Bivariate integer-autoregressive process with an application to mutual fund flows (Q2274940)

The paper is focused on bivariate nonnegative integer-autoregressive (BINAR) models with applications in monitoring of the liquidity risk of an open-ended mutual fund. This can be of great interest to understand clients' behavior and the manager reaction to exogenous liquidity shock. After presenting the previous results about similar models, in this paper the existing BINAR(1) model is generalized in a form which allows intuitive interpretations, as well as tractable aggregation and stationarity property. Also it is shown that in this generalized family of BINAR models, the predictive distributions at various horizons are easily computable via a matrix-based algorithm. The obtained extended BINAR model allows to introduce higher order dependent BINAR(p) and BINAR(\(\infty\)) processes which also have for closed-form predictive distributions at any horizons and for any values of p, and can better capture slowly decaying serial correlation patterns. A BINAR(\(\infty\)) model with memory persistence is applied to open-ended mutual fund purchase and redemption order counts, on a daily basis, not on a weekly one, which permits a better control of the vulnerability to liquidity risk and a daily monitorization in parallel the outflow and inflow. The paper ends with some conclusions about bivariate models and possibility to extend to higher dimensions.