Arthur Robson's excellent comment introduces a new formulation of the problem of consistent conjectures. My reply uses the new formulation in order to resolve Robson's troubling example of nonexistence, and to make the problem clearer. It also shows why my original approach was sensible only for the linear demand curve case. A slightly more general approach used here has consistent conjectures in Robson's quadratic inverse demand example.
The centenary of Schumpeter's birth coincides with a revival of Schumpeterian economics. Could the third quarter of this Century justly be called the age of Keynes (Hicks, 1974), the fourth quarter has a fair chance of becoming the age of Schumpeter. Before giving substance to this proposition, I shall present a Short introduction to Schumpeter's life, work and paradigm.
In a recent contribution to this Review, Alan Blinder provides an interesting analysis of the relationship between optimal inventory decisions and the process of price adjustment. Arguing that inventory permits separation of production and sale decisions, and allows the firm greater flexibility in responding to anticipated demand disturbances, Blinder shows that the responsiveness of relative prices depends critically upon the cost of storage and the temporal structure of the demand shocks. Few economists would fault the intuitive substance of Blinder's basic proposition, and the proposition is justified with characteristic precision. The analysis, however, is considerably less general than Blinder maintains. The formal argument of his paper relies upon a simple model of a monopolist capable of backlogging orders and storing unsold goods. As the model is constructed, order backlogs assume a role of fundamental importance in any stationary equilibrium. This feature of the model vitiates Blinder's extension of his result to situations in which order backlogs are prohibited. The basic monopoly model developed in Sections I and II of Blinder's paper places no restriction upon the sign of inventory. Negative inventory is interpreted as net unfilled orders, while no distinction is drawn between reductions in gross inventory and additions to the order backlog. As a result, the cost of holding inventory is symmetric around an arbitrary minimum point. The only role of inventory is to permit the time paths of production and sales to differ: gross inventory allows production to antedate sales, while order backlogs generate revenue before production costs are borne. Under these circumstances, if future costs and revenues are discounted, it is natural to expect any stationary equilibrium level of net inventory to be negative. This can be shown to be true of Blinder's model. It is useful to distinguish in general between the nonstochastic stationary level of inventory (n-) and the stochastic stationary level (n). The former is realized when the current expected demand disturbance (eo) is zero and production equals sales; the latter, which differs from customary usage, represents the level of inventory expected to prevail when adjustment to the current (nonzero) expected demand disturbance is complete. If the demand disturbance is not a random walk, so that p < 1 in Blinder's (5), the two equilibria coincide. Relatively little attention is devoted to either stationary solution, because Blinder's main interest is in the optimal current solution level (n1). The stochastic stationary level is crucially important, however, since it represents the goal toward which the current choice must provide an optimal first step. To verify that both nand n are negative whenever production is strictly positive, note that Blinder's necessary condition (7) with the coefficient cl set equal to zero, together with the definition of XA as the deviation of the shadow value of inventory (q,) from its nonstochastic stationary value (q), implies
One of the most obvious facts of recent monetary history is that high inflation is associated with high nominal interest rates. This association has been interpreted by many as supporting a superneutrality hypothesis: that an increase in inflation will not affect real interest rates in the long run.' However, the bulk of the evidence contradicts superneutrality. Beginning with Irving Fisher (1896, 1930), most empirical investigations have found that fully anticipated inflation has less than a unit effect on nominal interest rates, and thus reduces real interest rates even in the longest of runs. This has been shown under the assumption that expectations are formed rationally (Douglas Pearce, 1979), that they take the form of an arbitrary distributed lag on past inflation rates (Fisher, 1930; William Gibson, 1970), or that they are accurately represented by the Livingstone expectations data (Pearce; Kajal Lahiri, 1976). Lawrence Summers (1983) attempted to measure the long-run effect without an explicit theory of expectations. Regressing long swings in various nominal interest rates against long swings in inflation over various subintervals from 1860 to 1979, he found coefficients consistently less than unity. For the post-World War II era as a whole the coefficients were in the range of 0.5 to 0.75, with standard deviations of 0.08 to 0.33. A few studies have found coefficients close to unity (William Yohe and Denis Karnosky, 1969; Martin Feldstein and Otto Eckstein, 1970; Gibson, 1972; Lucas). But, as several authors have observed (Thomas Sargent, 1976; Robert Shiller, 1980; John Wood, 1981; Summers), these findings are limited to a particular period of U.S. history, approximately 1953-71. Furthermore, even a unitcoefficient would contradict superneutrality of the after-tax real interest rate, which would require a coefficient substantially greater than unity. Even taking into account the other inflation distortions in the tax system, Summers calculated that the coefficient ought to lie in the range of 1.3 to 1.5, far higher than observed. These