Six Month Lag

Inflation, finance, economics.

Browsing:

Tag: nowcasting

One month centered inflation vs annualized 3, 6, and 12 month lagged inflation

Centered 12 month inflation and 3, 6, and 12 month lagged inflation. Also the probability that inflation rose over the past 6 month. 2019-2024.

I showed a chart plotting errors for various annualized lagged percentage changes in a previous post. Here we can get an idea of what they look like. This chart from my paper shows annualized lagged percentage changes relative to underlying contemporaneous inflation as measure by the one year centered percentage change shown in the thick blue line. The red dashed line shows three month lagged percentage changes: they move around too much to provide insight into the underlying trend. The one year lagged percentage change, which is generally reported as the contemporaneous inflation rate, provides a somewhat outdated perspective (six months outdated) to existing underlying inflation. The sweet spot is in the six or seven month range: the thin green line shows the annualized six month percentage change. It fits the thick blue line visibly better than the 2 alternatives shown.

The bottom half of the chart indicates inflation’s direction, whether the centered thick blue line is rising or falling over the previous 6 months, something which is only known retrospectively. Formally, it is the probability that inflation is rising: when it is above say 80% it predicts rising inflation and when it’s below say 20% it predicts falling inflation. The probabilities are based on the price data alone. Such lack of sophistication is helpful to the extent that at least the forecast process (or rather the nowcast process) can be readily understood, unlike models where there are many contributing variables.

The data goes through July 2024 and reflects the late August report. I hope to post something more up to date soon.


Estimating one year inflation centered on the current month rather than six months prior

Media discussions of inflation releases like this one or this one highlight one year and one month percentage changes in prices. The one year figure is stale: half of it reflects economic developments that are over six months old. The one month figure is up to date but too volatile to be useful.

What we would like is a one year percentage change centered over the current period, but naturally that won’t be available for another six months. However, we can estimate that figure with annualized percentage changes taken over various lags. The best estimate for one year centered inflation turns out to be six or seven months long. Here’s the first chart measuring the errors for various lags: we want errors to be low:

For both 1959-2023 and a subset of that era when inflation fluctuated a lot, annualized six month lagged percentage changes approximated one year centered inflation best. One month inflation annualized inflation produces errors about 2 and a half times as high. The one year lagged percentage change, emphasized in most inflation reports, has higher errors than any lag between 4 and 11 months.

Memo to reporters: calculate the annualized 6 month lagged percentage change! And download my paper! The next post will discuss annualization in greater detail and provide a formula.

 


First post: six month lag

Glasses on Calendar

Greetings and welcome to Six Month Lag, a blog focusing on inflation, finance, and economics. I started the blog to promote my paper, Forecasting the Present: Optimal Lags for Contemporaneous Core PCE Inflation. The optimal lag for nowcasting core PCE inflation turned out to be 6 or 7 months at the time of this post. I make no promises regarding what future research will show for other time series. That said, the internet appendix to my paper shows that the six month annualized percentage change has lowest errors relative to centered one year inflation for headline CPI, core CPI, and unadjusted PCE inflation as well. So I am hopeful that I won’t have to change my website name.

Here’s a summary of the paper:

Percentage changes bracketing a given date gauge inflation best but can only be measured retrospectively. Geometrically annualized percentage changes over the previous 6 or 7 months provide the best timely approximation of underlying core PCE inflation according to a range of criteria. Decimal lags permit the optima to be stated with greater precision. Probabilistic modeling of inflation’s level and direction can be conducted using estimates of the error’s variance and kurtosis. A review of Federal Reserve statements during the COVID and post-COVID era suggests that greater focus on the geometrically annualized 6 month percentage change could have provided earlier warning of incipient inflation.

Future posts will march through some of the points of the paper in a conversational style. I also plan occasional commentary on monthly inflation data releases.