How to test any trading strategy with AI (no coding needed) — Transcript
Full transcript
- 0:00How do you find out whether a trading
- 0:01strategy works or not? There are two
- 0:04main ways. One costs you a few hours,
- 0:07the other costs you your account. Most
- 0:09traders pick the second one because the
- 0:11first one means either weeks of manual
- 0:13backtesting or coding skills they don't
- 0:15have. So, they do what everyone does.
- 0:18They find an idea on YouTube or come up
- 0:20with their own. Then they check four or
- 0:22five examples on TradingView, and if it
- 0:24looks good, they put money on it. Then
- 0:26the losses come, and they wonder, is
- 0:28this a normal losing streak? Does the
- 0:30strategy just not work? Most traders
- 0:33never find out. They move on to the next
- 0:35strategy, and the cycle starts again. No
- 0:38wonder over 70% of people lose money in
- 0:40trading, and only about 7% of people get
- 0:43a payout with prop firms. But for you,
- 0:46that ends today because in this video,
- 0:48I'm going to show you how to test any
- 0:50trading strategy, yours or one you
- 0:52learned on YouTube, without writing a
- 0:54single line of code. Next time someone
- 0:56shows you a strategy, you won't have to
- 0:58wonder. You'll check it yourself before
- 1:01you risk a dollar.
- 1:06To backtest our trading strategy, we are
- 1:07going to use an approach called Vibe
- 1:09Coding. Basically, you describe what you
- 1:11want in plain language, and the AI
- 1:13writes the code for you. But before
- 1:15backtesting our first trading strategy
- 1:17together, I want you to understand how
- 1:19the setup works because many people just
- 1:21open a chat in ChatGPT or Claude and
- 1:23type, "This is my strategy. Can you
- 1:25build the code for it?" Here, we are
- 1:27going to take things one step further so
- 1:29you can start exploring the potential of
- 1:31Vibe Coding. And you can use this
- 1:33approach beyond trading, whether you
- 1:35want to build a business or make
- 1:36something in your everyday life easier.
- 1:39The setup I'm going to show you has
- 1:40three main parts. The AI, which can be
- 1:43ChatGPT, Claude, Gemini, or whatever you
- 1:46prefer. Then we have Hermes, and then we
- 1:49have a service provider. Think of the AI
- 1:51as the brain. Hermes is the system that
- 1:53lets you put that brain into agents.
- 1:56Think of them as your AI employees.
- 1:58These employees need a place to work.
- 2:00The service provider gives them an
- 2:01office with the setup already taken care
- 2:04of.
- 2:04Now, everything you'll see in this
- 2:06video, you can replicate with just the
- 2:08first two parts for free. Hermes itself
- 2:10is free, and for the AI, you can use a
- 2:13free model, or if you already pay for a
- 2:16subscription like ChatGPT, you can
- 2:18connect it to Hermes at no extra cost.
- 2:21The third part, the office, is not free.
- 2:24Later, I'll show you the advantages of
- 2:26it and how much it costs, so you can
- 2:28decide for yourself whether it's worth
- 2:29it.
- 2:30So, what's [snorts] the advantage of
- 2:32working with an agent instead of a
- 2:33normal chat? There are many, but if I
- 2:36had to pick the most important one, I
- 2:38would choose continuity.
- 2:40A chat can store files and remember
- 2:42things about you, but everything stays
- 2:44inside that chat. And a chat can only
- 2:47hold so much. When it gets long, it can
- 2:49lose track of earlier details, or you
- 2:52have to start a fresh one. With an
- 2:53agent, all the data, the code, the
- 2:56reports, and the working procedures live
- 2:58in a workspace, which is a folder that
- 3:00exists outside any single chat. When we
- 3:03start a new chat to test a different
- 3:05strategy, the agent can reuse the
- 3:07relevant material from that workspace
- 3:09instead of rebuilding the whole process
- 3:11from scratch. And the best thing is that
- 3:13it keeps an internal notebook of what
- 3:15worked and what went wrong, so it learns
- 3:17from the previous work and can
- 3:19self-improve over time, just like a real
- 3:21employee getting experience working for
- 3:23you.
- 3:24On top of that, agents can also be
- 3:26coordinated and work at the same time,
- 3:28sharing the same workspace. One can
- 3:30research trading ideas, one can prepare
- 3:33reports, one can analyze your trading
- 3:35journal. You can build your own trading
- 3:37desk with AI employees working for you.
