Where Economic Data Comes From: Agencies, Schedules, Revisions and Rate Conventions

What you will learn

• Which agencies publish the market-moving releases in the US, Germany and the euro area, and how schedules work.
• Why first estimates are revised and how a careful reader handles revisions.
• What seasonal adjustment does and when to look at unadjusted data instead.
• How to convert between month-on-month, annualised and year-on-year rates and recognise a base effect.

Explanation

The course Macro Basics covered which reports matter and why the surprise against consensus drives the reaction. This course goes deeper: how each number is built, so you can judge whether a surprise is information or noise.

Who produces the data. In the United States the Bureau of Labor Statistics (BLS) publishes the Consumer Price Index (CPI) and the jobs report; the Bureau of Economic Analysis (BEA) publishes gross domestic product (GDP) and the Personal Consumption Expenditures (PCE) price index; the Census Bureau publishes retail sales and durable goods. In Germany the Statistisches Bundesamt (Destatis) publishes CPI, GDP, industrial production and retail sales. Eurostat aggregates national figures into euro-area statistics such as the Harmonised Index of Consumer Prices (HICP). Private providers such as S&P Global and the ifo Institute publish surveys. Agencies publish release calendars a year ahead with fixed times, for example 08:30 Eastern Time for most US releases.

Release lag and vintages. US CPI appears about two weeks after month-end; the German preliminary CPI about two working days before the end of the reference month, with the final figure in the middle of the next month. The first published number is called the first vintage. Later vintages add late survey responses and better source data; a revision is the difference between vintages. Revisions are the normal price of early publication.

Seasonal adjustment. Many series have a predictable yearly pattern: retail sales jump in December, construction falls in winter. Seasonal adjustment (SA) removes the estimated regular pattern so that month-to-month changes show the trend; non-seasonally-adjusted (NSA) data show what actually happened. Rule of thumb: use SA data for month-on-month and NSA data for year-on-year comparisons, where the pattern cancels out. Shocks such as the 2020 lockdowns can distort seasonal factors for years (residual seasonality).

Three rate conventions. A month-on-month (m/m) rate compares this month's index with last month's. A year-on-year (y/y) rate compares with the same month a year earlier. An annualised rate asks: if this month's pace continued for twelve months, what yearly rate would result? The formula is (1 + m/m)^12 − 1 (where ^12 means "to the power of 12"), not 12 × m/m. US GDP is reported as an annualised quarterly rate; German and euro-area GDP as plain quarter-on-quarter rates, so a US print of 2.0 percent and a German print of 0.5 percent describe roughly the same pace.

Base effects. A y/y rate is a ratio of two index levels. When an unusually large monthly change from twelve months ago drops out of the window, the y/y rate moves even if nothing happened this month. Forecasters know these shifts in advance, so they are not news.

Worked example

Hypothetical example: a price index stood at 120.0 in March of last year and at 123.6 in March of this year. The index rose 0.8 percent in April of last year. Assume this April it rises only 0.2 percent. Rounding to two decimals.

1. March y/y rate: 123.6 / 120.0 − 1 = 0.030 = 3.00 percent.
2. Annualised April pace: (1.002)^12 − 1 = 2.43 percent. For comparison, a 0.3 percent monthly pace annualises to 3.66 percent, not 3.60 percent.
3. April index levels: last year 120.0 × 1.008 = 120.96; this year 123.6 × 1.002 = 123.85.
4. April y/y rate: 123.85 / 120.96 − 1 = 2.39 percent.

The y/y rate fell 0.61 points, but most of that is the 0.8 percent month leaving the window (the base effect), not fresh news. A three-month annualised rate is a useful middle ground: monthly changes of 0.1, 0.2 and 0.4 percent give (1.001 × 1.002 × 1.004)^4 − 1 = 2.84 percent. Fees, spreads and taxes play no role in index arithmetic.

Common mistake

Practice activity

On paper: an index is 104.0 in June of last year and 107.1 in June of this year. In July of last year the index rose 0.6 percent. Assume it rises 0.1 percent this July. Calculate (a) the June y/y rate, (b) the annualised July pace, and (c) the July y/y rate. Then state how much of the change in the y/y rate is a base effect.

Key takeaways

• Know the producer, lag and release time of each series; calendars are published a year ahead.
• First estimates are revised; compare like with like and expect the first print to change.
• Use adjusted data for m/m and unadjusted data for y/y comparisons.
• Annualise with (1 + m/m)^12 − 1, and check for base effects before reading a y/y move as news.

Next step

Next we apply these conventions to the most watched release of all: the consumer price index, in its US and European forms.