Numerous contexts in macroeconomics, finance, and quality control require real-time estimation of trends, turning points, and anomalies. We formulate the real-time signal extraction problem as a multivariate linear prediction problem, present the optimal solution in terms of a known model, and propose multivariate direct filter analysis to address the more typical situation where the process' model is unknown. We show how general constraints - such as level and time shift constraints - can be imposed on a concurrent filter in order to guarantee that real-time estimates have requisite properties, and apply the methodology to petroleum, capitalization, and construction data.