Analysing the US Panel Study of Income Dynamics, we present a new empirical method to investigate the extent to which households reduce their financial risk exposure when confronted with background risk. Our novel modelling approach – termed a deflated fractional ordered probit model – quantifies how the overall asset composition in a portfolio adjusts with background risk, and is unique in recovering for, any given risky asset class, the shares that are reallocated to a safer asset category. Background risk exerts a significant impact on household portfolios, resulting in a 'flight from risk', away from riskier to safe assets.