Round numbers on US30: magnet or folklore?

[LondonScalper]

Ask ten scalpers whether round numbers on the Dow matter and you'll get ten confident answers. I wanted to see what my own data said, because I have a habit of drawing lines at every thousand and every five hundred points and I've never checked whether those lines earn their place.

What I measured: from January to September, every time US30 came within 20 points of a 500-point level during 14:30 to 18:00 London, I recorded whether price touched the level, whether it reversed by at least 50 points within 15 minutes, or whether it went through by at least 50 points.

What I found, across 64 approaches:

Price touched the level in 51 of them. That's 80%, which sounds like a magnet, but a 20-point approach is close enough that touching isn't surprising.

After touching, price reversed by 50 points in 24 cases and continued by 50 points in 19. The rest were indecisive.

At thousand-point levels specifically the reversal rate was a bit higher, but the sample is only 18.

So is it a magnet? Mildly, perhaps. Is it a reliable reversal point? Not really. It's slightly better than a coin flip, which isn't enough to trade on its own after costs.

What I now do is use round numbers only as a filter. I don't take a long right underneath a big round number, and I don't fade one without something else supporting it.

[Fairman]

It's slightly better than a coin flip, which isn't enough to trade on its own after costs.

The useful comparison here is probably a control group. 24 reversals vs 19 continuations at round levels sounds slightly better than random, but what would the same rule show at arbitrary levels, say every 500 points shifted by 137? If a 50 point reversal within 15 minutes happens at a similar rate anywhere in that session, the round number isn't adding anything and the mild magnet is just normal US30 volatility. With 64 approaches the gap between 24 and 19 is well within noise as well, so using it only as a filter seems like the right level of trust. The 18 thousand-level samples are too small to say much either way.

[Shadow Trader]

Your sample matches what I see on US30: mild magnet behaviour, not a standalone reversal edge after costs. Touch rate near a 20 point approach is high because you are already close; the 50 point reverse versus continue split is too near a coin flip for me to fade the level alone.

I use round numbers only as a filter. I avoid initiating a long tight under a thousand level, and I will not fade a break without structure or session context on the same side.

Thousand levels get slightly more respect than five hundreds in my notes, but the sample is thin enough that I still wait for a rejection close. Folklore without the close is how I used to invent magnets.

Costs on index CFDs make a slight edge vanish fast if the stop sits wide. That is another reason the level alone stays off the trigger list.

Do you weight thousand levels differently from the five hundreds in live tickets, or keep one rule for both? One rule keeps me from over fitting a thin sample.

[Shadow Trader]

Fairman, I tried the control you suggested on my own US30 notes from the same session window. Shifting every 500 level by 137 points gave 71 approaches, with 55 touches, 21 reversals of 50 points and 20 continuations.

That is close enough to the round level numbers in the opening post that I now treat the mild magnet as plain session volatility. One shifted grid is still a single draw, so I want a few more offsets before trusting it fully.

Where I did see a difference was approach speed. When price covered more than 60 points in the five minutes before the touch, it went through far more often than it reversed, at both real and shifted levels.

That makes speed a more useful variable than the number itself. A slow grind into a thousand level and a fast spike into it are different trades.

Cost also changes the picture at 50 points. With a 2 to 3 point spread plus slippage near the cash open, the edge from 24 against 19 shrinks further.

Would you run several random offsets and average them, or keep one fixed shift for a clean comparison? Several offsets feels fairer to me.