Abstract
In this chapter, we shall analyze a family of multi-sector dynamic models of the quantity adjustment process in which firms determine their productions and raw material orders based on inventories and demand forecasts under constant prices and final demands. Special attention is paid to the role of demand forecast by a moving average of past demands. Section 4.1 describes the assumptions common to all the models and explains how and in what sequence firms determine productions and raw material orders. For the demand forecast method, we assume the simple moving average and the geometric moving average. Sections 4.2 and 4.3 investigate the dynamic properties of the process generated through the interactions between sectors under the above two demand forecast methods. A series of theorems will elucidate how the stability of the process depends on the input structure, the extent of averaging, and the buffer inventory coefficients. It will be shown that averaging of past demands in the demand forecast formation is essential for stability if the input matrix has negative or complex eigenvalues reflecting the interactions between sectors. Section 4.4 closely examines the mechanism of stabilization by averaging of past demands in forecast formation. Stability conditions corresponding to the forecast by the finite geometric moving average will indicate that stability depends on the allocation of weights to each past demand in addition to the number of past periods referred for averaging.
Access this chapter
Tax calculation will be finalised at checkout
Purchases are for personal use only
Notes
- 1.
- 2.
- 3.
For simplicity, we assume here that the buffer inventory coefficients for various raw materials are uniform within the firm.
- 4.
The corresponding stationary values of product and raw material inventories are z * = \( {\overline{s}}^e_{i}\left(1+k_{i}\right)-{x}_{i}^{\ast } \) and \( {v}_i^{\ast}=\left(\left(1+{l}_i\right){\overline{s}}_i^e-{x}^{\ast}_{i}\right){a}_i \), respectively. Note that the convergence to these values requires that \( {\overline{s}}_i^e\ge {x}_i^{\ast}{\left(1+{k}_i\right)}^{-1} \) and \( {\overline{s}}_i^e\ge {x}^{\ast}_i{\left(1+{l}_i\right)}^{-1} \) are satisfied for any i. If one of these conditions do not hold, convergence is hindered by the depletion of inventories.
- 5.
The policy, which was referred to in Subsect. 3.2.3 of the previous chapter, can be regarded as a way of adjustment based on a fixed forecast. When each firm follows this policy in its decision on the production, (4.3) is replaced by
$$ {x}_i(t)={Q}_i-{z}_i(t)\kern0.5em \mathrm{if}\kern0.5em {z}_i(t)<{q}_i\kern0.5em \mathrm{and}\kern0.5em {x}_i(t)=0\ \mathrm{if}\kern0.5em {z}_i(t)\ge {q}_i, \vspace*{-4pt} $$where Q i and q i correspond to S and s, respectively. Let Q and q be the vectors of Q i and q i, and for simplicity, let us assume, in place of (4.4), that the raw material orders are simply determined by m i(t) = x i(t)a i. Then, in matrix notation, it holds s(t) = x(t)A + d. In this case, as long as z i(t) < q i holds for any sector and in any period, we have
$$ x(t)=Q-z(t)=x\left(t-1\right)A+d. \vspace*{-4pt} $$Therefore, if x(0) = d(I − A)−1 and z(0) < q, then the system is stable against a small perturbation of final demand d. It should be noted that Taniguchi’s model in Chap. 6, Subsect. 6.3.3 of this book, is based on different assumptions about the time sequence of events. Under those assumptions, the production is determined by
$$ {x}_i(t)={Q}_i-{z}_i\left(t-1\right)\kern0.5em \mathrm{if}\kern0.5em {z}_i\left(t-1\right)<{q}_i\kern0.5em \mathrm{and}\kern0.5em {x}_i(t)=0\ \mathrm{if}\kern0.5em {z}_i\left(t-1\right)\ge {q}_i, \vspace*{-4pt} $$and (4.1) is modified to z i(t) = z i(t − 1) + x i(t − 1) − s i(t). Accordingly, we have
$$ x\left(t+1\right)=x(t)\left(I+A\right)-x\left(t-1\right)+d. $$As Taniguchi remarks, this system is unstable if the economy consists of more than three sectors. A cause of such instability lies in that firm i does not consider x i(t − 1), which is already known at the beginning of period t, in its decision on x i(t).
- 6.
Lovell obtained a similar result (1962, p. 274–275).
- 7.
