Accelerating the distribution of financial products through classification and regression techniques

Background context. To date, financial products mostly consists in instruments used by households to prepare for retirement (consumption smoothing across the various stages of and one's life-cyle) and/or to transfer wealth across generations. However, the current financial system is not completely efficient in doing so as the level of participation of households remains low. This calls for potentially new tools and techniques aimed at accelerating the distribution of the said products. Specific knowledge gap the work aims to fill . The current studies available on Fintechs & the distribution of financial products mostly revolve around notions of robo-advisors. Use cases usually orbit around automatically structuring \& managing households portfolios under risk (incl. diversification) and performance constraints (e.g. what is the mix of bond and equity to use within a retirement plan to optimize the performance measured as the Sharpe ratio?). However there is no available study on the choices made by households on the products themselves, the associated drivers and the level of investments, a gap this study aims to bridge. Methods used in the study. This study leverages a private data-set from a French Fintech describing at a macro level the structure of 1.5K+ households and their wealth. Post feature selection and data cleansing, the information is fed to standard classification algorithms (e.g. random forest, support vector machine...) and regression techniques (linear regression, multiple kernel estimators...) to predict whether or not households are likely to subscribe to a life insurance or a retirement plan or a real estate program over the forthcoming year and to forecast the associated level of investment . Key findings. The current data-set shows that macro levels indicators on French household structure and wealth can be used to predict their subscription behavior over the next 12 months towards core investments products with a high level of performance (A.U.C >90%). However, macro level information appears insufficient to predict the level of investments on those products (R^2