---
title: 'I Automated My Trading Strategy — Here''s What Went Wrong'
source: 'https://youtube.com/watch?v=sAteGzzysq8'
video_id: 'sAteGzzysq8'
date: 2026-09-06
duration_sec: 493
channel: 'The Moving Average'
---

# I Automated My Trading Strategy — Here's What Went Wrong

> Source: [I Automated My Trading Strategy — Here's What Went Wrong](https://youtube.com/watch?v=sAteGzzysq8)

## Summary

The video documents an experiment where the creator automated a manual RSI divergence and Fibonacci retracement trading strategy into a TradingView indicator and then a MetaTrader 5 (MT5) expert advisor (EA). The goal was to test whether the automated bot would replicate the exact trades the creator would take manually. The experiment revealed significant discrepancies between the indicator and the bot, highlighting the challenges of translating a visual, discretionary strategy into precise computer code.

### Key Points

- **Experiment Setup** [00:00] — The creator took a manual trading strategy, turned it into a TradingView indicator, and then converted the same rules into an automated MT5 bot to compare their outputs.
- **Unexpected Bot Behavior** [00:44] — The bot initially appeared to work—taking trades and showing profitable backtests—but on closer inspection, it wasn't always following the intended strategy, revealing critical errors.
- **Strategy Description** [01:13] — The strategy combines hidden RSI divergences with a Fibonacci retracement, looking for a specific relationship between price, RSI, and the 0.618 level.
- **Build an Indicator First** [01:44] — The creator advises using AI to build a TradingView indicator first, not a bot, so the computer can visually show its interpretation of the strategy for comparison.
- **Iterative Refinement** [02:11] — By comparing the indicator's signals to manual ones, the creator iteratively adjusted the rules until the indicator closely matched his visual interpretation.
- **MT5 Translation** [03:04] — Translating the rules to MT5 for longer backtesting initially seemed successful, but the bot's lack of visual context made it hard to verify its logic.
- **Trade Discrepancies** [03:54] — Comparing trades between TradingView and MT5 revealed mismatches—some trades were taken by one but not the other, despite supposedly identical rules.
- **Swing High/Low Ambiguity** [04:12] — A key issue was how the computer defined swing highs and lows, especially with equal highs on adjacent candles, which affected RSI comparisons and Fibonacci levels.
- **Hidden Errors** [05:05] — The bot's first version had a 75% win rate over 8 years, but inspection showed winners were smaller than losers, and the structure didn't match the original indicator.
- **Recommendation to Learn and Visualize** [06:29] — The creator emphasizes mastering a strategy visually first, then using AI to create an indicator that shows its work, before attempting full automation.
- **Mastery and AI Prompting** [07:27] — He advises explaining the strategy simply (like to a five-year-old) and using that detailed explanation to prompt AI tools like ChatGPT or Grok.

### Conclusion

The experiment underscores the importance of validating automated trading systems against manual visual checks. The creator plans to release a detailed tutorial on his exact process for structuring and prompting AI to build accurate indicators.

