Macroeconomic Theory II (G31.

1026) Spring 2010

Heterogeneity in Macroeconomics
Gianluca Violante

General Information
Lecture Times and Location: Monday and Wednesday 9:45-11:45 am, in room 517. The last class is on May 3rd. Oce Hours: Tuesday 11:30-12:30 in my oce, room 712. You can always contact me by phone at extension 29771 or by email <glv2@nyu.edu> to arrange an alternative time. TA and Tutorials: The TA is Shengxing Zhang who can be reached at extension 87578 or by e-mail <shengxing.zhang@nyu.edu>. He sits in room 818. His oce hours are Wednesday 3:30-5:30 pm. Weekly tutorials are held on Friday 9:30-11:30 am in Room 517. Homework: There will be weekly problem sets that are required for a passing grade. You are allowed to cooperate with other students in the class, but every student has to hand in his/her own uniquely written assignment. Examination: There will be a written nal examination on Wednesday May 5th, 9:3011:30 am. The nal exam does not include material in the last week of class (but the core exam will). Course website: The webpage of the course, where announcements, homeworks, lecture notes and additional readings will be posted is http : //www.econ.nyu.edu/user/violante/macrotheory.htm Please, make sure to check it regularly.

Course Summary and Objectives

Summary: This section of the course is devoted to studying economies where agents are heterogeneous. These models are very helpful to analyze questions pertaining to distribution of resources, inequality, and the eects of policies. We will start with some aggregation theorems in complete markets economies where a representative agent still exists. Next, we will move towards economies with incomplete markets where agents can only borrow and save through a risk-free bond. We begin by characterizing in detail the individual problem. Next, we proceed to the description of the stationary equilibrium. Then, we study an incomplete-markets model with aggregate shocks. The last classes are devoted to dening economies where there is default in equilibrium, and economies with heterogeneous rms. 1

 Objectives: The aim of this section of the course is twofold: 1) to become familiar with this important class of macroeconomic models, and 2) to learn how to solve numerically for the equilibrium of these model economies, in order to perform quantitative research.

Reading Material
Textbooks: The main textbook is Recursive Macroeconomic Theory, by Lars Ljungqvist and Tom Sargent, MIT Press, second edition, 2004 (denoted by LS below). You will also use Recursive Methods in Economic Dynamics, by Stokey, Lucas and Prescott, Harvard University Press, 1989. In the recitation of Friday 26th, Shengxing will teach some basic concepts of measure theory from SLP, chapters 7, 8.1,11.1,11.2 and 12.4. We need them from class 6 onwards. Other readings: I will make lecture notes and links to papers available online on my webpage, as we go along.

Course Outline (by class)

1. Heterogeneity in the Neoclassical Growth Model with Complete Markets (LS, 8.5.3) We discuss the assumptions on fundamentals under which, although households are heterogeneous in preferences and endowments, a representative agent exists. And we apply these results to the neoclassical growth model. We discuss quickly (since you have already seen it with Tom) the Negishi method. This methodology allows to calculate the competitive equilibrium prices and allocations of complete markets economies (in particular, economies for which the rst welfare theorem holds) with heterogeneous households. This method proves to be particularly useful for those economies where aggregation does not hold, hence we cannot use the representative agent. We present one illustration of this method based on a paper by Maliar and Maliar. Chatterjee, Satyajit (1994) "Transitional dynamics and the distribution of wealth in a neoclassical growth model", Journal of Public Economics Maliar, Lilia and Serguei Maliar (2001) "Heterogeneity in capital and skills in a neoclassical stochastic growth model", Journal of Economic Dynamics and Control Maliar, Lilia and Serguei Maliar (2003), "The Representative Consumer in the Neoclassical Growth Model with Idiosyncratic Shocks", Review of Economic Dynamics 2. The Income Fluctuation Problem I (LS, 16.1-16.4, 17.13) We discuss the empirical implications of full-insurance for consumption. We review the Permanent Income Hypothesis and we apply it to characterize the consumption-saving problem of a single-agent who faces a stochastic income stream and can trade only a risk-free bond. We introduce the notion of precautionary savings and relate it to the convexity of marginal utility (prudence). 2

