Economy vs. Trend

The pre-COVID end point. Every trend stops learning at the last reading before the pandemic: February 2020 for monthly series, the fourth quarter of 2019 for quarterly ones. Nothing after that point shapes any trend; each is then carried forward to the latest month or quarter published.

Fitted trends (2, 5 or 10 years, or from a chosen year). A straight line through the logarithm of the series over the window, by least squares — the constant growth rate that best fits those years. Because the line is fitted, it need not pass through the last pre-COVID reading: if the economy was running hot in 2019, the series starts the pandemic above its fitted trend. Short windows follow the late-2010s pace; long ones average over more of the recovery from the 2008–09 recession, when growth in jobs and hours was faster than it could last. "Fit from a chosen year" is the same line with a start year you set on the slider, ending at the same pre-COVID point.

Peak to peak (2007–19 and 2001–19). The constant growth rate between business-cycle peaks — December 2007 (2007 Q4) or March 2001 (2001 Q1) to February 2020 (2019 Q4), the National Bureau of Economic Research's dates — extended forward. Measuring from peak to peak nets out whole recessions and recoveries, the usual way to read an underlying growth rate. The history loaded in this tool for the inflation-adjusted GDP components begins in 2007 (BEA publishes earlier years; they are not loaded here), so the 2001–19 line is missing for them.

Hodrick-Prescott trend. The standard smoothing filter (smoothing weight 1,600 for quarterly data, 129,600 for monthly) run on the data from 2000 to the pre-COVID end point, then extended at the trend's growth rate over its final year. Unlike a straight line it lets the trend's growth rate drift, so it reflects how fast the series was growing just before COVID without hinging on any single month. A Kalman-filter trend would add little: the Hodrick-Prescott filter is exactly what a Kalman smoother gives for a trend whose growth rate wanders, with a fixed ratio of noise to signal, and estimating that ratio instead would let the 2008–09 recession dominate it.

CBO January 2020 projection. The Congressional Budget Office's last forecast before the pandemic, the same path as the CBO January 2020 Explorer draws: for levels and price indexes it keeps CBO's growth path but is scaled so 2019 Q4 matches today's revised data; rates are as CBO projected them. Monthly series interpolate CBO's quarterly path. Real ECI wages are CBO's ECI divided by CBO's CPI, since CBO did not project a real wage. Only the series CBO projected have this line.

Rates. The unemployment rate and the participation and employment rates do not grow, so for them a "trend" is the average over the window — the two or five years before COVID — and gaps are in percentage points. The 10-year, chosen-year and peak-to-peak windows are not used for rates: those averages take in the high unemployment of the 2008–09 recession and its slow recovery, and would make any normal labor market look unusually tight. Nor is the Hodrick-Prescott trend: unemployment fell steadily through the 2010s, so the filter's pre-COVID level (3.2%) sits below where unemployment ever settled. Population aging lowers participation over time; a flat average ignores that and CBO's line does not.

Gaps. For everything but rates, the gap is the percent difference between the latest reading and the trend at the same date. A price level above trend is not an inflation rate: it is the cumulative excess since 2019, and it does not shrink unless prices actually fall relative to trend. Real wages are each nominal wage measure divided by the CPI, so their gap is roughly the wage gap less the CPI gap. The Atlanta Fed wage tracker is left out because it is a growth rate of individual workers’ wages, not a level that can sit above or below a trend. There was no October 2025 CPI (the government shutdown canceled it), so the fourth quarter of 2025 in the quarterly real-wage series averages November and December.