Alina Tänzer, Lufthansa
Developing ideas for research projects on the flight
Alina Tänzer joined the IMFS in April 2017, supporting the Macroeconomic Model Comparison Initiative team and working as a research assistant. Just recently, she completed her Ph.D. in economics at Goethe University. Before focusing on economic research, Alina terminated her airline pilot training with Deutsche Lufthansa. Since 2018, she continued her career as First Officer, flying for Lufthansa alongside her Ph.D. studies.
In her research, she deals with different methods of artificial intelligence and applies them to solve monetary policy related questions. In a forecast comparison, artificial neural networks are trained to predict core macroeconomic variables and forecasts are compared to conventional prediction methods. The effectiveness of unconventional monetary policy is analyzed in a model comparison analysis on the one hand, and within a nonlinear model economy based on artificial neural networks on the other hand. Further, optimal monetary policy reaction functions are derived making use of another branch of machine learning called reinforcement learning. Alina's research shows evidence for artificial intelligence to provide powerful and flexible tools which can improve upon conventional methods and which are therefore very useful in the field of economics.
How would you describe your job to other people?
Being a pilot is a very exiting occupation, which combines several dimensions of technical knowledge with practical skills. Before each flight, you make the fuel calculations, check the weather along the route, and the technical condition of the airplane. It is important to process this information and to consider possible threats which might be present. Hence, it is required to be focused and to notice every detail. During the flight, the main task is to steer the airplane of course, which is done by giving the correct commands to the autopilot and by checking their execution. Continuously, the surrounding weather and terrain conditions are evaluated. Especially during takeoff and approach, good manual flying skills are required to keep the aircraft in the desired operating envelope. Hence, while being trained to handle emergency situations, a normal working day is in most cases rather uneventful.
What do you like most about your job?
The view from my "office". Seeing the world with a bird's eye view is very impressive and often times puts personal struggles in a different light. Furthermore, I especially like the connection to the meteorological and geographical environment and the communication with air traffic controllers and handling staff of different countries. Last but not least, the execution of one flight is always based on team work. I very much like this interpersonal aspect as well.
What was the main focus of your research at the IMFS?
In my research, I contrast novel methods of artificial intelligence against conventional approaches with the goal to answer macroeconomic and monetary policy related questions. In one research project, I provide a forecast comparison of a parsimonious and fundamental artificial neural network (ANN), a dynamic stochastic general equilibrium model (DSGE), a Bayesian VAR, as well as official forecasts. Training or estimating either model with U.S. real-time data in an expanding window framework reveals a superior forecasting performance by the ANN. This superiority increases with the forecasting horizon and with the actuality of the data. As even this benchmark ANN yields promising results, further applications of ANNs for forecasting but also in a modeling-context is recommended.
Two of my research projects deal with the effectiveness of unconventional policy, i.e. the central banks' purchases of treasury securities. A structural model comparison of several DSGE models aims at comparing QE effects across models to tackle the problem of model uncertainty. The results provide evidence for QE having rather little impact on the economy, which questions large investment volumes and motivates balance sheet normalization.
To overcome the problem of model uncertainty, in another project I employ a novel empirical approach making use of a nonlinear ANN-based model economy, which does not require a-priori assumptions about the transmission channel of QE. The results point to asymmetric effects of this policy intervention which are contingent on the prior state of the economy. It can be said, that the deeper the crisis, the more powerful is QE. Nevertheless, also this project underlines rather small overall effects.
The idea for another research project was developed when I was thinking about the working mechanism of the autopilot and the underlying principles of control theory. As it relies on circular dependencies which can also be found in a macroeconomic context, the methods should be transferrable as well. Hence, a machine-learning technique called reinforcement learning, which is an enhancement of control theory, is employed to find optimal monetary policy reactions functions. The rules (linear and nonlinear) are trained in two different environments which are based on a structural VAR and a nonlinear ANN economy. The trained rules can be shown to reduce the central bank loss by over 43%, assuming an inflation target of 2% and an output gap target of zero.
The findings of my research projects allow to conclude that methods of artificial intelligence can improve upon standard approaches, especially when nonlinearities are present or underlying functional interrelations are unknown. The promising results motivate further research in this interdisciplinary field.
Can you benefit in any way from your research at IMFS in your job at Lufthansa? / How can your research help you being a good pilot at Lufthansa?
There is clearly no direct link between research on monetary policy and the occupation as a pilot. However, there are some relevant skills that I developed through doing research such as stress resistance, the ability to put confuse information in order or to have an eye for the detail. Besides that, the teamwork within the IMFS initiative allowed to train many interpersonal skills which are a prerequisite for any job-related and also private relationship.
Furthermore, my plan is to complement my job as a pilot by another function which will take more advantage of Ph.D.-related knowledge as regards to content. I would especially like to exert and strengthen my skills in artificial intelligence as I see many use cases of these techniques in business companies and policy agencies.
What did you enjoy most regarding your time at the IMFS?
Working at the IMFS offered a lot of opportunities which supported my research path and also my future career plans. On the one hand, working on a great project as the macroeconomic model comparison initiative widened the knowledge about project management and it offered several touch points with an international research community. On the other hand, we had a lot of free space to work on projects of our own which facilitated to progress with the doctoral thesis. I would say working for the IMFS was a welcome chance to start working in a research-related field and to conduct own thesis-related research at the same time.