Computational Social Science

• Algorithms in the public sector: Why context matters
  

(Prof Dr Anja Achtziger in cooperation with Georg Wenzelburger, Pascal D. König, Julia Felfeli) Algorithms increasingly govern people's lives, including through rapidly spreading applications in the public sector. This paper sheds light on acceptance of algorithms used by the public sector emphasizing that algorithms, as parts of socio-technical systems, are always embedded in a specific social context. We show that citizens' acceptance of an algorithm is strongly shaped by how they evaluate aspects of this context, namely the personal importance of the specific problems an algorithm is supposed to help address and their trust in the organizations deploying the algorithm. The objective performance of presented algorithms affects acceptance much less in comparison. These findings are based on an original dataset from a survey covering two real-world applications, predictive policing and skin cancer prediction, with a sample of 2661 respondents from a representative German online panel. The results have important implications for the conditions under which citizens will accept algorithms in the public sector.

• Analysis of Electoral System Reform Elements using Simulations
  

(Prof Dr Joachim Behnke) For more than 20 years, various reform proposals and possibilities of reform regarding the electoral system of the German Bundestag have been discussed. The focus of the discussion is on ”Überhangmandate” (surplus seats) and the associated enlargement of the Bundestag. For the assessment of the effects of various reform proposals and their comparison, simulations are used as a means to provide estimates for the effects of certain reforms and to elaborate the sensitivity by modifying the model parameters. One of the more specific aims of the project is to illuminate how the size of the Bundestag is affected by various options of institutional reform.

• Analysis of Processes of Party Competition and Political Opinion Dynamics using Agent-based Models
  

(Leonie Geyer & Prof Dr Joachim Behnke) The political competition between parties is often described using so-called spatial models.
These focus on the consequences and implications of voters’ and parties’ locations on one or several issue dimensions. In these models, parties are seen as being in a economic competition for the votes of citizens as their “customers”. As voters’ choice of party depends on the political proximity of their position to that of the parties, parties can use their choice of position strategically to improve their outcome. Since voters choose parties by comparing parties’ positions on several dimensions and the effect of one party’s changing position depends on the positions of other parties, who in turn can adjust their own positions, the interaction patterns of parties’ positional changes are very complex and can hardly be modelled analytically. Agent-based models are one possibility to calculate the results of complex analytical calculi using simple adaptation strategies that are ”more realistic” in this respect and therefore may provide valuable information about the actual dynamics of party competition.

• Evaluations within international organisations
  

(Prof Dr Steffen Eckhard in cooperation with Dr. Vytautas Jankauskas, Elena Leuschner & Ian Burton) Funded by the Deutsche Forschungsgemeinschaft (DFG), duration 2017-2020/21-2023, (projekt number EC 506/1-1)

While the first phase of the project (2017-2020) analysed the political use of evaluation in international organisations (IOs), the second phase (2021-2023) delves into evaluation reports themselves, scrutinizing their content and potential political biases. Following a mixed-method research design, we conducted field research with over 70 interviews at 19 international organizations, and even participated in training exercises for professional evaluators, to gain qualitative insights into evaluation processes. At the same time, creation of an original databank with 2,000 evaluation reports for the first time allowed to quantitatively scrutinize the actual content of IO evaluation reports, aiming at the identification of systematic political biases.

• Transfer project on computational text analysis of evaluation reports with GIZ
  

(Prof Dr Steffen Eckhard in cooperation with Dr. Vytautas Jankauskas, Daniel Baumann & Rita Sevastjanowa) Since November 2022, the research team has been cooperating with the German Society for International Cooperation (GIZ) to support its Evaluation Unit in setting up the infrastructure and methodology for computer-based text analysis of evaluation reports. The organization has been conducting evaluations of its projects for many years and also has a large amount of additional text-based data, such as reports. These documents provide a valuable source for future evaluations and are to be made available for future analysis within the framework of the project.

• Monte Carlo Analysis of Estimators in Multilevel Analysis

(Prof Dr Martin Elff) Multi-level analysis is often used in comparative social research to trace the influence of the social and political context on individual behavior. This method is often used on the data of multinational surveys (such as the Eurobarometer or the European Social Survey), which are characterized by a large number of units of investigation at the individual level (respondents) but a small number of units at higher aggregation levels (states). To what extent the multilevel analysis still delivers reliable results under these circumstances and to what extent accurate statistical inference can be achieved are some of the questions that this project deals with.

• Party Competition and the Nature of Political Ideologies
  

(Prof Dr Martin Elff) The reconstruction of the positions of political actors is essential for a variety of political science issues. In analyzing the formation of government coalitions, as well as understanding changes in patterns of electoral behavior, it is essential to consider the political positions of parties. However, the reconstruction of these positions turns out to be anything but easy. There are two basic approaches for this: on the one hand, ratings by experts (i.e. country experts assign positions to the parties on predetermined scales), on the other hand, the quantitative analysis of the parties' election programs (i.e. based on the statistical analysis of prepared texts and with the aid of spatial models the positions estimated). The research project follows the second approach. As part of the project, on the one hand, existing procedures are to be improved and examined with regard to their reliability; on the other hand, they are to be used to create a comprehensive database of party positions with which the patterns of party competition can be determined. An application for funding has been submitted to the Deutsche Forschungsgemeinschaft (DFG) to support the project.

