First supervisor: Prof. Irwin Collier
Second supervisor: Prof. Jonathan Fox
Third supervisor: Dr. Till Strohsal
The global financial crisis of 2007 resulted in the creation of macro-prudential oversight bodies at central banks and ministries around the globe. A main component of their purpose is to predict financial crises based on historical economic data (early-warning models). For this, advanced econometrical, and more recently also machine learning models are trained based on relatively recent data. In the past ten years, economics has benefited greatly from the ever growing availability of data. This trend is also reflected in the rise of interest in quantitative economic history (cliometrics) which I seek to harness and apply to the field of financial stability.
Modern macro-prudential oversight is, almost exclusively, an empirical exercise that seeks to predict (and prevent) financial crises by employing sophisticated tools trained with relatively recent economic data. Most models are fed with time series that start no earlier than in the 1970s for most economies. The coverage for some other, smaller countries may even be shorter. On the theoretical side, modern macro-prudential oversight lacks fundamental underpinning. There is no explicit resorting to theory and only implicit assumptions about the workings of the economy through the selection of variables and parameters as model inputs.
In my dissertation I seek to address the question of how historical considerations, be it in the form of historical data series or economic theories, can enhance macro-prudential oversight. There are two main questions I want to answer specifically. First, I assess how prolonging the time-dimension of input data affects the outcome of financial crisis prediction models and, second, I like to contribute to the neglected theoretical side of macro-prudential oversight by aligning modern model results with crises theories from the realm of the history of economic thought.
To implement the above agenda, I will use the United States as a laboratory to see whether prolonging the time dimension yields better or otherwise valuable prediction results for financial crises. Additionally, theories of US-based scholars such as Hyman P. Minsky will be used to attempt an alignment of empirical results with theory in the realm of macro-prudential oversight. For the empirical exercises I construct a standard early-warning input data set with extended time dimension for the Unites States (and potentially for other countries as well) with a variable selection in accordance to the literature.