Tl;dr - using monthly vs yearly data for making estimates gives me significantly different results. Which should I pick?

A person recently told me they expect double-digit unemployment in the US in the next two years. Being a good little forecaster, I went to the historical rate for guidance:

https://fred.stlouisfed.org/series/UNRATE

I count 16 months (1.33 years) out of the last 74 years having double-digit unemployment. This is a base rate of 0.018 months/year having double-digit unemployment.

Plugging this into https://stattrek.com/online-calculator/binomial.aspx , I get a 27% chance that by end of 2023 we'll have a month with double-digit unemployment.

This seems high to me.

If we look at this from a yearly perspective rather than a monthly perspective, we can say 4 out of the last 74 years had a month with double-digit unemployment (=5.4% of years). Then, with the same calculator, the chances of one of the next 2 years being a year where there was double-digit unemployment becomes 10.4%.

I'm confused about which of these base rates is the "right" base rate in this case.

Is the cause of the differences that the monthly approach implicitly assumes an even distribution of double-digit unemployment months across years, where the 16 months are assumed to randomly happen across years (rather than clustering), and therefore the calculation becomes more like (16 random months)/(74 total years) = 0.216 base rate per year, which over two years becomes 34%?

Similarly, if we were to spread this out over days instead of months, we'd get an even higher estimate (0.018 days per year, over the course of 660 days, means a chance of essentially 100%).

So does this mean that your choice of granularity contains an implicit assumption about the clustering vs uniformity of positive occurrences over the time period? What other considerations are there when choosing the granularity for making the binomial estimate?