[Victerius]

For the mathematically inclined among us, dp/dt is still positive. It's d^2p/dt^2 that is slightly negative.

[contravariant]

Though the mathematically inclined should also know that we're not looking at dp/dt, we're looking at p(t) - p(t-T).

The difference is important, especially because we're looking at a yearly increase every month. The derivative of the annual inflation is not d^2p/dt^2 but (dp(t)/dt - dp(t-T)/dt).

[queuebert]

Anyone who's ever had to numerically estimate second and higher derivatives from noisy data can appreciate how difficult the Fed's job is.

[redox99]

Not really.

It means CPI_oct22 - CPI_oct21 < CPI_sep22 - CPI_sep21

d^2p/dt^2 isn't necessarily negative.

[OscarCunningham]

'Inflation is slower now than it was this time last year.'

[Victerius]

Oh, right. I mixed it with month-on-month inflation.

[compumike]

For the mathematically inclined, here is a graph of 1/p over the past year:

https://totalrealreturns.com/s/USDOLLAR?start=2021-11-10&end...

or over a longer time period: https://totalrealreturns.com/s/USDOLLAR

(this site is my side project)

If "p" is the relative price level index (CPI-U in this case), then "1/p" represents the relative purchasing power of a single dollar over time -- explained on homepage in more detail.