Bifurcations in Regional Migration Dynamics

Munich Personal RePEc Archive Bifurcations in Regional Migration Dynamics Berliant, Marcus and Kung, Fan-chin Washington University in St. Louis, City University of Hong Kong, East Carolina University 28 January 2009 Online at https://mpra.ub.uni-muenchen.de/13053/ MPRA Paper No. 13053, posted 29 Jan 2009 05:07 UTC Bifurcations in Regional Migration Dynamics Marcus Berlianty Department of Economics, Washington University in St. Louis Fan-chin Kungz Department of Economics and Finance, City University of Hong Kong and Department of Economics, East Carolina University January 2009 Abstract The tomahawk bifurcation is used by Fujita et al. (1999) in a model with two regions to explain the formation of a core-periphery urban pattern from an initial uniform distribution. Baldwin et al. (2003) show that the tomahawk bifurcation disappears when the two regions have an uneven population of immobile agricultural workers. Thus, the appearance of this type of bifurcation is the result of assumed exogenous model symmetry. We provide a general analysis in a regional model of the class of bifurcations that have crossing equilibrium loci, including the tomahawk bifurcation, by examining arbitrary smooth parameter paths in a higher dimensional parameter space. We …nd that, in a parameter space satisfying a mild rank condition, generically in all parameter The authors thank Yuri Mansury for helpful comments but retain responsibility for any errors herein. y Department of Economics, Washington University, Campus Box 1208, 1 Brookings Drive, St. Louis, MO 63130-4899 USA. Phone: (314) 935-8486. Fax: (314) 935-4156. E-mail: [email protected]. z Department of Economics and Finance, City University of Hong Kong, 83 Tat Chee Avenue, Kowloon, Hong Kong. Phone: 852-27887407. Fax: 852-27888806. E-mail: [email protected]. 1 paths this class of bifurcations does not appear. In other words, conclusions drawn from the use of this bifurcation to generate a core-periphery pattern are not robust. Generically, this class of bifurcations is a myth, an urban legend. Keywords and Phrases: Bifurcation; genericity analysis; migration dynamics JEL Classi…cation Numbers: C61, R23, F12 1. Introduction Economic activities are not distributed uniformly in space. Manufacturing often concentrates in a few regions, resulting in a core-periphery pattern. How does one region come to dominate others and become a manufacturing core? The literature often considers a two region system. Beginning with a uniform distribution of immobile agricultural workers or farmers, Fujita et al. (1999) explain the emergence of the core-periphery urban pattern using the dynamics of a tomahawk bifurcation when transportation cost varies and other parameters are …xed (see also Fujita and Mori, 1997). When transportation cost is high, the symmetric equilibrium, where both regions have the same mobile manufacturing population, is the only equilibrium and it is stable. When transportation cost is moderate, two other stable equilibria emerge; when this happens, one of the two regions attracts all of the manufacturing, resulting in a core-periphery pattern. When transportation cost is low, the symmetric equilibrium becomes unstable and the only stable equilibria are the two core-periphery equilibria. Is this type of bifurcation robust? Baldwin et al. (2003) examine the case where one region has slightly more immobile agricultural workers than the other. The model still preserves the feature of catastrophic agglomeration but the tomahawk bifurcation disappears. This means that the tomahawk bifurcation results from exogenous model symmetry. In addition, they show that, in the footloose entrepreneur model, the tomahawk bifurcation appears in the case of symmetric immobile population in regions but disappears with asymmetric populations (see also Forslid and Ottaviano, 2003). To illustrate how the underlying exogenous parameters, such as the location of immobile population, a¤ect bifurcation patterns, let’s consider the following one2 dimensional dynamical system with two parameters: x_ = a + bx x3 where x 2 < and parameters (a; b) 2 <2 . This system exhibits the standard pitchfork bifurcation when a = 0 (see Figure 1; the solid and dashed lines indicate