Abstract
Turnpike theory, as originally conceived by Samuelson, pertains to optimal programs over a large but finite time horizon with given initial and terminal stocks. In this paper, we present two turnpike results in the context of a model proposed by Robinson, Solow and Srinivasan, and the subject of extensive recent analysis as the RSS model. Our results are classical except that they are phrased in terms of (i) approximately optimal programs, and (ii) golden-rule stocks rather than their parent facet, and they underscore the distinction between the original theory and the asymptotic stability of optimal infinite horizon programs. Our results, and the arguments used to prove them, go beyond the RSS model to contribute to the general theory.
Received: September 22, 2008
Revised: October 20, 2009
JEL classification: C62, D90, Q23
Mathematics Subject Classification (2000): 49J99, 54E52
The authors are grateful above all to Paul Samuelson: they were sustained by a letter from him during the rather long pre-publication travails of this paper. The authors are also grateful to Alex Ioffe: he waived anonymity as a referee, and to the extent that this version is easier to read than the previous one – negotiates better the “thicket of words” and the “jungle of mathematics” – it is due to him. Ali Khan acknowledges enlightening correspondence with Tapan Mitra, and thanks Adriana Piazza, Debraj Ray, and two wildly differing referees of TE for stimulating discussion regarding both substantive and methodological issues.
This paper is dedicated to the memory of David Cass: one of the authors learnt (in 1969–1970) the rudiments of capital theory and Pontryagin’s principle at his hands, and to his everlasting regret, was not afforded the chance of also working with him on his Yale Ph.D. dissertation.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
- 3.
In [23, § 5] McKenzie writes “Almost all the attention to asymptotic convergence has been concentrated on convergence to balanced paths, although it is not clear that optimal balanced paths will exist. This type of path is virtually impossible to believe in... I made this point in articles published in 1974 and 1976.” Also see a repetition of this rather basic point in [24, p. 390].
- 4.
The phraseology is Tapan Mitra’s; the terms good and bad have precise technical meaning going back to Gale [8], as in the definitions below.
- 5.
- 6.
The phrase “turnpike property of good programs” already serves as a counterpoint. Note that in his recent survey, Mitra uses the phrase “turnpike property” synonymously with asymptotic stability, [27, first paragraph]. Also see [26] and the reference to the “often called turnpike property” in [28, Introduction and Conclusion].
- 7.
Such a trajectory will be of particular relevance to what follows in the sequel.
- 8.
The question has to do with whether the von-Neumann growth model serves as a possible discontinuity in the evolution of the application of techniques of maximization and of the calculus of variations to capital theory; whether “maximization of an objective function had no part in his [von-Neumann’s] theory.” In his Nobel Lecture, [33], Samuelson motivates his work on the turnpike by recounting this episode, and McKenzie also refers to it in his Ely Lecture, [23]. Whether von-Neumann’s position is to be granted half or full credit is another interesting shade of difference between McKenzie and Samuelson.
- 9.
- 10.
- 11.
- 12.
See [16, § 1] for references to this literature from the viewpoint of the RSS mode.
- 13.
- 14.
- 15.
See Examples 2 and 3 in [11]. We also note here that Example 1 in this paper shows the existence of a two-period optimal cycle, thereby negating Stiglitz’ monotonicity result in the continuous case.
- 16.
See [15] for the surprising differences between the linear and strictly concave theories. Note, however, that throughout this paper, we limit ourselves to an undiscounted setting, and do not comment on work, by now substantial, on the RSS model with a positive discount factor. It is also for this reason that in the discussion of the Ramseyian turnpike, there is no reference in the sequel to the theorems of Cass and Koopmans. We also draw the reader’s attention to work with non-concave felicities as in [42].
- 17.
To quote McKenzie again, but now in the context of this more delimited turnpike theory, “The real ground for the result is the tendency for optimal paths to bunch together in the middle time, and this tendency is preserved even in models that are time-dependent;” see [21, p. 843]. In [19], McKenzie refers to the middle-turnpike for the von-Neumann model as the “Samuelson Turnpike”.
- 18.
- 19.
- 20.
Whether methods of non-smooth analysis, and in particular, the subdifferential of the felicity function w( ⋅) at the golden-rule stock \(\hat{x},\) can be substituted for the differential, remains a hope, though a distant one.
- 21.
The one cautionary flag in the context of the earlier work on linear models relates to the fact that rather than the Ramseyian setting of the maximization of aggregate utility, the works relate to the von Neumann objective of the maximization of the utility of terminal capital stocks; see [18] and the references therein.
- 22.
- 23.
For the reader innocent of economic theory, as we shall see below, the model is a rather standard one in discrete optimal control theory in infinite time.