The Frobenius roots calculated from the input-output tables of the Japanese economy in 1995-2011 are 0.5242 (93 sectors, 1995), 0.5230 (104 sectors, 2001), and 0.5527 (190 sectors, 2005).
- 8.
“It would be too complicated to work out the expectation de novo whenever a productive process was being started; and it would, moreover, be a waste of time since a large part of the circumstances usually continues substantially unchanged from one day to the next” (Keynes 1936, p. 51, emphasis added).
- 9.
Simon (1959, p. 269) suggests that forecasts formed by the assumption that “the next period will be a weighted average of recent past periods” are sufficiently satisfactory from a practical perspective. He further adds, “The elaboration of the models [of expectations] beyond the first few steps of refinement does not much improve prediction.”
- 10.
- 11.
For more on the indecomposable matrices, refer to Nikaido (1961, pp. 80-89).
- 12.
For the properties of cyclic (or imprimitive) matrices, refer to Nikaido (1961, pp. 103–113).
- 13.
In fact, for τ = 3, (4.26) is equivalent to μ < μ 1, where μ 1 is the only positive root of μ(3b + b 2 μ + b(3 + 2b)μ 2) < 9 (μ = λ F(A)).
- 14.
- 15.
Matrix \( \left[\begin{array}{cc}{a}_{11}& {a}_{12}\\ {}{a}_{21}& {a}_{22}\end{array}\right] \) has a negative eigenvalue when a 11 a 22 < a 12 a 21. This condition implies that the coefficients corresponding to the intersectoral input-output relationship are relatively large in comparison with the ones corresponding to self-consumption.
- 16.
This is shown from \( {\left(\frac{b\mu}{\tau}\right)}^{\frac{1}{\tau +1}}<\overline{\phi}<1 \) for \( \tau >\left(\frac{2b}{\tau }+1\right)\mu \) and \( \underset{\tau \to \infty }{\lim }{\left(\frac{b\mu}{\tau}\right)}^{\frac{1}{\tau +1}}=1 \).
- 17.
- 18.
- 19.
For any w 1, …, w τ such that w 1 + … + w τ = 1, the sum of squares \( {w}_1^2+\dots +{w}_{\tau}^2 \) is minimized when \( {w}_1=\dots ={w}_{\tau }={1}/{\tau } \). Therefore, the variance of averaged demand is minimized by equal allocation of weights among past demands.
- 20.
- 21.
Theorem 9 can be extended for \( -{1}/{2}\le r<0 \). Since \( {\mu}_2\left(-{1}/{2},b\right)<{\mu}_2\left(0,b\right) \), demand forecast by \( {s}_i^e(t)={s}_i\left(t-1\right)+ h \left({s}_i\left(t-1\right)-{s}_i\left(t-2\right)\right)\ \left(0< h <1\right) \) is more likely to cause instability than the static expectation.
- 22.
Assumptions in our models about decisions by individual firms satisfy the following three prerequisites set by Simon (1959, p. 275): “(1) The models must be formulated so as to require for their application only data that are obtainable. … (2) The models must call only for practicable computations. … (3) The models must not demand unobtainable forecast information.”
- 23.
As remarked in the previous chapter, this fact was first discovered by Taniguchi (1991) using numerical computations.
- 24.
- 25.
On this theorem, see Nikaido (1961, p. 114).
- 26.
By Descartes’ rule of signs, a real algebraic equation whose coefficients change their signs from plus to minus only once has one and only one positive root. See Takagi (1965, p. 101)
- 27.
Concerning this theorem, refer to Takagi (1965, p. 104–105).
- 28.
Concerning Schur’s condition, refer to Takagi (1965, pp. 373–376).
- 29.
When the characteristic equation of a second-order linear difference equation has a pair of conjugates roots \( \left\{\lambda, \overline{\lambda}\right\} \), its solution is written as
$$ x(t)={\left|\lambda \right|}^t\frac{\sqrt{x^2(0){\left|\lambda \right|}^2-2x(0)x(1)\operatorname{Re}\lambda +{x}^2(1)}}{\left|\operatorname{Im}\ \lambda \right|}\cos \left( t\theta +\omega \right), $$$$ \tan \theta =\frac{\operatorname{Im}\lambda }{\operatorname{Re}\lambda },\tan \omega =\frac{x(0)\operatorname{Re}\lambda -x(1)}{x(0)\operatorname{Im}\lambda }, $$where x(0) and x(1) are initial values.
