:14versions. Off means no higher time frame
7:17bias at all. Pair is the original rule.
7:20Then we have the weekly used as a bias,
7:22the daily, and the 4-hour. Rule two is
7:25that the trend on the chart where we
7:27trade needs to be in line with the
7:28direction of our trade. Rule three is
7:31that we take the trade on one of the
7:33candlestick patterns in Alex's list.
7:36Rule four is the retest logic. Something
7:38Alex only shares in a few of his videos
7:41and something some viewers pointed out
7:43in the comments. I decided to test it.
7:45I'm not going much into the details of
7:47this concept here because, spoiler
7:49alert, it turned out not to add anything
7:52meaningful. Rule five, the trading
7:54session. According to Alex, the best
7:57session to trade is the overlap between
7:59the London and New York session. For
8:01this rule, instead of just giving or
8:03removing the pill, we test five
8:06different settings. The Asia session,
8:08the London session, the New York
8:10session, the overlap between London and
8:12New York, and all the sessions. Every
8:15combination of these five rules gives us
8:17200 different configurations of the
8:20strategy. For now, I aggregated all the
8:22time frames together. We are going to
8:24dive deeper into time frames later. So,
8:27let's see the results of ablation.
8:31These are all 200 configurations ordered
8:34by profit factor from the highest to the
8:36lowest.
8:37Look at the winner.
8:39It has everything off.
8:41No higher time frame bias, no trend
8:44evaluation on the time frame where the
8:46trade is taken. It doesn't consider
8:48candlestick patterns to take the trade.
8:50It doesn't care about retest. And the
8:52best session was the overlap between
8:54London and New York.
8:56And credit where it's due, this is the
8:59session Alex recommends.
9:01This means that if you draw areas of
9:03interest on the daily and weekly time
9:05frames, and you wait for the overlap
9:07between the London and New York session,
9:10and then you simply enter the trade as
9:12soon as the price closes inside an area
9:14of interest, you outperform any other
9:17combination of the strategy.
9:19But, that's just the beginning of the
9:21analysis, because this table is pure
9:23gold in terms of findings.
9:26Another interesting thing that you might
9:28immediately spot, the bias doesn't seem
9:31to help at all. Most of the top
9:33combinations have the bias turned off.
9:36When a higher time frame bias is used at
9:38all, it's taken from the 4-hour chart.
9:41The first configuration using the bias
9:43Alex teaches, the weekly daily or daily
9:464-hour pair, is at number 38, and its
9:50profit factor is barely above one.
9:53This is very interesting, especially
9:55considering how popular higher time
9:57frame bias is among YouTube traders.
10:01This is the second strategy we dig deep
10:03into, and once again, the data says the
10:05higher time frame bias is hurting rather
10:08than helping.
10:09Food for thought.
10:11Another interesting thing to notice is
10:13that the variation that turns everything
10:15off consistently scores very high, not
10:18just in the overlap session.
10:21Yeah, sure, the overlap session steals
10:23the spotlight with its number one
10:25position, but the same variation during
10:27the Asia session ranks number five.
10:30During London, number eight. And
10:32considering all the sessions together,
10:34number 16 out of 200.
10:38All of them with a profit factor above
10:40one.
10:41So, yes, I could have gone digging into
10:44all the other variations, but it felt
10:46like we had a clear winner in front of
10:48us.
10:49Turning everything off and just taking
10:51the trade when price closes inside an
10:53area of interest seems to beat
10:56everything else.
10:57The question is, do we take the trades
11:00only in the overlap between London and
11:02New York or in all the sessions?
11:05These for me were the two variations of
11:07the strategy worth pursuing.
11:10Now, before we dive deeper into these
11:12two variations, it's worth mentioning
11:14something. What I showed you here are
11:16the figures for the whole period, from
11:18January 2016 to May 2026.
11:23In reality, in my work behind the
11:24scenes, I was working with a training
11:26data set that stopped at the end of
11:282021, while all the data from January
11:312022 to May 2026 stayed locked in a test
11:36data set.
11:37This is a common separation in
11:39backtesting. You never want to select
11:41your strategy on the whole period,
11:43especially when you're running 200 tests
11:46because some of them will come out
11:47profitable by pure luck.
11:50The thing is that the results were
11:51already clear in the training set. The
11:53combination with everything off during
11:56the Asia session came second.