- 24.
Note that in restricting d to the open unit interval, we rule out the case of circulating capital. There is little doubt that our analysis can be extended to cover this case. See [7] for an analysis of the optimal policy function with circulating capital in a model that includes the two-sector RSS model.
- 25.
- 26.
This paragraph is a response to a referee’s insistence that the choice of notation does not (i) contradict minimal aesthetical requirements, and (ii) present a serious obstacle to understanding of the material. We hope that these criteria are now being adhered to as a result of the explanation in this paragraph.
- 27.
As a referee pointed out, “it is customary in modern literature to use different alphabets for vectors and numbers” and that “inner product is not a group operation.” On the other side, changing the notation now when it has been established in more than fifteen published papers on the RSS model would cause confusion in another branch of the literature.
- 28.
- 29.
As emphasized in their introduction and abstract, this is the point of departure of Khan–Mitra [11].
- 30.
It bears emphasis that Stiglitz did not offer sufficient condition for the existence of an optimal program, and that these were furnished only recently by [12, 41]. But more to the point, his results applied only to the case of w( ⋅) being linear, and the problem of characterizing transition dynamics with concave functions remains open even in continuous-time models; see [6, 37]. In the discrete time setting, the problems has not be fully solved even for the case of one machine (n = 1), the two-sector RSS model, leave alone the multi-sectoral one being considered in this paper; see [11, 15].
- 31.
To be able to prove the turnpike results reported here from the above suggested optimal control theory point of view has remained an open problem right from the inception of the subject.
- 32.
- 33.
We shall come back to this lemma in the discussion in § 4 below.
- 34.
Note that the conditions furnished in Theorem 2.2 are only sufficient conditions; it is entirely conceivable that the convergence of good programs to the golden-rule stock, (the, so-called, turnpike property of good programs) holds with felicity functions that are only strictly concave only in the neighborhood of the golden-rule stock. See [8] for this assumption; it is highlighted in [24, p. 389].
- 35.
- 36.
As emphasized in the introduction, our primary concern in this paper is with optimal, and approximately optimal, large but finite programs.
- 37.
- 38.
If we endow the parameter space \((a,d) \subset ({R}_{+} \times (0, 1)) \subset {R}_{+}^{2}\) with Lebesgue measure, the set {(a, d) : (1 ∕ a) = 2 − d} has zero measure.
- 39.
- 40.
The terminology is McKenzie’s [23, § II and III respectively,] as already referred to in Footnote 17.
- 41.
The parameter Γ is necessitated by a particular structural characteristic of the RSS model. The reader can refer to Fig. 3 to check out that in the two-sector RSS model, the maximal sustainable capital stock (1 ∕ ad) can never be attained by a finite program starting from an initial stock less than it. Note that in this special case, n = 1, and therefore a is a scalar.
- 42.
Also see Inada [10] and his emphasis on uniformity considerations in turnpike theory. We leave for future work a detailed comparison of our results to those in these papers. It is clear that the RSS model is a particular case of the Leontieff model without joint production and many (but a finite number) techniques for the production of the consumption good.
- 43.
An anonymous referee brought Theorem 5 in Chap. 4 of Arkin–Evstigneev [2] to our attention in the context of approximate optimality. This theorem establishes in a stochastic setting a version of the “middle turnpike theorem (in the spirit of Theorem A) for a family of uniformly good finite programs, and since optimal finite programs are uniformly good, the same is true for a family of approximately optimal programs.”
- 44.
See [24, p. 389] for this point in the context of the general theory.
- 45.
- 46.
The form that McKenzie chooses as final version suitable for a text, [25, Theorem 3] referred to above, obviously relies on [18, 19] in its evolution. However, since these latter references are phrased in terms of the maximization of a function of the terminal stock configuration, the Samuelson turnpike, so to speak, we do not discuss them in any detail.
- 47.
See page 845 in [20] for the first quote, and pages 850, 851 for the second.
- 48.
- 49.
It is of interest in the light of Theorem 2.3(ii) above, that McKenzie’s theorem 9.3 also furnishes an existence result that asserts the optimality of such a converging path, if it is maximal, (weakly maximal in the terminology used in this paper).
- 50.
- 51.
An anonymous referee pointed out that methods of proof of Theorem 1 in [1] could possibly be used to provide a direct and unified proof of the corollaries. We leave this for future work.
- 52.
In the succeeding line, y(t) stands for \(\bar{y}(t)\) and \(\tilde{y}(t)\) alternatively.
- 53.
In the succeeding line, y(t) stands for \(\bar{y}(t)\) and \(\tilde{y}(t)\) alternatively.
- 54.
Thus Radner’s emphasis on Karlin’s proof of the von-Neumann theorem is hardly incidental from this substantive point of view.