- 30.
The maximum solution (measured by absolute value) of an algebraic equation is not smaller than that of its derivative function. On this theorem, see Takagi (1965, pp. 102–104).
- 31.
This result can be shown also by differentiating with respect to . In fact, is differentiable with respect to any of its arguments except at . However, it is a tedious work to prove this and to examine the sign of the derivatives. See Morioka (2005, pp. 280–282).
- 32.
Like the case of , this result can be shown also by differentiating with respect to . In fact, is differentiable with respect to any of its arguments unless . See Morioka (2005, pp. 283–284).
- 33.
Concerning this point, refer to Takagi (1965, pp. 48–49)
References
Brown, R. G. (1959). Statistical forecasting for inventory control. New York: McGraw-Hill.
Foster, E. (1963). Sales forecasts and the inventory cycle. Econometrica, 31(3), 400–421.
Hawkins, D., & Simon, H. (1949). Note: Some conditions of macro economic stability. Econometrica, 17(3-4), 245–248.
Hines, W. G. S. (2004). Geometric moving average. In Encyclopedia of statistical sciences (Vol. 4, 2nd ed., pp. 2784–2788). New York: Wiley-Interscience.
Keynes, J. M. (1936). The general theory of employment, interest and money. London: Macmillan.
Lovell, M. C. (1962). Buffer stocks, sales expectations and stability: A multi-sector analysis of the inventory cycle. Econometrica, 30(2), 267–296.
Metzler, L. A. (1941). The nature and stability of inventory cycle. The Review of Economic Statistics, 23(3), 113–129.
Morioka, M. (1991–1992). Two types of the short-term adjustment processes (in Japanese: Tanki choseikatei no niruikei). Keizai Ronso (Kyoto University), 148(4-6), 140–161, 149(1–3), 79–96.
Morioka, M. (2005). The Economic Theory of Quantity Adjustment: Dynamic analysis of stockout avoidance behavior (in Japanese: Suryo chosei no keizai riron). Tokyo: Nihon Keizai Hyoronsya.
Muth, J. F. (1960). Optimal properties of exponentially weighted forecasts. Journal of American Statistical Association, 55, 299–360.
Nikaido, H. (1961). Linear Mathematics for Economics (in Japanese: Keizai no tameno senkei sugaku). Tokyo: Baihukan.
Simon, H. A. (1959). Theories of decision making in economics and Behavioral science. American Economic Review, 49(3), 253–283.
Simonovits, A. (1999). Linear decentralized control with expectations. Economic Systems Research, 11(3), 321–329.
Takagi, T. (1965). Lectures on Algebra (In Japanese: Daisugaku kogi). Kyoritsu Shuppan.
Taniguchi, K. (1991). On the traverse of quantity adjustment economies (in Japanese: Suryo chosei keizai ni okeru iko katei nitsuite). Keizaigaku Zashi 91(5):29-43.
Author information
Authors and Affiliations
Appendix
Appendix
1. Proof of Theorem 1 Footnote 24
(i) Let ϕ ≠ 0 be an eigenvalue of Φ, and let ξ = [ξ τ, ξ τ−1, …, ξ 1, η] be an associated eigenvector, where ξ i and η are 1 × n vectors. Then, from ϕξ = ξΦ we obtain
Therefore, ϕ is also an eigenvalue of Q(ϕ)A. Conversely, if ϕ is an eigenvalue of Q(ϕ)A and η is the corresponding eigenvector, then it is also an eigenvalue of Φ and the above ξ is the corresponding eigenvector. Consequently, ϕ ≠ 0 is an eigenvalue of Φ if and only if ϕ is an eigenvalue of Q(ϕ)A.
