We propose a risk-management strategy for portfolio allocation based on volatility scaling. The strategy involves decomposing realized volatility according to the magnitude and sign of a given return and, then, using part of the realized variance to design volatility-scaled versions of traditional portfolios. By applying our method to four risk-portfolios (namely, market, small minus big, high minus low, and winners minus losers), we show that scaling according to an appropriate criterion (i.e. the realized volatility of the largest negative returns) increases the profitability of the original strategies, while it simultaneously reduces other risks related to market crashes. The better economic performance of our method – the inter-quantile semivariance model – lies in its better adjustment to the market liquidity of our statistics, and more accurate modeling of the risk-return relationship and of the asymmetric impacts on consumption, production and asset prices, generated by a different fragment of the market realized variance.