empirical findings pose a challenge to traditional monetary theory, much of which implies that superneutrality should hold at least approximately. For example, the model of Miguel Sidrauski (1967) implies that the real interest rate should equal the marginal product of capital, which in the long run should equal the representative household's marginal rate of time preference. If this rate of time preference is a constant, then in particular it will be independent of the rate of inflation. If it is positively related to the household's wealth, or utility, then inflation can reduce the marginal product of capital somewhat through what are commonly called Mundell-Tobin effects. That is, higher inflation can reduce the demand for real balances, which reduces real wealth, which lowers the rate of time preference and leads to further capital accumulation.2 But this real balance effect on saving is commonly recognized to be too small to make *Department of Economics, Social Science Centre, University of Western Ontario, London, Canada N6A 5C2. We are grateful to Charles Adams, Norman Cameron, John Chilton, Jacob Frenkel, David Laidler, Ben McCallum, Baldev Raj, Brad Reid, Jack Weldon, John Whalley, Ron Winrck, and two anonymous referees for helpful discussions and comments on earlier drafts. All errors are attributable to transaction costs. 'Robert Lucas (1980) finds empirical support for the hypothesis, which he calls one of the central implications of the quantity theory of money. It has also been adopted by a wide range of more eclectic economists, as evidenced by its endorsement in two of the most popular macroeconomics textbooks (Rudiger Dornbusch and Stanley Fischer, 1981, pp. 454-58; Robert J. Gordon, 1978, pp. 289-91). 2These implications of variable time-preference can be drawn almost immediately from the work of Hirofumi Uzawa (1968). A graphical analysis is presented by David Laidler (1969b), who focuses on the logically equivalent question of the effects of paying interest on money.
Because the main objectives of the Federal Reserve System (the Fed) concern what Milton Friedman (chapter 2, p. 12) describes as the Holy Trinity — full employment, economic growth, and stable prices — scholarly evaluations of the central bank’s performance understandably center most often on the behavior over time of macroeconomic variables such as the aggregate price level, interest rates, and national income. Recently, however, Friedman and others have been drawn to the view that an explanation for the Fed’s monetary policy record may be found in the bureaucratic incentives faced by central bank officials.
In 1974, Robert Barro argued that if individuals in successive generations were linked by bequests, changes in the stock of government debt or in Social Security programs would have no effect on the steady-state capital stock. Barro assumed that individuals held static expectations and that the size of the population was constant. Martin Feldstein (1974, 1977) had concluded that Social Security would reduce the capital-labor ratio in models that did not admit a bequest motive, and, in his 1976 comment on Barro, argued that the introduction of government debt or Social Security into a perfect foresight, dynamic growth model with bequests would still reduce the capital-labor ratio. Barro replied that Feldstein's criticisms were invalid if the steady-state capital-labor ratio were smaller than the Golden Rule level, but would be correct if the opposite were true. Barro did suggest, however, that the latter possibility, with overaccumulation of capital, might be ruled out in his model, as it was in Miguel Sidrauski (1967), but he was unable to demonstrate that the required behavior would be consistent with utility maximization by finite-lived individuals (1976, p. 345). Although recent work by Willem Buiter (1979), Jeffrey Carmichael (1979, 1982), Truman Bewley (1981 a, b), and others has helped to elucidate the nature of overlapping-generations models with bequests and gifts, and the issues in the Barro-Feldstein debate, they remain far from clear. The purpose of this paper is to shed a little more light on the subject. In what follows, I contend that some of the authors cited above either have specified the individual's optimization problem in an asymmetric way, or have failed to impose necessary conditions for a sensible optimization problem, or both. There is a natural specification of the Barro model1 that emphasizes the similarity between this model and optimal growth models (see, for example, David Cass, 1965; Peter Diamond, 1973). Here, in steady-state equilibrium, the (aftertax) interest rate, r, must equal the rate at which individuals discount the utility of their heirs, p (r = p has become known as the modified Golden Rule) and a meaningful individual optimization problem requires that p exceed n, the natural growth rate of the economy. My results confirm Barro's hunch about his model; in a Barro model, r must exceed n. Much of the Barro-Feldstein 1976 interchange, which assumed that r could have any relation to n, is wrong-headed. Barro and Feldstein reach different conclusions about the effects of government debt and Social Security because they assume different individual utility functions. Also, my results contrast sharply with those of Buiter and Carmichael. For example, Carmichael concluded that with intergenerational transfers from parents to children (bequests), r must exceed n, but with transfers in the opposite direction, n must exceed r (1982, pp. 205-06). Indeed, I believe that the Buiter-Carmichael representation of the Barro model is logically faulty (see Section II). In Section I, I outline an overlapping-generations model with gifts and bequests, and derive its steady-state properties. In the next section, I discuss the shortand long-run effects of introducing government debt into the model and contrast my results with those