stable and unstable equilibria respectively). To show that this bifurcation is not robust, we perturb a to 0:005 and obtain Figure 2 instead. The general contour is still the same and the stable and unstable regions change slightly, but the equilibrium loci do not cross each other. This is a saddle-node bifurcation. The same pattern appears when we perturb a to 0:005 in Figure 3. The full equilibrium diagram against the two-dimensional parameter space (a; b) is plotted in Figure 4. Consider all one dimensional paths in (a; b) space and the equilibrium diagram generated by taking a slice of the three-dimensional picture along any path. We can see that only in some paths passing through (0; 0), there is a pitchfork bifurcation. [Figure 1 Here] [Figure 2 Here] [Figure 3 Here] [Figure 4 Here] This paper provides a general analysis of the class of bifurcations having crossing equilibrium loci in a two region model. This class includes, for example, the tomahawk, the pitchfork, and the transcritical bifurcations. It is well-known that such bifurcations are not stable: “all bifurcations of one-parameter families at an equilibrium with a zero eigenvalue can be perturbed into saddle-node bifurcations” (Guckenheimer and Holmes, 1997, p. 149). Baldwin et al. (2003) demonstrate exactly this by adding a slight population asymmetry while letting the dynamical system change along the transportation cost axis. We examine equilibrium dynamics along arbitrary smooth (C r ) parameter paths in a higher dimensional parameter space. We show that in a parameter space satisfying a mild rank condition, generically1 in all parameter paths this class of bifurcations does not appear. 1 Thus, these kinds of Here, a generic property means a property satis…ed by parameter paths in an open and dense set. 3 bifurcations are not robust, and their appearance relies on the strategic choice of very speci…c parameter values. The rank condition just mentioned requires that the Jacobian matrix of the dynamical system with respect to endogenous variables and parameters has full rank at every equilibrium for all parameter values, and is standard in the general equilibrium literature on smooth economies. We show that it is easy to …nd such a parameter space. Section 2 introduces the benchmark model and extends it to more exogenous parameters. Section 3 discusses migration dynamics and presents the main result. Section 4 concludes. 2. The Model The core-periphery model features a two-region economy with the same resources in both regions. The same populations of immobile farmers in both regions produce a homogeneous agricultural good under constant returns to scale. A population of mobile manufacturing workers can migrate between regions. These manufacturing workers move to the region where they enjoy a higher utility level. The transportation of manufactured goods across regions bears a cost while transport of the agricultural good does not. Manufacturing …rms produce di¤erentiated products under increasing returns to scale technologies, competing monopolistically. There are two types of pecuniary externalities that generate forces causing agglomeration. These forces imply positive feedback that comes from …rms locating near each other. First, manufacturing production will concentrate where there is a large market with many workers consuming manufactured goods. Second, workers will move to the region where production concentrates because the manufactured goods are cheaper there. The benchmark model is introduced formally next. We then expand the model by adding three more exogenous parameters to conduct genericity analysis in a higher dimensional parameter space. The Benchmark Model There are two regions in the economy indexed by i 2 f1; 2g. There are two types of commodities: a homogeneous agricultural good and horizontally di¤erentiated manufactured goods. There is a continuum of manufactured goods of size n 2 <+ , determined endogenously. Each manufactured good is denoted by j 2 [0; n]. 