Last week I wrote on birth of the most common measure of how the economy is doing. This measure - the Gross Domestic Product (GDP) - is produced by every country and tries to estimate how big is the total output of a country. You would imagine all economists should know how it is calculated then. My impression is that is not true. So here is my contribution today: How is GDP calculated?*
It all starts with some (almost) complete and detailed information on every activity behind the wheels of the economy: census information. These surveys almost all of the businesses with paid employees, hence covering over 95% of expenditures included in GDP. This information allows the people at the Bureau of Economic Analysis (or your favorite country's economic statistics office) to calculate their GDP very precisely for the year of the census. This estimate is usually called "benchmark" or "best-level." The problem then is that it is expensive to run a census as frequently as needed for politicians to "manage" the economy or for entrepreneurs to decide on their investments (nowadays GDP data is usually published in quarterly manner). Consequently, census data is complemented with higher frequency (and cheaper) surveys. These are less precise and not as complete as the census, but allow the statisticians to build models on how the total economy is doing based on this data. For example, a very simple model could be that if your information in the survey says the sampled businesses are producing 10% more, you could estimate the whole economy is doing 10% better.
This survey information is usually available either monthly, quarterly or annually. For example, monthly survey data covers around 35 thousand units (or less than 0.5% of businesses). The annual one covers a bit over 1.5%. And all these complement each other. Hence, the farther you are from the period of interest the more information you have on it. Therefore, GDP estimates are updated - hopefully improved - from time to time (in the US they are updated at least 3 times). One problem with survey data is that it can have significant measurement errors. For example, you may wonder why isn't consumer survey data used to estimate household consumption? Well, it is well known that people surveyed tend to report less than 25% of how much they really spend on alcohol. So this sector seems a lot smaller when you add up what people claim they spent on liquor than when you sum how much liquor stores sold. Hence consumer survey data is usually avoided. Business surveys also have problems (smaller ones) and so the models have to fix for these.
On top of measurement error, survey data is usually not enough since some sectors are not covered. For these, past trends and related data are used to estimate. For example, electricity and gas consumer spending are extrapolated using past temperatures. Only 45% of the first estimate of quarterly GDP is from components based on the monthly surveys. All the rest comes from extrapolations. Later, the share based on extrapolations is reduced to less that 6%.
Every 5 years new census data is available in the US, allowing for a big re-estimation of both previous years estimates as well as the models to reflect changes in the economy (to be used in future survey-based estimations). For example, as computers are used more as entertainment, the models are updated to reflect that when they allocate sales of computers across different sectors (home consumption versus fixed investments for example).
How good are the initial estimates relative to the final ones? Remember there is no number provided by god telling us with certainty the true state of the Economy. Hence, in serious countries you have to trust the final estimate of the statistics office (In countries like my dear Argentina, you just don't trust anyone's statistics). Initial estimates of (real) GDP correctly indicate the direction 98% of times. And its major components accuracy is as high as 88%. Taking into account the size of the project at hand (basically measuring everything produced in the country), the result is quite good.
*If you are an economist, you may know that GDP can be estimated in three different ways. I focus here on the final demand or expenditures approach. The methodology for the other two is very similar to the one described above. But if you are interested in the details you should check the paper below!
Based on an article by Landefeld, Seskin and Fraumeni.
Follow me on twitter at @diedaruich
News and posts for an active mind.