History is full of broad long-term trends with many local short-term exceptions. People seem to me a bit too eager to try to learn from those exceptions, compared to learning from the trends. A more complex trend model might better fit the exceptions, but beware over-fitting.