- 55.
Note that it is only the assumption of strict convexity that does not hold in the RSS model; the von-Neumann ray is indeed unique by virtue of the fundamental assumption embodied in 2.1. Note also that in this statement, as well as loose treatment of the basic ideas in this section, we are using the golden-rule pair \((\hat{x},\hat{p})\) of the RSS model as the relevant analogue of the von-Neumann balanced growth path and the associated shadow prices. The assumption of exogenously given labor supply in each period obviously vitiates the constant returns to scale assumption in the von-Neumann model. It is of course the constant returns to scale assumption that necessitates the particular metric that Radner uses.
- 56.
In addition to [31], see [10]. Also see the reference to Atsumi in [22] and [25, Lemma 4, p. 252]. A preliminary statement for the two-sector RSS model with strictly concave felicities, but one that ignores the complications of initial and terminal capital stocks, is available in [15, Footnote 4]. For the non-stationary case, see [21, Lemma 8.1].
- 57.
In any case, as the two-sector RSS case brings out, the turnpike result does not hold in the case ξ = 1, and so it can hardly be proved by any method.
- 58.
See Footnote 37 above and the accompanying text.
- 59.
- 60.
See [23, Concluding paragraph of § 4]. It is of interest that both Samuelson and Koopmans avoid the use of Pontryagin’s principle and couch all their arguments in the classical phraseology of the calculus of variations.
- 61.
- 62.
To our knowledge, a result of this form is not available in the economic literature though it is well-known to students of optimal control and the variational calculus; see [40].
- 63.
The distinction between the two kinds of trajectories was referred to in the introduction and illustrated in Fig. 2.
- 64.
- 65.
See Footnote 3 and the accompanying text.
- 66.
The reader is invited to diagram the program in Fig. 3.
- 67.
References
Anoulova, S.V., Evstigneev, I.V., Gundlach, V.M.: Turnpike theorems for positive multivalued stochastic operators. Adv. Math. Econ. 2, 1–20 (2000)
Arkin, V.I., Evstigneev, I.V.: Stochastic models of control and economic dynamics. New York: Academic 1987
Atsumi, H.: Neoclassical growth and the efficient program of capital accumulation. Rev. Econ. Stud. 32, 127–136 (1965)
Brock, W.A.: On existence of weakly maximal programmes in a multi-sector economy. Rev. Econ. Stud. 37, 275–280 (1970)
Cass, D.: Optimum growth in an aggregative model of capital accumulation: a turnpike theorem. Econometrica 34, 833–850 (1966)
Cass, D., Stiglitz, J.E.: The implications of alternative saving and expectations hypotheses for choices of technique and patterns of growth. J. Polit. Econ. 77, 586–627 (1969)
Fujio, M.: Undiscounted optimal growth in a Leontief two-sector model with circulating capital: the case of a capital-intensive consumption good. J. Econ. Behav. Organ. 66, 420–436 (2008)
Gale, D.: On optimal development in a multi-sector economy. Rev. Econ. Stud. 34, 1–18 (1967)
Hahn, F.H., Matthews, R.C.O.: The theory of economic growth. Econ. J. 74, 779–902 (1964)
Inada, K.: Some structural characteristics of turnpike theorems. Rev. Econ. Stud. 31, 43–58 (1964)
Khan, M. Ali, Mitra, T.: On choice of technique in the Robinson–Solow–Srinivasan model. Int. J. Econ. Theory 1, 83–110 (2005)
Khan, M. Ali, Mitra, T.: Optimal growth in the two-sector RSS model: a continuous time analysis. In: Proceedings of the seventh Portuguese conference on automatic control, Electronic publication 2006
Khan, M. Ali, Mitra, T.: Undiscounted optimal growth in the two-sector Robinson–Solow–Srinivasan model: a synthesis of the value-loss approach and dynamic programming. Econ. Theory 29, 341–362 (2006)
Khan, M. Ali, Mitra, T.: Optimal growth in a two-sector model without discounting: a geometric investigation. Japanese Econ. Rev. 58, 191–225 (2007)
Khan, M. Ali, Mitra, T.: Growth in the Robinson–Solow–Srinivasan model: undiscounted optimal policy with a strictly concave welfare function. J. Math. Econ. 44, 707–732 (2008)
Khan, M. Ali, Zaslavski, A.J.: On a uniform turnpike of the third kind in the Robinson–Solow–Srinivasan model. J. Econ. 92, 137–166 (2006)