4 Let pA i 2 <++ denote the local price of the agricultural good, and let pi (j), where pi : [0; n] ! <++ is a measurable function, denote the local price of each manufactured good j in region i. There are two types of consumers: immobile farmers of M population LA who i in region i 2 f1; 2g, and mobile workers of total population L migrate between regions. Each worker is endowed with one unit of labor, supplied inelastically. Let A 2 <+ denote the quantity of the agricultural good, and let m (j), where m : [0; n] ! <+ is a measurable function, denote the quantity of manufactured good j. All consumers have the same utility function u (m; A) = M A1 where M = Rn 0 ; 1 and 0 < ; < 1. A consumer in region i with income Y m (j) dj solves the following problem. M ax s:t: The demand functions are A; m(j)2<+ Rn A pi A + 0 p i u (m; A) ; A^i (Y ) = (1 ) Y =pA i ; m ^ i (j; Y ) = Y Gi1 =pi (j) 1 where Gi = hR n 0 pi (j) 1 dj i (1) (j) m (j) dj = Y: 1 ; 1 is the manufacturing price index. There are two types of workers, skilled workers who work in the manufacturing sector and unskilled workers or farmers who work in the agricultural sector. Skilled workers can move between regions, whereas unskilled workers cannot move between regions. Neither type of worker can change type to move to the other sector. Each worker is also a consumer, and supplies one unit of labor inelastically to the sector in which they are employed. The agricultural good is produced by farmers with a one-to-one (labor input)output ratio. For simplicity, the transportation of the agricultural good is assumed to bear no cost. Thus, the equilibrium agricultural commodity price is the same in A A both regions by no arbitrage; let pA 1 = p2 = p . Farmers retain all the revenue and they have income pA . 5 Manufactured goods are produced by …rms that employ mobile workers. Labor is the only input required. All …rms have the same inverse production function l = F + cq; where F; c > 0 are the …xed and the marginal input requirements in terms of labor, whereas l units of labor are required for q units of output. The production technology exhibits increasing returns to scale due to …xed costs. There is free entry into the market that is subject to the …xed cost. Because of increasing returns to scale, each j-good is produced by and is the only product of an operating …rm. Operating …rms choose locations and engage in Chamberlinian monopolistic competition. Each …rm chooses a location and charges a uniform free-on-board price for its product. Firms make decisions simultaneously. Let wi 2 <++ denote the wage rate in region i. Suppose a …rm locates in region i, charges price p, pays wage wi , and sells output q (p), where q : <++ ! < is the demand of consumers. Its pro…t is i (p) = pq (p) wi [F + cq (p)] : A …rm in region i solves the following problem. M ax i p2<++ (p) : (2) It is well-known that because of the assumed constant elasticity utility function and the iceberg transportation cost (to be de…ned shortly), the elasticity of demand faced by a …rm is independent of the locations of its consumers. A monopolistically competitive …rm charges a price marked up from the marginal cost. The pro…t-maximizing price for a …rm in region i is pi = cwi = . Its maximized pro…t is i = 1 cwi q F (1 )c : The transportation cost of manufactured goods takes the Samuelson iceberg form. If one unit of good is shipped across regions, the fraction 1=T of the unit arrives (T > 1). Since …rms are identical and their behavior di¤ers only in location, we label …rms and their products with their locations. This simpli…es the notation to j 2 f1; 2g. We replace pi (j) with pji , which denotes the price of region j products in region i, and replace m ^ i (j; Y ) with m ^ ji (Y ), which denotes the demand for region j 6 products by region i consumers (with prices an implicit argument). We denote the utility function by u (m1i ; m2i ; Ai ), replacing the function m with scalars m1i and m2i representing the quantity of manufactured goods consumed by a region i worker, and letting Ai represent the agricultural commodity consumption of a region i worker. Let m1Ai , m1Ai , and AAi denote the analogous quantities for region i farmers. The superscript denotes the region in which commodities are produced. Let ni denote the number of …rms in region i. The total number of operating …rms equals the total i 1 h 2 1 1 1 + n2 (pi ) . variety of products; n1 + n2 = n. Note that Gi = n1 (pi ) A region i …rm charges a free-on-board price pi = cwi = . Thus, pii = pi and pji = pj T for j 