Estonia forecasts its economic activity in the short term

Printer-friendly versionEnviar a un amigo

Researchers from the IT Impact Studies Center of the University of Tartu in Estonia created an application that makes short-term economic forecasts (from 1 to 12 months) based on real-time company tax returns.

According to the Estonian Tax and Customs Board (ETCB), 98% of tax revenue in Estonia comes from voluntary payments and only 2% is collected through the intervention of the tax authorities. These voluntary payments are reported in the monthly tax returns of approximately 166,000 private companies and public entities subject to taxes. The ETCB wanted to use the real-time information contained in these statements to produce automatic forecasts for the entire economy and its subsectors.

A team of researchers and analysts from the University of Tartu at the Center for Impact Studies of IT (CITIS) took on the task of developing prediction models and automating the emission of predictions. "The economic time series tend to be very seasonal and can be predicted with standard approaches, but the challenge here was to create models that predict from 1 to 12 months in the future the total turnover of the companies, the number of employees and the volume of Export for all sectors separately, at all levels of aggregation and all that without too much human intervention on the part of the analyst, "said Andrés Võrk, economist and professor of econometrics affiliated with the CITIS team.

UT researchers devised an application that allows the user, be it the tax authority or any other actor interested in the forecast, to simply select the indicator to be forecast, the sector and the time window and a set of algorithms that will adjust to the models and issue a numerical forecast.

User interface of the forecasting application

The application uses a series of methods that fit a set of very different forecast models that are applied to the data and then uses machine learning to arrive at a compendium of relative proportions that are used to combine the different models into a single economic forecast

"Given the need for flexibility and the very different nature of how different sectors behave, as well as the fact that the overall economy changes over time, we had to find a modeling solution that is capable of self-adjusting to different sectors and the changing circumstances of economic conditions on the fly, "explained Taavi Unt, the author of the weighting algorithm. "Ensemble models tend to be more accurate in terms of forecasts than any model in itself, apart from being flexible by design, as the structure of the economy changes," continued Unt, who is also currently working on his doctorate in mathematical statistics in UT.

The forecast application is only a solution that the UT researchers built for the tax authority and is part of a broader impulse towards the predictive economy, that is, to use the data produced by the economic actors to have a complete vision of the economy nationwide in real time and have quick and accurate predictions of likely changes in the future. Solutions like this have the potential to make economic policy faster and smarter, allow real-time impact assessment of policy changes, and also reduce the administrative burden on companies when submitting corporate reports.