To prove the contraposition, assume ρ(Φ) ≥ 1 and ϕ is an eigenvalue of Φ such that ϕ ≥ 1. Then, from the above fact, it follows \( 1\le \rho \left(Q\left(\phi \right)A\right).\vspace*{-3pt} \) Let q i be the i-th diagonal element of matrix Q(ϕ), then
Let C + denote a matrix obtained from a matrix C by replacing its elements with their absolute values. It is known that ρ(C) ≤ λ F(C +) holds for any complex square matrix C.Footnote 25 Since Q(ϕ) is diagonal and A is nonnegative, (4.47) leads to
Since ϕ is an eigenvalue of QA, these inequalities lead to
(ii) To prove the contraposition, assume that ω λ F(A) is an eigenvalue of A satisfying (4.22) and \( {\lambda}^{\ast } \) = λ F((2T −1 B + I)A) ≥ 1. Let ϕ ≠ 0, ψ = ω −1 ϕ, and \( \tilde{A}={\omega}^{-1}A \); then since \( {\omega}^{\tau_i}=-1 \) for any i, ϕ is an eigenvalue of Φ if and only if ψ is an eigenvalue of \( F\left(\psi \right)\tilde{A} \), where
For any ψ > 0, matrix F(ψ)A is a nonnegative and identically cyclic with matrix A; hence ωλ F(F(ψ)A) is an eigenvalue of F(ψ)A, and hence λ F(F(ψ)A) is an eigenvalue of \( F\left(\psi \right)\tilde{A} \). Let
then h(ψ) is an eigenvalue of \( \psi I-F\left(\psi \right)\tilde{A} \); hence h(ψ) = 0 implies that ϕ = ωψ is an eigenvalue of Φ. It follows from F(λ \( ^{\ast} \)) ≤ F(1) that \( {\lambda}^{\mathrm{F}}\left(F\left({\lambda}^{\ast}\right)A\right)\le {\lambda}^{\mathrm{F}}\left(F(1)A\right)={\lambda}^{\ast }; \)hence
\( h\left({1}\right)=1-{\lambda}^{\mathrm{F}}\left(F(1)A\right)=1-{\lambda}^{\ast}\le 0,\hspace*{2pt} \) \( h\left({\lambda}^{\ast}\right)={\lambda}^{\mathrm{F}}\left(F(1)A\right)-{\lambda}^{\mathrm{F}}\left(F\left({\lambda}^{\ast}\right)A\right)\ge 0.\vspace*{8pt} \)
Since polynomial h is continuous with respect to ψ, there exists an α such that
hence αω is an eigenvalue of Φ. Therefore,
2. Proof of Theorem 2
(i) The characteristic equation of Φ′ is
Let Λ be the Jordan normal form of matrix A, then there exists a regular matrix X satisfying XAX −1 = Λ and
where ν is the number of A’s different eigenvalues and κ i is the multiplicity of an eigenvalue \( {\lambda}_i={\mu}_i{\mathrm{e}}^{\mathrm{i}{\theta}_i} \) (thus \( {\kappa}_1 + \cdots + {\kappa}_n=n \)). Thus, ρ(Φ′) < 1 is equivalent to \( \overline{\phi}\left({\mu}_i,{\theta}_i,\tau, b\right)<1 \) for any i ∈ {1, …, ν}.
(ii) Let ξeiω = ξ(cosω + i sin ω) (ξ ≥ 0) be a root of (4.24). Then, it satisfies
Let \( {\theta}^{\prime }=\theta +{2\pi}/{\tau },{\omega}^{\prime }=\omega +{2\pi}/{\tau },{\theta}^{{\prime\prime} }={2\pi}/{\tau }-\theta, {\omega}^{{\prime\prime} }= {2\pi}/{\tau }-\omega, \) then
The second equation is obtained by observing that f(ξ, −ω, −θ) is the conjugate of f(ξ, ω, θ). Therefore, when μeiθ rotates by 2π/τ radians on the complex plane, each of the solutions of (4.24) also rotates by 2π/τ radians. Furthermore, if θ 1 and θ 2 are symmetric on the complex plane about the direction π/τ, two sets of the roots of (4.24) corresponding θ 1 and θ 2 are also symmetric about this direction.