6= i by no arbitrage. Substituting Y with wi , we have region i manufacturing workers’ indirect utility: vi = (1 )1 wi Gi for i 2 f1; 2g : Manufacturing workers are freely mobile. They choose a region that o¤ers the highest utility level. Extended Parameters Above is the standard model of the new economic geography. The model is usually studied with varying transportation cost. In order to facilitate analysis in a higher dimension, we augment the system with three more exogenous parameters. These parameters do not change the model signi…cantly, but they do accommodate asymmetric parameterizations. Let = ( 1; 2; ), where i be an open subset of <3 ; its elements are denoted by 2 ( F; 1) (for i = 1; 2) and 2 ( 1; 1). These parameters enter the model in the following way: (i) i parameterizes “regional …xed inputs”: The …xed labor input of a region i …rm is F + i. Note that although …rms’ pro…t function is changed to i (p) = pq (p) wi [F + i + cq (p)] ; their chosen price cwi = is not a¤ected. (ii) parameterizes “regional amenity”: Workers have preferences over regions as follows. If a worker lives in region 2, her utility function is unchanged. If she lives in region 1, her utility is factored up by (1 + ). The new utility function of region 1 workers is (1 + ) u m11 ; m21 ; A1 : 7 This captures regional di¤erences such as the weather and the landscape. Note that region 1 workers’ indirect utility is changed to v1 = )1 (1 (1 + ) w1 G1 : An economy is speci…ed by a vector of exogenous parameters 2 model is parameterized at . The standard = (0; 0; 0). The basic structure of the extended model and its equilibrium remain the same as those of the standard model, but there are many other interesting ways to extend the standard model to more parameters; we view this set of extended parameters as a natural example. Equilibrium To facilitate the analysis, we present the de…nition of equilibrium in a general equilibrium format. Let LM i denote the worker population in region i, and let Ai , (m1i ; m2i ) denote the consumption of agricultural and manufactured goods, respectively, in region i. Let AAi , m1Ai , m2Ai denote the consumption of farmers in region i. Let q i denote the output level of each region n i …rm. An allocation in2 the economy o2 j j i ; A ; A ; m ; m is described by the following list of variables: LM ; n ; q . i Ai i i i Ai j=1 i=1 A feasible allocation satis…es the following constraints: M M LM 1 + L2 = L : (3) 1 A 1 M 1 A 1 LM 1 m1 + L1 mA1 + L2 m2 T + L2 mA2 T q 1 = 0: (4) 2 A 2 M 2 A 2 LM 1 m1 T + L1 mA1 T + L2 m2 + L2 mA2 q 2 = 0: (5) A M A LM 1 A1 + L1 AA1 + L2 A2 + L2 AA2 LA = 0: (6) Equation (3) balances the total manufacturing worker population, each providing one unit of labor inelastically, and the total labor used. Equations (4) and (5) balance the consumption of each manufactured good and the amount produced. Equation (6) balances consumption of agricultural commodity and the amount produced. Facing prices pA , p1 , p2 , w1 , and w2 , the following conditions are satis…ed in equilibrium. (Note that we have already imposed no-arbitrage on the transportation of goods.) The entry of new …rms drives the pro…t of operating …rms down to zero. 1 = 2 8 = 0: (7) Workers in the manufacturing sector are identical and freely mobile; they migrate to the region with a higher utility level. Let M = LM denote region 1’s share 1 =L of manufacturing worker population. In equilibrium, manufacturing workers’ utility levels must be the same in both regions if there are manufacturing workers in both regions. Thus, the migration equilibrium condition is v1 = v2 , if 0 < < 1: (8) Note that manufacturing workers’ utility vi is not de…ned if there are no manufacturing workers in region i. For completeness, we de…ne the potential manufacturing wage of a region as the limit of the equilibrium manufacturing wage when worker population approaches zero. Then, the potential utility is derived accordingly. Having all manufacturing workers in one region constitutes a (boundary) equilibrium if the potential utility in the other region is not higher. However, since the crossing part of the bifurcation is interior, we focus on 2 (0; 1). An equilibrium consists of a list of prices and a feasible allocation such that conditions (1), (2), (7), and (8) are satis…ed. We simplify the system as follows. First, by (1), the demand by workers in region i for the agricultural good and manufac1 pji ) wi =pA and mji = wi Gi1 tured goods are Ai = (1 , respectively, and the 1 demand by farmers in region i for the two types of goods are AAi = (1 ) and 1 mjAi = pA Gi1 = pji . By (2), pi = cwi = . Then by (7), 1 (F + 1 ) ; c (1 ) LM i qi = ni = (F + i) +c (F + i ) c(1 ) LM (1 = i F+ ) : i Plugging the results above into equations (4) and (5), we have LM w1 G11 1 + A 1 LA 1 p G1 1 cw1 + 1 cw1 ) LM w2 G21 T (1 1 1 cw1 + A 1 LA 2 p G2 T 1 T 1 cw1 T 1 (F + 1 ) = 0; c (1 ) (9) L M w1 G 1 T 1 1 cw2 T 1 A LA 1 p G1 T 1 + 1 cw2 T 1 + (1 ) L M w2 G 2 1 1 cw2 1 A LA 2 p G2 1 + 1 cw2 1 (F + 2 ) = 0; c (1 ) (10) 9 Equation (8) can be replaced with (1 + ) w1 G1 w2 G2 = 0: (11) Finally, normalizing the agricultural price to pA = 1, we have a system of three variables and three equations. The three equations are the last three above, and the three variables are w1 , w2 , and . Let w = (w1 ; w2 ) and let f1 , f2 , g denote the left-hand side functions of (9), (10), and (11), respectively. Let F (w1 ; w2 ; ; ) = (f1 ; f2 ; g), F : <3++ ! <3 . We will focus on the parameter space ; F (w; ; ) = 0 de…nes the reduced form static equilibrium concept for a parameterized economy. Since the focus is on migration dynamics, the adjustment of market prices is assumed to take no time. Once all workers choose a region to live in, commodity markets reach an equilibrium instantaneously given the population distribution. For …xed parameters, let w ( ; ) denote the equilibrium price under population , which is derived from ffi (w; ; ) = 0gi=1;2 . In Proposition 1, we will show that this solution is unique. With this structure, the migration balance condition (11), after solving for w ( ; ) but with as the remaining endogenous variable, is g (w ( ; ) ; ; ) = 0: Note that F (w; ; ) = 0 if and only if g (w ( ; ) ; ; ) = 0. Let f = (f1 ; f2 ). This approach is valid if there exists a unique solution w ( ; ) to ffi (w; ; ) = 0gi=1;2 for any …xed and any …xed admissible parameters 2 . We use the following su¢cient condition for existence and uniqueness of equilibrium: the system f (w; ; ) satis…es the index condition if j Dw f (w; ; )j > 0 at every equilibrium for all and for all 2 2 (0; 1) (as in Mas-Colell, 1995, De…nition 17.D.2). This index condition implies that the equilibrium of F (w; ; ) = 0 is unique by the Index Theorem (see Mas-Colell, 1995; and Kehoe, 1998). To explain further, the index condition is a standard condition from the smooth economies literature that implies existence and uniqueness of equilibrium. In simple terms, it uses the mathematical theory for an index of a …xed point to force uniqueness of equilibrium. For example, in the classical exchange economy with two commodities, the condition tells us that the slope of the derivative of aggregate excess demand has the same sign at all equilibria, namely whenever aggregate excess demand is zero, 10 so by continuity aggregate excess demand crosses zero at most once. Existence of equilibrium also follows from the index theorem. Proposition 1. f (w; ; ) satis…es the index condition j Dw f (w; ; )j > 0. Proof. 1 @f1 = @w1 c @f1 = @w2 cw1 1 @f2 = @w1 cw2 1 @f2 = @w2 c 1 (w1 ) 2+ 1 cw1 B1 + 1 1 LM G11 1 ) LM G21 T 1 (1 1 LM G11 T 1 1 1 (w2 ) 2+ 1 B2 + cw2 1 1 ) LM G21 (1 where B1 = LM w1 G11 B2 = LM w1 G11 T 1 ) LM w2 G21 T 1 + (1 ) LM w2 G21 + (1 A 1 1 + LA 2 p G2 T A 1 + LA 1 p G1 A 1 1 + LA 1 p G1 T A 1 + LA 2 p G2 >0 > 0: So, j Dw f (w; ; )j = = + + @f1 @f2 @w1 @w2 @f1 @f2 @w2 @w1 2 c 1 c 1 c 1 c 1 (w1 w2 ) 2+ 1 B1 B2 + c 2 1 (w1 ) 2+ 1 (w2 ) 1 1 B1 (1 2 (w2 ) 2+ 1 (w1 ) 1 1 B2 LM G11 2 (w1 w2 ) 1 1 (1 ) LM 2 G11 G21 (1 ) LM 2 G11 G21 T 1 : 2 (w1 w2 ) 1 1 2 The …rst three terms are all positive, and the fourth and …fth terms become c 2 1 (w1 w2 ) 1 1 (1 ) LM 11 2 G11 G21 1 T1 2 > 0: ) LM G21 Since T > 1 and 2 1 < 0, we have T 1 2 < 1 . Therefore, j Dw f (w; ; )j > 0. This property holds for all values of endogenous variables, not just equilibrium values. Corollary 1. 