GDP released after 30 days for the first time
How much has the economy really shrunk? This is a question that affects markets and keeps politicians busy. To answer the question, a specific figure is of special importance - the development of the gross domestic product, in short the gross domestic product (GDP). For the first time, the Federal Statistical Office is releasing the GDP already 30 days after the end of the quarter. And it is done exactly for the quarter when the corona crisis had its greatest impact. Why are we changing this? And why now? An interview with Albert Braakmann, head of the “National Accounts, Prices” department at the Federal Statistical Office, and Michael Kuhn, head of the “Domestic Product, Input-Output Accounts” division at the Federal Statistical Office and responsible for the calculation and publication of the gross domestic product.
Mr. Braakmann, in the past, first results of the gross domestic product were released only after 45 days. Why exactly at that point in time?
Albert Braakmann: The gross domestic product is calculated on the basis of facts, that is, we evaluate mainly official basic data and economic indicators. For the short-term indicators that are included in the gross domestic product, generally enterprises are surveyed that have to transmit their data to us or to the statistical offices of the Länder. In addition, we evaluate administrative data, for example, on foreign trade or on the labour market. This means that the data have to be collected, checked and processed. This takes some time. Usually, the first short-term indicators are available after slightly more than 40 days.
Even after 45 days, however, robust results are not available for all economic sectors. This is why the first calculation of the gross domestic product generally includes a proportion of estimates. The earlier the GDP is to be calculated, the larger the proportion of data to be estimated will be. After 45 days, we have major results that allow a robust early estimate to be produced. In other words, in the past, the best possible and well-balanced trade-off between timeliness and accuracy was reached after 45 days.
Mr. Kuhn, Eurostat releases an early estimate for the euro area - that includes the German GDP - already after 30 days. Many other European countries, too, publish their results after 30 days. Why did Germany not do so?
Michael Kuhn: For a number of years already, we have been testing the GDP early estimate at t+30 days, and we have carefully analysed the differences from the figures calculated after 45 days under various framework conditions. The gross domestic product and its change rate are of enormous importance for the economic sentiment in the various markets. To put it more strongly, a few tenths of a percentage point can have a marked impact on the economic sentiment and make the difference between stagnation and recovery. Reliability and accuracy are highly important quality criteria for us! We cannot afford to make errors in GDP calculation - there is just too much at stake. Therefore it is important that we are very careful in considering an earlier release of this important figure and avoid hasty decisions. And, of course, we have other indicators to monitor the short-term economic development. Irrespective of when the first GDP early estimate is released, our monthly short-term indicators allow an earlier assessment of the situation in individual branches. Just over 30 days after the end of the month, we release data on retail turnover, on new orders in manufacturing, on industrial production and on foreign trade. All those indicators are also included in the early estimate of the gross domestic product. Our earliest short-term indicator, however, is the truck toll mileage index, which is published in cooperation with the Federal Office for Goods Transport and Deutsche Bundesbank.
Now you are changing over to t+30. Why now?
Michael Kuhn: Already at the beginning of the year, we decided to release the gross domestic product figure for the second quarter of 2020 after 30 days for the first time. The main reasons at the time were that, first, the data for the first two months of a quarter are nearly identical to those of the t+45 estimate and, second, our early estimates had improved, one of the reasons being more rapid evaluation of basic data. This has been combined with an adjustment of our estimation models and with the inclusion of expert knowledge. So we are now able to meet the quality criteria already at t+30 days.
Of course, we did not know at the time that the coronavirus pandemic would push the German economy into a serious crisis. The corona pandemic differs from earlier crises in that the slump has been massive and very abrupt; also, the crisis affects countries all over the world. The measures taken to contain the pandemic have had a massive impact both on the domestic economy and on foreign trade. In this situation, it is difficult to do an early estimation of the GDP. On the other hand, users require up-to-date and fact-based economic data, especially in times of crisis. Considering all this, we have decided to release the GDP figure at t+30 days despite the coronavirus crisis. This involves the risk that our first GDP estimate will have to be revised more extensively than usual at t+45 when additional data have become available. However, we have robust short-term indicators and a good estimation model and we use economic expertise to keep the revision within narrow limits.
