We have argued that a major use of software metrics is in the forecasting problem for software projects. By analogy with weather forecasting, we may characterize the current state of knowledge in software forecasting as the gathering of portents. While these may be useful and sometimes decisive in project management, they are prescientific and qualitative. Further, it seems very unlikely that the portents can be developed into a useful theory of forecasting. To develop scientific forecasting tools, a rational way of predicting the future from historical primary data is required. It is also important that the primary data and the measurements used to obtain it satisfy some basic methodological requirements -- for example, the hypotheses developed from the measurements should be meaningful in the sense implied by measurement theory. The statistical approach, seeking to predict future events on the basis of historical patterns, seems to be an attractive short range approach to the forecasting problem. The goal of the exact method is to be able to apply largescale computation to many micropredictions to synthesize a quantitative forecast.