The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
orKing raper series Chaos, Sunspots, a n d Automatic Stabilizers Lawrence J. Christiano and Sharon G. Harrison 3 Working Papers Series Macroeconomic Issues Research Department Federal Reserve Bank of Chicago August 1996 (WP-96-16) 1' B R A R V N0\ 0 5 1996 KtuERAL RESERVE BANK OF CHICAGO FEDERAL RESERVE B A N K OF CHICAGO August 1996 Chaos, Sunspots, a n d A u t o m a t i c Stabilizers* Lawrence J. Christiano Northwestern University, NBER, and the Federal Reserve Banks of Chicago and Minneapolis Sharon G. Harrison Northwestern University ABSTRACT W e study a one-sector growth model which is standard except for the presence of an externality in the production function. The set of competitive equilibria is large. It includes constant equilibria, sunspot equilibria, cyclical and chaotic equilibria, and equilibria with deterministic or stochastic regime switching. The efficientallocation ischaracterized by constant employment and a constant growth rate. W e identify an income tax-subsidy schedule that supports the efficient allocation as the unique equilibrium outcome. That schedule has two properties: (i) itspecifies the tax rate to be an increasing function of aggregate employment, and (ii)earnings are subsidized when aggregate employment is at its efficient level. The first feature eliminates inefficient, fluctuating equilibria, while the second induces agents to internalize the externality. *We have benefitted from discussions with Fernando Alvarez,Jess Benhabib, Michele Boldrin, V.V. Chari, Russell Cooper, Martin Eichenbaum, Ian Domowitz, Chris Gust, Nicola Persico, Michael Woodford, Randall Wright, and Michelle Zaharchuk, and we are grateful to Victor Valdivia for research assistance. Christiano isgrateful to the National Science Foundation for financial support. The views expressed herein are those of the authors and not necessarily those of the Federal Reserve Bank of Chicago or the Federal Reserve System. 1. In tro d u ctio n Interest is growing in business cycle models with multiple, self-fulfilling rational expectations equilibria,1 These models offer a new source of impulses to business cycles— disturbances to expectations— and they offer new mechanisms for propagating and magnifying the effects of existing shocks, such as shocks to monetary policy, to government spending, and to tech nology.2 Although initial versions of these models appear to rely on empirically implausible parameter values, recent vintages are based on increasingly plausible empirical foundations.3 The policy implications of the new models differ sharply from those of current main stream equilibrium models, which emphasize shifts to preferences and technology as the basic impulses to the business cycle. These models have been used to articulate the notion influential early papers include Azariadis (1981), Bryant (1983), Cass and Shell (1983), Cooper and John (1988), Diamond (1982), Farmer and Woodford (1984), Shleifer (1986), and Woodford (1986b). The first paper to take seriously the quantitative predictions of a business cycle model with self-fulfilling expectations isWoodford (1988). Rational expectations models with multiple equilibria have attracted attention in other areas too. See Benhabib and Perli (1994), Krugman (1991), and Matsuyama (1991a) for a discussion in the context of international trade and growth. See Cole and Kehoe (1996) for an analysis of the Mexican debt crisis. See Bryant (1981) and Diamond and Dybvig (1983) for discussions in the context of models of banking. See Boldrin, Kiyotaki, and Wright (1993) and Mortensen (1989,1991) for discussions on dynamic models of search and matching. 2An extensive literature documents the inadequacy of propagation in standard business cycle models. See, for example, Burnside and Eichenbaum (1995), Christiano (1988, p. 269), Cogley and Nason (1995), Rotemberg and Woodford (1996), and Watson (1993). An early study showing how models with indetermi nate equilibria provide increased magnification and propagation of monetary shocks isFarmer and Woodford (1984). More recent studies include Beaudry and Devereux (1994), Benhabib and Farmer (1996), Guesnerie and Woodford (1992), and Matheny (1994). For a recent argument that macroeconomists are short on shocks for accounting for the business cycle, see Cochrane (1995). 3For example, the models of Benhabib and Farmer (1994), Farmer and Guo (1994) and Gali (1994a,b) rely on increasing returns in production. (See Schmitt-Grohe 1995.) The results of subsequent empirical research suggested that the degree of increasing returns required for the Benhabib and Fanner and Fanner and Guo models to display sunspot equilibria is too high. (See Burnside, Eichenbaum, and Rebelo 1995 for one such effort, and see also Benhabib and Fanner 1995 for a review of several others.) These empirical results in turn stimulated further theoretical work: recently, Benhabib and Fanner (1995) showed that a twosector version of their model requires a much lower degree of increasing returns to guarantee the existence of sunspot equilibria. that economic fluctuations represent the economy’s efficient responses to shocks and to cast doubt on the desirability of macroeconomic stabilization policy.4 In contrast, the new mod els suggest that institutional arrangements and policy rules designed to reduce fluctuations in output may be desirable. (See Grandmont 1986, Guesnerie and Woodford 1992, Shleifer 1986, Woodford 1986b, 1991, and the articles in the symposium summarized in Woodford 1994.)56 This paper pursues these ideas in a particular business cycle model. The model studied is a version of the one-sector, external increasing returns model recently studied by Baxter and King (1991), Benhabib and Farmer (1994,1995), and Farmer and Guo (1994,1995).® Our version of this model has a significant advantage relative to the versions analyzed in the literature. In those, analysis of the global set of equilibria is typically quite difficult, and so researchers confine themselves to studying the set of equilibria that is local to the steady state. By contrast, the structure of our model is such that the global set of equilibria is transparent and can therefore easily be analyzed. It turns out that this set of equilibria is remarkably rich, and itincludes sunspot equilibria, regime switching equilibria, and equilibria 4See Kydland and Prescott (1980) and Sargent (1979, p. 393) for a statement of the case that output sta bilization is undesirable. The “preference and technology” literature on macroeconomics did not completely rule out the possibility that some forms of stabilization might be desirable. Researchers who incorporated frictions like price rigidities did see some role for activist policy. (See Fischer 1980.) 5An important early exsimple of the potential stabilizing role of institutional arrangements occurs outside the area of business cycle analysis and is provided by the work of Bryant (1981) and Diamond and Dybvig (1983). They showed that a spontaneous burst of pessimism on the part of depositors could trigger so cially inefficient bank runs and that a government policy— deposit insurance— could be designed that would eliminate this source of instability. 6As in these papers, we do not formally articulate what the source of external increasing returns is. Examples of analyses that are explicit about the nature of external effects include Diamond (1982), Howitt and McAfee (1988), and Romer (1986). Benhabib and Farmer (1994) suggest the possibility that there is a way of reformulating our model so that the source of increasing returns is internal to the firm, while leaving our basic analysis unaffected. The analysis in Romer (1987) suggests yet another possibility: that the increasing returns may actually reflect gains from specialization. W e have not yet explored these possibilities. 2 which appear chaotic. Our analysis illustrates the potential pitfalls of focusing only on the equilibria that are local to some steady state. W e establish that the set of bounded solutions to a particular expectational difference equation corresponds to equilibria for our model. This set is simple to characterize because the difference equation is only first order and has a simple analytic representation. This is true, despite the fact that capital accumulation and employment are endogenous in the model. A n important feature of the difference equation is that for every initial condition, it has two solutions. This two-branch feature of the Euler equation is an important reason the set of equilibria for the model includes re g im e s w it c h in g equilibria of the type studied by Hamilton (1989) and equilibria that appear chaotic.7 Even the efficient allocations in our model are straightforward to determine, despite the lack of convexity in the aggregate resource constraint set due to the externality. The efficient allocations are unique and involve no fluctuations. W e examine the operating characteristics of two a u t o m a tic s t a b iliz e r tax regimes. Each has the property that the income tax rate rises if the economy moves into a boom and falls if it goes into a recession. Under each tax regime, the economy has a unique interior equilibrium, in which output is constant. However, one tax regime stabilizes output on an inefficient level of output, and the other stabilizes output on the efficient allocations. W e show that implementing the first tax regime may increase, or even decrease, welfare.8 W e 7 For other examples of a “branching” Euler equation in infinite horizon growth models, see Benhabib and Perli (1994) and Benhabib and Rustichini (1994). 8This possibility has been discussed by Guesnerie and Woodford (1992, pp. 383-388), Shleifer (1986), and Woodford (1991, p. 103) in other contexts. 3 establish two things about designing a tax system that supports the efficient allocations as the unique interior equilibrium. First, such a system must specify that the tax rate vary with the level of aggregate economic activity. W h e n the tax rate is specified to be a fixed constant, then there is more than one equilibrium, with the efficient one being only one of them. Second, the efficient equilibrium in this case is determinate, so that a standard local analysis of the set of equilibria would falsely conclude that only one equilibrium is possible. These results draw attention to the importance of the proper design of automatic stabilizer tax systems and point to a potential pitfall in the traditional approach to policy design, which tends to focus on minimizing output variance.9 These results also illustrate the potential dangers of the standard practice of focusing exclusively on local equilibria. Finally, our model provides a convenient vehicle for articulating some econometric issues that arise in the analysis of models with multiple equilibria. As emphasized by Woodford (1991, p. 77), there is a widespread perception that “anything goes” with sunspot models— any set of facts can be explained. The model in this paper can be used to illustrate that sunspot models in principle do impose discipline on an empirical analysis.10 With one ex ception, the econometric procedures used to analyze standard models with unique equilibria and driven by exogenous shocks can be used to analyze and test sunspot models too. The exception is that procedures which select parameter values by equating model first moments and corresponding sample first moments may no longer be well-defined. This is because 9An influentialexample isthe analysis ofPoole (1970), who argues that the appropriate choice of monetary policy regime depends on whether shocks emanate from financial markets or investment decisions. The criterion driving the policy design in Poole’s analysis is minimization of output variance. 