McKenzie, L.W.: Accumulation programs of maximum utility and the von Neumann facet. In: Wolfe J.N. (ed.): Value, capital and growth. Edinburgh: Edinburgh University Press, 1968 pp. 353–383
McKenzie, L.W.: Capital accumulation optimal in the final state. Zeitschrift für Nationalökonomie Supplement 1, 107–120 (1971)
McKenzie, L.W.: Turnpike theorems with technology and the welfare function variable. In: Loś J., Loś M.W. (eds.): Mathematical models in economics. New York: American Elsevier, 1974 pp. 271–287
McKenzie, L.W.: Turnpike theory. Econometrica 44, 841–866 (1976)
McKenzie, L.W.: Optimal economic growth, turnpike theorems and comparative dynamics. In: Arrow K.J., Intrilligator M. (eds.): Handbook of mathematical economics, vol. 3. New York: North-Holland, 1986 pp. 1281–1355
McKenzie, L.W.: Turnpike theory. In: Eatwell J., Milgate M., Newman P.K. (eds.): The New Palgrave, vol. 4. London: MacMillan, 1987 pp. 712–720
McKenzie, L.W.: Turnpikes. Am. Econ. Rev. Pap. Proc. 88, 1–14 (1998)
McKenzie, L.W.: Equilibrium, trade and capital accumulation. Japanese Econ. Rev. 50, 369–397 (1999)
McKenzie, L.W.: Classical general equilibrium theory. Cambridge: The MIT Press 2002
Mitra, T.: On optimal economic growth with variable discount rates: existence and stability results. Int. Econ. Rev. 20, 133–145 (1979)
Mitra, T.: Characterization of the turnpike property of optimal paths in the aggregative model of intertemporal allocation. Int. J. Econ. Theory 1, 247–275 (2005)
Mitra, T., Zilcha, I.: On optimal economic growth with changing technology and tastes: characterization and stability results. Int. Econ. Rev. 22, 221–238 (1981)
Morishima, M.: Theory of economic growth. Oxford: Oxford University Press 1964
Nikaido, H.: Persistence of continual growth near the von Neumann ray: a strong version of the Radner turnpike theorem. Econometrica 32, 151–162 (1964)
Radner, R.: Paths of economic growth that are optimal with regard only to final states; a turnpike theorem. Rev. Econ. Stud. 28, 98–104 (1961)
Samuelson, P.A. A catenary turnpike theorem involving consumption and the golden rule. Am. Econ. Rev. 55, 486–496 (1965)
Samuelson, P.A.: Maximum principles in analytical economics. Les Prix Nobel en 1970. Stockholm: The Nobel Foundation, 1971 pp. 273–288
Samuelson, P.A.: In: Merton R.C. (ed.): The collected scientific papers of Paul A. Samuelson, vol. III. Cambridge: MIT Press 1972
Solow, R.M.: Neoclassical growth theory. In: Taylor J.B., Woodford M. (eds.): Handbook of macroeconomics, vol. 1. New York: Elsevier, 1999 pp. 637–667
Stiglitz, J.E.: A note on technical choice under full employment in a socialist Economy. Econ. J. 78, 603–609 (1968)
Stiglitz, J.E.: Recurrence of techniques in a dynamic economy. In: Mirrlees J., Stern N.H. (eds.): Models of economic growth. New York: Wiley 1973
von Weizsäcker, C.C.: Existence of optimal programs of accumulation for an infinite horizon. Rev. Econ. Stud. 32, 85–104 (1965)
Zaslavski, A.J.: Optimal programs in the RSS model. Int. J. Econ. Theory 1, 151–165 (2005)
Zaslavski, A.J.: Turnpike properties in the calculus of variations and optimal control. New York: Springer 2005
Zaslavski, A.J.: Optimal programs in the continuous time RSS model. In: Proceedings of the seventh Portuguese conference on automatic control, Electronic publication 2006
Zaslavski, A.J.: Good programs in the RSS model with a nonconcave utility function. J. Ind. Manage. Optim. 2, 399–423 (2006)
Author information
Authors and Affiliations
Corresponding author
Editor information
Editors and Affiliations
Rights and permissions
Copyright information
© 2010 Springer
About this chapter
Cite this chapter
Khan, M.A., Zaslavski, A.J. (2010). On two classical turnpike results for the Robinson–Solow–Srinivasan model. In: Kusuoka, S., Maruyama, T. (eds) Advances in Mathematical Economics. Advances in Mathematical Economics, vol 13. Springer, Tokyo. https://doi.org/10.1007/978-4-431-99490-9_3
Download citation
DOI: https://doi.org/10.1007/978-4-431-99490-9_3
Publisher Name: Springer, Tokyo
Print ISBN: 978-4-431-99489-3
Online ISBN: 978-4-431-99490-9
eBook Packages: Mathematics and StatisticsMathematics and Statistics (R0)