A carregar...
Publicação
Forecasting sovereign bonds markets using machine learning: forecasting the portuguese government bond using machine learning approach
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 1.74 MB | Adobe PDF |
Orientador(es)
Resumo(s)
Financial markets, due to their non-linear, volatile and complex nature turn any type of forecasting
into a difficult task, as the classical statistical methods are no longer adequate. Many factors exist
that can influence the government bonds yields and how these bonds behave. The consequence of
the behaviour of these bonds are extended over geographies and individuals.
As the financial markets grow bigger, more investors are trying to develop systematic approaches
that are intended to predict prices and movements. Machine Learning algorithms already proven
their value in predicting and finding patterns in many subjects. When it comes to financial markets,
Machine Learning is not a new tool. It is already widely used to predict behaviours and trends with
some degree of success.
This dissertation aims to study the application of two Machine Learning algorithms - Genetic
Programming (GP) and Long Short-Term Memory (LSTM) - to the Portuguese Government 10Y Bond
and try to forecast the yield with accuracy. The construction of the predictive models is based on
historical information of the bond and on other important factors that influence its behaviour,
extracted through the Bloomberg Portal.
In order to analyse the quality of the two models, the results of each algorithm will be compared. An
analysis will be presented regarding the quality of the results from both algorithms and the
respective time cost. In the end, each model will be discussed and conclusions will be taken about
which one can be the answer to the main question of this study, which is “What will the Yield of the
Portuguese Government 10Y Bond be on T+1?”.
The results obtained showed that Genetic Programming can create a model with higher accuracy.
However, Long Short-Term Memory should not be ignored because it can also point to good results.
Regarding execution time, velocity is a problem when it comes to Genetic Programming. This
algorithm takes more time to execute compared to LSTM. Long Short-Term Memory is considerably
quicker to get results. In order to take the right decision about which model to choose one must keep
in mind the priorities. In case accuracy is the priority, Genetic Programming will be the answer.
Nevertheless, when velocity is the priority Long Short-Term Memory should be the choice.
Descrição
Dissertation presented as the partial requirement for obtaining a Master's degree in Statistics and Information Management, specialization in Risk Analysis and Management
Palavras-chave
Portuguese Government Bonds Machine Learning Genetic Programming Long-Short-Term Memory Financial Markets SDG 8 - Decent work and economic growth