10See Dagsvik and Jovanovic (1994), Farmer and Guo (1995), Imrohoroglu (1993), Jovanovic (1989), Sargent and Wallace (1987), and Woodford (1987,1988,1991) for farther development of this point. 4 there may be a set of possible first moments associated with any parameter configuration, depending on which equilibrium the economy is in. The intuition underlying the dynamics in our model is essentially the same as that de scribed by Benhabib and Farmer (1994) and Farmer and Guo (1994). The economy is perfectly competitive, and individual producers have linearly homogeneous production func tions in capital and labor, which are strictly concave in each. However, economywide average output operates as an externality in front of each firm’s production function, shifting it up when average output is high. The latter is the key to why there are multiple equilibria in our model and to why expectations can act as an independent source of fluctuations. If all households act on the conjecture that the current period’s wage rate is high by supplying more labor services to the market, then the market-clearing wage is high because of the externality on labor. Similarly, ifhouseholds act on the conjecture that next period’s rental rate on capital is high by buying more investment goods today, then their conjecture will be validated.11 A regime which specifies that the tax rate rises with aggregate employment has the potential to stabilize output by defeating the mechanism that gives rise to multiple equilibria. Conjectures that the rate of return on market activity is high cannot be self-fulfilling ifthe 11This is the case for two reasons. First, the externality on capital prevents next period’s increase in capital from directly reducing the marginal product of capital. (That the externality is strong enough for the aggregate capital stock not to enter the marginal product of capital is the reason the difference equation mentioned above is first order, which in turn is the reason the global set of equilibria is transparent in our model.) Second, the externality on labor helps ensure that the increase in next period’s wage rate, occasioned by the rise in capital next period, stimulates a large increase in employment. This indirectly helps drive up next period’s rental rate on capital. 5 proceeds are taxed away.12 The outline of the paper is as follows. Our model and equilibrium concept are presented in Section 2. Section 3 establishes our characterization result for the set of competitive equilibria. Sections 4 and 5 analyze the deterministic and stochastic equilibria of the model, respectively. Econometric issues are addressed in the context of the analysis of stochastic equilibria. Section 6 considers the impact of an automatic stabilizing tax policy and reports the socially optimal allocations. Section 7 concludes. 2. T h e M od el Let st denote the realization of the exogenous shocks at date t. In models with shocks to fundamentals, st would include shifts to preferences or technology, or to government spending. In this paper, we do not consider such shocks. In the stochastic versions of our model, st represents disturbances which influence equilibrium outcomes, but which do not affect fundamentals. W e let s* — ( s o ,s i, denote a history of realizations up to and including date t. For simplicity, we only consider environments in which the number of possible values of s t is finite for each t. The probability of history s‘ is denoted /xt(s‘). To conserve on notation, from here on we delete the subscript t on /x. This should not cause confusion: that the functions /xt (s‘) and p r (sr), r ^ t are different is evident from the fact that the quantity of elements in s* and sr is different. W e adopt this notational 12Our argument is related to the one in Sdimitt-Grohe and Uribe (1996). They show that a procyclical tax policy (designed to support a balanced budget) can destabilize the economy by making possible sunspot equilibria. 6 convention for all functions of histories. The probability of st+1 conditional on f i ( s t+1 s* is denoted |s*) = /i1+1(s t+1)//it (s t). W e now discuss the agents in our model and our concept of equilibrium. 2 .1 . Households W e assume a large number of identical households. At each s4 and t, the representative household values consumption and leisure henceforth according to the following utility func tion: I] £ (2.1) n(«0 ] aJ|st where /3c(0,1) is the discount rate, |s l denotes histories, sJ,that are continuations of the given history, s‘,and c(s7), n(sJ) denote consumption and labor, respectively, conditional on history s j . The household must respect the following sequence of budget constraints: c(*0 + k{s>) - (1 - 6 ) k ( s * - 1) [1 - -r(s^)][r(s^)A:(s,“1) + w(^)n(^)] + T(^), all where r(sj ) tively. Also, and t w (s^ ) ($*) s\ j (2.2) = > t denote the market rental rate on capital and the wage rate, respec is the tax rate on income, T (si ) denotes lump-sum transfers from the government, and k(s*) denotes the stock of capital at the end of period j , given history s^. The household also takes fc(st_1) as given at s*. Finally, the household must satisfy the 7 following inequality constraints: k(si ) > for all s7 |s‘ and j > t O.c^) > 0 ,0 < n(s^) < 1 (2.3) and takes as given and known the actual future date-state contingent prices and taxes: { r(« 0 , w M T (« 0 ,r(« * ); J > all s> |s % (2.4) W e assume that u(c, n) = log c + <7log(l — where a (2.5) n) > 0. Formally, at each s* and t, the household problem is to choose (c(s^), n(s^), k ( s j ); j > t, all s* | s*} to maximize (2.1) subject to (2.2), (2.3), (2.4), and the initial stock of capital, fc(st_1). The intertemporal Euler equations corresponding to this problem axe uc(s^) = all s> |s*, j > 0 M si+1 I5i)ttc(«i+1){[l - r(sJ+1)]r(sj+1) + 1 - 6} (2.6) t, and the intratemporal Euler equations are uc(sJ) = t1 ” T(s,) M si)> all ^ |s‘, j > t. (2.7) Here, uc(sJ) and U n { sj ) denote the partial derivatives of u with respect to its first and second 8 arguments, evaluated at c ( s j ) , Urn n ( s j ). PT Finally, the household’s transversality condition is K sT (2.8) Is t) u c ( s T ) k ( s T ) = 0. S*>‘ The sufficiency of the Euler equations, (2.6) and (2.7), and transversality condition, (2.8), for an interior solution to the household problem may be established by applying the proof strategy for Theorem 4.15 in Stokey and Lucas with Prescott (1989). 2 .2 . Firms W e assume a large number of identical firms, each of which solves a static problem at every s*. As a result, we can, without risking confusion, simplify the notation by deleting the s1 notation. The representative firm faces the following technology relating its output, its capital, K , and labor, Y N , = inputs and to the economywide average level of output, f{ y , K ,N ) = y7fCaM 1-a\ 0 < 7 ,a < 1- Y , to y: (2.9) W e assume that a = 1- 7. (2.10) The relation between the economywide average level of output and the economywide average stock of capital, k , and labor, n, is obtained by solving y = f( y , k, n) for y : (2.11) 9 given (2.10). The linearity of this function in terms of A: is essential for simplifying the analysis to come. In addition, as discussed by Rebelo (1991), linearity allows for growth to occur endogenously. The firm takes y , r, and w as given and chooses K ,N to maximize profits: Y - r K - w (2.12) N subject to (2.9). The firms’first-order conditions for labor and capital are I where /k and fs n = w, I k = r are the derivatives of / with respect to its second and third arguments, respectively. W e assume firms behave symmetrically, so that consistency requires y K ,n = N . (2.13) = Y ,k = Imposing these, we get } n = (1 - c*)nA:, /* = cm2 (2.14) with 7 = 2/3. With this value of 7 ,the model implies that labor’s share is 2/3, which is close to the value estimated using the national income and product accounts (Christiano 1988). 10 2.3. Government The income tax rate policy, r(st), is specified exogenously, and we require that the following budget constraint be satisfied for each s*: r(st)[r(st)k(st X) + (2.15) = T(s‘). 2.4. Equilibrium The resource constraint for this economy is c(s‘) + k(sl) — (1 — 6 )k(st x) < 1)n(s*)2 = y(s‘). (2.16) W e then have13 Definition 2 .1 . A s e q u e n c e - o f - m a r k e t s e q u ilib r iu m i s a s e t o f p r i c e s (r(st), tu(s*); a l l s*, a l l > 0}, q u a n t i t i e s (y(«*), c(s*), Ac(«*),n(a*); a l l s t , a l l t > 0}, a n d a t a x p o l i c y {r(st),T(st); a l l s * ,t > 0} w i t h t h e f o l l o w i n g f o u r p r o p e r t i e s f o r e a c h t, s l : t • Given t h e p r ic e s , t h e q u a n t i t i e s s o lv e t h e h o u s e h o l d ’s p r o b le m . • t h e p r i c e s a n d g iv e n G iv e n (y(s*) = fe(st-1)n(st)2}, t h e q u a n t i t i e s s o lv e t h e f i r m ’s p r o b le m . • T h e g o v e r n m e n t ’s b u d g e t c o n s t r a i n t i s s a t is f ie d . • T h e r e s o u r c e c o n s t r a i n t i s s a t is f ie d . W e find it useful to define an i n t e r i o r equilibrium. This is a sequence-of-markets equilibrium in which a < n(s*) < 6 for all s* for some a and b satisfying 0 < a < b < 1. 13It is easily verified that the analysis would have been unaltered had we instead adopted the date 0, Arxow-Debreu equilibrium concept. In this case, households would have had access to complete contingent claims markets. 11 3. C h aracterizin g E quilibrium In the next section of the paper, we study deterministic equilibria in which prices and quan tities do not vary with st and stochastic (sunspot) equilibria in which prices and quantities do vary with s t . The analysis of these equilibria is made possible by a characterization result, which is presented next. Substituting (2.13) and (2.14) into the household’s intertemporal Euler equation, (2.6), we get g(P) = 1s,)^ ) {11 - ’V +1) M » ‘+1>2 + 1 - «> (3-D where Substituting (2.14) into the household intratemporal Euler equation, (2.7), we get c(s*) = [1 - r(st)]^n(s*)[l - n(s‘)]. (3.3) The resource constraint implies that c(s*) = n(s*)2 + 1 — 6 — A(s‘). (3.4) Combining the two Euler equations, (3.1) and (3.3), and the resource constraint, (3.4), our 12 system collapses into a single equation in current and next period’s employment: ^2 /*(s *+1 IstM n ( s t),n(st+1);r(st+1)] = 0, all s‘, t > 0 (3.5) * t+ l where v is v ( n , n '; r') 1 /? [(l-T > (n ')2 + l - 6 ] A[(n')2 + 1 - n2 + 1 - 6 - X 6 - (3.6) X' ] with A = n2 + 1 — 6 — (1 — r)— n(l — n). <7 (3.7) Here, a ' denotes next period’s value of the variable. The transversality condition, (2.8), is equivalent to v v- oT + 1“ , T\ ) l1 - r(®T)]2"(ar)[l - 6 - n { s T )}] [1 — r(sT)]^n(sT )[l — n(sr )] n °' (3'8) The basic equilibrium characterization result for this economy is Proposition 3.1. S u p p o se th a t r ( s ‘ ) = 0. If, fo r a ll s t a n d t > 0, {n(s*)} satisfies (3 .5 ) and a < th e n {n(s‘)} n(s*) < b fo r so m e 0< a < b < 1 c o r r e s p o n d s t o a n e q u il ib r i u m . Proof. To establish the result, we need to compute the remaining objects— prices and quantities— in an equilibrium and verify that they satisfy (2.6), (2.7), (2.8), (2.13), and (2.16). A candidate set of objects is found in the obvious way. The sufficiency of the firstorder and transversality conditions for household optimization and the sufficiency of the first-order conditions for firm optimization guarantee that these are an equilibrium. 