I suspect we are just seeing the randomness. And perhaps observation bias. And surely false accuracy.

One of the days of the month is going to be best. Statistically there is a 3/21 chance that one of the last 3 trading days will show as the best. We may be just seeing what we want to see.

Here's the false accuracy bit.

From GTR1 for the RS52 5 HTD12 screen 21 day holding period.
The CAGRs for the 21 cycles 2/1/1985 - 11/28/2025 range from 24.1% to 29.6%. That is a 5.5% range.

For 9/1/2006 - 11/28/2025 the CAGRs range from 19.5% to 26.1%. That's a 6.6% range.

I bet that any day-of-month benefit is lost in this range of returns.

Heck, any of us would be ecstatic to get the lowest of these -- 19.5% CAGR for ten years.

What does Chatgpt say about the best improvement you'd get by trading the optimal day of the month? Is it just 2 numbers that are in the above range?

If so, we are probably just looking at a blip in the random noise.

I have never purposely used Chatgpt, from what google AI (yes, I appreciate the irony) says, it "is trained by fed massive amounts of data from the internet (books, articles, websites, and code)."

So what is the chance that it is simply echoing what it has read from internet posts, articles, and papers?