11:58During the London New York overlap,
12:00sixth, and considering all sessions,
12:03ninth out of 200 combinations, all with
12:07a profit factor above 1.1.
12:10Asia session looks strong, but I didn't
12:12want to build the strategy around the
12:14session with the lowest liquidity. So,
12:16the London New York overlap looked more
12:18interesting to me.
12:20So, I had already locked these two
12:22variations based on the training set
12:24alone, and then I looked at what
12:25happened in the test set.
12:27The all sessions variation continued to
12:30deliver stable profitable results, and
12:33the overlap variation actually improved
12:35its edge in recent years. So, we had two
12:38serious candidates. One version that
12:40trades every session, and one version
12:43that trades only the London New York
12:45overlap.
12:46Both removed all the rules that were
12:48supposed to make Alex's strategy
12:50smarter. Both looked promising. Both
12:53looked like they might have a real edge,
12:55and not just be profitable by pure luck.
12:58In the next chapter, we dig deeper to
13:00find out which one is our winner.
13:06The first thing I wanted to dissect is
13:08the time frames. Not just to find out if
13:10one time frame performs better than the
13:12others, but also as a proof that we are
13:14not over-fitting. If most of the time
13:17frames are profitable, we are likely
13:19dealing with a real edge. If one or two
13:22time frames are massively out-performing
13:24while all the others are negative, we
13:26are probably dealing with overfitting.
13:29I analyzed the time frames based on
13:31where the area of interest was taken.
13:34For areas of interest taken on the
13:35weekly chart, I analyzed the 4-hour, the
13:382-hour, and the 1-hour time frames.
13:41For areas of interest taken on the daily
13:43chart, I analyzed the 4-hour, the
13:452-hour, the 1-hour, the 30-minute, and
13:48the 15-minute.
13:50The first good news is that out of 16
13:52combinations, the worst profit factor is
13:550.95,
13:57not far from break even. It's likely we
13:59are dealing with a real edge.
14:02Two patterns stand out. The first,
14:05levels drawn on the daily time frame
14:07outperform levels drawn on the weekly
14:09time frame.
14:10I aggregated the results to give you a
14:12better view. In both versions, trades
14:14taken on weekly areas of interest are
14:17barely around break even, while trades
14:19taken on daily areas of interest are
14:21clearly profitable.
14:23The second pattern that stands out is
14:25that for daily areas of interest, the
14:27lower we go with the time frame, the
14:29better the strategy gets.
14:31From around break even on the 4-hour
14:33time frame, step-by-step up to a profit
14:36factor of 1.35 on the 15-minute.
14:39The same happens with the overlap
14:41version of the strategy.
14:43At this point, both all sessions and
14:45overlap still look good, so they pass
14:48the test. But, I would promote daily
14:50levels to the main engine of the
14:51strategy. Drop the weekly levels and
14:54also drop the 4-hour time frame from the
14:56daily areas of interest.
14:58With this, we go to the next piece of
15:00the exploration, results by currency
15:03pairs.
15:04The goal is the same. We want to make
15:06sure the results are good across most
15:08currency pairs, not just on two lucky
15:11pairs.
15:12Since we have a lot of currency pairs, I
15:14summarize the data here. The all
15:16sessions version is profitable on 14 out
15:19of 16 pairs. The overlap version on 13
15:22out of 16.
15:24Once again, both versions look solid.
15:26So, we proceed to the next step, risk
15:29management. Alex's risk management is a
15:31simple 2% per trade. This sometimes
15:35creates a problem. You might end up
15:36losing money even if your strategy is
15:38profitable.
15:40This has to do with a concept called
15:41volatility drag. In simple words, the
15:44more you lose, the harder it gets to
15:46recover.
15:48A loss of 50% of the account needs a
15:50performance of plus 100% just to go back
15:53to break even.
15:55A loss of 80% of the account needs a
15:57plus 400% to break even.
16:00To overcome this problem, we will
16:01introduce the risk management technique
16:03called high watermark.
16:06Here's the idea.
16:07Normally, when you risk 2% per trade,
16:10you risk 2% of whatever is in the
16:12account right now. So, every loss makes
16:15your next trade smaller. And since your
16:17next trade is smaller, it is harder to
16:19recover.