13 The characterization result indicates that understanding the equilibria of the model re quires understanding the v function. It is easily confirmed that v = function in n ' for each fixed n , u . (Later, we refer to a; as the E u l e r n ,u ) there are two possible n ’ : n ' — f u ( n ,u j ) f u (n,u>) and = |( 6 (71, 0 /) + n! u defines a quadratic e r r o r .) Hence, for each = /t(n,u;), where ^ b ( n ,u ) 2 - 4 e ( n ,u j ) } (3.9) fi( n ,u > ) = 5 (6(11,a/) - y j b ( n ,u j ) 2 - 4 c ( n ,o j ) } Here, b ( n ,u ) (p (n ) = = y ( " ),(7 l; 1 .)C(n,u>) a + q ( n ,u>)<p(n) n 2 + 1 — 6 — ^ti(1 — n) 0n( 9Ka;) = 1- 1 ---n(l — 1— 6 a + (3.10) q ( n , u i) ip ( n ) (3.11) n) n)u ). (3.12) The function v has the shape of a saddle, as can be seen in Figure la. The intersection of v and the zero plane (u; = 0) is depicted in Figure la as the boundary between the light and dark region of v . This intersection defines the curves /„(•,0) and /t(-,0), which are shown in Figure lb. W e refer to these as the upper and lower branches of the function v. The lower branch intersects the 45-degree line at two points, which are denoted n 1 and n2.These intersection points cannot be seen in Figure lb, but can be seen in Figure lc, which displays ■ n! — n for n near the origin. It is easy to see from Figure la that with higher values of a>, f i increases and fu decreases. The figure also indicates that for these functions to be 14 real-valued, u> must not be too big. T he branches in th e figure are com puted using our baseline param eterization, <7 = 2, 13 = 1.03~«, 6 = 0.02. Here, n 1 = 0.02 and n2 = 0.31. The gross growth rates of capital (that is, A) at these two points are 0.973 and 1.004, respectively. Our assigned value of /? is often used in the real business cycle literature. The value of a causes the m odel’s im plication for the share o f incom e going to capital and labor to coincide w ith one estim ate of this quantity based on the national income and product accounts. (See, for exam ple, Christiano 1988.) In addition, this param eterization o f a facilitates som e of th e analytic results described above. However, we have verified that the shape of the v function is not very sensitive to the perturbations in a . The assigned value of 6 can be justified based on our m odel’s capital accum ulation equation and on U.S. capital stock and investm ent data. (See, for exam ple, Christiano 1988 and Christiano and Eichenbaum 1992.) The issue of how data may be brought to bear to determ ine a value of a is addressed below. 4. D e t e r m i n i s t i c E q u i l i b r i a We begin by considering determ inistic equilibria, in which prices and quantities depend on t, but not on s t . To sim plify th e presentation, we drop the history notation, and we use th e conventional tim e subscript notation instead. As we shall see, the set of determ inistic equilibria is quite rich. For exam ple, any constant sequence {n t}, w ith tt* = n 1 or nt = n2, satisfies th e conditions of th e characterization result and so is an equilibrium. Similarly, any sequence w ith no € (n 1, n) and nt+1 = f i ( n t ,0 ), t > 0 is also an equilibrium, w ith n t —►n2. 15 Here, h satisfies n > n2 and n l = fi(n , 0). Figure 2 exhibits two equilibrium paths, one starting w ith no = 0.4 and the other w ith no = 0.2. Each path converges m onotonically to n 2. Other determ inistic equilibria are more exotic and display a variety o f types o f regim e switching. For exam ple, the equilibrium employm ent policy function could be tim e non- stationary, w ith em ploym ent determ ined by the lower branch for, say, six periods, followed by a single-period jum p to the upper branch, followed by another six-period sojourn on the lower branch, and so on. T he m odel has another type of regime sw itching equilibrium too, in which th e employm ent policy function is discontinuous. As an exam ple of the latter, consider equilibria in which em ploym ent, n', is determ ined by the upper branch for n over one set o f intervals in (0 ,1 ) and by the lower branch over the com plem ent of these intervals. One exam ple of this is given by 0) for n' = f ( n ), where /( n ) = /i(n >0) for n < n1 n1 < n < m 1 (4.1) / u(n,0) for m 1 < n < m 2 fi(n, 0) for m2 < n where m l < n2 and m 2 are a chosen set of numbers. B y considering different values o f a, (4.1) defines a fam ily of maps. As we shall see, there are elem ents in this fam ily o f maps which exhibit characteristics that resemble chaos.14 There are several concepts o f chaos in 14For other discussions of chaos, with economic examples, see Boldrin and Woodford (1990) and Mat suyama (1991b). 16 the literature. We consider two. 4 .1 . T o p o lo g ic a l C h a o s We consider th e topological concept of chaos as discussed in Devaney (1989). We require two definitions first: D e fin itio n 4 .1 . T h e m a p f : J —* J is said to b e topologically tran sitive i f fo r a n y p a ir o f open sets, U, V c J, th ere ex ists k > 0 such th a t f k(U) fl V 0. Here, f l {n) = /( n ) , / 2(n) = /[ /( n ) ] , and so on. Loosely, the above definition says that for alm ost all initial conditions, iterations on the map, / , produce an orbit (that is, n, /( n ) , / 2(n ),...) th at visits every region, no m atter how sm all, of J. An exam ple of a map that violates th is condition is the policy function o f the standard one-sector growth m odel. For any initial capital stock, iterations on the policy function generate a sequence that converges m onotonically to the steady sta te .' If the initial capital stock is below th e steady state, then capital stocks sm aller than the initial condition and above steady state are not visited. A second definition that is important is D e fin itio n 4 .2 . T h e m a p f : J —* J has sen sitive depen den ce on in itial co n d itio n s i f th ere ex ists S > 0 such th a t, for an y n € J an d a n y neighborhood N o f n, th ere e x ists y € N an d m > 0 such th a t | / m(n) — / m(y)| > 6. T his says th at for any initial condition, n, and any neighborhood, no m a tte r how sm all, around n, there is at least one other initial condition whose orbit eventually differs by at least <5 from the orbit o f n. N ote that th e parameter 6 is chosen as a function of th e map, but it is not a function o f n or the size of the neighborhood around n. A sequence generated by a map th at exhibits sen sitivity to initial conditions is difficult to forecast for tw o reasons. 17 First, th e slightest measurement error in th e initial conditions may result in a substantial error o f forecast. Second, even if th e initial conditions are measured accurately, then any slight rounding error in com puting an orbit is likely to be magnified. Then, we have D e fin itio n 4 .3 . L e t J b e a set. T h e m a p f : J -* J is sa id to b e ch aotic on J i f • f h as sen sitive depen den ce on in itial conditions. • f is to p o lo g ica lly tran sitive. • p e rio d ic p o in ts o f f are dense in J. B y a periodic poin t, n, we mean one for which there is some k > 0 such that n = /* (n ). Theorem s exist th at establish conditions under which a given map is chaotic. Unfortu nately, these theorem s require either that / be continuous (see, for exam ple, chapter 1.1 of Devaney, 1989) or that it be piecewise continuous and differentiable w ith derivative greater than unity in absolute value. (See Lasota and Mackey, 1985, chapter 6.) We are not aware o f theorem s that include maps of the kind considered here. Instead, we follow th e strategy pursued in Dom owitz and El-Gam al (1993,1994) and develop sim ulation-based evidence on w hether our map is chaotic. Consider sen sitivity to initial conditions first. For this, we com pute the Lyapunov coef ficient, L (n ), associated w ith the map, / , defined in (4.1). The function, L , m aps n € (0 ,1 ) into th e real line. For any fixed n € (0 ,1 ), L(n) = l i m i ^ t o g t= l 18 dfiO dn (4.2) where n,+1 = /(n * ), i — 1, ...,T — 1, n i = ft, and d f ( n i ) / d n denotes the derivative o f / w ith respect to n, evaluated at n = rij. To see why L is of interest, note th at the sum is equivalent to log l^ n ^ j- If n is a periodic point o f any finite order k, then f k{n) — h. If it is a stable periodic point, then < 1- If T — m k , where m is an arbitrary positive integer, then d f T {n) dn f t d f k(n,) L \ dn d f k(n) dn <0 (4.3) where nj = n, and nj+1 = /* (n (), 1 = 1 ,..., m —1. This suggests th at if ra is a stable periodic point of any period, then L (n ) m ust be negative. But if / has a stable periodic point, or a point whose orbit intersects w ith such a point, then it violates sensitivity to initial conditions. Thus, a negative value of L(n) indicates that one of the conditions necessary for / to be chaotic fails. A positive value of L(n) is a necessary condition for chaos. Figure 3 shows L(n ) for a in the range 1.25 to 2.20, w ith n fixed at 0.255.15 We truncated the infinite sum in (4.2) at T = 2,000. We set m 1 = 0.33, m2 = 0.70. N ote th at L(h) is positive for values of <r less than 1.5. Hence, for values of a in this range, there is evidence that / is characterized by sen sitivity to initial conditions. To investigate this further, consider a = 1.25. Figures 4a and 4b show two sequences of 400 observations on hours worked, sim ulated using / . In one case th e initial condition is 0.455, and in the other th e initial condition is 0.454. D espite th e fact th at the initial conditions are very close, the tw o orbits are quite different. In fact, th ey are eventually as dissim ilar as they would have been had the 15 In computational experiments not reported here, we found that the graph in Figure 3 is insensitive to the value of fi. This insensitivity is consistent with results in Figure 5, discussed below. 19 initial conditions been far apart. B y the 70th observation, th e two series are com pletely out o f phase. Then, by around th e 310th observation, they are alm ost identical again. A lthough th e am plitude of th e tw o series varies som ewhat, m ost of th e differences betw een th e tw o series has to do w ith phase. N ote how strikingly different these equilibria are from th e ones shown in Figure 2. Now consider topological transitivity. One way to investigate this is to com pute th e histogram o f orbits associated w ith different initial conditions. To com pute th is histogram , we divide th e unit interval into 3,000 equal-width intervals and approxim ate th e histogram o f an orbit by th e number of tim es hours worked lands in each interval in a sim ulation o f length 50,000. Figure 5 shows histogram s for orbits associated w ith four initial conditions drawn from very different parts of the unit interval. These histogram s are sim ilar in tw o respects. First, consistent w ith topological transitivity, each orbit appears to cover th e sam e region of th e unit interval. In particular, let J denote the union of intervals w ith positive m ass in Figure 5. Then J appears to be independent o f th e initial conditions. T his is consistent w ith th e notion th at orbits associated w ith alm ost all n 6 J wander through every sm all neighborhood of J. T he second striking feature of the histogram s in Figure 5 is th at th ey appear to have th e sam e shape. Thus, the histogram s are consistent not only w ith th e notion th at alm ost all orbits in J visit each subinterval in J w ith positive probability ( topological tra n sitivity), but they are also consistent w ith th e notion th at th e probability of visitin g each subinterval is the sam e across orbits. 