(iii) This statement is obvious for μ = 0, so let us assume μ > 0. Let
and be the unique positivesuch that .Footnote 26 By a theorem about the upper bound on roots of an algebraic equationFootnote 27, it follows \( \overline{\phi}\left(\mu, \theta, \tau, b\right)\le {\phi}_{+}\left(\mu, \tau, b\right) \), where the equality sign holds if and only if θ = π/τ. Hence, \( \overline{\phi}\left(\mu, {\pi}/{\tau },\tau, b\right)<1 \) holds if and only if
where \( \overline{\mu} \equiv {\tau }/{(2b+\tau)} \). Therefore, \( \overline{\phi}\left(\overline{\mu},{\pi }/{\tau },\tau, b\right)=1 \) and
Let , then . According to Schur’s condition,Footnote 28 a necessary and sufficient condition for is that all the following inequalities hold:
Whereis the unit matrix of order, , ( is the zero row vector of order , is the zero matrix of order ), and denotes transposition. Let , . Then, since
satisfies the following recursion formula:
Similarly, from
we have
It can be shown that satisfies the same recursion formula as that of . Let , and
then can be written as
where satisfies the recursion formula:
Assume \( \mu >\overline{\mu} \), then \( -\alpha +\beta =\mu /\overline{\mu}>1 \), hence
Let ε = α 2 + β 2 − 1, \( \eta = \sqrt{{4\alpha}^{2}-\delta^{2}} \). From (4.49) and (4.50), we haveFootnote 29
where
and takes values in interval . Under the assumption and , and can be simultaneously positive if and only if
Note that , . Since
(4.51) guarantees , , …, Thus, providing that and, (4.48) holds if and only if
Let , then . Thus, differentiating with respect to , we have
for\( \kern0.5em \overline{\mu}<\mu <{\mu}^{\dagger } \). Since increase from to as increases from to , there exists a uniquesuch that
Let us denote this unique μ by μ ∗(θ, τ, b), then (4.51) is equivalent to μ < μ ∗(θ, τ, b) for \( 0\le \theta < {\pi }/{\tau } \). When \( \theta = {\pi }/{\tau } \), (4.51) is never satisfied, and from the above-mentioned fact, (4.48) is equivalent to \( \mu <\overline{\mu} \). However, by definition we have \( F\left({\mu}^{\ast}\left({\pi}/{\tau },\tau, b\right),\tau, b\right)=-{\pi }/{2} \), hence, \( {\mu}^{\ast}\left({\pi }/{\tau },\tau, b\right)=\overline{\mu} \), and consequently, (4.48) is equivalent to \( \mu < \mu^{\ast}\left(\theta, \tau, b \right) \) for \( 0\le \theta \le {\pi }/{\tau }. \)
(iv) For \( 0\le \theta <{\theta}^{\prime }< {\pi }/{\tau } \), it holds
which implies μ ∗(θ′, τ, b) < μ ∗(θ, τ, b).
(v) The first part of this statement directly follows from statement (iii). As for the second part, μ ∗(0, 1, b) and μ ∗(0, 2, b) can be directly calculated from (4.48).
(vi) Let \( {\mu}_1\equiv\sqrt{\frac{\tau \left(1+\tau \right)}{2{b}^2+\left(2b+\tau \right)\left(\tau -1\right)}} \) and \( {\xi}_{1 }\equiv{\left(\frac{b\tau +b}{\tau +b}\right)}^{\frac{1}{\tau }} \), then we have
here we used the fact that x arctan \( a/x \geq \text{arctan}\ a\ \text{holds \ for}\ \text{any} \ x \geq 1 \ \text{ and}\ a > 0\). Hence μ 1(τ, b) ≥ μ ∗(θ, τ, b), where the equality sign holds if and only if τ = 1, θ = 0 (note that \( \mu_1 < \mu^{\dag} \)). Treating τ as a continuous variable and differentiating F with respect to τ, for \( \overline{\mu} < \mu < 1\) we obtain
Here we used the fact that for x . Assume , then , hence at . If \( {\mathit{\mu }}^{{}_{\ast}} (\theta, \tau, b) < (\tau + 1)/(2b + \tau + 1) \), then \( {\mathit{\mu }}^{{}_{\ast}} (\theta, \tau, b) < {\mathit{\mu }}^{{}_{\ast}} \left({\pi}/{(\tau+1)}, \tau + 1, b\right) \leq {\mathit{\mu }}^{{}_{\ast}} \left({\tau\theta}/{(\tau + 1)}, \tau + 1, b\right) \). Otherwise, since \( \mu_1 \) is increasing with respect to \( \tau \), we have
Thus, \( \mu^{\ast} (\theta, \tau, b)<\mu^{\ast}(\tau \theta/(\tau + 1), \tau + 1,b), \) and this holds also for \( \theta = \pi/\tau \).