- 3:40But why Hermes? Can't we build agents in
- 3:43ChatGPT or Claude? Absolutely, you can.
- 3:47But imagine you build your agents in
- 3:48ChatGPT, and then Claude releases a much
- 3:51more powerful model, or maybe ChatGPT
- 3:54raises the price of its subscription.
- 3:56You want to move, but all your work is
- 3:58in ChatGPT, and moving it takes time and
- 4:01knowledge you may not have. With Hermes,
- 4:04you can use ChatGPT or Claude or any
- 4:06other model and swap them without
- 4:08starting over because all your work is
- 4:10saved in a workspace outside the AI.
- 4:14The third part of the setup is the
- 4:15service provider, which gives your
- 4:17agents the office and the setup. This
- 4:19brings us to the sponsor of this video,
- 4:21Hostinger. But, I'm not going to bore
- 4:23you with the usual 1-minute talk trying
- 4:25to brainwash you about how great
- 4:27Hostinger is. I prefer to show you along
- 4:29the way why I personally use Hostinger
- 4:32and why I think it's great value,
- 4:34especially for beginners. By the end of
- 4:36this video, you will have a clearer idea
- 4:38of whether Hostinger is for you or not.
- 4:41Hostinger provides a product called
- 4:42managed Hermes. It runs Hermes on
- 4:45Hostinger's server and gives you a
- 4:46beginner-friendly interface to manage
- 4:48it. You can install Hermes, update it,
- 4:51restore a backup, connect it to Telegram
- 4:53so you can talk to your agents from your
- 4:54phone, and much more. And Hostinger
- 4:57takes care of the setup so you can
- 4:58manage all these features with a few
- 5:00clicks rather than working from your
- 5:02terminal. So, with Hostinger's managed
- 5:04Hermes, no technical expertise is
- 5:06needed.
- 5:07Let's install Hermes so we can start
- 5:09vibe coding our first trading strategy.
- 5:11Here, we're inside Hostinger. On the
- 5:14left side, we can find Hermes agent and
- 5:16all we have to do is click on setup. It
- 5:19will ask you which app you want to
- 5:20install and we choose Hermes agent web
- 5:22UI.
- 5:24Now, it lets you choose the brain you
- 5:25want to give to your agent. You can
- 5:27always change this later. For now, I'm
- 5:29going to select nexos.ai,
- 5:31but we will come back to this later on.
- 5:34Here, it asks if I want to connect
- 5:35Telegram so the agents can send me
- 5:37updates on Telegram, but I'm going to
- 5:39skip this for now.
- 5:41Done. Hostinger created your access to
- 5:43Hermes. This is the random name it
- 5:45assigned to it and before clicking on
- 5:47open app, I recommend you copy the
- 5:49password and store it somewhere safe.
- 5:51You will need it whenever you want to
- 5:53log in to Hermes. After that, we can
- 5:55open the app, insert the password, and
- 5:57we are in. Welcome to Hermes.
- 6:00It looks very similar to most AI
- 6:02interfaces. At the bottom, you can
- 6:04choose the model you want to use. There
- 6:06is a list of popular models from Claude,
- 6:08Deep Seek, Gemini, GPT, and more, but
- 6:11you can also import other models if you
- 6:13want.
- 6:14Then, you can select the effort, just
- 6:16like most modern AIs.
- 6:18Now, how to Vibe code our first trading
- 6:21strategy? What do we need to let our
- 6:23employee code the trading strategy for
- 6:24us? We need precise instructions. I have
- 6:28made a text file with instructions for
- 6:30the agent. You can download it for free
- 6:32on GitHub. The link is in the
- 6:33description of this video. Once you
- 6:35click on the link, you can either copy
- 6:37the prompt or download it. I'm going to
- 6:39copy it, paste it into my Hermes chat,
- 6:42and I can finally start the process of
- 6:44testing my trading strategy.
- 6:46Now, in a normal chat, you ask
- 6:48something, the AI replies, then you
- 6:50continue. It's you driving the
- 6:52conversation. With the agent, you
- 6:54immediately see that it's different. You
- 6:57give instructions about what you want to
- 6:58get done, and the agent starts to ask
- 7:00you questions to get the information it
- 7:03needs to complete its job. So, it's the
- 7:05agent driving the conversation. I'm
- 7:07going to fast forward this so I can show
- 7:09you the relevant parts.
