A Comment on James M. Poterba's ‘Demographic Structure and Asset Returns’

Abstract

THE last twenty years have seen dramatic changes in U.S. equity valuations. Measured in relation to dividends or earnings, current valuations are wildly different from their long-term historical averages. Figure 1, taken from Campbell and Shiller (2001), illustrates this point. The bottom-left panel shows the dividendprice ratio (D/P) for the SP the previous peak was 28 in 1929. The decline in the SP because of the large size of the generation, this has increased national savings and bid up the prices of financial assets. The difficulty with this argument is that the overall U.S. savings rate does not appear to be unusually high; also, an increased propensity to save should bid up the prices of all financial assets, including bonds, and thus should drive down real interest rates. Other versions of the argument emphasize risk taking rather than overall saving. One story says that investors are more willing to take equity risk early in middle age; thus, a large cohort of early middle-aged investors should drive up equity prices. Another story says that baby boomers are intrinsically more tolerant of equity risk and less tolerant of bond market risk than earlier cohorts were, perhaps because they have direct memories of the 1970s inflation, with its bad effects on nominal bonds, but not of the 1930s depression, with its bad effects on equities. These stories are quite different, both in their mechanisms and in their implications for the future. The first story relies on an age effect, which will reverse as the baby boomers age; the second story relies on a cohort effect, which is a permanent attribute of a generation. Thus, the first story leads to concerns that stock prices may decline as the baby boomers enter old age. James Poterba’s paper (2001) correctly emphasizes that these arguments can be evaluated rigorously only if one can identify time, age, and cohort effects on asset demands. The difficulty is that time, age, and cohort effects are linear combinations of one another and can never be separately identified without theoretical restrictions. Poterba emphasizes empirical results that restrict time effects to be zero and allow for freely estimated age and cohort effects, on the grounds that this restriction can reconcile a flat age-wealth profile at any point in time with a hump-shaped effect of age Received for publication June 12, 2001. Revision accepted for publication June 12, 2001. * Harvard University.