Factor models seek to reduce the impact of spurious in-sample correlations on the estimation of a covariance matrix of security returns. Among U.S. equities, market value is concentrated in a small number of assets both across the market and within industry sectors. Over 50% of market value, for example, is captured by 1% of assets. This can unwittingly compromise factor model design and result in a significant misprediction of portfolio risk. Correlations may be induced between the specific returns of high-cap securities as an artefact of regression procedures. There are solutions to avoid such mistakes and thus improve the efficacy of risk predictions.