library(tidyverse)

Apparently my colleagues are receiving some questions about the GDP growth in 2020 Q3, which some analysts have independently derived as being slightly higher. I am not sure what difference it makes 12.4% instead of 12.6% for analyzing the economic outlook but I will try to explain why there might be differences between their estimates and Eurostat’s.

One difference is that they may be using different data and incorporating revisions from some countries, like Spain or France. The GDP estimate of Eurostat for past quarters is frozen between t+100 of a quarter and t+60 days of the next quarter. Data received at t+30 or t+45 is only taken into account for calculating the latest quarter not the previous. I don’t know if it is the right thing to do (I think so), but it is like that.

But the difference I want to explain is the aggregation method to derive the European aggregates. Most analysts use annual country weights from the latest available year. There can be some discussion about the annual data to use, which can be based on annual data (or theoretically consistent quarterly raw data) or the sum of the seasonally and calendar adjusted quarterly data.

Eurostat, instead, aggregates by summing up quarterly series measured in previous year prices. These series are not reported by most countries but can be easily derived by multiplying the quarterly chain linked series by the annual deflator of the preceding year. Here, again, a decision has to be made about which annual deflator to use, with or without seasonal and calendar effects. This imply an implicit quarterly weighting system.

In normal times, with quarterly growth rates in a normal range (+1%/-1%), the weights will not imply a visible difference on the result of the aggregation, maybe at the third digit. But with a growth rate of 12% it does. Let’s plot the 2019 weights and the quarterly weights for some quarters in a chart for some selected countries.

weights<- readxl::read_xlsx("weights_EA19.xlsx")
weights<- weights %>% 
  gather(weights,value,c(2:9))

ggplot(weights %>% filter(geo %in% c("DE", "FR", "IT", "ES", "NL", "IE")), aes(reorder(weights,value),value, colour=geo))+
   geom_point(size=2)+
  scale_colour_viridis_d()+
  coord_flip()+
  theme_minimal()+
  xlab("")+ylab("")+
  labs(title="Different GDP Country weights", caption= "Luis Biedma")

Germany for example has an annual 2019 weight of 28.9% but in 2020 Q2 the weight had increased to 29.4%, and will be even higher in 2020 Q3, 30.0%, which will be the one used to calculate 2020Q4. Something similar happens with the Netherlands and Ireland. On the opposite side, France had an annual 2019 weight of 20.3%, which decreased to 19.8% in 2020Q2 and will be 19.4% in 2020Q3.

So, use with quarterly weights from quarterly previous year prices series if you want to get a very precise estimate of the Euopean aggregates.

Disclaimer: I do work in Eurostat but I do not work with these numbers, although I did in the past. They can be easily derived from the data publicly available in the Eurostat database.