- 7:11The first thing the agent asks me is my
- 7:13programming level. If I'm a beginner, it
- 7:16will handle all the technical work and
- 7:17routine fixes. If I'm intermediate, it
- 7:20will also show me relevant code and
- 7:22recommend solutions. And if I'm
- 7:24advanced, it shows me the code and lets
- 7:26me approve it or change it before it
- 7:28goes ahead. Since we are doing this from
- 7:30a beginner's perspective, I'm going to
- 7:31select beginner.
- 7:33Which asset or market do you want to
- 7:35test? Let's say I want to test my
- 7:37strategy on the NASDAQ. I already have
- 7:40the data from Dukascopy, so I specify
- 7:42that, too.
- 7:43Now, the time frame. Let's work on the
- 7:4515-minute time frame.
- 7:47Now, it asks me about the period I want
- 7:49to test. Let's do from January 2016 till
- 7:52today, since I usually use this period
- 7:54in my other backtesting videos.
- 7:57It gives me a quick summary of the
- 7:58choices so far, and it asks me to upload
- 8:01the data since I said I already have it.
- 8:03Next, it starts asking about my
- 8:05strategy. What makes me enter a trade?
- 8:08What makes me exit? And whether I would
- 8:10like to test the whole period or split
- 8:12it into a training period and a test
- 8:14period, which is a very good practice
- 8:16when testing and optimizing a trading
- 8:18strategy.
- 8:19I'm not going to spend much time on the
- 8:21strategy itself because the goal of this
- 8:23video is not to share the best trading
- 8:25strategy you'll ever see.
- 8:27The goal is to understand how to vibe
- 8:29code your own trading strategy and how
- 8:31to test it. I'm going to use a simple
- 8:33strategy that buys the Nasdaq every time
- 8:36the price crosses the 50 period simple
- 8:38moving average from below. For the exit,
- 8:41I'm going to have a stop loss at minus
- 8:43two ATR and then trail my stop loss
- 8:45every time the price goes up by three
- 8:47ATR. And I'm going to test the whole
- 8:50period since this test is just to show
- 8:52you how it works.
- 8:53So, it's like an interview about your
- 8:55trading strategy. Next, it's going to
- 8:57ask about the risk per trade, the
- 8:59trading costs, and so on. We can fast
- 9:01forward this part. It's really just
- 9:03about answering questions about the
- 9:04trading strategy you want to test.
- 9:07After this, it gives me a final strategy
- 9:09recap for me to approve. It also
- 9:11includes two examples of trades from
- 9:13real charts, so you can verify whether
- 9:15it replicates what your strategy does.
- 9:18You can either verify these examples on
- 9:20your charts like TradingView or you can
- 9:22open the workspace here on the right
- 9:24side and follow the path the agent gave
- 9:26you. So, trading research, strategies,
- 9:30Nasdaq, examples, and let's check the
- 9:33image of the first example. I can open
- 9:35it in the browser or download it so I
- 9:37can see it at full size.
- 9:39Yeah, it looks like it does what it's
- 9:41supposed to do. It entered at the
- 9:42crossover between the price and the
- 9:44moving average. It set an initial stop
- 9:46loss two ATR below the entry. Then,
- 9:49after the price went up three ATR, it
- 9:51raised the stop, and then we got stopped
- 9:54out at the new stop loss. So, I'm going
- 9:56to tell the agent that everything looks
- 9:58good and it can proceed with coding the
- 9:59strategy and testing it.
- 10:02And here are the results. $100,000
- 10:05turned into $2,128,888.30
- 10:11with an annualized return of 32.65%.
- 10:15The strategy looks good until we see the
- 10:17drawdown, which is 74.07%
- 10:21too high to justify this kind of return.
- 10:24And you can see that the profit factor
- 10:26is barely above one.
- 10:28The agent also gives you a summary of
- 10:30what seems to work in the strategy and
- 10:32what doesn't. But you can also retrieve
- 10:34the full report from your workspace.