Next, let. Note that F(μ 2(1, b), 1, b) = π/2 holds because . Since
we have (the last inequality sign can be confirmed by some mechanical calculation). Therefore,
which implies \( \mu_{2}(\tau, b) > \mu^{\ast}(0,\tau, b) \) for \( \tau > 1 \)
(vii) Differentiating with respect to , we have
for . Hence, if , then
which implies μ ∗(θ, τ, b′) < μ ∗(θ, τ, b).
3. Proof of Theorem 3
(i) By virtue of Theorem 4.2(ii), we can assume 0 ≤ θ ≤ π/τ without losing generality. Let \( {\overline{\psi}}_m\left(\mu, \theta, \tau, b\right) \) denote, for \( m> 0 \), the dominant root of the equation:
Obviously, ψ is a solution of (4.52) if and only if mψ is a solution of (4.24), therefore, \( {\overline{\psi}}_m\left(\mu, \theta, \tau, b\right)\lessgtr 1 \) is equivalent to \( \overline{\phi}\left(\mu, \theta, \tau, b\right)\lessgtr m \). Let
Note that . Define similarly to the case of , then requires
By induction, it can be shown that is a product of
where is a -th order polynomial of and the double sign corresponds to odd and even in this order. Since for , implies
By the Gauss’ theorem about the bound on the solution of an algebraic equationFootnote 30, also requires . Since , becomes more restrictive than \( \mu_- \) when
Like the case of , if , then , which ensures . If , then , and holds if and only if and (hence ), where
Note that \( \mu_+ < {\mathit{\mu }}^{{}_{\prime}} \) and \( F^m (\mu_+, \tau, b) = \pi/2 \) for any \( m >0 \). Let be the partial derivative of with respect to . If , then , \( F^m ({\mathit{\mu}}^{{}_{\prime}}, \tau, b) \geq \pi/2 \), and
If , then is equivalent to , and by definition, we have. Accordingly,
for . Since we have \( {\mathit{F}}^{\mathit{m}}\left({\mathit{\mu }}_{ - },\mathit{\tau },\mathit{b}\right) = \mathit{\pi }/2. \) Therefore, there exists a unique such that \( \mu_+ \leq \mu \leq \mu^\prime \mbox{ for } M \leq X, {\mathit{\mu }}_{ + } \leq \mathit{\mu } \leq \min\left\{{\mathit{\mu }}_{ - },\tilde{\mu}\right\} \mbox{ for } X < M, \mbox{ and }\)
Let us denote this unique by, then ; therefore, like the case of , is equivalent to
Because of the continuity of with respect to its arguments, is equivalent to . Assume and , then , which implies .Footnote 31
(ii) Assume \( 0\le \theta <{\theta}^{\prime}\le {\pi }/{\tau } \) and \( \overline{\phi}\left(\mu, {\theta}^{\prime},\tau, b\right)=m \). Then, since
we have \( \overline{\phi}\left(\mu, \theta, \tau, b\right)<m=\overline{\phi}\left(\mu, {\theta}^{\prime},\tau, b\right). \)
(iii) Assume m > ξ 0 \( (M >X)\) and \( Z \equiv X\sqrt{2\tau+1}. \) Note that \( X < Z < Y. \) Differentiating F m with respect to \( \tau \), we have
First, we consider the case . If , then . Assume , then since we have
for . Therefore, in these cases, we have
Because and \( \mu_0 - \mu_{-} = \frac{m(M-Y)}{B(M-X)}\), we obtain , hence for On the other hand, since
we have , which implies . Therefore, at .
Next, we turn to the case . In this case, it follows that .If , then because , hence If , then and because . Thus,
Since , we have . Therefore, is increasing in. Hence,
where the second inequality sign holds because
for . Consequently, holds for any . Hence, in this case alsoat .
In sum, at for any, and it also holds for any \( \mu <\mu_2 \) (when \( X < M < Y\)) and \( \mu < \mu_{-}\) (when \( Y \leq M \)). Let . Then, since and both \( \mu_2 \) and \( \mu_{-} \) are increasing with respect to \( \tau \) in their relevant ranges, we have
which implies
(iv) Differentiating with respect to , we have
.