• Book Project: “Data Management with R: A Guide for Social Scientists”

(Prof Dr Martin Elff) The data analysis software R is on the one hand a comprehensive infrastructure for data analysis and data management, on the other hand it is a programming language for statistics and graphics. Due to this dual character, it can be used in a variety of ways. Their field of application ranges from the analysis of experimental data, through the analysis of survey data, to the analysis of massive data sets of behavioral traces. However, the data structures relevant for this are not yet very familiar in the social sciences; on the other hand, the support for survey data sets that are typical for social research through R packages can still be expanded. The book project is intended to build the corresponding bridges: on the one hand, the common data structures are presented and discussed, on the other hand, it addresses the special packages relevant for the analysis of survey data.

• Don't miss out on NFTs?! A sentiment-based analysis of the early NFT market

(Prof Jarko Fidrmuc and Florian Horky in cooperation with Lili Dubbick and Franziska Rhein) This study investigates the impact of Twitter sentiment on the Non-Fungible Token (NFT) market. Using a dataset of over 5 million English-language tweets on NFTs, we calculate a daily sentiment index and link it to NFT sales and trading volume. Applying wavelet analysis and DCC-GJR-GARCH models, we analyze the NFT market, characterized by multiple bubbles and high volatility. The findings reveal Twitter's significance as a primary source of information for a broad audience. Moreover, the study contributes to the literature by examining the role of Twitter sentiment in the NFT market's development. Additionally, the study indicates weak links between established cryptocurrencies and the NFT market. Based on our findings, we recommend that traders and policymakers use social media activities to monitor new digital markets.

• Fintech and Artificial Intelligence in Finance - Towards a Transparent Financial Industry
  

(Prof Dr Florentina Paraschiv) Funded by the European Union, EU Grant COST ACTIONS 2020 to 2025, duration 2020-2024

The financial sector is the largest user of digital technologies and a major driver in the digital transformation of the economy. Globally, more than $100 billion of investments have been made into FinTech companies and Artificial Intelligence (AI) since 2010, and continue growing substantially. In early 2018, the European Commission unveiled (a) their action plan for a more competitive and innovative financial market, and (b) an initiative on AI with the aim to harness the opportunities presented by technology-enabled innovations. The Action will investigate AI and Fintech from three different angles: (1) Transparency in FinTech, (2) Transparent versus Black Box Decision-Support Models in the Financial Industry, and (3) Transparency into Investment Product Performance for Clients.

• Intra-daily Volatility Measurement on Stock Markets

(Prof Dr Franziska Peter) Funded by the Deutsche Forschungsgemeinscahft (DFG), duration 2018-2020, (project number 389577820)

This project examines volatility in connection with risk forecasting and risk management on intraday frequency. Current financial market literature shows that the options-based measure of implied volatility contains information related to future stock market volatility that cannot be derived from historical stock prices. Prominent examples of such implicit volatility measures are the VIX or its German counterpart, the VDAX. In this research project, an implicit volatility measure for stocks of individual companies is developed, which is based on high-frequency option prices. This measure is calculated for a sample of European and US American companies and makes it possible to examine several important aspects relating to the assessment of stock market risk.

• Intraday Volatility Forecasting with Neural Networks and Pattern Recognition

(Prof Franziska Peter and Thomas Heil) Research in “Computational Sciences” is largely a result of the availability of new data. Not only extraordinarily large data sets (“Big Data”) require new analysis methods and new statistical procedures for their analysis. Especially high-frequency traders with a focus on "algorithmic trading" need input and analysis at a very high frequency. It is therefore of great importance to be able to predict the volatility structures within a day. In addition, this allows us to filter and understand consistent structures (“Pattern Recognition”) within one trading day. The analysis of these structures and their prediction is carried out using algorithms from the field of machine learning. These new findings make it possible to create new statistical models and key figures for intra-day volatilities.

• Value at Risk/Expected Shortfall calculation from Neural Network Density Forecasts

(Prof. Franziska Peter and Thomas Heil) The calculation of the Value at Risk or Expected Shortfall requires the estimation of the expected return as well as the estimation of the future volatility. However, greatly simplifying assumptions about the latent process of volatility often lead to inaccurate calculations of the value at risk and thus the expected shortfall. In principle, classical statistical models offer the possibility of predicting a density function from time series of returns, but these models are subject to severe restrictions, which leads to a poor prediction of the non-negative densities. The applicability of machine learning (neural networks, SVM) is proving to be very promising in the prediction of densities, since as the ultimate function approximators it is superior to classical statistical models when dealing with non-linear data. Based on such an estimate or prediction of the return densities, a direct calculation of the value at risk and the expected shortfall can be made. In addition, it is possible to increase the frequency of observations in order to be able to identify patterns in the volatility structure within a day. The prediction of the densities is done with "Mixture Density Networks".