- 10:36Hermes gives you the path here. In the
- 10:38workspace, you won't just find the
- 10:40report, but also all the data, the code
- 10:43the AI built, the trading journal with
- 10:45all the details of every single trade
- 10:47taken during the backtest, and more.
- 10:50So, where is all of this stored?
- 10:52Without Hostinger, it's on your
- 10:54computer. With Hostinger, it's on their
- 10:57server. And that's why I personally use
- 10:59it. Having these files on Hostinger
- 11:02rather than on my computer means three
- 11:04things. First, I can work from anywhere.
- 11:07Today, I'm on my home computer.
- 11:09Tomorrow, I can log in from my laptop at
- 11:11the airport or at a coffee shop and pick
- 11:13up exactly where I left off because the
- 11:15work is on Hostinger, not on my home
- 11:17computer. I don't need to download or
- 11:19transfer anything. Second, my computer
- 11:22stays clean. All these files in the
- 11:25workspace can add up to several
- 11:26gigabytes, and I don't want them to take
- 11:28up space on my computer. Managed Hermes
- 11:31comes with 25 GB of storage included in
- 11:33the price. Third, if I want to leave a
- 11:36task running, I don't have to leave my
- 11:38computer on. My agent can continue
- 11:40working on Hostinger even when my
- 11:42computer is off. So, how much does all
- 11:44of this cost? They're currently running
- 11:46a Black Friday sale with up to 85% off
- 11:49selected plans. The 24-month option goes
- 11:52for $4.99 a month. There's a 30-day
- 11:55money-back guarantee, and if you click
- 11:57the link in the description and apply
- 11:59the promo code Revelio, you get an
- 12:01additional 10% discount on top of the
- 12:04Black Friday deal. And you support the
- 12:06channel, too. So, thank you very much,
- 12:08and thank you, Hostinger, for sponsoring
- 12:10this video. So, back to our report. It
- 12:13gives you a first summary of the
- 12:14strategy. Then, scrolling down, you can
- 12:17find more detailed results and metrics,
- 12:19then an equity chart, a drawdown chart,
- 12:21and more.
- 12:23All of this is customizable. As I said
- 12:25earlier, the agent keeps an internal
- 12:27notebook. So, when you tell it something
- 12:29that you don't like, it will remember it
- 12:31and will apply your changes next time.
- 12:33For example, I would like to compare my
- 12:35equity curve with a simple buy and hold
- 12:38equity curve, so I can see the
- 12:39difference on the chart. I'll tell my
- 12:41agent later, so the next report will
- 12:43include it.
- 12:44The more time you spend with the agent,
- 12:46the more it learns your workflow and
- 12:48your preferences.
- 12:50The report also includes results year by
- 12:52year. There's an assessment of what
- 12:54helped and what hurt the strategy, so
- 12:56you can try to improve it, and more.
- 12:59Now, let's move to the next part,
- 13:01optimization. I always recommend
- 13:03splitting the period into a training
- 13:05sample and a test sample, so you
- 13:07optimize on the training sample before
- 13:10you open the test sample. But, since
- 13:12this is just a demonstration, I have
- 13:14used the whole period. So, I'm also
- 13:16going to run the optimization on the
- 13:17whole period. The agent suggests three
- 13:20potential areas of improvements we can
- 13:22test. I'm going to take two of them,
- 13:24because I didn't like the drawdown. It
- 13:26was too high. And, I also didn't like
- 13:29that we paid a lot in overnight fees.
- 13:31So, I'm going to ask my agent to test
- 13:33what happens if we reduce the risk from
- 13:351% to 0.5% of high watermark, and if we
- 13:39close the trades right before paying
- 13:41overnight fees, and reopen them right
- 13:44after.
- 13:45And, here we have the report of our
- 13:47optimized strategy. It makes much less
- 13:49money, but with less than half the
- 13:51drawdown we had before. Overall, not a
- 13:54trading strategy I would use in my
- 13:56portfolio. Of course, we can keep
- 13:58optimizing based on the suggestions the
- 14:00agent gives us, or testing our own
- 14:02ideas. But, that's not the point of this
- 14:04video. I wanted to show you how to run
- 14:06the whole framework without knowing how
- 14:09to code.
