As at 30 December 2019
Official early indicators facing the conflicting goals of timeliness and accuracy
First official results of the overall economic development in Germany are currently available when the gross domestic product figures are released 45 days after the end of the quarter. For some economic branches, the Federal Statistical Office publishes monthly short-term indicators, some of which are available slightly earlier.
In the context of the advancing digitalisation, there is a growing interest on the part of users that data should be available earlier but should still be reliable. Especially reliable data on the current economic development in Germany are often the focus of public interest. However, collecting official basic data and processing them to obtain official short-term indicators is time-consuming. Consequently, the first calculation of a short-term overall economic indicator generally includes some proportion of estimates as not all basic data are available at that time of release. The earlier such a short-term overall economic indicator is to be calculated, the larger the proportion of data to be estimated will be.
To satisfy the need for earlier but still reliable data, the Federal Statistical Office started first test calculations in 2018 regarding an early overall economic indicator based on the gross domestic product and other short-term indicators. The goal is to give a first estimate of the economic development of a quarter shortly after its end. Such an early indicator faces the conflicting goals of timeliness and accuracy and can be referred to as a nowcast. The idea of experimental statistics is to give an insight into the continuous development of the overall economic nowcast. The offer is meant as work in progress, its aim being to encourage discussion about the data and methods used and the quality of the nowcast.
Forecast, nowcast, flash estimate - which is which?
An early estimate of an economic key figure can be referred to as forecast, nowcast or flash estimate, depending on when it is produced.
Forecast
An estimate that is produced at a time when few, or no official basic data are available, is generally a forecast.
Nowcast
A nowcast (from "now" and "forecast") is a very early estimate produced during or shortly after the end of a reference period.
Flash Estimate
A flash estimate is typically calculated only some time after the end of a reference period.
For a nowcast or a flash estimate, first official basic data are available for the reference period. The two together may also be referred to as early estimates. The share of estimates required for a nowcast is higher than that for a flash estimate.
The transitions between forecast, nowcast, flash estimate and calculation are fluid. Timeliness decreases when moving from forecast, nowcast and flash estimate through to calculation, whereas accuracy typically increases over time. The figure below illustrates the terms mentioned here, their chronological order and the conflict between timeliness and accuracy. Here, t refers to the end of a reference period.
Availability of major official short-term indicators
The figure below shows a selection of major official short-term indicators to be used for the calculation of a nowcast of the overall economic development. The reference period of most official short-term indicators is a month. The gross domestic product and short-term statistics in the service sector are calculated and released every quarter. The overall economic nowcast will also be calculated on a monthly basis. Taking the example of the third quarter of 2019, the figure illustrates when the official short-term indicators were released. For example, the truck toll mileage index for reference period July 2019 was released on 15 August 2019, the gross domestic product for the third quarter of 2019 on 14 November 2019. The figure reveals how many monthly values of the official short-term indicators are available when. If an overall economic nowcast is to be calculated at t+10 days after the end of the third quarter of 2019, for instance, the only monthly value of the index of turnover in accommodation and food services is that for July. The missing monthly values for August and September would have to be estimated at that time. For an overall economic flash estimate at t+30 days, however, two of three monthly values of the relevant quarter would be available.
Possible methodology of a nowcast of the overall economic development
In a first project phase, setting up an early indicator of the overall economic development will be based on the structure of the first release of the gross domestic product. The figure below presents common features of, and major methodological differences between the planned early overall economic indicator and the GDP flash estimate.
First of all, the early indicator of the overall economic development is designed as a purely model-based estimation based on econometric models. This means that, unlike the GDP flash estimate, only one of the pillars of the three-pillar model successfully applied in the GDP flash estimate will be used (see also the paper "Schnellschätzungen für das Bruttoinlandsprodukt" in the WISTA journal). As the early indicator of the overall economic development uses a purely model-based method, it is not envisaged to implement the expert estimates and the coordinated estimate. In the first project phase, ARIMA models (autoregressive integrated moving average models) using various short-term indicators as external regressors will be applied, as is the case in the GDP flash estimate. Also, modelling will be bottom up. This means that the early indicator of the overall economic development will be obtained by aggregating the estimates of various sub-aggregates of the gross domestic product.
Most official short-term indicators that are available early refer to the supply side, as is shown in the table. This is why the focus was on modelling the production approach. The use approach is still being set up. Another major difference between the overall economic nowcast and the GDP flash estimate refers to the short-term indicators used. In the GDP flash estimate, the focus is on official indicators. In the early overall economic indicator, additional indicators will be used: non-official short-term sentiment and survey indicators (for example ifo business climate index) as well as indicators based on new digital data (for example electricity production data or satellite data). Also, official indicators are included which so far have not been used for the GDP flash estimate (for example truck toll mileage index). Further methodological details on the overall economic nowcast are contained in a paper in the WISTA journal.
First provisional estimates of an early indicator of the overall economic development
The figure below shows first provisional estimates of the overall economic nowcast for 14 quarters from 2016 Q1 to 2019 Q2. They have been calculated on the basis of historical real-time data, where available. This means that they are simulated estimates based on historical data shortly after the end of the relevant quarter. The focus here is on modelling the production side of the gross domestic product. The uses side models are still being set up. As soon as the uses side is available, ideas should be developed on how the process of weighting or of coordination with the production side can best be modelled. As the underlying econometric models of both calculation sides are continuously enhanced, the estimates shown here are meant as work in progress. They will change also for earlier quarters along with the further development of the methodology and the data sources used. The data presented in the figure are overall economic percentage changes on the same quarter of the previous year. The relevant change rates of the GDP flash estimate and the GDP first release, which is published 45 days after the end of a quarter, are also shown for comparison.
The further development of the overall economic nowcast could focus on these issues:
- How can the econometric models of the uses side be modelled?
- What is a meaningful method regarding the coordination or mechanical weighting (combined forecasts) of the two calculation sides?
- What alternatives to the ARIMA model category could be investigated (for example dynamic factor models or machine learning models)?
- What are the benefits of top-down modelling, what are the advantages of a bottom-up approach?
- What is a good benchmark to assess the quality of the overall economic nowcast?