8 2 , 8 2 (0; 1), equilibrium in commodity markets de…ned by equations (9) and (10) exists and is unique. The index condition implies a unique equilibrium for a system of excess demand functions (see, for example, Mas-Colell, 1995; and Kehoe, 1998). It can easily be veri…ed that f satis…es the properties of excess demand functions such as: Walras’ Law holds; f is bounded from below; and if there is a sequence of prices with a component approaching zero, then the excess demand approaches in…nity. Notice that since the Index Condition for our model relies on derivatives with respect to endogenous variables, it is veri…ed for the parameter space that is a product of ours and transport cost T , for example. 3. Migration Dynamics The free migration condition requires that at an interior equilibrium (0 < < 1), skilled workers receive the same utility level in both regions. Various migration dynamics can be added, in a consistent manner, on top of this migration equilibrium condition. Given some parameters describes the dynamics of 2 , a C 2 vector …eld h ( ; ), h : (0; 1) !R after solving for w ( ; ): _ = h( ; ): The dynamics are consistent with the migration condition if the following properties are satis…ed for all ( ; ) 2 (0; 1) : (D1) If h ( ; ) = 0, then g (w ( ; ) ; ; ) = 0. (D2) If D g (w ( ; ) ; ; ) has full rank (equal to 1), then D h ( ; ) has full rank (equal to 1). Condition D1 says that stationary points of h select from solutions to the migration equilibrium condition g (w ( ; ) ; ; ) = 0. Moreover, condition D2 says that the 12 dynamics of h preserve the rank of the Jacobian matrix of g in the parameter space. The function g is the di¤erence in indirect utility for the two regions. A stronger condition on the derivatives, which is not needed in our analysis, would be: when an exogenous change in parameters (keeping endogenous variables …xed) makes utility higher in a region, population wants to move there. Conditions D1 and D2 rule out strange dynamics that alter the nature of the economy. Our genericity analysis in fact applies to all C 2 dynamics that satisfy conditions D1 and D2. A common example of dynamics satisfying our assumptions is replicator dynamics (Weibull, 1995; Fujita et al., 1999; and Baldwin et al., 2003).2 The population change in a region is proportional to the di¤erence between the local utility level and the average utility level: h( ; ) = [v1 ( ; ) ( v1 ( ; ) + (1 De…nition 1. A dynamic equilibrium of an economy ) v2 ( ; ))] : 2 is a population ratio 2 (0; 1) such that h ( ; ) = 0. Under D1, implicit in this de…nition is the fact that commodity markets clear, since w ( ; ) is an argument of g. Parameter Paths The vector …eld h ( ; ) for dynamics is de…ned over the whole parameter space . Previous literature has examined dynamics when the transportation cost is changed, keeping other parameters …xed. This is a very special parameter path that fol- lows along the transportation cost axis. The general case is when many parameters change simultaneously, resulting in a one dimensional smooth path through the multidimensional parameter space . Therefore, we proceed to examine the dynamics along arbitrary “parameter paths” in the parameter space. A parameter path is a C r map : [0; 1] ! where r 2. In other words 2 C r ([0; 1]; ), where we impose the standard C r topology on this space of parameter paths. The path de…nes a one-parameter family of vector …elds h ( ; (t)), where 2 The replicator dynamics satisfy conditions D1 and D2. 13 t 2 [0; 1] is used to index the parameter path. Let E ( ) = f( ; t) 2 (0; 1) [0; 1] j h ( ; (t)) = 0g denote the set of dynamic equilibrium points. Given this structure, we can de…ne bifurcations. An equilibrium locus from an equilibrium point ( ; t) 2 E ( ) is the image of a continuous map e : [0; 1] ! (0; 1) [0; 1] such that e (0) = ( ; t) and e (z) 2 E ( ) for z 2 [0; 1]. The equilibrium locus takes as its domain the unit interval purely for convenience. A parameter path has a bifurcation with crossing equilibrium loci at ^ ; t^ 2 E ( ) if for any neighborhood around ^ ; t^ there are more than two distinct equilibrium loci from ^ ; t^ . This type of bifurcation includes the tomahawk, the pitchfork, and the transcritical bifurcations. Next, we claim that a necessary condition for having crossing equilibrium loci at ^ ; t^ is that D( ;t) h ^ ; t^ does not have full rank. It is easy to see this as follows. D( ;t) h ^; t^ , a vector with