20 4 .2 . S ta tis tic a l C h a o s A second concept o f chaos, closely related to the first, is statistical (Lasota and Mackey 1985). Here, we follow the treatm ent in Domowitz and El-Gam al (1993,1994) and El-Gam al (1991). Let g be a density function defined on J. That is, f j g ( n ) d n = 1 and g(n ) > 0 for all n 6 J. If we draw from g and apply the map f* : J —►J to each draw, we have a new distribution of points on J. D enote this distribution by Then following Dom ow itz and El-Gam al (1993,1994), we say that / exhibits sta tistica l chaos if /* exhibits th e ergodic property or th e m ixing property. D e fin itio n 4 .4 . T h e m a p f : J —* J is ergodic if fan ^ for aU 9 € G. For regularity conditions on the lim iting density function, q, and the set of density func tions, G , see Dom owitz and El-Gam al (1993,1994). (Obviously, G cannot include density functions which place mass exclusively on a single periodic point.) For / to be ergodic does not actually require th at f* settle down for large i. The m ixing property does require this. D e fin itio n 4 .5 . T h e m a p f : J —*■J is m ixin g if lim f l — q, for all g € G. t—*oo 9 T he properties of ergodicity and m ixing are closely related to the notions of topological transitivity and sensitive dependence. For exam ple, consider a density function, g, which 21 assigns positive probability to an arbitrarily selected and extrem ely sm all interval o f initial conditions. M ixing requires that the orbits of these points eventually cover the sam e range in J as if the initial conditions were instead drawn from a density that assigns positive probability to every subinterval o f J. We adopted the sim ulation-based approach of Dom owitz and El-Gam al to investigate w hether our / map exhibits statistical chaos. Thus, we considered two g functions. One places a uniform distribution on th e interval [0.16,0.32] and the other places a uniform distribution on the interval [0.58,0.71]. In each case, we drew 1,000 tim es from th e g function and com puted / 3,00° for each draw. The resulting histogram s are shown in Figure 6. There are two interesting features of these histogram s. First, to the unaided eye th ey appear very sim ilar to each other, consistent w ith th e notion that / satisfies th e mixing condition. Still, the differences can be reasonably substantial, as the bottom graph in Figure 6 shows. Second, th e histogram s in Figure 6 closely resemble the orbit histogram s shown in Figure 5. T his suggests that our / map approxim ately satisfies concepts of ergodicity in standard econom etrics textbooks (for exam ple, H am ilton 1994, pp. 46-47), in w hich sta tistical properties o f individual sam ple realizations (that is, histogram s of orbits) coincide w ith q = lim ^oo f g. 5. S u n s p o t E q u i l i b r i a In th is section, we study equilibria of our m odel in which prices and quantities respond to s t . We construct tw o equilibria to illustrate the possibilities. T he first, which we call 22 a conventional sunspot equilibrium, uses fi only. This equilibrium is constructed near the determ inistic steady state, n2, which, as noted above, has a continuum o f determ inistic equilibria which converge to it. Our choice of name reflects th at th is type o f equilibrium is standard in the quantitative sunspot literature.16 The second equilibrium considered, which we call a regim e switching sunspot equilibrium, involves stochastically sw itching betw een ft and /„ . Our analysis o f these equilibria focuses on their welfare and business cycle properties. For this analysis, we find it useful to use the business cycle properties o f U .S. data and o f a standard real business cycle model as benchmarks. We conclude this subsection by making som e observations about the econometrics of sunspot m odels in general and by discussing the em pirical plausibility of our model. 5.1. C onventional Sunspot Equilibrium In this equilibrium , s € R is independently distributed over tim e, w ith s = —0.06 and s = 0.06 w ith probability 1 /2 each. These values for s were chosen so that th e equilibrium ’s im plication for the standard deviation of Hodrick-Prescott detrended, logged equilibrium output coincides w ith th e corresponding figure in th e data. G iven any n, next period’s hours worked, n \ is com puted by first drawing s and then solving n' = /,( n ,s ) (5.1) 16 Because a continuum of other nonstochastic equilibria exists near the steady state equilibrium, n2,this equilibrium is said to be indeterminate (Boldrin and Rustichini 1994, p. 327). For a general discussion of the link between indeterminate equilibria and sunspots, see Woodford (1986a). Examples of quantitative analyses that construct sunspot equilibria in the neighborhood of indeterminate equilibria include Benhabib and Farmer (1994,1995), Farmer and Guo (1994,1995), and Gali (1994a,b). 23 where ft is defined in (3.9). We set the initial level of hours worked, no, to n2. R ecall th at n2 is th e higher of th e tw o determ inistic steady states associated w ith th e lower branch, fi. T hat is, o f the two solutions to x = f i ( x , 0), n2 is the larger of the two. To establish th at th is stochastic process for employm ent corresponds to an equilibrium , it is sufficient to verify that the conditions of the characterization result are satisfied. The first condition is satisfied by construction, and the second is satisfied because n (s4) remains w ithin a com pact interval th at is a strict subset o f th e unit interval. That is, let a be the sm aller o f th e tw o values of n th at solve a = /j(a , —0.06), and let b > a be the unique value o f n w ith th e property o = ft(b, —0.06). Here, a and b are 0.0249 and 0.9509 after rounding. We verified th at if a < n < b, then a < vl < b for n' = /j(n , —0.06) and n' = /j(n , 0.06). Thus, prob[ a < n' < b | a < n < 6] = 1. It follows that a < n { s l ) < b for all histories, s ‘, w ith > 0. T he conditions of the characterization result are satisfied, and so we conclude that n(s*) corresponds to an equilibrium. T he first-m om ent properties o f this equilibrium are reported in Table 1. T hey are sim ilar to the corresponding properties o f the U .S. data and o f th e real business cycle m odel. T he second-m om ent properties of this equilibrium (see Table 3) also compare favorably w ith th e corresponding sam ple analogs, at least relative to the performance o f th e real business cycle m odel (see Table 2). In th is context, three observations are worth stressing. F irst, note th e equilibrium 's prediction that consum ption is sm ooth relative to output and that productivity is roughly as volatile as hours worked. In th e latter respect, th e conventional sunspot equilibrium actually conforms more closely to the data than does th e real business 24 cycle m odel. The real business cycle m odel im plies th at productivity is about 65 percent more volatile than hours worked, whereas th e conventional sunspot equilibrium im plies that productivity is about as volatile as hours worked. In th e data, productivity is about 30 percent less volatile than hours worked. Second, hours and productivity are both procyclical in the equilibrium , as they are in the data. The equilibrium’s im plication th at productivity is procylical reflects the increasing returns in the model. Procyclical productivity helps account for the fact th at equilibrium hours worked and consum ption are both procyclical in the m odel. Finally, the model inherits a shortcom ing of standard real business cycle models in overpredicting the correlation between productivity and hours worked. In th e data, this quantity is essentially zero. Some o f these properties can also be seen by exam ining the plots in Figure 7. T hey are graphs o f th e logged and Hodrick-Prescott filtered data from the equilibrium described above." Consum ption is sm ooth and investm ent is volatile in these graphs. In addition, hours worked and productivity are seen to be procyclical. Overall, this sunspot equilibrium compares quite well to th e real business cycle m odel in its ability to m im ic key features of postwar U .S. business cycles. 5.2. R egim e Sw itching Sunspot Equilibrium For this equilibrium , s — [s (l),s(2 )] 6 R 2, w ith s ( l) € {u, 1} and s(2) = € {— 0.06,0.06}. We set the date 0 value of hours worked to n(s<j) = n 2. We use th e following reclusive procedure to assign a level of employment to each history, s ‘, th at is logically possible, given the specified sq. For any history, s ‘, and associated level of em ploym ent, n (s‘), let n (st+1) be 25 as follows: n (st+ l) = f , t+l(l) (n(s*), s t+ 1(2)) (5.2) for t = 1 ,2 ,.... We construct an equilibrium by devising a sequence o f probabilities, /*(s‘), th at assigns positive probability only to histories, s*, for which a < n (s‘) < 6, for som e a, b such th at 0 < a < b < 1. W hen the value of s ( l) changes along a history, we say there has been a regim e switch. Consider the following probabilities for s t+1(l): 0.9, n 1 < n (s‘) < n 2, n 1 = 0.0370, n2 = 0.9279 { (5.3) 1, otherwise. Let s(2 ) have th e sam e distribution as 5 in th e previous equilibrium . the tw o elem ents of s are independently distributed. We assum e that Let a and b be defined as in the conventional sunspot equilibrium. We verified num erically th at, under these circum stances, if a < n (s‘) < b, then prob[a < n ( s t+l) < 6] = 1. It follows that, for all s l such th at /i(s ‘) > 0, a < n (s‘) < 6.17 This establishes the second of th e two conditions of th e characterization result. To establish th e first condition, note that by (5.2), v (n (s* ),n (st+1)) = st+i(2 ), for all s* (5.4) 17Oux specification of n1 and n2 is crucial for guaranteeing the second condition of the characterization result. For example, with n 1 = a and n2 = b, histories, s', in which hours worked fluctuate between values that approach 0 and 1 occur with high probability. With /i(s‘) specified in this way, the second condition of the characterization result fails. 