 Mace, Barbara (1991); "Full Insurance in the Presence of Aggregate Uncertainty," Journal of Political Economy Cochrane, John (1991); "A Simple Test of Consumption Insurance," Journal of Political Economy Hall, Robert (1978); "Stochastic Implications of the Life Cycle-Permanent Income Hypothesis: Theory and Evidence", Journal of Political Economy Leland, Haynes (1968); "Saving and Uncertainty: the Precautionary Demand for Saving", Quarterly Journal of Economics Sibley, David (1975); "Permanent and Transitory Income Eects in a Model of Optimal Consumption with Wage Income Uncertainty", Journal of Economic Theory (only section VI) Blundell, Richard and Ian Preston (1998); "Consumption Inequality and Income Uncertainty," Quarterly Journal of Economics 3. The Income Fluctuation Problem II (LS, 16.5-16.8, 17.3-17.5) We introduce borrowing constraints and show that precautionary savings can arise even without prudence as long as borrowing constraints may bind in some state of the world. We then derive an important condition on the interest rate that guarantees that the optimal individual consumption sequence is bounded above, in presence of income uncertainty. 4. Numerical Techniques to Solve the Income Fluctuation Problem We present a set of simple numerical techniques to solve for the consumption and saving policy functions in the recursive formulation of the income-uctuation problem for the single-agent who self-insures by saving/borrowing through a risk-free bond. In particular, we study a new very fast numerical method, called endogenous grid method. Tauchen, George (1986); "Finite State Markov Chain Approximations to Univariate and Vector Autoregressions", Economic Letters Suen, Richard and Kopecki, Karen (2010); Finite State Markov-Chain Approximations to Highly Persistent Processes, Review of Economic Dynamics Aruoba B., Fernandez-Villaverde Jesus, and Rubio-Ramirez, Juan (2006); "Comparing Solution Methods for Dynamic Equilibrium Economies", Journal of Economic Dynamics and Control Carroll, Chris (2006). The Method of Endogenous Gridpoints for Solving Dynamic Stochastic Optimization Problems, Economics Letters Barillas, Francisco and Fernandez-Villaverde Jesus (2007). A Generalization of the Endogenous Grid Method, Journal of Economic Dynamics and Control Judd, Ken (1998); Numerical Methods in Economics, MIT Press, chapters 6-10 Marimon, Ramon and Scott, Andrew (1999); "Computational Methods for the Study of Dynamic Economies", Oxford University Press 3

 Heer, Burkhard and Alfred Maubner (2005); "DGE Modelling, Computational Methods and Applications," Springer 5. The Neoclassical Growth Model with Incomplete Markets (Bewley Models) (LS 17.1-17.2, 17.6-17.12) We analyze the equilibrium of a neoclassical growth model populated by a continuum of agents who face idiosyncratic labor income risk and trade only a risk-free asset. We use the tools we learned to characterize (as much as possible...) the existence and uniqueness of the invariant distribution. Imrohoroglu, Ayse (1989); The Costs of Business Cycles with Indivisibilities and Liquidity Constraints, Journal of Political Economy Huggett, Mark (1993); The Risk-Free Rate in Heterogeneous-Agent IncompleteInsurance Economies, Journal of Economic Dynamics and Control Aiyagari, Rao (1994); Uninsured Idiosyncratic Risk and Aggregate Saving, Quarterly Journal of Economics Hopenhayn H. and E. Prescott (1992); Stochastic Monotonicity and Stationary Distributions for Dynamic Economies, Econometrica 6. Some Applications of Bewley Models We illustrate how to use this class of self-insurance models to analyze questions related to the wealth distribution and to scal policy. Budria Santiago, Javier Diaz-Gimenez, Vincenzo Quadrini, Victor Rios-Rull (2002); Updated Facts on the US Distributions of Earnings, Income and Wealth, Minneapolis Fed Quarterly Review Floden Martin and Jesper Linde (2001); Idiosyncratic Risk in the U.S. and Sweden: Is there a Role for Government Insurance?, Review of Economic Dynamics Aiyagari, Rao and Ellen Mc Grattan (1998); The Optimum Quantity of Debt, Journal of Monetary Economics 7. Constrained eciency in the Aiyagari model We discuss the dierence between the rst-best allocations and the constrained ecient allocations in the Aiyagari model with self-insurance. We argue that the planner, through saving decisions, will manipulate prices in order to raise wages (if the income of the poor is labor intensive), hence redistributing from the lucky-rich to the unlucky-poor. We also debate whether macroeconomists should rene the neoclassical growth model with incomplete markets by adding observed channels of insurance (family, bankruptcy laws, public insurance), or whether they should think about the fundamental reasons that limit full insurance (moral hazard, adverse selection, imperfect enforcement). Hong, Jay, Julio Davila, Per Krusell, and Jose-Victor Rios-Rull (2006), Constrained eciency in the neoclassical growth model with uninsurable idiosyncratic shocks. 4

8. Transitional Dynamics in the Neoclassical Growth Model with Incomplete Markets We study how to compute the transitional dynamics and how to measure correctly the welfare changes associated to a tax reform. Floden, Martin (2001), The Eectiveness of Government Debt and Transfers as Insurance, Journal of Monetary Economics (especially, section 3 on welfare decomposition) 9. Adding Aggregate Risk: A Near-Aggregation Result (LS 17.14.2) We extend the model to add aggregate uctuations in productivity. We explain how to solve this model and present the near-aggregation nding of Krusell-Smith. Krusell, Per and Tony Smith (1998), Income and Wealth Heterogeneity in the Macroeconomy, Journal of Political Economy Heathcote, Jonathan (2004), Fiscal Policy with Heterogeneous Agents and Incomplete Markets, Review of Economic Studies 10. Economies with Default We rst study an incomplete-market economy where agents face borrowing constraints that are tight enough so that they never have the incentive to default in the equilibrium. Then, we formalize a model where agents can default and the nancial sector takes into account the default probability and increases the prices of loans accordingly. Zhang Harold (1997), Endogenous Borrowing Constraints with Incomplete Markets, Journal of Finance Livshits Igor, Jim McGee and Michele Tertilt (2006), Consumer Bankruptcy: A Fresh Start, American Economic Review 11. Life-Cycle Economies with Incomplete Markets We study a life-cycle version of the standard incomplete-markets model with overlapping generations, and an application to optimal Ramsey-style taxation. Imrohoroglu Ayse, Imrohoroglu Selo, Joines, Douglas (1995), A life cycle analysis of social security, Economic Theory Huggett, Mark (1996), Wealth distribution in life-cycle economies. Journal of Monetary Economics Rios-Rull Jose-Victor (1995), Models with heterogeneous agents. In Frontiers of Business Cycle Research, edited by Cooley TF. Princeton, NJ: Princeton University Press. Conesa Juan-Carlos, and Dirk Krueger (2006), On the optimal progressivity of the income tax code, Journal of Monetary Economics 5

