Video Transcript:
This is nonsense. Nonsense. Nonsense. I don’t know. Is this random or not? Completely wild. The win rate is insane. Perfect reaction to the pip. Beautiful reaction. You know the logic behind this trade. I don’t know where it comes from. If you asked me to guess the outcome before we even started, I would have said it would be random.
Hello everyone, it’s Dale here, and I decided to test whether ChatGPT can predict the market and provide solid support and resistance levels. I wanted to try this with the free ChatGPT version. That means it has no history and no training. And the result? Well, let’s take a look.
Here’s how we do this. First, I upload a random screenshot of a chart. ChatGPT then gives me support and resistance levels. After that, we fast forward the chart to see how those levels worked. We’ll repeat this a couple of times to see how good ChatGPT is at predicting the market. So, first, I grab a random screenshot of a chart. What you see here is the NinjaTrader 8 platform, a 30-minute chart of the EUR/USD. I’ve loaded some data and picked a random spot to capture. Now I upload the screenshot to ChatGPT, using the free version, which has no history, no training, no idea who I am, nothing at all. It’s basically like an anonymous user. Then I give it a very simple prompt:
“You are an expert Volume Profile trader. Take a look at this chart and give me strong support and resistance levels to trade from in the future. Provide exact values. Give a short explanation behind each support or resistance level. Be brief. No fluff.”
ChatGPT gives me the levels. I plot them on my chart: green for supports, red for resistances. Now let’s see how the price reacted to those levels in the future. Right now, we’re at support 1.1110. The price should go up, right?
And believe it or not, it does. First trade: winner. Honestly, a bit surprising to me. The reasoning ChatGPT gave was “local point of control” and “consolidation.” That’s true, but personally, I wouldn’t have picked this trade. Still, it worked. One winner. Let’s delete this level. Next, price hits the first resistance. It should go down. But it doesn’t. And that doesn’t surprise me because the level isn’t based on any volumes—nothing to trade from. Looks random. Loser. So now we have one winner, one loser. Let’s continue.
Next resistance hit—another loser. Price blew right through it. Though honestly, this level wasn’t terrible; there was a heavy volume zone followed by a sell-off. I might have considered it myself, though likely passed on it due to the weak high. Still, not a completely bad idea. That makes one winner, two losers. Next level—hit. Perfect reaction! The reasoning said “high volume node,” but that doesn’t quite make sense. The area had already been tested, and there wasn’t really a proper node where it pointed. Still, the reaction was there. Winner. So now we’re at two winners, two losers.
Some supports remained untested. Funny enough, one of them was picked below the visible chart area, meaning ChatGPT couldn’t even see the price action there. It still called it a “low volume rejection zone from May 1st.” That’s wild because the data wasn’t visible. Shooting in the dark. Random, but also accurate historically. So, we need more testing.
I delete untested levels and pick another random area. Upload another screenshot, same prompt. It gives new levels. I plot them again: red resistances, green supports. This time, one resistance was above the price, which doesn’t make sense, so I deleted it. The rest: price hits first resistance, drops about 20 pips. I’ll count that as a winner, even if it feels more like luck. Four winners, two losers. Next resistance hit—reaction again, surprisingly strong. I didn’t see heavy volume there, but the price reacted. Winner. Next, price goes to support. Random looking level, but strong reaction. Winner. Now we’re at six winners, two losers. The logic behind the calls doesn’t make sense, but the reactions keep coming.
Fast forward again. Another test, another random chart spot. More levels. One resistance gets broken—loser. Another, minor reaction then break—loser. But support hit—beautiful bounce. Winner. Reasoning said “major high volume node,” but honestly, the placement didn’t quite line up. Still, reaction worked. Then the last support—perfect reaction to the pip. Winner. Now I’m really surprised. Some reactions are unreal. Win rate at this stage looks great.
We repeat more tests, and by the end of the session, I counted 10 winners and 10 losers. Exactly 50%. Random. Some levels worked beautifully, to the pip, while others completely failed. And the reasoning? Mostly nonsense. It made up things about “nodes” or “defenses” that didn’t exist.
So here’s my conclusion: free, untrained ChatGPT gives you random levels. Sometimes they work, sometimes they don’t. That’s no edge. But this experiment showed me something important: with training, ChatGPT can probably be turned into a useful tool that provides logical, high-quality levels with a real edge.
This test version wasn’t trained, but you can train it. That’s where the potential is. If this video gets at least 500 likes, I’ll make a follow-up video showing how to actually train ChatGPT to provide proper levels that make sense and produce solid results.
I hope you enjoyed this little experiment. See you in the next one. Until then—happy trading.
Comments
Post a Comment