16:20Imagine you are trading with $10,000 and
16:23you risk 2% per trade.
16:25You may think that after one loss and
16:27one win, you are breaking even.
16:30Wrong. When you lose, you lose 2% of
16:33$10,000. So, you lose $200 and your
16:36account is now $9,800.
16:40Now, you risk 2% of $9,800,
16:44which is $196.
16:46You win and your account goes to $9,996.
16:51You are still losing $4 despite being at
16:54one win and one loss.
16:56With the high watermark, you risk a
16:58percentage of the highest value your
16:59account has ever reached, not a
17:02percentage of the current value.
17:04So, following the previous example, we
17:06risk 2% of $10,000 for the first trade.
17:09We lose and the account goes to $9,800.
17:14For the second trade, we risk 2% of the
17:17highest value the account has reached.
17:19So, we don't risk 2% of $9,800,
17:22but 2% of $10,000.
17:25We win the trade, we win $200, and the
17:28account goes back to $10,000.
17:31So, we will run three tests. The
17:33original 2% per trade, 1% high
17:36watermark, and 0.5% high watermark.
17:41Results are positive for all the
17:42versions, but let's not celebrate yet.
17:45We still need to account for trading
17:47costs. So, let's run the same test, but
17:50this time with spreads, commissions, and
17:52swap rates taken from IC Markets.
17:56Well, still positive for all the
17:58versions.
18:00In this final table, we can declare our
18:02winner. After costs and risk management,
18:05the all sessions version had the best
18:07balance between return and drawdown, and
18:10also the higher profit factor.
18:13But, which exact configuration of risk
18:15management wins here?
18:17There's no winner. There's only personal
18:19preference. The 2% risk per trade looks
18:22appealing, but with a drawdown of minus
18:2474%,
18:26the risk of blowing your trading account
18:28is quite high. Personally, I would feel
18:31uncomfortable even with minus 57.6%
18:35as a drawdown. So, I would not choose
18:37the 1% of high water. I would settle for
18:40the 0.5% of high water, which gave a
18:43return of plus 493%
18:47over the past 10 years. Annualized
18:49return of 18.7%
18:52per year with a drawdown of minus 28.8%.
18:57Looking at the equity curve and return
18:59by year, we discover that the 18.7%
19:02annualized return is actually not evenly
19:05distributed. The edge was not that big
19:08till 2020, then it seems it's increasing
19:11in recent years, which is actually a
19:14good thing as it signals that market
19:16conditions are recently more favorable
19:18to trade this strategy.
19:20If we check the annualized return since
19:222021, it is about 39%
19:26much higher than 18.7%.
19:30I put a recap of the final strategy here
19:32on the screen. So, if you want to spend
19:34some time looking at all the final
19:35rules, you can pause the video.
19:38One last thing. I also tested the
19:40strategy on other assets like gold,
19:43Bitcoin, and indices.
19:45The results were not great and the
19:47number of trades was quite small. If we
19:49check the NASDAQ on a weekly chart, it's
19:52easy to understand why. We don't have
19:55many areas of interest because the price
19:57has a long-term upward drift and doesn't
20:00retest the same levels very often.
20:03So, I leave this work confined to Forex
20:05trading and it's time to answer the
20:07question we had at the beginning of the
20:09video. Did we improve this trading
20:11strategy?
20:12My opinion is yes, we did improve the
20:16original trading strategy and we
20:18actually built something that seems to
20:20have a real edge. But, I hope this
20:23research work gave you something more
20:24than just results. Something that can
20:26hopefully improve your own trading.
20:29There were many interesting points, but
20:31if I had to choose my key takeaway from
20:33this video, it would be this.
20:35Every rule must earn its place. Stacking
20:38rule after rule may look professional,
20:41but unless you have evidence that a rule
20:43improves results, don't add it.
20:46Complexity doesn't create an edge. It
20:49only creates the illusion of control.
20:52This work took a few weeks of obsessive
20:55non-stop daily research. So, if you made
20:58it to the end of the video, thank you
21:00very much for your support and I'll see
21:02you in the next one.
21:05>> I'm a fool.
21:12I'm a fool.

About this transcript

This page contains the full transcript of I Tried to Improve fxalexg’s Trading Strategy by Revelio Trading, generated from the public captions YouTube serves with the video. The transcript has 3,368 words across 532 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.