two components, has full rank if and only if it is not zero. Say Dt h ^ ; t^ is nonzero. By the implicit function theorem, h ( ; (t)) = 0 can be locally solved as a C 1 function of . This means that E ( ) is a C 1 curve in a neighborhood of ^ ; t^ . Therefore, in a small neighborhood, there can be only two distinct equilibrium loci from ^ ; t^ . An analogous argument applies if D h ^ ; t^ is nonzero. Therefore, if a path has D( ;t) h ( ; (t)) with full rank at all of its equilibria (namely where h ( ; (t)) = 0), it does not have bifurcations with crossing equilibrium loci. The next proposition says that if the parameter space is chosen properly, generically in all paths there is no such kind of bifurcation. More precisely, the set of paths without such bifurcations is open and dense. We say that parameter space b satis…es the rank condition for h if D( ; )h ( ; ) has full rank whenever h ( ; ) = 0 (such parameter spaces are used in Debreu, 1970; Dierker, 1974; and Mas-Colell, 1985). This condition is standard in the smooth economies literature of general equilibrium theory, and is satis…ed by an open set of economies. In the language of that literature, it is called a regular parameterization. Proposition 2. For any h satisfying D1 and D2, parameter space rank condition for h. 14 satis…es the Note that parameter spaces with more exogenous variables but containing as a subspace also satisfy the rank condition. Thus, it is easy to …nd such parameter spaces as long as they contain a minimum set of parameters that have a full rank Jacobian matrix with respect to endogenous variables and exogenous parameters at equilibrium (see also Berliant and Zenou, 2002; and Berliant and Kung, 2006). Proof of Proposition 2. Using the de…nition of F = (f1 ; f2 ; g) and = ( 1; 2; ), it is straightforward to calculate 0 B D F (w; ; ) = B @ 0 0 0 c(1 0 0 0 c(1 ) ) w1 G1 By the index condition and the Implicit Function Theorem, D w( ; ) = 1 C C: A D f (w; ; ) [Dw f (w; ; )] 1 : Then, D g (w ( ; ) ; ; ) = D w ( ; ) Dw g (w; ; ) + D g (w; ; ) = = D f (w; ; ) [Dw f (w; ; )] 1 Dw g (w; ; ) + D g (w; ; ) D f1 (w; ; ) D f2 (w; ; ) [Dw f (w; ; )] 1 Dw g (w; ; ) + D g (w; ; ) : This expression is a linear combination of three vectors D f1 , D f2 and D g, that are linearly independent whenever F (w; ; ) = 0 since D F has full rank. Thus, we can conclude that D g (w ( ; ) ; ; ) 6= 0 and has full rank whenever g (w ( ; ) ; ; ) = 0. By conditions D1 and D2, we know that D h ( ; ) has full rank whenever h ( ; ) = 0. Note that D( ; )h ( ; ) having full rank does not imply that D( ;t) h ( ; (t)) has full rank, but rather implies a generic property of the parameter paths : Proposition 3. For dynamics h satisfying D1, the set of parameter paths that do not have bifurcations with crossing equilibrium loci is open and dense for any open parameter space b satisfying the rank condition for h, for example any b with a lower dimensional subspace of b . 15 We will use the following Theorem in the proof of Proposition 3. For a C r map : A ! B between manifolds A and B, we say that b 2 B is a regular value of if Da (a) has full rank whenever (a) = b. We cite the following theorem (see Guillemin and Pollack, 1974, p. 68; and Mas-Colell, 1985, p. 320): Transversality Theorem. Suppose that S ! <n is a C r map where : X X and S are C r boundariless manifolds with r > max f0; dim (X) (x) = f (x; s), s s : X ! <n . If c 2 <n is a regular value for in a set of measure zero in S, c is a regular value for ng, and let , then except for s s. The proof of Proposition 3 follows closely the proof of Mas-Colell (1985, Proposition 8.8.2, p. 345). Proof of Proposition 3. The set of paths such that D( ;t) h ( ; (t)) has full rank whenever h ( ; (t)) = 0 is open because of continuity.3 To show that this set is also dense, for any path , we construct a path 0 that is arbitrarily close to and does not have bifurcations with crossing equilibrium loci. For any path , de…ne a map : (0; 1) [0; 1] <3 ! <, ( ; t; a) = h ( ; (t) + a) where a 2 R3 and full rank whenever (t) + a 2 b . Then, Da ( ; t; a) = D h ( ; ); the latter has ( ; t; a) = h ( ; (t) + a) = 0 (using the rank condition). the Transversality Theorem, D( ;t) ( ; t; a) has full rank whenever By ( ; t; a) = 0 for 0 almost all a. So, we can pick any a with this property arbitrarily close to zero and set 0 (t) = (t) + a0 . Therefore, D( whenever h ( ; 0 (t)) = ;t) h ( ; ( ; t; a0 ) = 0. Then, 0 0 (t)) = D( ;t) ( ; t; a0 ) has full rank (t) is the path we want. Evidently, Proposition 3 holds for any open parameter space b satisfying the rank condition for h, including all b that contain as a subspace. For example, b could include all of the parameters in but also the transport cost parameter T . simply easier to exposit our analysis with fewer parameters. 