Today's, the new economic and financial world, is more often oriented toward processes automation, which until few year ago were under the human control. This work has as objective, to examine the birth and the evolution of those processes and the so-called "human-machine interactions", under the point of view of both regulation and automation. In particular, this work is oriented on analysing those techniques regarding the cluster analysis and the neural networks, in order to better predict those risk classes associated to a set of financial products, in this case a group of 52 ETFs, comparing them to the classes associated to the relative customers, established through the actual MiFID directive. So, this work first will analyse the FinTech world, in which Robo-Advisory is collocated, and then the procedures linked to the regulamentation, through its major firms operating in this field, and its financial authorities. Finally, it will exploit clustering techniques, in order to determine the actual risk classes for every financial product in object, trying to verify their correct matching through the use of machine learning, in particular the neural networks. The best result and the most predictive is the one obtained dividing those 52 ETFs into 4 clusters, and then giving, as inputs for the neural network, a train set of 31 days, composed by ETFs' returns, exploiting a first layer composed by 5 neurons and the a second one composed by 3.
Sempre più spesso, al giorno d’oggi, anche il mondo economico e finanziario sta puntando verso l’automazione di processi che fino a qualche decennio fa erano di esclusiva competenza umana. Questo lavoro si pone come obiettivo quello di esaminare l’instaurarsi di tali nuove relazioni “uomo-macchina”, sia dal punto di vista regolamentativo che dal lato dell’automazione in sé. In particolare, si pone come interesse quello di analizzare le tecniche di analisi dei gruppi, i.e. clustering, e le reti neurali, i.e. neural networks, al fine di meglio predire le classi di rischio associate ad un insieme di asset o prodotti presi in analisi, in questo caso un gruppo di 52 ETF, paragonandole a loro volta con le classi di rischio dei relativi clienti, stabilite secondo la direttiva MiFID vigente. Perciò, questo lavoro è impostato analizzando per primo l’ambito di collocazione del Robo-Advisory in sé, ovvero la Financial Technology, per passare poi alla sua regolamentazione tramite l’analisi delle maggiori compagnie che operano in tale ambito e i principali organi di vigilanza. Infine sfrutta le tecniche di clustering per determinare le classi di rischio per ogni singolo prodotto finanziario in oggetto, verificandone tale correttezza di classificazione tramite l’utilizzo del machine learning e in particolar modo tramite lo sfruttamento di reti neurali. Il risultato migliore e più predittivo avviene dividendo in 4 clusters i 52 ETF in esame, per poi passare come input per la rete neurale un train set di 31 giorni, composto dai ritorni dei relativi ETF, sfruttando un primo layer di 5 neuroni e un secondo di 3.
Clustering and Network models for Robo-Advisory
RE, RICCARDO
2017/2018
Abstract
Today's, the new economic and financial world, is more often oriented toward processes automation, which until few year ago were under the human control. This work has as objective, to examine the birth and the evolution of those processes and the so-called "human-machine interactions", under the point of view of both regulation and automation. In particular, this work is oriented on analysing those techniques regarding the cluster analysis and the neural networks, in order to better predict those risk classes associated to a set of financial products, in this case a group of 52 ETFs, comparing them to the classes associated to the relative customers, established through the actual MiFID directive. So, this work first will analyse the FinTech world, in which Robo-Advisory is collocated, and then the procedures linked to the regulamentation, through its major firms operating in this field, and its financial authorities. Finally, it will exploit clustering techniques, in order to determine the actual risk classes for every financial product in object, trying to verify their correct matching through the use of machine learning, in particular the neural networks. The best result and the most predictive is the one obtained dividing those 52 ETFs into 4 clusters, and then giving, as inputs for the neural network, a train set of 31 days, composed by ETFs' returns, exploiting a first layer composed by 5 neurons and the a second one composed by 3.È consentito all'utente scaricare e condividere i documenti disponibili a testo pieno in UNITESI UNIPV nel rispetto della licenza Creative Commons del tipo CC BY NC ND.
Per maggiori informazioni e per verifiche sull'eventuale disponibilità del file scrivere a: unitesi@unipv.it.
https://hdl.handle.net/20.500.14239/7443