Is more data always better? Optimal data usage in non-stationary systems

Starting from the premise that risk is a notion that changes over time this paper introduces a statistical methodology that is able to identify the amount of data yielding minimal bias estimators. I formalise the issue of... [ view full abstract ]

Starting from the premise that risk is a notion that changes over time this paper introduces a statistical methodology that is able to identify the amount of data yielding minimal bias estimators. I formalise the issue of representativeness and introduce a trade-off between estimator convergence, incentivising us to use as much data as possible and, and the aforementioned issue of representativeness doing the opposite. As an application an impact study of a paragraph in the banking regulation framework is adressed.