## Transcript

I ran an experiment. I took one of my trading strategies, turned it into a trading view indicator, made sure it was finding the exact trades that I wanted to, so it had a one-to-one visual representation of what I see when I look at the chart and what the indicator is telling me.
And then I took the exact same rules and turned them into a completely automated trading bot on MetaTrader. Same strategy, same entries, same stop losses, same take profits. At least,
That's what I saw. Because when I started testing the bot, something really strange happened. It worked. It was taking trades. The backtest was profitable. Some of the numbers actually looked really good. But when I started going through the individual trades, I realized the
bot wasn't always trading the strategy I had built. And one of the mistakes it made was so ridiculous that it completely changed the way I think about automating trading strategies. Now, let me show you what happened. The strategy itself isn't really the important aspect of this video,
but I'll explain it quickly so you can understand what the computer was trying to do. I made a previous video about this, and I've traded RSI divergences for years. One particular setup that I like combines hidden RSI divergences with a Fibonacci retracement.
Basically, there's a move in the market, price pulls back, and I'm looking for a very specific relationship between price, RSI, and the 0.618 Fibonacci level. When everything lines up, that's my trade. Now, normally when I'm trading this myself, it's incredibly visual. I can look
at the chart and say, that's the move. That's the high, that's the low, that's the retracement, and that's the divergence. I look at how the candle looks at the Fibonacci level, and that's my entry. And that's important because before I automate anything, I first built it as an indicator
on trading view. And if you're thinking about using AI to build your own trading system, I really think that you should do this first. Don't immediately ask AI to build you a trading bot. Ask it to build you an indicator that shows what it thinks your strategy is.
Because what you tell it and how it interprets it two completely different things Now the computer has to show its work This was basically my first test I explained my strategy and had it plot what I see visually on the chart
Whenever the computer thought there was a setup, it showed it to me. And I could sit there looking at the chart and compare what the computer saw against what I saw. Sometimes I'd look at a signal and think, yep, that's perfect. That's exactly what I would have traded.
Then I'd find another one and think, no, no, that's not right. So I'd figure out why it was showing me that, change the rules, run it again, and slowly the computer started seeing the chart more like I did. That's one of the reasons I find AI to be extremely, extremely, incredibly useful.
I'm not necessarily asking the AI to find me a profitable strategy. I'm saying, here is what I already do. Help me explain it precisely enough so that a computer can do it for me. Those are two very different things.
Eventually, I got the trading view indicator pretty damn close to what I see. So now came the actual experiment. I wanted to see what would happen if I took those rules and gave them to MT5, because now I could turn it into an expert advisor, an EA,
and test the strategy mechanically longer across more years of market data. So I translated everything over, and initially, it looked like it worked. MT5 started taking trades.
I had entries, stops, targets, winners, losers, an equity curve, all the stuff you normally look at when backtesting a strategy. But there was one massive difference. I couldn't necessarily see its thought process nearly as clearly anymore.
It just showed me entries. On TradingView, when I saw the trade, I could see why the trade existed. It had the start of the impulse move, the end of the impulse move, the Fibonacci retracement level, the level where it broke the previous structure,
the take profit, and the stop loss, all on a single indicator. On NT5, I basically saw sell, and then it trades. And that's a problem because I started comparing the individual trades between what TradingView was giving me and what MT5 was giving me,
and some of them matched almost perfectly. Same setup, almost identical entry, same stop, same target, beautiful. But then TradingView would take a trade and MT5 wouldn Or MT5 would take a trade and TradingView didn And I remember I given the exact same strategy So how is this possible
It turned out one of the problems was something ridiculously small. The swing highs and the swing lows. To a human, this seems obvious. You look at a chart and there's a swing high. But the computer doesn't see a chart like you do.
Imagine two candles next to each other with the exact same high point. Which one counts? The first candle? The second one? Both? Neither? that tiny decision matters. Because if the computer chooses a different high, now it's comparing RSI to a different point.
That can change completely where the divergence exists. That can change where the impulse is measured. And that changes the Fibonacci level. One tiny difference at the beginning can produce an entirely different trade at the end.
And that's when I started realizing how dangerous this was all becoming. One tiny little fraction decimal point of an error blows the strategy. And the issue is I wouldn't necessarily know what's happening. Because if I had just looked at the MP5 results, I wouldn't necessarily know any of this was happening.
And this wasn't even the biggest problem. My first version entered trades at a market execution once the setup was confirmed. And then I tested that version for like eight years. It had about a 75% win rate. And now imagine that's where I stopped.
Imagine I posted the screenshot online, trading bot gives 75% win rate, looks incredible. Except I started inspecting the actual trades. the winners were much smaller than the losers. Nothing lined up to the original trading view indicator
where it had perfect structure, perfect placement for the stop loss, perfect placement for the take profit. Everything was ideal. And this was taking the same concept,
but completely skewing the structure of where the impulse move was measured to and from. And this is why I wanted to run this experiment and make this video. Because a lot of you guys, are new into trading. You are
interested in making a trading view indicator You want to create the perfect setup for your daily trading And the issue is you must first learn that specific setup so that you can visually see it on the chart Then once you have a concrete idea of what it looks like do your best to translate it into something a computer can understand and have it make version 1 version 2 version 15 until it matches exactly what you see
And it shows its work on the indicator. It shows you the exact levels, what your structure is, what the price movement is, how the RSI divergences are working, what's bullish structure look like.
All of those things need to be visually represented on the chart. That's why I always recommend start with a trading view indicator. Because a lot of you guys are going to be tempted to immediately go and do an MT5 EA, automate everything, walk away.
That is going to ruin your life entirely and blow account after account after account. Within the next month or so, because this is extremely important to me, I'm going to do a tutorial video that's like an hour long.
My exact process of how I'm going about structuring this and how I'm prompting AI to get what I'm thinking into its brain. But for now, what I want you guys to learn, whatever strategy you're working on, or if you haven't learned one yet, learn one.
Then, master it. Absolutely master it. Get something that makes sense to you, so much so that you can explain it to a five-year-old. Then, make an extremely wordy version of that explanation to the five-year-old and start talking to your AI. Whether you're using ChatGPT, whether you're using Grok, it doesn't matter.
what I want you to do is use TradingView, use AI, and use your knowledge in a cohesive environment where you're getting almost a one-to-one depiction on your TradingView chart of what your eyes
visually see. Let me know in the comment section if you actually want to see that hour-long tutorial. If this video was helpful for you, please share it with your friends. Thank you so much for watching and we will see you in the next one.