Considering that the sign of is equal to that of , and
holds for . Let , then since \( \mu = \mu_m^* (\theta, \tau, b^\prime) \) and both \( \mu_2 \) and \( \mu_- \) are increasing with respect to \( \tau \) in their relevant ranges, we have
which implies
4. Proof of Theorem 4
(i) To prove the contraposition of statement (i), let us assume ρ(Ψ) ≥ 1. Let ψ be an eigenvalue of Ψ such that |ψ| ≥ 1 and ξ = [ξ I, ξ II] be an associated eigenvector, where both ξ I and ξ II are 1 × n vectors. Then, from ψξ = ξ Ψ we obtain
Therefore, ψ is also an eigenvalue of matrix Q(ψ)A, hence 1 ≤ |ψ| ≤ ρ(Q(ψ)A). Let q i be the i-th diagonal element of matrix Q(ψ), then
Since Q is diagonal and A is nonnegative, we have
Because is an eigenvalue of , these inequalities imply
(ii) To prove the contraposition of statement (ii), assume that –λ F(A) is an eigenvalue of A and \( \lambda^* = \lambda^{\rm F} ((2\Gamma B(2I - \Gamma)^{-1} + I)A) \geq 1 \). Let ψ ≠ 0, then ψ is an eigenvalue of Ψ if and only if
For any ψ > 0, matrix Q(−ψ)A is nonnegative and identically cyclic with matrix A. Hence, −λ F(Q(−ψ)A) is an eigenvalue of Q(−ψ)A. Let
for ψ > 0, then h(ψ) is an eigenvalue of ψI + Q(−ψ)A = − ((−ψ)I − Q(−ψ)A). Hence, h(ψ) = 0 implies that −ψ is an eigenvalue of Ψ. Since \( Q(-\lambda^\ast) \leq Q(-1) \), we obtain \( \lambda^{\mathrm{F}} (Q(-\lambda^\ast)A) \leq \lambda^{\mathrm{F}} (Q(-1)A) = \lambda^\ast \). Consequently,
Because of the continuity, there exists an α such that
Therefore, −α is an eigenvalue of Ψ, hence ρ(Ψ) ≥ 1.
5. Proof of Theorem 5
(i) Like the case of \( \Phi^\prime \), the characteristic equation of \( \Psi^\prime\) is
where ν is the number of A’s different eigenvalues and κ i is the multiplicity of an eigenvalue λ i = \( {\mu}_i{\mathrm{e}}^{\mathrm{i}{\theta}_i} \). Therefore, ρ \( (\Psi^\prime) < 1 \) is equivalent to \( \overline{\psi}\left({\mu}_i,{\theta}_i,\gamma, b\right)<1 \) for any i ∈ {1, …, ν}.
(ii) Let ξeiω = ξ(cosω + i sin ω) (0 ≤ ξ, 0 ≤ ω < 2π) be a root of (4.33), then
Let f(ξ, ω, θ) be the left-hand side of this equation; then f(ξ, −ω, −θ) = 0, because it is the conjugate of f(ξ, ω, θ). Therefore, \( \overline{\psi}\left(\mu, \theta, \gamma, b\right)=\overline{\psi}\left(\mu, -\theta, \gamma, b\right) \).
(iii) From Schur’s condition, holds if and only if
where . Note that
and E 1 > 0 is equivalent to Dμ = (1 − γ + γb)μ < 1. Let
then μ 1 < μ 2 < μ 3 because
By direct calculation, we have
Let , then Differentiating with respect to , we have
Since has the same sign as that of , is equivalent to either or , is a quadratic expression ofand
where . Thus, is decreasing in . Hence, there exists a uniquesuch that and . Letbe this unique, then (4.53) is equivalent to .
(iv) From the above results, for 0 < θ < θ′ ≤ π we have
which implies The latter part of the statement directly follows from .
(v) This statement directly follows from statement (iii).
(vi) Assume γ < γ′ < 1. Differentiating H with respect to γ, we have
Note that the sign of \( {H}_{\gamma } \) is the same as that of L(μ) for μ 2 < μ < μ 3. From
we have \( {H}_{\gamma }<0 \) for μ 2 < μ < μ 3. Assume \( \mu = \mu^\ast (\theta, \gamma,b) \) and \( 0 \leq \theta < 2\pi \). If \( \mu_3 (\gamma^\prime) =1/(1-\gamma^\prime+ \gamma^{\prime} b) < \mu \), then \( \mu^\ast (\theta, \gamma^\prime, b) \leq \mu^\ast (0, \gamma^\prime, b) \leq \mu^\ast (\theta, \gamma, b) \). Otherwise, since
we have μ ∗(θ, γ, b) = μ > μ ∗(θ, γ′, b), and this also holds for θ = 0.