26 and by construction of the Euler error, s t+i(2 ), 5 3 /i(s ‘+1 | s‘)s t+i(2) = 0, for all s ‘. (5.5) T his establishes that the conditions of the characterization result are satisfied, and we con clude th at n (s‘) corresponds to an equilibrium. We now consider the dynam ic properties of the regime sw itching sunspot equilibrium. First-m om ent properties are reported in Panel C o f Table 1, while second-m om ent properties are reported in Panel B o f Table 3. Regime switching is the key to understanding the dynam ics of this equilibrium. Periodically, the economy switches to the upper branch, / u, where employm ent is very high. The economy typically stays on the upper branch only briefly, and when it switches down again, employment drops to a very low level: near o. Employm ent then rises slow ly until another sw itch occurs, when the econom y jum ps to the upper branch, and the process continues. The fact that the economy spends much tim e in the left region of the lower branch explains why average employm ent in this equilibrium is so low. This also explains why investm ent is, on average, negative. Regarding th e secondmom ent properties, output is substantially more volatile than it is in the data. A lso, output displays very little serial correlation. The positive serial correlation produced by sojourns on th e lower branch is offset by the negative serial correlation associated w ith transient jum ps to th e upper branch. These observations are supported by th e tim e series plots o f th e logged, H odrick-Prescott filtered data from this equilibrium, presented in Figure 8. T he regime switching equilibrium nicely illustrates a type of sunspot equilibrium that 27 is possible. However, in contrast w ith the conventional sunspot equilibrium , th e secondm om ent properties o f th is equilibrium do not match the corresponding quantities in the data. 5.3. E m pirical E valuation o f th e M odel A variety o f other econom etric m ethods can be used to assess th e em pirical plausibility of this m odel.18 One test o f the m odel analyzes the fitted values of the sunspot shocks, s. G iven values for th e m odel param eters, these shocks can be recovered using em ploym ent data.19 For this test, we used th e data on per capita, quarterly hours worked covering the period 1955Q3-1984Q 1 studied in Christiano (1988) and Christiano and Eichenbaum (1992). T he data, shown in Figure 9a, were converted into fractions o f available tim e worked under the assum ption th at households’ available tim e is 15 hours per day (1,369 hours per quarter). T he fitted values of s im ply that all quarterly U .S. observations on hours worked lie on //, th at is, 3(1) = l throughout th e sam ple. T his complements th e findings of the second-m om ent analysis reported above, which indicates that—w ithin the confines of th is m odel—regim e sw itching does not improve our understanding of the aggregate data. T he tim e series on th e fitted Euler equation error, 3(2), are shown in Figure 9c. T he m odel requires th at th is shock satisfy (5.5). A ll dynamic m odels have at least one orthogonality 18For a formal statistical approach to the moment comparison strategy for testing undertaken in the previ ous subsection, see the method based on the work ofHansen (1982) developed in Christiano and Eichenbaum (1992). This analysis integrates parameter uncertainty into evaluations of the “distance” between model and data second moments. ,9For any two consecutive observations on employment, n and n', s(2 ) = v(n, n1). Then, given s(2 ), one finds the two values of i, ij < iu such that s(2) = v(n, x). Ifn' = xi, then s(l) = l, and s(l) = u otherwise. 28 condition like this. Generalized m ethod of moments (GMM) procedures for testin g it have been developed and applied extensively, beginning w ith the work o f Hansen (1982) and Hansen and Singleton (1982). These tests focus on a model’s im plication th at date t + 1 Euler errors be orthogonal to all inform ation available at date £, including a constant. The evidence in Figure 9c indicates that this test fails: the sam ple mean of the fitted Euler error is significantly negative ( —0.27), indicating that a nonzero constant is useful for predicting this variable. The second-m om ent properties of the fitted values o f 5(2) are more consistent w ith the theory. Figure 9b shows the scatter plot of consecutive values of fitted Euler errors, and it suggests that the first-order autocorrelation of 5(2) is not significantly different from zero. (T he point estim ate is —0.18, w ith standard error 0.0920.) Figure 9c shows th e scatter plot of the empirical measure of hours worked at date t, n t , against th e date t + 1 fitted Euler error, 5e+1(2). Here too, the evidence does not imply a strong relationship between th ese variables. T he point estim ate of the correlation between these two variables is —0.20, slightly more than tw ice the standard error of 0.09. These results are subject to tw o caveats. First, they do not take into account sampling uncertainty in th e estim ated values of a and S. However, th e results in Christiano and Eichenbaum (1992) suggest th at th is is very sm all and unlikely to change the results. Second, they are based on arbitrarily settin g a = 2. A conventional GMM approach to this would select a value for <r to ensure th at sam ple analogs of the population orthogonality properties of the Euler errors are satisfied. For exam ple, the value of a th at sets the sam ple average of the fitted Euler errors to zero is a = 2.72. Apart 20The standard error is l/\/115, where 115 is the number of observations in the sample. 29 from changing th e sam ple mean, this change in the value of a does not alter the properties of the Euler errors reported above. We conclude th at there is little evidence in the Euler errors against th e m odel and that the upper branch appears not to be operative in the data. Interestingly, conditional on ruling out the upper branch, / u, th e m odel im plies a re duced form relation very much like the one found in standard m odels driven by fundam ental shocks. For exam ple, equation (5.1) closely resembles th e equilibrium relation for employ m ent im plied by the general equilibrium model analyzed in Sargent (1979, p. 377). T he only qualitative difference is that in th e latter, the shock variable, a, is a com bination of distur bances to preferences and technology, while here it is a sunspot shock, s(2 ). An im plication is th at th e m odel can be estim ated and tested using the sam e maximum likelihood strategies pursued in A ltug (1989), Christiano (1988), Hall (1996), and M cGrattan, Rogerson, and Wright (1996). This observation is consistent w ith the notion that sunspot m odels sim ply offer a new source o f shocks. From an econometric perspective, they are not qualitatively different from m odels w ith fundam ental shocks. Our final test of th e m odel focuses on its im plications for the aggregate production technology, (2.11). To assess the plausibility of this form ulation, we plot th e log o f output per unit o f capital versus th e log of per capita horns worked in Figure 10. (See th e observations marked *.) T he output and capital stock data used are the quarterly data covering th e period 1955Q 3-1984Q 1 studied in Christiano (1988) and Christiano and Eichenbaum (1992). T he data do indicate a positive relation, but it is not as strong as the one im plied by th e m odel, in 30 which l o g ( y / k ) = 2 log(n). One way to assess th e plausibility of the m odel is to compare this line w ith th e least squares line fitted through the U .S. data points. A factor com plicating the comparison w ith this data is th at the m odel does not contain a theory o f th e error term in this relation—clearly one is needed, given the wide dispersion of the U .S. data points. Now, suppose the errors reflect technology shocks, which could easily be incorporated into the analysis. Then, assum ing equilibrium labor responds positively to technology shocks, a standard sim ultaneity bias argument im plies that the slope of th e least squares line is biased upward, as an estim ate of the power on hours worked in the aggregate production function. (See Klenow 1992.) Thus, conditional on interpreting the dispersion o f data points in Figure 10 as reflecting the effects of technology shocks, we conclude that the data in th at figure constitute a rejection of the very high power on hours worked in the production function of this paper.21 To summarize the results so far, the analysis in this section shows that— at least in the context o f our model— it is far from true that “anything goes” em pirically w ith sunspot m odels. T he m odels can be tested using standard econometric m ethods— GMM procedures for comparing sam ple and m odel-based second mom ents, GMM procedures for testin g Euler equations, and standard maximum likelihood procedures. A lthough th e conventional sunspot equilibrium does a surprisingly good job of accounting for business cycle phenom ena, in the end, its strong increasing returns assum ption is rejected by the data.22 2 1 This complements findings in Burnside, Eichenbaum, and Rebelo (1995) and in the references they cite. For an attempt to extend our analysis to a version of the model with a smaller externality, see Guo and Lansing (1996). 22 31 5.4. W elfare A n alysis We approxim ated th e expected discounted u tility for our equilibria using a M onte Carlo sim ulation m ethod. For th e conventional sunspot equilibrium and th e regim e sw itching sunspot equilibrium , th e expected present discounted utilities axe —378.21 (0.24) and —570.58 (1.77), respectively (numbers in parentheses are M onte Carlo standard errors).23 To understand the im pact on u tility o f variance in the Euler error, s(2 ), we also com puted expected u tility for a high variance version o f our conventional sunspot equilibrium . In this case, s(2) € { —0.55,0.55}. T he expected present value of u tility for this equilibrium is —363.35 (2.14). T he present discounted level of u tility associated w ith the constant em ploym ent determ inistic equilibrium at ri2 is —378.49. We refer to this equilibrium as th e con stan t em ploym en t equilibrium. To compare these welfare numbers, we converted them to consum ption equivalents. T hat is, we com puted the constant percentage increase in consum ption required in th e constant em ploym ent equilibrium to make a household indifferent between th at equilibrium and an other given equilibrium. T he results are shown in Table 4. They indicate th at going from the constant em ploym ent equilibrium to the regime sw itching sunspot equilibrium is equivalent to a 289 percent perm anent drop in consum ption. Going to the conventional sunspot equi librium is equivalent to a 0.9 percent permanent rise in consum ption, and going to th e high 23 For each equilibrium, we drew 1,000 histories, s‘,each truncated to be of length 2,500 observations. Subject to the initial level of employment being n2 always, we computed consumption and employment along each history. For each equilibrium, we computed 1,000 present discounted values of utility, vi,...,vioooOur Monte Carlo estimate of expected present discounted utility, v, is the sample average of these: v = iwo Ei^° vi- The fact that we use a finite number of replications implies that v is approximately normally distributed with mean v and standard deviation <r</v/l,0 0 0 ,where <7 < isestimated by the standard deviation of vi,viooo* W e refer to 0i/\/1,000 as the Monte Carlo standard error. 32 variance version o f th at equilibrium is equivalent to an 11.2 percent rise in consum ption. An interesting feature o f these results is th at, despite concavity in th e u tility function, increasing volatility in s(2) raises welfare. This reflects a trade-off betw een two factors. First, other things being the sam e, a concave utility function im plies th at a sunspot equilib rium is welfare-inferior to a constant, determ inistic equilibrium (concavity effect). However, other things are not the same. T he increasing returns means that by bunching hard work, consum ption can be increased on average w ithout raising the average level of em ploym ent (bunching effect). W hen the volatility of the model economy w ith initial em ploym ent n2 is increased by raising the volatility of s(2 ), then the bunching effect dom inates th e concavity effect. W hen volatility is instead increased by allowing regime switches, then th e concavity effect dom inates. In interpreting these results, it is im portant to recognize th at they say nothing about th e nature of the efficient allocations. A ll of the equilibria th at we consider are inefficient, because o f th e presence of the externality in production. 6. P o l i c y A n a l y s i s We now consider the im pact of various policies on the set of equilibria. We consider two countercyclical tax policies th at reduce the set of interior equilibria to a singleton in that output is a constant. We refer to the first as a pure stabilizer because it does not distort margins in equilibrium. T he second tax policy introduces just the right distortions so that the equilibrium supports the optim al allocations. We show that, for a tax policy to isolate the efficient allocations as a unique equilibrium, it is necessary that the tax rate vary in the 33 right way w ith th e state o f the economy. For exam ple, under a constant tax rate policy, th e equilibrium is not unique. Interestingly, th e equilibria are isolated in this case, so th at th ey would escape detection under the usual procedure of analyzing local equilibria. 