Forecasting for large changes in demand should benefit from different estimation than that used for estimating mean behavior. We develop a multivariate forecasting model designed for detecting the largest changes across many time series. The model is fit based upon a penalty function that maximizes true positive rates along a relevant false positive rate range and can be used by managers wishing to take action on a small percentage of products likely to change the most in the next time period. We apply the model to a crime dataset and compare results to OLS as the basis for comparisons as well as models that are promising for exceptional demand forecasting such as quantile regression, synthetic data from a Bayesian model, and a power loss model. Using the Partial Area Under the Curve (PAUC) metric, our results show statistical significance, a 35 percent improvement over OLS, and at least a 20 percent improvement over competing methods. We suggest management with an increasing number of products to use our method for forecasting large changes in conjunction with typical magnitude-based methods for forecasting expected demand.