The permanent income hypothesis implies that people save because they rationally expect their permanent income to decline; they save "for a rainy day." It follows that saving should be at lease as good a predictor of declines in labor income as any other forecast that can be constructed from publicly available information. The paper tests this hitherto ignored implication of the permanent income hypothesis, using quarterly aggregate data for the period 1953-84 in the United States. By contrast with much of the recent literature, the results here are valid when income is stationary in first differences rather than levels.