6.1. A P ure A u tom atic Stabilizer In this section, we display a particular procyclical tax rate rule which reduces th e set o f equilibria to a singleton w ith a* = n2 for all t (the constant em ploym ent equilibrium ). T he tax policy has th e property th at in equilibrium, the tax rate is always zero and thus does not distort any margins. Given our previous results for the constant em ploym ent equilibrium , this tax rate rule improves welfare relative to the regime sw itching sunspot equilibrium , but actually reduces welfare relative to the conventional sunspot equilibrium . T he possib ility th at stabilization of a sunspot by government policy m ight reduce welfare should not be surprising, given that both the sunspot equilibrium and the n2 equilibrium are inefficient. Consider th e following tax rate: r(n ) = 1 - — n (6.1) where n denotes econom ywide average employment and n2 is th e higher o f th e tw o n on sto chastic steady state em ploym ent levels. (See Figure lb .) N ote th at this tax rate is zero w hen aggregate em ploym ent is n2. It turns positive for higher levels of em ploym ent and negative for lower levels. Let v ( n , n ' ) denote (3.6) after substituting out for r(n ) from (6.1). It is easily verified 34 th a t, fo r each v alu e o f n , th e re is a t m o st one n ' th a t solves v ( n , n ') = 0. T h is is g iv en b y n' = /( n ) = n2 - AT(n)(l - 6) n2[1 -f aif(n )] where K (n ) = -^r(l a [ ti) n), A(n) = n2+ 1 - 6 - ^n2(l - n). a T he function, / , and its derivative, / ' , have the property that at n = 1, /(1 ) = 1, / '( ! ) = 0 a n 2 + 1 —6 2 - 6 < 0 since a n 2 < 1. Figure 11 shows / under our baseline parameter values. For convenience, th e two branches o f v = 0, f u and //, are also displayed. There are three things worth em phasizing about / . First, it cuts the 45-degree line from below at n = n2, and it intersects the horizontal axis at a positive level of em ploym ent. T his im plies th at there is no infinite sequence, nt, t = 0 ,1 ,2 ,..., w ith no < n 2 and n t — f { n t - 1 ), such th at > 0 for all t. Since satisfaction of th e Euler equation, v = 0, is a necessary condition for an interior solution to the household problem, it follows that there is no interior equilibrium w ith no < n2. Second, a sequence of employm ents, nt, t = 0 ,1 ,..., which has th e property nt = /(n t_ i) and no > n2, has the property nt —►1 as £ —►oo. Appealing again to th e necessity of the Euler equation, we conclude th at there is no interior equilibrium w ith no > n 2. Third, nt = n2 for all £ satisfies the Euler and transversality conditions and so corresponds to an interior equilibrium. Thus, th e only determ inistic interior equilibrium is 35 the one that corresponds to n t = n2 for t = 0 ,1 ,.... That sunspot equilibria are also ruled out follows from the fact that the Euler equation cuts the 45-degree line from below and from the arguments in Woodford (1986a). These remarks establish P rop osition 6.1. For t h e b a selin e p a r a m e te r iz a tio n a n d u n d e r th e t a x p o lic y in (6 .1 ), th e r e is a u n iq u e in te r io r e q u ilib r iu m w ith n t = n 2 fo r a ll t. Note that under the tax rate policy considered here, rt = 0 in equilibrium. Evidently, the mere threat to change tax rates is enough to rule out other equilibria. This feature of fiscal (and monetary) policies designed to select certain equilibria is common in models with multiple expectational equilibria. (See, for example, Boldrin 1992, p. 215 and Guesnerie and Woodford 1992, p. 380-382.) 6.2. O ptim al A llocation s The efficient allocations correspond to a fictitious planner’s choice of investment, employ ment, and consumption to maximize discounted utility subject to the resource constraint. We reproduce the utility function here for convenience: ° t=0 ** togl1 - n (3*)]}- (6-2) The resource constraint is c(s*) + fc(s‘) - (1 - S W s * - 1) < k i s '- 'f t n i s * ) ) 2, for all t, s*. 36 (6.3) T h is p ro b lem sim plifies g reatly . T h u s, u sin g th e ch ange o f v aria b le in (3.2) a n d th e id e n tity OO 5ZY .P t=0 st 1 (6.4) ^ s‘) !<>**(»'') |lO g ko + /? £ p W P loS *(«*) | the objective function can be written (6.5) In (6.5), consumption has been substituted out using the (scaled) resource constraint after replacing the weak inequality in (6.3) by a strict equality. Notice that the objective in (6.5) is separable across dates and states. This has two implications. First, unsurprisingly, the efficient allocations are insensitive to sunspots. Second, the efficient levels of employment and capital accumulation do not exhibit cycles. It is trivially verified that this result is independent of the curvature on leisure in the utility function, the degree of nonconvexity on labor in the production function, and the degree of homogeneity on capital in the resource constraint.24 Thus, for example, increasing the gains from bunching production, by raising the power on labor above 2, and reducing the associated costs, by making utility linear in leisure, still does not imply that the efficient allocations exhibit cycles. 24 Lack of cycling in the efficient allocations also obtains for utility functions which are homogeneous of degree 7 ^ 0 in consumption. See the Appendix for further discussion. 37 With our specification of preferences, optimizing (6.5) requires that the planner maximize, for each t , s*, log[n2 + 1 - 6 —A] + 0 log A + <xlog[l —n] (6.6) 1 - 0 by choice of n and A, subject to 0 < A < n 2 + l —6, 0 < n < 1. (6.7) The objective, (6.6), is not concave, because of the nonconcavity in the production function. However, for fixed n, (6.6) is strictly concave in A, and its optimal value is readily determined to be A = 0(n 2 + 1 —6). Substituting this into (6.6), the criterion maximized by the efficient allocations becomes Y ~ g loS(n2 + 1 - 6) + & log(l - n) (6.8) after constant terms are ignored. The constraint on this problem is 0 < n < 1. There are two values of n that set the first-order condition associated with maximizing (6.8), and the larger of the two is the global optimum. This is given by n°, where n 2 2 + o { \ — /3 )’ . <r(l - /?)(! - 6) 2 + a ( l —($) (6.9) With the baseline parameter values, n° = 0.98, which implies that the optimal value of A is 1.94, or 94 percent per quarter. The fact that equilibrium employment is so high reflects the 38 fact th at th e efficient allocations internalize the externality in the production function. It is easily verified th at th e tax rate which supports n° as an equilibrium is r = —2. It is not surprising th at this involves a subsidy, since the tax m ust in effect coax individuals into internalizing the positive externality associated w ith production. Consider first th e case in which th e tax rate is sim ply fixed at r = —2 for every n. Let v (n, n') denote (3.6) after substituting out for r = —2. In effect, reducing r from zero to -2 pushes the saddle in Figure la down, so th at the w = 0 plane now covers the seat of the saddle. T he consequences can be seen in Figure 12a, which displays the values of n! that solve v(n, n') = 0 for n € (0 ,1 ). N ote the region of values for n for which there are no values of n' that solve v(n , n') = 0. In the other regions, there are generally two values of n' th at solve this equation for each n. Interestingly, th e unique intersection of these points w ith th e 45-degree line, at n°, is associated w ith a slope greater than one. A s a result, the equilibrium associated w ith n ° ,n ° ,n 0, ... is determ inate. However, there is at least one other equilibrium, fi,n 0,n 0, .... (See Figure 12a for n .) Evidently, the constant tax rate policy does not guarantee a unique equilibrium. One way to construct a tax regime that selects only the desirable equilibrium follows the strategy taken in the previous subsection. Thus, consider Evidently, w ith this policy, r(n °) = —2, so that there is an equilibrium associated w ith th is tax policy which supports the efficient allocations. Also, it is easily verified that— following 39 the sam e reasoning as in th e preceding subsection—th e Euler equation has only one branch. In addition, we found for th e baseline parameter values th at th is branch is m onotone, and it cuts th e 45-degree line from below. It follows by th e logic leading to Proposition 6.1 th at there is a unique interior equilibrium. 7. C o n c l u s i o n a n d D i r e c t i o n s for F u t u r e R e s e a r c h We have displayed a m odel environment which rationalizes im plem enting a tax regim e which is procyclical in the sense that if aggregate employment were to rise, th e governm ent stands com m itted to raise the tax rate. In this sense, th e environment seem s to rationalize th e im portance assigned by m acroeconom ists before the 1970s to devising autom atic stabilizer tax system s.25 B ut, since a properly constructed tax regime elim inates fluctuating equilibria, the actual tax rate is constant. We expect this basic result to survive in versions of our m odel w ith fundam ental shocks. Thus, if there were technology shocks, we conjecture th at the optim al tax rate regim e would move procyclically w ith sunspot shocks, but would not vary w ith technology shocks. Assum ing an efficient tax regime elim inates sunspot equilibria, the optim al “autom atic stabilizer” tax rate would then no t be procyclical in equilibrium . An interesting question for future research would be to investigate w hat happens when th e tax regim e cannot respond differently to fluctuations due to sunspots and to fluctuations due to technology shocks. Possibly, under these circum stances an efficiently constructed ta x regim e w ould exhibit procyclical behavior in equilibrium. Another interesting question for 25 For a discussion of “automatic stabilizers,” see Christiano (1984). 40 future research would explore the robustness of our result th at a properly constructed tax regim e necessarily stabilizes fluctuations. We have shown th at th is is so under a particular hom ogeneity assum ption on the resource constraint. B ut, standard m odels do not satisfy this condition. 41 Table 1: First-M om ent Properties n k/y c/y i/y growth in k growth in y Panel A: U .S. D ata 0.23 10.62 0.73 0.27 1.0047 1.0040 Panel C: Real Business C ycle M odel 0.23 10.64 0.73 0.27 1.0040 1.0040 Panel B: Conventional Sunspot 0.309 0.745 10.46 0.255 1.0045 1.0046 Panel D: Regime Switching Sunspot 0.094 5.17 298 -4 .1 7 0.989 4.74 Note: Entries in the table axe the mean of the indicated variable. U .S. data results are taken from Christiano (1988). Statistics based on m odel econom ies are com puted using 100 artificial data sets of length 114 each. Entries in th e table are an average of 100. 