Manipulating Correlations Through Latent Drivers

The ability to manipulate correlation matrices is useful for a number of applications in finance, including stress testing and portfolio construction. We outline a flexible framework that enables managers to manipulate correlations based on their beliefs and preserve the positive semi-definiteness of the matrix. The framework links the returns for which correlations are captured in the matrix through unobservable factors or drivers, called latent drivers. Changes in correlations can then be introduced through these drivers. Users can specify levels of exposure to a driver based on their prior beliefs. The framework can be applied to any correlation matrix - asset, factor, or asset class