1 Introduction
The folllowing sections first describe the crucial problem that the Open Time Series address, then explain how our background provides intrinsic motication and how our approach solves the problem. Our background in economic research and time series analysis combined with our experience in the regular publication of reliable economic indicators helps us understand both – the perspective of a data consumer and provider.
1.1 Problem Statement
Even though most official statistics in Switzerland are publicly available, data publications from regional and federal public data providers often lack machine readability. Therefore, these data publications are not convenient for academic researchers and other groups that ingest public data frequently.
In the field of economics, for example, intertemporal comparisons play an important role. This type of research implies regular and timely sourcing of data. Traditional data publications are often unsuitable1 for the scientific consumer, as modern economic (forecasting) models make use of hundreds or thousands of variables stemming from a large range of datasets. Particularly, researchers who source a large amount of time series data from different providers are impacted by the substantial difference in the pace of the digital transformation across providers or the differences that exist even between departments of the same providers.
Though application programming interfaces (APIs) from the official providers have led to great improvements in the machine readability of public data, machine readability is not homogenous across datasets. The proposed project mitigates the machine readability issue by not only publishing time series data created from a fully transparent process, but also by providing an open technical processing framework. Our framework eases the data sourcing and quality control burden and, at the same time, encourages researchers to contribute time series datasets of their own. In addition, the Open Time Series project establishes a steady and hands-on exchange between regional and federal data providers and the community of academic users of public data.
1.2 Background and Motivation
Although we aim at extending the approach suggested in this proposal to other fields in the future, this Explore proposal focuses on our ‘home court’ disciplines of public statistics and empirical economics. We are confident that our strong network in both fields will leverage our initial push sufficiently to give the impact of the Open Time Series project a great chance to spill over to other communities. The following sections introduce our background and explain how the proposed project helps the economics, forecasting and open data communities.
Time Series Analysis, Forecasting and Data Science in Economics
Time Series analysis continues to play an important role in many strands of modern empirical economics: Economic forecasting and its evaluation, studies of shock absorption, and many other econometric modeling exercises require data that allow intertemporal comparison. Modern models often use hundreds, or even thousands, of time series.
Even though machine readability of publicly available data has improved in Switzerland, thanks to great efforts led by the Swiss Federal Statistical Office (FSO), researchers are confronted with a data publication mindset that is focused on the wider public and hence on the most recent wave of data. As a consequence, variable names or features may change over time, and minor errors may simply be corrected on an ad hoc basis or with the next publication - even in datasets that are already available through an application programming interface. In combination with different degrees of machine readability across different providers of public data, the issues mentioned above can lead to two very heterogeneous situations when it comes to automated data sourcing, particularly for those researchers who need large amounts of data from various providers. Another aspect in which the focus of public data providers differs from academic research is the importance of readily available, so-called real-time vintages. Vintages can be seen as versions of time series that allow researchers to work with past publications even after the official version has been revised. While versioned time series are inevitable for many research exercises, such as the evaluation of forecasts based on the data that had been available when the forecast was made, multiple official versions are harder to communicate to the broader public and induce higher data management costs.
Our motivation is to help the time series community immediately while larger organizations focus on different challenges in their own digital transformation and often have different migration paths and schedules. We intend to do so by not only open-sourcing a framework to create scientific grade time series from official statistics, but also publishing our framework in the most transparent way possible:
Motivation through Immediate Impact
We focus on thorough documentation that involves the community.
We provide comprehensive application examples facilitating transfer to one’s own needs, making many thousand proven time series publicly available.
Machine-readable data and the Open Time Series automation framework substantially facilitate hosting a versioned time series database.
We have an infrastructure-as-code approach and share precise information on how to set up an environment that is capable of running our framework.
Official Statistics and Public Data
Our commitment to regular publication of data-driven indicators designed for intertemporal comparison and our leading role in the community ensure the sustainability of the Open Time Series project. The strong empirical focus of the academic research at the KOF Swiss Economic Institute provides great synergies with the production of forecasts and indicators, many of which rely on publicly available data provided by Swiss federal and regional administrations. The Open Time Series project bridges the described gap between official statistics and academic economic research and strenghtens the connections between these communities in Switzerland.
In addition to its popular indicators for the Swiss economy, KOF makes a wide array of fine-grained, sector-specific time series publicly available. Per January 2024, the KOF time series explorer made more than 31’000 time series publicly available in human-browsable and machine-readable fashion2. This experience with the regular publication of economic data, the production of indicators, economic forecasting and KOF’s task of monitoring the Swiss economy makes us relate further to the needs and challenges of official statistics. Through our steady exchange with key data providers and data consumers such as the Swiss National Bank (SNB), the Swiss Federal Statistical Office (FSO), the Secretariat of Economic Affairs (SECO), the State Secretariat of Education, Research, and Innovation (SERI) and the Swiss cantonal statistical offices, we are aware that other institutions have often cumbersome, semi-automatic data sourcing processes to process publicly available data that are highly redundant across institutions. Hence, we will not only open-source our implementation but also commit to put in substantial additional effort to make our solution inclusive and transparent to others. This will impact the community in three sustainable ways:
Motivation through Sustainable Impact
i.) We help reduce redundant implementation of data sourcing routines across publicly-financed institutions
ii.) We establish source code as an unambiguous channel that is understood across institutions and fields in order to communicate the needs of the research community in a highly dynamic environment
iii.) We develop a blueprint for data consumption in a collaborative fashion, increasing the commitment of involved collaborators
The feedback generated in the process is a core strategy of the Open Time Series project that we validated in talks with official data providers in advance of this proposal. Communication through code is unambiguous and reproducible and helps to communicate needs across fields, perspectives, and focuses.
Further concrete steps could be the integration of FSO experts into teaching or the use of Swiss data in academic teaching and open-source software packages.
The Open Time Series project can build on an established network for great outreach and at the same time, produce an entirely new group of publicly available, scientific use grade time series datasets. While the vast majority of the existing, aforementioned KOF time series stem from KOF’s own business tendency surveys, the Open Time Series project facilitates research use of additional data stemming from other providers, most notably the Swiss Federal Statistical Office (FSO). In addition to such major sources of macroeconomic data, the Open Time Series project helps to make the maintenance of time series from smaller, independent providers more sustainable. Independent community maintainers can use the Open Time Series framework to process datasets of their needs stemming from smaller providers. The modular approach of our framework does not only adding new datasets easily but also to exchange maintainers at the dataset level in case certain a previous dataset stakeholder cannot keep do their maintenance effort. In addition, our approach allows researchers to publicly archive data and reproduce results more easily.