42 Table 2: Second-M oment Properties xt Correlation of y t w ith x t+T Ox/Oy r = 2 r = 1 T = 0 r = —1 T = —2 Panel A: U .S. D ata y 0.02 0.65 0.86 1.00 0.86 0.65 c 0.46 0.48 0.66 0.78 0.76 0.61 i 2.91 0.33 0.56 0.71 0.68 0.57 n 0.82 0.69 0.81 0.82 0.66 0.41 y /n 0.58 0.12 0.32 0.55 0.55 0.53 y/n,n 0.70 -0 .1 7 -0 .0 7 -0 .0 3 0.21 0.33 Panel B: Standard Real Business Cycle y 0.02 0.51 0.74 1.00 0.74 0.51 c 0.55 0.59 0.78 0.98 0.69 0.44 i 2.37 0.45 0.70 0.99 0.76 0.55 n 0.38 0.40 0.67 0.98 0.77 0.57 y/n 0.63 0.57 0.78 0.99 0.71 0.47 y/n,n 1.65 0.61 0.77 0.94 0.61 0.33 Note: R esults axe taken from Christiano and Todd (1996, tables 2 and 3). Panel B results are based on 2,000 artificial observations sim ulated from a standard real business cycle m odel. In both panels, prior to com puting statistics, data were logged and HodrickPrescott filtered. T he m odel corresponds to the one in this paper, w ith a = 3.92, 7 = 0, 6 = 0.021, a = 0.344, and a production function has the form Y = K a (zn)^l ~a\ w ith z — z_iexp(A ), and A ~ 7/iV (0.004,0.0182). The last two rows o f each panel report the standard deviation of productivity (y /n ) relative to that of hours (n ). T he correlations reported there are ccrrr[(y/n)t , nt_T]. 43 Table 3: Second-M oment Properties, Sunspot Equilibria xt II r = 2 H-* Correlation o f y t w ith x t+T T = 0 T = —1 T = -2 Panel A: Conventional Sunspot y 0.02 0.35 0.63 1.00 0.63 0.35 c 0.33 0.58 0.72 0.87 0.44 0.13 i 3.13 0.26 0.57 0.99 0.66 0.40 71 0.51 0.22 0.54 0.98 0.66 0.42 y /n 0.52 0.46 0.69 0.98 0.57 0.27 y/n,n 1.02 0.49 0.68 0.91 0.44 0.11 Panel B: Regime Switching Sunspot y 0.78 -0 .0 7 -0 .0 7 1.00 -0 .0 7 -0 .0 7 c 0.06 0.25 0.30 0.35 -0 ,4 2 -0 .3 5 i na na na na na na n 0.54 -0 .1 1 0.11 0.99 -0 .0 1 -0 .0 3 y/n 0.47 -0 .0 3 -0 .0 2 0.99 -0 .1 3 -0 .1 2 y/n,n 0.88 0.01 0.02 0.96 -0 .1 9 -0 .1 7 Note: M odel statistics axe com puted in th e sam e wav as for Panel B in Table 2. T he notation “na” appears in the investm ent colum n of Panel B, because gross investm ent is often negative. 44 Conventional Sunspot I 0.9% Conventional Sunspot II Regime Switching 11.2% -289% Note: This is the constant percentage decrease in consumption required for households in the indicated equihbrium to be indifferent between that equilibrium and the constant equilibrium at n = n2. Let v denote the discounted utility associated with the constant employment level. Let v denote the discounted utility associated with one of the other equilibria. Then, the number in the table is 100[exp((l —0 ) ( v — v ) ) —1]. 45 A . A p p e n d i x : Linearity of Policy R u l e s U n d e r H o m o g e n e i t y In this appendix, we establish efficiency for a policy of the form, fct+i = A*kt and n* = n*, where A*,n* are fixed numbers. We do this for a class of economies in which the resource constraint is homogeneous in capital and in which preferences are homogeneous in consump tion. Our result parallels that in Alvarez and Stokey (1995), except their environment does not explicitly allow for variable hours worked. Consider the following planning problem: max V j9‘«(c(1nt) ,0 < / ? < 1 (A.1) t_ o subject to the following feasibility constraints: fco > 0 is given, 0 < c* < F(fct,k t + lf nt), 0 < n* < 1, k t +1 > 0, for £ = 0 ,1 ,2,.... We assume that F is homogeneous: F ( k , k', n ) = n)> where /(A , n) = F ( 1, A, n), A = xj;> 0. (A.2) In terms of A and n, the constraints on the planner axe: B = {A,n : 0 < n < 1, 0 < A, and /(A ,n ) > 0}. That is, the planner’s feasible set is the set of infinite sequences, {At, Xt , n t e B for each t > 0. We place the following assumptions on / : / : B —*■R , continuous, decreasing in A, and increasing in n. such that (A.3) Also, there exists a largest value of A, A > 0, such that /(A, 1) > 0 *pr<i. (A.4) 46 and there exists 0 < h < We place th e following assum ptions on u(c, n) = <Pg{n)/ 7 , 7 ^ 0 1 such th at / ( 1 , n) > 0 . (A .5) u: , g (n ) > 0 , g is continuous and decreasing. (A .6 ) We have the following proposition: P r o p o s itio n A .l . I f (i) th e functions F an d u s a tisfy (A .2)-(A .3), an d (A .6 ), (ii) (A .4 ) holds when 7 > 0 , a n d (A .5) h olds when 7 < 0 . then, a p o lic y o f th e follow ing form solves (A .l): kt+i = A*fct, n t = n * , t > 0, for fixed (n*, A*) G B . P r o o f . W rite u {c,n ) = ( f ( X , n ) ) y g ( n ) / q f . A lso, kT" * = 1,2,. = Sim ple su b stitu tio n establish es v(ko) = max = k'tf’w {*i+i,nt}£o t=0 where: 00 w (/(A t,rtt))7 ff(ut) max {(At,nt)€B)t«0 t=0 (A .7) 7 We establish —00 < w < 0 0 . W hen 7 < 0, then u is bou n ded a b o ve b y zero an d so trivially, w < 0 0 . For th e case 7 > 0 , consider th e (infeasible!) p o lic y o f a p p lyin g th e en tire tim e en dow m en t b o th to la b o r effort and to leisure, an d o f a p p ly in g all o f o u tp u t b o th to con su m p tio n and to in vestm en t. T h e value o f th is p o lic y is w = (/(0, l ))7 <7(0) / £7 ( 1 —/JA7^ ) ! . We h ave w < 0 0 , since w < w < 0 0 . To establish — 0 0 < w when 7 > 0 n o te s im p ly th a t 47 u is b o u n d ed b elo w b y zero in th is case. For th e case 7 < 0, n o te th a t th e feasible policy, At = 1 , n* = n, for t > 0 has retu rn k ^ w , w here w = f ( \ , h ) ' 1g ( n ) / [7 ( 1 —/?)], so th a t — 0 0 < U) < w . W e h a ve esta b lish ed th a t w is a fin ite scalar. B y w ritin g (A .7 ) o u t explicitly, on e verifies th a t w satisfies th e follow ing expression: w = (Ao,no)€5 {^(/(* o ." o )) 7 * (" o) h + 0 (Ao) * 1 J • (A .8 ) L e t A* a n d n* d en o te values o f Ao an d n 0 th a t so lve th e above m a x im iza tio n p ro b lem . T h e resu lt follow s from th e fa ct th a t th ese so lve a p ro b lem in which th e o b je ctiv es an d co n strain ts are in d ep en d en t o f ko. Q .E .D . R em a rk 1 . T h e p r o o f for th e class o f u tility fu n ction s u(c, n) = log(c) + g(n ) is a trivia l p e rtu rb a tio n on th e argum ent in th e tex t. R em a rk 2 . When 7 > 0 , then th e fixed p o in t p ro b lem in (A .8 ) can b e show n to be th e fixed p o in t o f a contraction m apping. In th is case, w in (A .7) is th e o n ly solu tion to (A .8), an d th e con traction m a p p in g th eorem p ro vid es an ite r a tiv e algorithm for co m p u tin g w , A*, and n*. W hen 7 < 0, th e m a p p in g im p lic itly defined in (A .8 ) is n o t necessarily a contraction. A lv a re z a n d S to k e y (1995) s u p p ly ite ra tiv e schem es for co m p u tin g w in th is case. 48 References [1 ] A ltug, Sumru, 1989, “Tim e to Build and Aggregate Fluctuations: Some New Evidence,” I n t e r n a t io n a l E c o n o m ic R e v ie w 30, pp. 889-920. [2 ] Alvarez, Fernando, and Nancy L. Stokey, 1995, “Dynam ic Programming w ith Homoge neous Functions,” manuscript, University of Chicago. [3] Azariadis, C ostas, 1981, “Self-Fulfilling Prophesies,” pp. 380-396. J o u r n a l o f E c o n o m ic T h e o ry 25, [4] Baxter, Marianne, and Robert King, 1991, “Productive E xternalities and Business Cy cles,” D iscussion Paper 53, Institute for Empirical M acroeconomics, Federal Reserve Bank of M inneapolis, November. [5] Beaudry, Paul, and Michael B. Devereux, 1994, “M onopolistic C om petition, Price Set tin g and th e Effects of Real and M onetary Shocks,” manuscript, U niversity o f British Columbia. [6 ] Benhabib, Jess, and Roger E. A. Farmer, 1994, “Indeterminacy and Growth,” o f E c o n o m ic T h e o r y 63, pp. 19-41. Jo u rn a l [7] Benhabib, Jess, and Roger E. A. Farmer, 1995, “Indeterminacy and Sector-Specific E xternalities,” manuscript, departm ent of economics, UCLA, May. [8 ] Benhabib, Jess, and Roger E. A. Farmer, 1996, “The M onetary Transm ission Mecha nism ,” manuscript. [9] Benhabib, Jess, and Roberto Perli, 1994, “Uniqueness and Indeterminacy: On the Dynam ics of Endogenous Growth,” J o u r n a l o f E c o n o m ic T h e o r y 63, pp. 113-142. [10] Benhabib, Jess, and Aldo Rustichini, 1994, “Introduction to the Sym posium on Growth, Fluctuations, and Sunspots: Confronting the D ata,” J o u r n a l o f E c o n o m ic T h e o r y 63, pp. 1-18. [1 1 ] Boldrin, M ichele, 1992, “Dynamic E xternalities, M ultiple Equilibria, and Growth,” J o u r n a l o f E c o n o m ic T h e o r y 58, no. 2, December. [1 2 ] Boldrin, M ichele, Nobuhiro K iyotaki, and Randall W right, 1993, “A Dynam ic Equilib rium M odel of Search, Production and Exchange,” J o u r n a l o f E c o n o m ic D y n a m ic s a n d C o n tro l [13] Boldrin, M ichele, and Aldo Rustichini, 1994, “Growth and Indeterm inacy in Dynam ic M odels w ith Externalities,” E c o n o m e t r ic a 62, no. 2 . [14] Boldrin, M ichele, and Michael W oodford, 1990, “Equilibrium M odels D isplaying En dogenous Fluctuations and Chaos,” J o u r n a l o f M o n e t a r y E c o n o m ic s 25, pp. 189-222. 49 [15] B ryant, John, 1981, “Bank Collapse and Depression,” B a n k in g 13, pp. 454-464. Jo u rn a l o f M o n ey, C r e d it a n d [16] B ryant, John, 1983, “A Sim ple R ational E xpectations K eynes-Type M odel,” J o u r n a l o f E c o n o m ic s 98, no. 3. Q u a r t e r ly [17] Burnside, Craig, and M artin Eichenbaum, 1995, “Factor Hoarding and th e Propagation o f B usiness C ycle Shocks,” Working Paper 4675, N ational Bureau o f Econom ic Research. [18] Burnside, Craig, M artin Eichenbaum, and Sergio Rebelo, 1995, “C apacity U tilization and Returns to Scale,” in Ben Bernanke and Julio Rotemberg, editors, M a c r o e c o n o m ic s Annual [19] C ass, D avid, and Karl Shell, 1983, “Do Sunspots M atter?” 91, pp. 193-227. J o u r n a l o f P o lit ic a l E c o n o m y [20] C hristiano, Lawrence J., 1984, “A Reexam ination of the Theory of A utom atic Stabiliz ers,” C a r n e g ie - R o c h e s t e r C o n fe r e n c e S e r ie s o n P u b lic P o lic y 20. [21] C hristiano, Lawrence J., 1988, “W hy D oes Inventory Investm ent F luctuate So Much?” J o u r n a l o f M o n e t a r y E c o n o m ic s . [2 2 ] C hristiano, Lawrence J., and M artin Eichenbaum, 1992, “Current R eal-business-cycle Theories and A ggregate Labor Market Fluctuations,” A m e r ic a n E c o n o m ic R e v ie w 82, pp. 430-450. [23] C hristiano, Lawrence J., and Richard Todd, 1996, “Tim e to Plan and A ggregate Fluc tu ations,” F e d e r a l R e s e r v e B a n k o f M in n e a p o lis Q u a r t e r ly R e v ie w 20, pp. 14-27. [24] Cochrane, John, 1995, “Shocks,” Cam egie-Rochester Conference on Public Policy. [25] Cogley, Tim othy, and James Nason, 1995, “O utput Dynam ics in R eal-Business-C ycle M odels,” A m e r ic a n E c o n o m ic R e v ie w 3, pp. 492-511. [26] Cole, Harold L., and Tim othy Kehoe, 1996, “Self-Fulfilling D ebt Crises,” Research D epartm ent Staff Report 210, Federal Reserve Bank o f M inneapolis. [27] Cooper, R ussell, and Andrew John, 1988, “Coordinating Coordination Failures in Key nesian M odels,” Q u a r t e r ly J o u r n a l o f E c o n o m ic s 103, A ugust, pp. 441-463. [28] D agsvik, John, and Boyan Jovanovic, 1994, “Was th e Great D epression a Low-Level Equilibrium ,” E u r o p e a n E c o n o m ic R e v ie w , December. [29] Devaney, Robert L., 1989, A n I n t r o d u c t io n to C h a o t ic tion , Addison-W esley: Redwood City, California. D y n a m ic a l S y s t e m s , second edi [30] D iam ond, D ouglas, and Philip Dybvig, 1983, “Bank Runs, D eposit Insurance, and Liquidity,” J o u r n a l o f P o lit ic a l E c o n o m y 91, no. 3, pp. 401-419. 50 [31] Diam ond, Peter, 1982, “Aggregate-Demand M anagement in Search Equilibrium ,” n a l o f P o lit ic a l E c o n o m y 90, no. 5, pp. 881-894. J o u r [32] D om ow itz, Ian, and Mahmoud A. El-Gam al, 1993, “A C onsistent Test of Stationary Ergodicity,” E c o n o m e t r ic T h e o r y 9, 589-601. [33] Dom ow itz, Ian, and Mahmoud A. El-Gamal, 1994, “Is Your M odel Chaotic? A Con sisten t Sim ulation-Based T est,” in V. Lakshmikantham and W . de Gruyten, editors, P r o c e e d in g s o f th e F i r s t W o r ld C o n g r e s s o f N o n lin e a r A n a ly s is , W alter de Gruyter Pub lishing Company. [34] El-G am al, Mahmoud, 1991, “Non-Paxametric Estim ation of D eterm inistically Chaotic System s,” E c o n o m ic T h e o r y 1, pp. 147-167. [35] Farmer, Roger E. A ., and Guo, J.