Title: Application of BigVAR: Forecasting fund shares using social variables
Authors: Viktoria Oellerer - PwC Vienna (Austria) [presenting]
Abstract: Issuers of funds need to forecast the purchase and disposal of fund shares of their clients. This is especially important for real estate funds, where selling/buying underlying assets takes a considerable amount of time. As a solution, one can use traditional time-series models relying on macroeconomic variables (e.g. vector autoregression models). In todays big data era, it has become quite common to use a large number of social factors trying to model the behavior of people. We want to combine these two ideas by forecasting the purchase and disposal of fund shares using both macroeconomic and social variables. The presented approach builds on the VARX-L, a regularized vector autoregression model allowing for exogenous variables. The regularization aspect of VARX-L improves forecasting accuracy, the introduction of exogenous variables facilitates stress testing.