Albert Braakmann: Generally, in such a situation of crisis, statistics are much in demand. Real data are needed to assess the situation and to take the right decisions regarding measures to manage the crisis. At the same time, the data must be available as early as possible. In statistics production, however, there is nearly always a conflict between the goals of timeliness and accuracy. The earlier we want to publish our data, the more estimates are required at the current end - especially when businesses can no longer report their data because they are busy coping with the crisis themselves or because they have simply shut down. All these factors made it difficult to produce an early estimate of the GDP for the second quarter of 2020, that is, the second quarter of the crisis. We had to make a hard decision between the currently very high demand for early data and the reliability of our early estimate, which at the moment is really hard to assess. Analysing the differences between t+30 and t+45 that occurred in the past cannot help us much because the framework conditions are just entirely different.
Mr. Braakmann, in the current debate, today’s situation is often compared with the financial and economic crisis. You said that estimation models no longer work, or work only to a very limited extent, in times of crisis. Isn’t there anything at all you could use from the experience of 2008/2009?
Albert Braakmann: The financial and economic crisis of 2008/2009 resulted from a real estate price bubble in the USA, which first of all affected the financial markets. Only later did this have an impact on the real economy and on production. During the financial and economic crisis of 2008/2009, there were in part larger revisions between t+30 and t+45 than in normal quarters. From the second quarter of 2008 to the fourth quarter of 2009, the differences averaged between 0.3 and 0.4 percentage points. In some quarters, there were differences of up to one percentage point. This is where we have to consider whether an early estimate provides sufficiently exact results. So we were very careful in considering whether we really wanted to release the early estimate after 30 days for the first time for the second quarter 2020, as had been planned - that is, the quarter that is expected to be affected most strongly by the corona crisis. But just now, there is a particularly great and important need for early indicators to be used for short-term economic monitoring. We are aware of this responsibility and we have decided to live up to it.
And what about the future? Is there potential for an even earlier estimate? Will it be possible some day to release a nowcast, that is, an estimate at the current end?
Michael Kuhn: We have been dealing with this question at the Federal Statistical Office since 2018. We are developing a nowcast, that is, an early indicator of the overall economic development, as a purely model-based estimate using econometric models. Similarly to the GDP early estimate, ARIMA models will be used in the first project phase, using various short-term indicators as external regressors. Also, modelling will be bottom up.
What exactly does that mean?
Michael Kuhn: It means that the nowcast is obtained by aggregating the estimates of various sub-aggregates of the gross domestic product. Of course it is desirable to have an early indicator of the overall economic development. However, it is not planned to do a purely econometric estimation of the missing data based only on the past. Nowcasting the gross domestic product in official statistics should include indicators of the actual development up to the current end, if possible. Digitalisation provides opportunities to move in this direction. Digital point-of-sale data of the retail sector, for instance - also referred to as scanner data - are a promising data source for price statistics. Thanks to digital process data, our earliest short-term indicator - the truck toll mileage index - is available not later than after nine working days and can contribute to obtaining an even better estimate. It will also have to be investigated whether the results of a nowcast can be improved by including non-official short-term sentiment and survey indicators, such as the ifo business climate index, and indicators based on new digital data, such as electricity production data or satellite data. Those indicators are included only marginally in the early estimates after 30 and 45 days. The main basis of calculation here is the official indicators that are calculated from collected data.
Albert Braakmann: But again - we said it before: the earlier we release the figure, the higher the proportion of estimates, and the higher the uncertainty or the potential inaccuracy. We must avoid results sending wrong signals, which would involve the risk that wrong decisions are taken on their basis. Of course, we always have to reconcile these conflicting goals. We always have to find the right balance between timeliness and accuracy.