Master Thesis Topics

On this page you can find a list of possible research topics for a master thesis. This list is not exhaustive, but can be supplemented with your own topic suggestions or extensions to our proposed topics. For an insight into the expertise and research interests of our team members, also take a look at the personal pages in the "Team" section of our website and feel free to directly get in touch with the person that best fits your suggested topic or interests.

Available projects

Can Data Integration Improve Consumer Demand Planning in the Photovoltaic Industry?

• Forecasting customer demand is a critical business process in most industries. Forecasting solutions in business practice often focus on manual forecasting (“expert forecasts”) or timeseries forecasting. However, these approaches neglect the explanatory power of important external factors for customer demand and may thus lead to poor forecast accuracy in the volatile business environment of the PV industry.

• The goal of this work is to assess whether integrating internal and auxiliary data (e.g., historical data, seasonalities, expert forecasts, time series forecasts, customer demand signals, market data, price developments, Google trends, …) can be leveraged to build a feature-driven demand forecasting model and to improve the demand forecast accuracy at BayWa r.e. Solar Energy Systems.

• This thesis requires solid knowledge of and experience with R / Python as well as experience with Machine Learning.

• Supervisior: Prof. Dr. Dominik Papies with Dr. David Scheuermann (BayWa r.e. Solar Energy Systems)

Can ChatGPT read and interpret the Terms of Service of other platforms and services

• Many platforms and services provide Terms of Service and Privacy Policy and that are many pages long and that would take hours and hours to read and understand. This leads to the undesirable situation that many users accept these policies and terms without reading them.

• The goal of this work is to assess whether we can leverage large language models (LLMs like ChatGPT) to understand, interpret, and summarize these terms and conditions for us.

• This thesis requires knowledge of and experience with R / python as well as experience with Machine Learning

• Supervisor: Prof. Dr. Dominik Papies

The problem of zero in log-log regressions

• The workhorse model in applied econometrics, economics, marketing, and other disciplines is a regression model that involves taking the log of one or more variables. This, however, is difficult if one or more values of the relevant variables involve zeros because the natural logarithm of zero is infinity. It is not obvious whether adding a (small) positive constant to zero is a generally valid solution.

• The goal of this work is to assess different approaches (e.g., Poisson regression) of how to deal with this problem on simulated and real data sets and to provide guidance and best practice recommendations to applied researchers.

• This thesis requires knowledge of and experience with R / python as well as good econometric knowledge.

• Supervisor: Prof. Dr. Dominik Papies

Competition in the German gasoline market

• The German gasoline market differs from other gasoline markets because it is characterized by a very high frequency of price changes, i.e., a typical gas station changes its prices multiple times a day. The question that arises is to what extent these price changes are predictable and whether dynamic cycles of price increases or decreases are always initiated by the same stations or brands, or whether these patterns are entirely unpredictable.

• To address these questions, the thesis will analyze a large data set that contains all price changes of all stations in the German gasoline market over several years. 

• This thesis requires knowledge of R and reasonable econometric knowledge.

• Supervisor: Jun.-Prof. Dr. Wiebke Keller

New (machine learning) methods for causal inference  

• In recent years, a multitude of new methods for causal inference from observational data have been developed by researchers from very different fields (econometrics, machine learning, computer science, epidemiology, ...). Examples include, among others, double machine learning, causal forests, deep instrumental variables, front-door adjustment, causal discovery, and targeted maximum likelihood. In many cases, these methods have laid the theoretical groundwork, but have not yet been applied widely to typical questions from business or economics. 
   
• The goal of this thesis is to select from these methods and a) compare multiple methods from different disciplines that have the same goal, and/or b) assess the applicability of one of the new methods to typical research questions and datasets from business and economics, and/or c) explore how results from influential studies in business and economics (using traditional methods) would change, when one of the new methods is applied.
   
• This thesis is suitable for students who have experience with empirical work, good programming skills in R or Python, and reasonable econometric knowledge.
   
• Supervisor: Prof. Dr. Dominik Papies with Jonathan Fuhr, M.Sc.