(vii) Differentiating H with respect b, we have
Considering that
we can apply the same argument as the case of (vi).
6. Proof of Theorem 6
(i) B, C, and D are the same as those in the proof of the previous theorem, and let \( \overline{\phi}_{m}\left(\mu, \theta, \tau, b\right) \) be the dominant root of the equation:
Then, it follows that \( \overline{\psi}\left(\mu, \theta, \gamma, b\right)=m\overline{\phi}_m\left(\mu, \theta, \gamma, b\right). \) Let
As before, by Schur’s condition, holds if and only if
By the assumption , it is sufficient to consider the case .Assume and let , then from
we have . If , then . If , then . Hence, (4.54) holds when .
Assume , then Since and , equationhas two positive roots. Let be its larger root, then because . Observe that ; for ; and for Let then
Differentiating with respect to , we have
Let . is a quadratic expression of, and
Hence, if then and in ; if , then and in . Thus, in , so there exists a uniquesuch that
Let us denote this uniqueby, then, it satisfies
and (4.54) is simplified into
Assume . Let then , hence , which implies Footnote 32
(ii) If , then and is equivalent to , which implies that holds irrespective of when .
Assume \( \mu_0 < \mu, 0\leq \theta < \theta^\prime \leq \pi \), and \( \psi\left(\mu, {\theta}^{\prime},\gamma, b\right)=m \).
Then, it follows that
which implies \( \overline{\phi}\left(\mu, \theta, \gamma, b\right)<m=\overline{\phi}\left(\mu, {\theta}^{\prime},\gamma, b\right). \)
(iii) Differentiating with respect , we have
Note that is a quadratic expression of . Since
holds for . Assume , and let , that is . Then, applying the same argument as in the proof of Theorem 5(vi), we have
which implies
(iv) Differentiation H m with respect to b, we have
Note that is a quadratic expression of . From
we have for . Thus, we can repeat a similar argument to the case of (iii).
7. Proof of Theorem 7
(i) Let Ψ be the transitive matrix of FGMA and ϕ be one of its eigenvalues such that |ϕ| ≥ 1. Then, like the case of GMA, ϕ is also an eigenvalue of QA, where Q is an n × n diagonal matrix whose i-th diagonal element is
Let ψ = ϕ −1, then
Let us denote this polynomial of ψ by . Since is a complex integer expression, for any complex number and, there exists a complex number such that Footnote 33 Therefore, under the constraint , takes the maximum value at a certain satisfying |. In this case, observing
we have . Consequently,
which implies
(ii) This can be shown in a similar manner to the proof of Theorem 4(ii).
8. Proof of Theorem 8
(i) By definition, it follows that
(ii) Differentiating with respect to , we have
Since and
can be written as:
9. Proof of Theorem 9
(i) By Schur’s condition, the dominant root of a cubic equation with real coefficient is less than unity if and only if
Therefore, holds if and only if
where . Let (do not confuse this \( \beta \) with the secondary coefficient of the above equation). Then, the pair of first two inequalities are equivalent to for , and
for . The third inequality is equivalent to for . Since for and these conditions are summarized into (4.44).
(ii) By replacing in (4.55) by , we have
as the necessary and sufficient condition for . (4.45) can be derived from these condition in the similar way to (4.44).
Rights and permissions
Copyright information
© 2019 Springer Japan KK, part of Springer Nature
About this chapter
Cite this chapter
Shiozawa, Y., Morioka, M., Taniguchi, K. (2019). Dynamic Properties of Quantity Adjustment Process Under Demand Forecast Formed by Moving Average of Past Demands. In: Microfoundations of Evolutionary Economics. Evolutionary Economics and Social Complexity Science, vol 15. Springer, Tokyo. https://doi.org/10.1007/978-4-431-55267-3_4
Download citation
DOI: https://doi.org/10.1007/978-4-431-55267-3_4
Published:
Publisher Name: Springer, Tokyo
Print ISBN: 978-4-431-55266-6
Online ISBN: 978-4-431-55267-3
eBook Packages: Economics and FinanceEconomics and Finance (R0)