Linear multifactor models are of great importance in portfolio construction and risk management since they provide market dimensionality reduction, which has numerous useful implications. In particular, factor models reduce the dimensionality of the asset covariance matrix, allowing for better estimation. Since estimation is traditionally based on historical time series of asset returns and other historical attributes, a model based on such a covariance matrix tends to suffer from a time lag, ie, it underestimates risk during a crisis period and overestimates it when the market panic subsides. Two novel methodologies for incorporating forward-looking market data into the classical multifactor fundamental models are presented in this paper. The first technique utilizes general market indicators such as the VIX index to dynamically adjust certain model parameters, allowing the model to become more responsive to current market conditions. The idea of having certain parameters of the model depend on time and market conditions is quite intuitive, resembling the "business time" concept, and can be applied in various other contexts when time series are used to estimate model parameters. The second methodology incorporates data from individual securities, such as option-implied and realized volatility. Both techniques are thoroughly back-tested on multiple portfolios using US market historical data. When compared with the traditional methodology, the new models exhibit significantly higher correlation between realized and model-predicted variance, and show reduced forecast bias. The new techniques incorporate and reflect market sentiment more accurately and prove to be of greater benefit to portfolio risk management.