-T ., 1994, “Real Business C ycles and th e Anim al Spirits H ypothesis,” J o u r n a l o f E c o n o m ic T h e o r y 63, pp. 42-73. [36] Farmer, Roger E. A ., and Guo, J.-T ., 1995, ‘T h e Econom etrics of Indeterminacy: An A pplied Study,” J o u r n a l o f M o n e t a r y E c o n o m ic s . [37] Farmer, Roger E. A ., and M ichael Woodford, 1984, “Self-Fulfilling Prophecies and the B usiness Cycle,” Center for A nalytic Research in Economics and Social Science, Working Paper 84-12, University of Pennsylvania, April. [38] Fischer, Stanley, 1980, “On A ctivist M onetary Policy w ith R ational E xpectations,” in Stanley Fischer, editor, R a t io n a l E x p e c t a t io n s a n d E c o n o m ic P o lic y , U niversity of Chicago Press. [39] Gali, Jordi, 1994a, “M onopolistic Com petition, Business C ycles, and th e C om position o f Aggregate Dem and,” J o u r n a l o f E c o n o m ic T h e o r y 63, pp. 73-96. [40] Gali, Jordi, 1994b, “M onopolistic Com petition, Endogenous Markups and Growth,” E u r o p e a n E c o n o m ic R e v ie w 38, pp. 748-756. [41] Grandmont, Jean-M ichel, 1986, “Stabilizing C om petitive Business C ycles,” in JeanM ichel Grandmont, editor, N o n lin e a r E c o n o m ic D y n a m ic s , Academ ic Press. [42] Guesnerie, Roger, and M ichael Woodford, 1992, “Endogenous Fluctuations,” in JeanJacques Laffont, editor, A d v a n c e s i n E c o n o m ic T h e o r y , Sixth World Congress, vol. 2 , Cambridge University Press. [43] Guo, Jang-Ting, and K evin Lansing, 1996, “State-C ontingent Fiscal Policy and Sunspot Equilibria,” unpublished manuscript, University of California, Riverside. [44] Hall, George, 1996, “Overtime, Effort and the Propagation of Business Cycle Shocks,” J o u r n a l o f M o n e t a r y E c o n o m ic s , forthcoming. 51 [45] H am ilton, Jam es, 1989, “A New Approach to th e Economic A nalysis o f N onstationaxy T im e Series and th e Business C ycle,” E c o n o m e t r ic a 57, pp. 331-339. [46] H am ilton, Jam es, 1994, N ew Jersey. T im e S e r ie s A n a ly s is , Princeton U niversity Press: Princeton, [47] H ansen, Lars Peter, 1982, “Large Sam ple Properties of Generalized M ethod of M oment M odels,” E c o n o m e t r ic a 50, 1269-1286. [48] H ansen, Lars Peter, and K enneth Singleton, 1982, “Generalized Instrum ental Variables E stim ation o f Nonlinear Rational E xpectations M odels,” E c o n o m e t r ic a 50, pp. 12691288. [49] H ow itt, Peter, and Preston M cAfee, 1988, “Stability of Equilibria w ith E xternalities,” Q u a r t e r ly J o u r n a l o f E c o n o m ic s 103, pp. 261-278. [50] Imrohoroglu, Selahattin, 1993, ‘T estin g for Sunspot Equilibria in the German Hyper inflation,” J o u r n a l o f E c o n o m ic D y n a m ic s a n d C o n t r o l 17, 289-317. [51] Jovanovic, Boyan, 1989, “Observable Im plications of M odels w ith M ultiple Equilibria,” E c o n o m e t r ic a 57, no. 6 , pp. 1431-1437. [52] Klenow, Peter J., 1992, “M ultiple Steady States and Business C ycles,” m anuscript, G raduate School of Business, U niversity of Chicago. [53] Krugman, Paul, 1991, “History Versus E xpectations,” 106, no. 2, pp. 651-667. Q u a r t e r ly J o u r n a l o f E c o n o m ic s [54] K ydland, Finn, and Edward C. Prescott, 1980, “A C om petitive Theory of Fluctuations and th e Feasibility and D esirability of Stabilization Policy,” in Stanley Fischer, editor, R a t io n a l E x p e c t a t io n s a n d E c o n o m ic P o lic y , University of Chicago Press. [55] L asota, Andrzej, and M ichael C. Mackey, 1985, S y s t e m s , Cambridge U niversity Press. P r o b a b ilis t ic P r o p e r t ie s o f D e t e r m in is t ic [56] M atheny, K enneth J., 1994, “Increasing Returns and M onetary Policy,” m anuscript, Purdue University. [57] M atsuyam a, Kim inori, 1991a, “Increasing Returns, Industrialization, and Indeterm i nacy o f Equilibrium ,” Q u a r t e r ly J o u r n a l o f E c o n o m ic s 106, no. 2 , pp. 617-650. [58] M atsuyam a, Kiminori, 1991b, “Endogenous Price Fluctuations in an O ptim izing M odel o f a M onetary Economy,” E c o n o m e t r ic a 59, pp. 1617-1631. [59] M cG rattan, E llen R ., Richard Rogerson, and Randall Wright, 1996, “A n Equilibrium M odel o f th e Business Cycle W ith Household Production and Fiscal Policy,” I n t e r n a t io n a l E c o n o m ic R e v ie w , forthcoming. 52 [60] M ortensen, D ale T ., 1989, “The Persistence and Indeterm inacy of Unem ploym ent in Search Equilibrium ,” S c a n d in a v ia n J o u r n a l o f E c o n o m ic s 91, pp. 347-370. [61] M ortensen, D ale T ., 1991, “Equilibrium Unemployment C ycles,” unpublished manu script, Northwestern University. [62] Poole, W illiam , 1970, “Optim al Choice of M onetary Policy Instrum ents in a Sim ple Stochastic Macro M odel,” Q u a r t e r ly J o u r n a l o f E c o n o m ic s , May, pp. 197-216. [63] R ebelo, Sergio, 1991, “Long-Run Policy Analysis and Long-Run Growth,” P o lit ic a l E c o n o m y 99, no. 3, pp. 500-521. [64] Romer, Paul, 1986, “Increasing Returns and Long-Run Growth,” E c o n o m y 94, no. 5, pp. 1002-1037. Jo u rn a l o f J o u r n a l o f P o lit ic a l [65] Romer, Paul, 1987, “Growth Based on Increasing Returns Due to Specialization,” ic a n E c o n o m ic R e v ie w P a p e r s a n d P r o c e e d in g s 77, no. 2, May. A m e r [6 6 ] Rotem berg, Julio, and M ichael Woodford, 1996, “Real Business C ycles and th e Fore castable M ovements in O utput, Hours, and Consum ption,” A m e r ic a n E c o n o m ic R e v ie w 8 6 , pp. 71-89. [67] Sargent, Thom as, 1979, M a c r o e c o n o m ic T h e o r y , Academic Press. [6 8 ] Sargent, Thom as, and N eil Wallace, 1987, “Inflation And T he Government Budget C onstraint,” in A. R azin and E. Sadka, editors, E c o n o m ic P o lic y in T h e o r y a n d P r a c t ic e , M acM illan Press: New York. [69] Schm itt-G rohe, Stephanie, 1995, “Comparing Four M odels of Aggregate Fluctuations D ue to Self-Fulfilling E xpectations,” Federal Reserve Board of Governors Finance and Econom ics D iscussion Series 95-17, W ashington, D.C. [70] Schm itt-G rohe, Stephanie, and Martin Uribe, 1996, “Balanced-Budget R ules, D istor tionary Taxes, and Aggregate Instability,” unpublished m anuscript, Federal Reserve Board of Governors, W ashington, D.C. [71] Stokey, Nancy, and Robert E. Lucas, w ith Edward Prescott, 1989, i n E c o n o m ic D y n a m ic s , Harvard University Press. [72] Shleifer, Andrei, 1986, “Implem entation C ycles,” 6 , pp. 1163-1190. R e c u r s iv e M e t h o d s J o u r n a l o f P o lit ic a l E c o n o m y [73] W atson, Mark, 1993, “Measures of Fit for Calibrated M odels,” E c o n o m y 101, no. 6 , pp. 1011-1041. 94, no. J o u r n a l o f P o lit ic a l [74] W oodford, M ichael, 1986a, “Stationary Sunspot Equilibria: The Case of Sm all Fluc tuations Around a D eterm inistic Steady State,” unpublished m anuscript, U niversity of Chicago, September. 53 [75] W oodford, M ichael, 1986b, “Stationary Sunspot Equilibria in a Finance Constrained Economy,” J o u r n a l o f E c o n o m ic T h e o r y 40, no. 1, October. [76] W oodford, M ichael, 1987, “Three Questions A bout Sunspot Equilibria as an Explana tion o f Econom ic Fluctuations,” A m e r ic a n E c o n o m ic R e v ie w P a p e r s a n d P r o c e e d in g s 77, no. 2, May, pp. 93-98. [77] W oodford, M ichael, 1988, “Expectations, Finance, and A ggregate Instability,” in M. Kohn and S-C. Tsiang, editors, F in a n c e C o n s t r a in t s , E x p e c t a t io n s , a n d M a c r o e c o n o m ic s . [78] W oodford, M ichael, 1991, “Self-Fulfilling E xpectations and Fluctuations in A ggregate D em and,” in N. Gregory M ankiw and David Romer, editors, N e w K e y n e s ia n E c o n o m ic s , v o l. 2 , C o o r d in a t io n F a ilu r e s a n d R e a l R ig id it ie s , MIT Press. [79] W oodford, M ichael, 1994, “Determ inancy of Equilibrium Under A lternative Policy Regim es,” E c o n o m ic T h e o r y 4, no. 3, pp. 323-326. 54 Figure la: The v(n ,n') Function Figure lb: Contour: v {n ,n ') =0 Figure 1c: Close-up of Figure 1b n (Hrs. T o d a y ) Note: Figure la is a three-dimensional view of the function v in equation (3.6), computed using the standard param eter values. The dark and light regions identify the parts o f v that are less than and greater than zero, respectively. Figure lb shows the values of n' that set v(n,nr) to zero, given n. H e re ,/ a n d /u denote the low er and upper branch functions defined in (3.9), respectively. Also, n 1 and n2 denote the points w here// crosses the 45-degree line. Figure 1c displays /(/z)-/z from Figure lb for values of n in a neighborhood of the origin. It shows t h a t / first cuts the 45-degree line from below, at n l , and then again from above, at n2. Figure 3: Testing Sensitivity to Initial Conditions Figure 4a: Fraction of Available Time Worked Figure 4b: Figure 4a, continued hours worked Figure 2: Two Equilibria on the Lower Branch occurrences per interval occurrences per Interval occurrences per Interval Figure 5: Histograms of Four Orbits 250.00 -r n (0) s 0.25 200.00 -150.00 100.00 50.00 0.00 ■«yOao — C0i0r-*0**— TT' OooOCMt Or ^ O — e O' Oa p QCN^ ' QO* — com d d d d d o d d d d d d d d d d d d d d d d d d d d d d d n (0) = 0.59 0.00 '^■vOoO'— O LCONri^wOM —c o’ »3 M ij ^O T P ^r O >0 — N COIO > ---CN(N n,cOoafOoO,7C ,« ^ 'i n~ iCnOi*oQfC) O;0Q' C0 M' 0^ < ) '0O* N N d d d d d o d d d d d d d d d d d d d d d d d d d d d d d 250.00 T occurrences per Interval d d d d d o d d d d d d d d d d d d d d d d d d d d d d d 250.00 T ’T c oCvSoCr '^ pI*N' (—0 5e '0C«a<o0 o,?cTNIi'o7r ^^ O T ''O^ —NcoiO ' -' O - c>O - '(— NW ^ l' — 0 lc O' A wQc i oDQCNT '0O'0 0'0 NN d d o d d d d d d d d d d d d d d d d d d d d d d d d d d Note: Four histograms of 50,000 iterates on/, defined in (4.1), using standard parameter values, except 0^=1.25. The iterates are differentiated according to the initial condition on n, as indicated. Figure 6: Density Functions Induced b y / 250.00 T | 200.00 c n (0) in interval (0.16, 0.32) & 150.00 CO © c© 100.00 K 50.00 0.00 ■<T>oO'-comcooCN'«3, >o^ — c o^,,!rCotOiO«OiO'0'0'0<)r^rvf^ maoQCN^r^O'-eooaoocsco — — •— CMCMCNCMcOCOCO«OCO^T^, d d d d d d d d d d d d d d d d d d d d d d d d o d d d TT'0«X> C O i O ^ O ' - ^ ' O O O O C MV f ^!T' <3r O ' —CO' Qc OQCN^ ' QC^ *— — — — cNCM<NCNCMcocoeoeo^r TrO ^T, mu5mu535' 0' 0' OOf ' N. r— ' >.com r'* d d d d d d d d d d d d d d d d d d d d d d d d d d d d o Difference Between the First Tw o Histograms in Figure 6 QCMTT'OO*'— M ri ^O cococococoMc ’ So Ti TO r o^ OT QmCmMm^mrv^OQ'^O---Oc'oOm'«OO^O rC ^ d d d d d d d d d d d d d d d o d Note: The top two graphs show the density of iterate number 3,000 on the map/defined in (4.1), with initial condition drawn uniformly from the indicated interval. F ig u r e 7 : L o g g e d a n d H P F ilt e r e d D a ta fro m C o n v e n t io n a l S u n s p o t E q u ili b r iu m Note: These graphs depict a realization of length 114, when we use our baseline parameterization. Figure 8: Logged, and H P Filtered Data from Regime Switching Sunspot Equilibrium Log HP Filtered y and n Log HP Filtered y and c Log HP Filtered y and y/n Note: These graphs depict a realization of length 114, when we use our baseline parameterization. Figure 9a: U.S. Data, Time Worked Figure 9b: Euler Errors Figure 9c: Scatter Plot Figure 9d: Scatter Plot 1 Euler error (t) * 0.5 * * X* * * , 0 s* * Ig * -0.5 * * * 2 <D v_ * * u . .32 LU * * -0.5 X ’ -1 -0.5 0 0.5 Euler error (t-1) 1 0.22 0.23 0.24 hours worked (t-1) 0.25 log, output per unit of capital Figure 10: Hours Worked Versus Output Per Unit of Capital Figure 11: Euler Equation, v(n,n')=0, for Taxed and Untaxed Economies n-n