- 14:10Now, let's talk about how much this
- 14:11costs. As mentioned, Hermes itself costs
- 14:15nothing, but using the AI inside Hermes
- 14:18usually does. There are three main ways
- 14:20to use an AI model inside Hermes. The
- 14:23first way is to connect Hermes to a free
- 14:25model, so you don't have to pay a cent
- 14:27to use it. There are several free models
- 14:30you can use, including Hermes' native
- 14:32ones. Hostinger once again makes things
- 14:35easy, and you can connect to Gemini in a
- 14:37few simple steps. You just need to click
- 14:39on Hermes agent, then manage
- 14:42AI provider, and you will find Gemini
- 14:44here. It will ask you to enter your API
- 14:47key, which is just a long password
- 14:49Google gives you, so Hermes can use
- 14:51Gemini. You can find it by clicking this
- 14:53link here.
- 14:55I already have one, and you can see it
- 14:57set on the free tier, which allows me to
- 14:59use Gemini's free models in Hermes. If
- 15:02you don't have one, you can click here
- 15:04and it will create one for you. Once you
- 15:06have it, you can copy the key and paste
- 15:08it into Hostinger. Click save and
- 15:10update, and you're ready to use Hermes
- 15:12with Gemini for free.
- 15:14Now, free doesn't mean unlimited. Free
- 15:17models have restrictions on the volume
- 15:19of requests you can send to the AI. For
- 15:21normal chatting and tasks, free models
- 15:24are great, and you rarely hit their
- 15:25limits. For an agent that has to
- 15:27understand your trading strategy,
- 15:29download the data, clean it, turn your
- 15:32strategy into precise rules, write the
- 15:34code, test it, optimize it, and create
- 15:37reports for you, you can understand
- 15:39that's a lot, and free models are not
- 15:42always capable of handling all of this
- 15:44before hitting their limit and stop
- 15:45working for the day.
- 15:47This brings us to the second way, which
- 15:49is to connect Hermes to your ChatGPT
- 15:51subscription. To do that, you need
- 15:54either a plus or pro subscription with
- 15:56ChatGPT. Once again, Hostinger makes
- 15:58things easy. You can connect to your
- 16:00ChatGPT with just a click, and then
- 16:03follow the simple instructions on
- 16:04screen.
- 16:05Once you connect it, you can use it
- 16:07within the limits of your ChatGPT
- 16:09subscription. You can still use ChatGPT
- 16:11normally in its own app. The two just
- 16:14share the same allowance.
- 16:16The last way is to use credits. You put
- 16:18some money in, select any model, and pay
- 16:21as you go. When we installed Hermes, I
- 16:23chose Nexo's.ai Hostinger AI router,
- 16:27which uses Hostinger AI credits. I
- 16:29tested different AI models with credits,
- 16:31and here is how much it cost me to do
- 16:33all this work with different models.
- 16:36Astra and Fable may look expensive, but
- 16:39they did an excellent job. Also, in
- 16:41terms of presentation and improvement
- 16:43suggested. And in any case, $34 is still
- 16:47much less than a programmer would charge
- 16:49you for coding and testing a strategy.
- 16:51GPT-6 Luna did a decent job. It made an
- 16:54assumption that didn't match the
- 16:56strategy I had in mind. So, the first
- 16:58backtest wasn't testing exactly what I
- 17:00wanted. After going back through the
- 17:02rules with the AI, we found the mistake
- 17:04and built the right code. For 45 cents,
- 17:07I can honestly say it provided good
- 17:09value for money.
- 17:11What I showed you in this video is only
- 17:14the beginning of what you can do with
- 17:15this setup. I moved to Hermes from Open
- 17:18Claw myself only last month. I'm still
- 17:21learning every day, and it feels like
- 17:23I've barely scratched the surface.
- 17:25I hope this video gave you what you
- 17:27needed to take action and start vibe
- 17:29coding your first trading strategy. If
- 17:31you decide to give Hostinger a try, you
- 17:33can support the channel by using my link
- 17:35in the video description and applying
- 17:37the promo code "revelio" for an
- 17:39additional 10% discount. Thank you,
- 17:41Hostinger, for sponsoring this video,
- 17:43and thank you all for supporting the
- 17:45channel. I'll see you in the next video.
About this transcript
This page contains the full transcript of How to test any trading strategy with AI (no coding needed) by Revelio Trading, generated from the public captions YouTube serves with the video. The transcript has 3,331 words across 490 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 choose a caption track available on the video.
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, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.