How do you identify and mitigate multicollinearity in a predictive model designed for stock price forecasting?