The aim of this thesis is to find a fixed income trading strategy basing on liquid future data. We will exploit Reinforcement Learning technique, with Fitted Q-Iterations algorithm and Tree Based Regressors. We will seek the most appropriate model among three different regressors, using Python as our main software tool. In doing so, we will investigate models' architectures and performances, and their stability across time. We will find something already well known in the literature, tree ensemble methods provide better results. Among them, the Extremely Randomized Trees method seems to perform the best.
Lo scopo di questa tesi è quello di trovare una strategia di trading a reddito fissobasata su contratti future liquidi. Sfrutteremo tecniche dell’Apprendimento conRinforzo, con l’algoritmo Fitted Q-Iterations e i Tree Based Regressors.Cercheremo il modello più appropriato tra tre diversi regressori, utilizzando Pythoncome nostro principale strumento software. In tal modo, esamineremo le architetturee le prestazioni dei modelli e la loro stabilità nel tempo.Troveremo qualcosa di già ben noto in letteratura, i metodi che creano forestedi alberi forniscono risultati migliori. Tra questi, il metodo degli ExtremelyRandomized Trees performa meglio.
Tree-based batch mode reinforcement learning : an application to European government futures
CRUCILLÀ, ALESSANDRA
2020/2021
Abstract
The aim of this thesis is to find a fixed income trading strategy basing on liquid future data. We will exploit Reinforcement Learning technique, with Fitted Q-Iterations algorithm and Tree Based Regressors. We will seek the most appropriate model among three different regressors, using Python as our main software tool. In doing so, we will investigate models' architectures and performances, and their stability across time. We will find something already well known in the literature, tree ensemble methods provide better results. Among them, the Extremely Randomized Trees method seems to perform the best.File | Dimensione | Formato | |
---|---|---|---|
2021_04_Crucilla.pdf
accessibile in internet solo dagli utenti autorizzati
Dimensione
6.32 MB
Formato
Adobe PDF
|
6.32 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/174913