Working Paper: NBER ID: w2359
Authors: James H. Stock; Kenneth D. West
Abstract: We use recent research on estimation and testing in the presence of unit roots to argue that Hall's (1978) t and F tests of whether consumption is predicted by lagged income, or by lags of consumption beyond the first, are asymptotically valid. A Monte Carlo experiment suggests that the asymptotic t and F distributions provide a good approximation to the actual finite sample distribution.
Keywords: Permanent Income Hypothesis; Consumption; Unit Roots; Cointegration
JEL Codes: C12; C22; D91
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| Consumption is predicted by lagged income (E21) | Hall's regression is legitimate (C29) |
| Consumption and income are difference stationary and cointegrated (E21) | Hall's regression is legitimate (C29) |
| Presence of time trends does not affect Hall's regression validity (C22) | Hall's regression is legitimate (C29) |
| Lack of including a lag of consumption leads to bias in estimating sensitivity of consumption to income (E21) | Mankiw and Shapiro (1985) results differ from Hall's regressions (E19) |
| Hall's tests involving lags of income are asymptotically valid (C22) | Evidence against the permanent income hypothesis in other circumstances is not precluded (D15) |