Predicting investor behaviour in European bond markets through machine learning
Auf einen Blick
- Projektleiter/in : Dr. Martin Hillebrand, Prof. Dr. Peter Schwendner
- Projektteam : Dr. Marko Mravlak, Dr. Martin Schüle, Bastien Winant
- Projektvolumen : EUR 200'000
- Projektstatus : abgeschlossen
- Drittmittelgeber : EU und andere Internationale Programme (European Stability Mechanism ESM)
- Projektpartner : European Stability Mechanism ESM
- Kontaktperson : Peter Schwendner
Beschreibung
ICMA Quarterly Report (11.7.2019), p.23:
Predicting investor behaviour in European bond markets through
machine learning
The quant team of ESM is developing, in cooperation with the Zurich
University of Applied Sciences, a machine-learning based
application to predict investor demand for syndicated bond
issuances.
A key reason for using machine learning algorithms is the ability
to analyse complex and high-dimensional data sets with widely
unknown structures, to capture complex dependencies and relations
of variables and identify any kind of patterns in the data.
The analysis comprises diverse datasets, including
transaction-related data such as orderbooks, internal and external
primary and secondary market data, including secondary market
transactions reported by primary dealers. In addition, it comprises
internal and external investor-specific data as well as
macroeconomic data. The applied ML methodology is promising. First
results show a prediction power of well above 50% of investor
demand by investor type (such as banks, brokers, fund managers,
pension funds or insurance). While results for individual investors
were overall less accurate due to smaller data sets, qualitative
information and behaviour patterns of specific investors could be
detected. These results can help better to understand and address
investor needs and consider this in the transaction planning and
execution. This machine-learning application to predict investor
demand is considered work in progress.
Further improvements of data quality, inclusion of further data
sources, and a refinement of the used ML algorithms are expected to
improve forecasts substantially. However, a key limiting factor is
the availability of data despite access to diverse data sources,
including primary dealer reporting. The inclusion of further
primary dealer data can play an important role. ML technology can
help solve potential confidentiality issues.
www.icmagroup.org/assets/documents/Regulatory/Quarterly_Reports/ICMA-Quarterly-Report-Third-Quarter-2019v2.pdf