Econometric Modelling of Changing Time Series

We model expenditure on food in the USA, using an extended time series.  Even when a theory is essentially 'correct', it can manifest serious mis-specification if just fitted to data, ignoring its observed characteristics and major external events such as wars, recessions and policy changes.  When the same theory is embedded in a general framework embracing dynamics and structural breaks, it performs well even over an extended data period, as shown using Autometrics with impulse-indicator saturation.  Although this particular illustration involves a simple theory, the point made is generic, and applies no matter how sophisticated the theory.