Big Data and measurement of GDP
Without fundamentally adjusting the measurement methods of economic activity there is a threat to economic policy to increasingly stumble in the fog of blur and inaccuracy. This is especially going to be dramatic once the positive consequences of new technologies like digitization – since not clearly visible – are underestimated and productivity growth is identified far too low. Consequently, real economic growth is underestimated, which negatively affects sentiments and expectations and therefore has an impact on economic forecasts. In case of incomplete, incorrect and misleading measurement, individuals might misjudge risks, over- or underestimate effects of business or economic decisions, and during the assessment of alternatives and their advantages and disadvantages, make wrong decisions. In order to reduce and avoid future measurement errors and to depict the essence of an economy of ‘Entdinglichung’, deterritorialisation and denationalization, the economy, politics and society should bid farewell to the old GDP concept more than ever. Instead, a new understanding in economy and economics and, based on this, a remeasurement of economic activities and transactions is needed. The research project “Big Data and GDP measurement” provides new methods for the measurement of economic activity in the age of digitization and data economy and thereby new concepts for economic diagnosis and prognosis.
Work in Progress
Extract from current research paper
Nowcast as Forecast – New Methods for predicting GDP in real-time with Big Data and Machine Learning
Fiscal or political decisions usually require assessments of current and expected future economic developments. For this purpose, decision-makers from politics, economic policy or central banks need timely economic data to respond optimally to the current economic situation. Since relevant economic indicators are only released every few months and with a delay, the majority of renowned indexes is not agile enough. Therefore, economists deeply address the improvement of macroeconomic monitoring in real-time, in order to develop a method to predict the present and the near past. A prediction of the present state or the near future or past, at the edge of available data is called ‚nowcast‘. This term consists of the words ‘now’ and ‘forecast’ and constitutes the observation of the current state of the economy in real-time through predicting the present, whereby the latest prediction is updated repeatedly.
The data economy and artificial intelligence (AI) open up new possibilities for the prediction of economic performance. As data source for generating daily predictions especially Big Data, as part of the new data economy, is suitable, since it is available for a variety of activities, in real time and in large amounts. For the best possible analysis of Big Data in the area of nowcasting especially the application of AI in form of machine learning (ML) proves to be efficient, since ML methods process the complexity and size of this data type best.
This chapter examines possibilities and limitations for the compilation of daily revised predictions (nowcasts) of the present state of the economy with ML as method of Big Data analysis. In order to compare the predictive performance of various nowcast and forecast methods, the statistical-empirical criterion root mean square error (RMSE) is applied. It determines, by how much the predicted values on average deviate from the actual values.
Maaß, Christina (2021): Nowcast als Forecast. Neue Verfahren der BIP Prognose in Echtzeit mit Big Data und Maschinellem Lernen. In: Thomas Straubhaar (Hg.): Neuvermessung der Datenökonomie. 1. Ausgabe (2021). Hamburg: Hamburg University Press, S. XX-XY.
For many innovations in digitization, detached from space and material, macroeconomic statistic measures are (still) missing. Information assets with network character are captured partially, at best. Therefore, GDP and its measurement methods have lower explanatory power than ever and an adjustment of statistics is necessary.
Furthermore, an alternative to traditional GDP forecasts, frequently discussed in current literature, is the so-called nowcast. This method can read data quasi in real-time and process them to predictions of the present. Which methods are most suitable, in which sense the application of Big Data can improve results and which role machine learning could play there, is under current investigation. In general, there is a tendency towards higher prediction accuracy with nowcasts, a result that further tests have to confirm.
- Straubhaar, T. (2020) “Das BIP hat ausgedient – so müssen wir jetzt unseren Wohlstand messen”, Die Welt on 23. September 2020.
- Straubhaar, T. (2019) “Das BIP in Zeiten der Datenökonomie: Neuvermessung der Ökonomie wir erforderlich”, Wirtschaftspolitische Blätter, Issue 3/2019.
Politics, (central) banks as well as individuals can directly apply the scientific results. The results thereby contribute to best-possible informed decision-making.
- On November 28, 2019, Professor Dr. Straubhaar moderated the Business Talk on the subject “Making artificial intelligence usable in practice for science, business and politics”. The business talk was organized by the Nordakademie Foundation and the Club of Hamburg.
- On June 18, 2019, Professor Dr. Thomas Straubhaar together with Professor Dr. Daniel Graewe took part in the Fishbowl Discussion on the subject "Big Data: between Big Business and Big Brother". The event was organized by the Nordakademie Foundation and the Club of Hamburg.