MulTi - Multiple Time Series Analysis in Economics

• Applicant: Professor Dr. Helmut Herwartz
• Subject Area: Statistics and Econometrics
• Term: from 2018 to 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 390996990

Within the recent decade, we observe a strongly growing demand for analysis tools unravelling dynamic relationships in (high dimensional) systems of socio-demographic, economic and financial processes. Considering more complex models which can be difficult to implement, the challenges for research software are also shifting. By meeting the needs of the scientific community, we contribute to modern multivariate time series analysis by means of developing an intuitive, comprehensive and consistent software framework: MulTi comprises a sophisticated research library and analysis tool for multiple time series research. In this project, we combine its transparent and web-based development with an innovative collaboration approach. The Java-based prototype JMulTi offers an extensive library of reference implementations promoting the use of time series methods. The software has proven successful in research, teaching and practice. However, we believe that the its potential can be largely increased. Incorporating a novel sustainability concept and recognizing recent trends in data science, we revise the software fundamentally in favor for its continuous usability. To this end, we adapt its structure to modern work flows and users preferences from the scientific community for distinct programming environments. Aiming for strongest possible contribution, we transfer JMulTis library from Java and GAUSS to Python. We publish the resulting module to distinct repositories from where it can be installed and employed freely. A new wizard-based graphical user interface is suitable for the training of young scientists. Meeting the versatile needs of an interdisciplinary community, we implement further interfaces to the programming environments R and Stata. We expand the library for models and methods by taking into account recent literature, such as Bayesian tools that mostly have not been included in any software. Jointly with the development of state-of-the-art frequentist methods, we design MulTi to become an important basis for replication studies in economics. Providing knowledge and training, we create interactive web-based teaching materials for workshops and seminars. Our key objective is to create a unique, consistent reference library for multivariate time series analysis. A number of internationally well-known time series researchers agreed to cooperate on this project. Throughout the project period, they support the development by means of methodical and technical contributions, supervision and dissemination. The entire development and a public documentation of the development progress take place transparently on the collaborative platform GitHub.

• DFG Programme: Research data and software (Scientific Library Services and Information Systems)