Thanks Finance project: financial sentiment analysis and profiling questionnaire to assess savers’ risk attitude

As a member of the European Union, Italy is one of the worst states in terms of financial knowledge. Italian students in this field are well below the averages of the OECD countries and adults are historically distrustful of the financial sector. However, in the last few years something is moving, COVID-19 pandemic gave a boost to the growing trend of online traders, authorized brokers and managed portfolios. The large amount of Italians' savings denounces the absence of software and tools able to fill this lack of knowledge and trust in the described context. In this thesis, Thanks Finance - a fintech innovative start-up based in Turin - aims at reassessing the perception of the Italian population towards finance, through a hybrid use of linguistics, machine learning and behavioral economics. The company is working on two software solutions: the first is a financial Sentiment Analysis web platform for traders and professionals, the second one is a platform to automate the collection and analysis of risk tolerance profiling questionnaires based on behavioural finance and MIFID II regulation. To meet the needs of the company this work aims at providing an extensive review of the literature on financial sentiment analysis through NLP techniques from text representation methods to the latest NLP transformers. In addition, it provides a state of the art analysis on risk tolerance profiling based on behavioural finance and likely classification algorithms such as logistics regression and clustering. The literature review is followed by the description of the two solutions: firstly, the sentiment analysis web platform is discussed by deepening backend development from data collection with scrapy and Reddit API, going through NLTK Vader Sentiment analysis and training stage on AWS Comprehend (NLP Cloud Service), ending with BE FastAPI development process and definition of functional and not functional software requirements plus basis of frontend development. Concerning the second solution, which involves the questionnaire based on behavioural finance designed by professors of UPO, the work describes how data collection has been performed through the modular approach of Google Sites and the development of a questionnaire application in Django plus ReactJS for future data collection. After gathering data a model to classify compilers out of 5 classes has been defined. Classes identify the level of risk tolerance and go from a very low to a very high one, namely “Very Low”, “Low”, “Average”, “High”, “Very High”. Statistical analysis is performed on 195 questionnaires collected in order to define 5 risk classes’ features. Results show relevance to literature and have been validated by the team of financial experts that will design the prototype portfolios in terms of macro-asset classes. The long-term objective of Thanks Finance is to produce a Software as a Service (SaaS) which integrates financial sentiment analysis and risk tolerance classification, in order to sell annual subscription plans to financial advisors. At the end of the section on the profiling questionnaire a short high level description of the SaaS is proposed.