3 A simple proof by contradiction works well here. 16 It is 4. Conclusion The study of bifurcations provides interesting insights into the complex dynamic behavior of a system. It is important to study an economic system in a one dimensional parameter space when the chosen parameter is the main force changing the economy. In the case of the new economic geography, that parameter is transportation cost. However, the real world has many parameters, and the choice of parameters a¤ects the equilibrium (bifurcation) diagram of a system. This raises the following question: Given enough parameters, what kind of dynamic behavior is typical? We characterize the generic pattern of dynamic regional systems along general smooth paths of parameter change in a higher (for example, 3) dimensional parameter space. We show that, in a parameter space satisfying the rank condition, there is a generic (open and dense) set of parameter paths that do not have bifurcations with crossing equilibrium loci. Thus, the use of such bifurcations, for example the tomahawk bifurcation, to generate core-periphery urban patterns from an initial uniform pattern is suspect because it relies on the strategic choice4 of very speci…c parameter values and paths of parameter change. This has led to an urban legend. It is easy, but notationally burdensome, to extend our results to more general models. For example, the two region framework can easily be replaced with n regions. The arguments are basically unchanged. References [1] Baldwin, R., Forslid, R., Martin, P., Ottaviano, G.I.P., Robert-Nicoud, F., 2003, Economic Geography and Public Policy, Princeton University Press: Princeton, NJ. [2] Berliant, M., Kung, F.-C., 2006, The indeterminacy of equilibrium city formation under monopolistic competition and increasing returns, Journal of Economic Theory 131, 101-133. 4 In particular, by the progenitors of the New Economic Geography. Point …nger here. 17 [3] Berliant, M., Zenou, Y., 2002, Labor di¤erentiation and agglomeration in general equilibrium, mimeo, Washington University and University of Southampton. [4] Debreu, G., 1970, Economies with a …nite set of equilibria, Econometrica 38, 387-392. [5] Dierker, E., 1974, Topological Methods in Walrasian Economics, Lecture Notes in Economics and Mathematical Systems, Springer-Verlag: New York, NY. [6] Forslid, R., Ottaviano, G. I. P., 2003, An analytically solvable core-periphery model, Journal of Economic Geography 3, 229-240. [7] Fujita, M., Krugman P., Venables A.J., 1999, The Spatial Economy: Cities, Regions, and International Trade, MIT Press: Cambridge, MA. [8] Fujita, M., Mori, T., 1997, Structural stability and evolution of urban systems, Regional Science and Urban Economics 27, 399-442. [9] Guckenheimer, J., Holmes, P., 1997, Nonlinear Oscillations, Dynamical Systems, and Bifurcations of Vector Fields, 5th ed., Springer-Verlag: New York, NY. [10] Guillemin, V., Pollack, A., 1974, Di¤erential Topology, Prentice-Hall: Englewood, NJ. [11] Kehoe, T.J., 1998, Uniqueness and stability, in: A. Kirman, ed., Elements of General Equilibrium Analysis, Blackwell: Oxford. [12] Mas-Colell, A., 1985, The Theory of General Economic Equilibrium: A Di¤erentiable Approach, Cambridge University Press: Cambridge, MA. [13] Mas-Colell, A., Whinston, M. D., Green, J. R., 1995, Microeconomic Theory, Oxford University Press: New York, NY. [14] Weibull, J.W., 1996, Evolutionary Game Theory, MIT Press: Cambridge, MA. 18 x 2 1.5 1 0.5 0 b -0.5 -1 -1.5 -2 -1 0 Figure 1: a = 0; 1 2 2 < b < 2: x 2 1.5 1 0.5 0 b -0.5 -1 -1.5 -2 -1 0 Figure 2: a = 0:005; 19 1 2 < b < 2: 2 x 2 1.5 1 0.5 0 b -0.5 -1 -1.5 -2 -1 0 Figure 3: a = 0:005; 1 2 2 < b < 2: 2 1 x 1 0 -1 -2 -2 0 -1 0 b 1 -1 2 Figure 4: 1:2 < a < 1:2; 20 2 < b < 2: a