Yeah, I would say the y-axis range on charts should be set at "3-sigma likelihood of observation" thresholds. Not everything that's charted can be framed as sampling from a distribution, but the principle of manually setting chart ranges would nonetheless still apply.

For instance, if we're charting someone's body temperature, we would likely fix our y-axis to 80-110.

This is correct. GP’s rule should be adhered to for bar charts though. Absolute values there are expected. For line charts continuous (ish) over time, people are usually more interested in relative change in a time period so ok if y axis starts closer to the minimum value of the series in that range.

That’s the exact point?

No, it isn't. People are interested in whether a stock has volatility, and whether it has moved x%. If a $10 stock loses $1 it should show a roughly matching pattern to a $200 stock losing $20. Intersecting at zero will show a very large movement for the $10 stock and a very small movement in the $200 stock, despite the effect being the same to stockholders.

> Intersecting at zero will show a very large movement for the $10 stock and a very small movement in the $200 stock...

How so? Surely both will show a 10% movement?

Adding to this, stock prices should be plotted on a log scale since it is log returns that are roughly normal, and then 0 really makes no sense.

Zero is not arbitrary when it comes to the stock price.

It very much is, because stocks don't start at zero and your entry point will also not be zero. You can be deep in the red long before the stock gets anywhere close to zero. Even when a stock hits zero it doesn't always mean something useful, look at the Hertz bankruptcy for a good example. As far as stocks are concerned, zero is arbitrary and pretty universally a useless reference point.