So your question is basically asking why a purely statistical model would be better at forecasting than a model based on theoretical relationships. For example, think about a simple form of the ARIMA model i.e. an AR model where you're just using lags of the process to explain...
Deciding between alternative approaches in macroeconomics
- Citation data:
International Journal of Forecasting, ISSN: 0169-2070, Vol: 34, Issue: 1, Page: 119-135
- Publication Year:
- Business, Management and Accounting
- Most Recent Tweet View All Tweets
Macroeconomic time-series data are aggregated, inaccurate, non-stationary, collinear and rarely match theoretical concepts. Macroeconomic theories are incomplete, incorrect and changeable: location shifts invalidate the law of iterated expectations and ‘rational expectations’ are then systematically biased. Empirical macro-econometric models are non-constant and mis-specified in numerous ways, so economic policy often has unexpected effects, and macroeconomic forecasts go awry. In place of using just one of the four main methods of deciding between alternative models, theory, empirical evidence, policy relevance and forecasting, we propose nesting ‘theory-driven’ and ‘data-driven’ approaches, where theory-models’ parameter estimates are unaffected by selection despite searching over rival candidate variables, longer lags, functional forms, and breaks. Thus, theory is retained, but not imposed, so can be simultaneously evaluated against a wide range of alternatives, and a better model discovered when the theory is incomplete.