Understanding Data Announcements
If you look at a data calendar (see here for a calendar and here for my essay) you can find out when certain data are released (for example, the Employment Report is typically released on the first Friday of the month at 8:30 am Eastern Time). The Employment Report essay (here) explains how the Bureau of Labor Statistics calculates the numbers that are included in the report; in this essay, we explore the announcement process, that is how analysts forecast the data, how the consensus forecast is obtained, what are “whisper” numbers, the lock-up and what happens at 8:30 am on the first Friday of the month. For the purposes of this essay I will use the monthly announcement of the Employment Report but the same analysis could be applied to other announcements.
Economic analysts regularly forecast upcoming economic numbers. These forecasts range from sophisticated statistical models (see here for a pdf of a recent academic paper that examines 30 variables that could be used to forecast the change in employment) to “back of the envelope” calculations (like the one described here). Companies that provide data services to traders (such as Bloomberg and Reuters) routinely ask economists for their forecasts and use them to find the average or median forecast.
There is a lengthy literature behind the idea (sometimes called the Delphi method) that the average or median forecasts typically outperforms any individual forecaster (see here for an online book with an extensive bibliography). But economic analysts work for profit making enterprises and do not wish to give away their best forecasts for free. So they play a game something like this: early in the process they give away to Bloomberg a free forecast, which may (or may not) reflect their actual thinking. The consensus forecast is formed on the basis of these early forecasts. But as more information becomes available (closer to the data of the announcement) the forecaster may only distribute his forecast to preferred clients; this forecast is referred to as a “whisper” number, that is not printed and distributed (there are a number of services that claim to provide whisper numbers for company earnings: see here, here and here but none that I know of that provide whisper numbers for economic data; please send me an email at ckendall ##at## umrkt.com if you know of any).
At 8:30 am the data are released and broadcast on the news wires and television; but before then there are approximately 20 journalists who have seen the data and written reports at the “lock-up” (see here for an excellent essay describing the lock-up). Accredited journalists from major news organizations receive the data release a half-hour early and prepare their articles that can be released at exactly 8:30 (they are “locked up” in a room where they are not allowed to contact the outside world before the official release). The data are usually available to senior government officials the night before the release so that the appropriate cabinet secretary can be prepared to speak just after the release.
Then at 8:30 the data are released to the public; typically the data are announced on television (CNBC and elsewhere), via paid data services to their clients, on news wires and on the web sites of the agencies that release the data. Traders often react to the initial announcements and analysis by the reporters in the lock-up before; later analysts will release comments on what the data really mean through to their traders, through wire services and in letters to clients (a service performed by the author of this article). But an hour later there is another announcement and a month later a revision and five years after that a benchmark revision and the data will look very different.