 Conesa Juan-Carlos, Kitao Sagiri, and Dirk Krueger (2008), Taxing Capital? Not a Bad Idea After All! American Economic Review 12. Industry Equilibrium We use what we have learned about heterogeneous agents economies to study the equilibrium of an industry with rms facing shocks to their productivity level, and with endogenous rm entry and exit. We analyze the impact of ring costs on the average productivity of the industry. Hopenhayn, Hugo (1992), Entry, Exit and Industry Dynamics in Long-Run Equilibrium, Econometrica Hopenhayn, Hugo and Richard Rogerson (1993), Job Turnover and Policy Evaluation: A General Equilibrium Analysis, Journal of Political Economy 13. Industry Equilibrium with International Trade We extend Hopenhyans model to an economy with monopolistic competition which is open to trade. Firms must pay a xed cost of exporting to access a foreign market. We examine the impact of trade openess on rm selection and aggregate productivity. Melitz, Marc (2003). The Impact of Trade on Intra-Industry Reallocations and Aggregate Industry Productivity, Econometrica

1
1.1

Heterogeneity in the neoclassical growth model with complete markets

Two preliminary results on aggregation of technologies and preferences

In what follows well talk about aggregation. What do we mean with this term? We say that an economy admits aggregation if the behavior of the aggregate equilibrium quantities (aggregate consumption, investment, wealth,...) and prices does not depend on the distribution of the individual quantities across agents. In other words, we can aggregate whenever we can dene a ctitious representative agent that behaves, in equilibrium, as the sum of all individual consumers. 1.1.1 Aggregating rms with the same technology

Consider an economy with M rms, indexed by i = 1, 2, ..., M which produce a homogeneous good with the same technology zF (ki , ni ) where z is aggregate productivity. Assume that F is strictly increasing, strictly concave, dierentiable in both arguments and constant returns to scale. Can we aggregate these individual rms into a representative rm? Suppose inputs markets are competitive. Then, each price-taking rm of type i solves i i i i max , n ( r + ) k zF k wn , i i zFk ki , ni = r + , zFn ki , ni = w, (1)

{k ,n }

with rst-order conditions

Recall that by CRS, Fk and Fn are homogenous of degree zero, hence: Fk (ki , ni ) fk (ki /ni ) = . Fn (k i , ni ) fn (ki /ni ) Dividing through the two rst-order conditions, we obtain fk (ki /ni ) r+ = , i i fn (k /n ) w

and using the fact that the left-hand side is a strictly decreasing function of (ki /ni ), we ki K r+ = , for every i = 1, 2, ..., M =g ni w N where capital letters denote averages: every rm chooses the same capital-labor ratio. Returning now to (1), we have that zFk k i , ni = zfK (K/N ) = r + zfK (K/N ) = r + zfN (K/N ) = w which implies the existence of a representative rm with technology zF (K, N ). 1.1.2 Aggregating consumers with the same preferences obtain

where the last equality is obtained by averaging over all i0 s. And similarly,

Consider a version of the neoclassical growth model with N types of consumers indexed by
i i = 1, 2, ..., N with the same endowments of capital k0 = for all i0 s and same preferences i i X U c0 , c1 , ... = t u ci t . t=0

Assume that markets are competitive, so that every consumer faces the same prices. Then, one would think that since all the N consumers make the same decisions, we can aggregate them into a representative agent, right? Not so quickly... Unless the utility function u is strictly concave, agents may not make the same optimal choices of consumption and leisure. Result 1.0: If every rm has the same CRS production function, consumers have the same initial endowments and same preferences and their utility function is strictly concave, then the neoclassical growth model admits a formulation with one representative rm and one representative household.

1.2

Heterogeneity in endowments

We begin by studying a version of the neoclassical growth model with heterogeneity where consumers are dierent only in terms of their initial endowments of wealth. There is no individual or aggregate uncertainty. 2

1.2.1

The economy

Demographics The economy is inhabited by N types of innitely lived agents, indexed by i = 1, 2, ..., N . Denote by i the number of agents i and normalize the total number of agents to one, so that averages and aggregates are the same. Preferences Preferences are time separable, dened over streams of consumption, given by U=
X t=0

where the period utility function u belongs to one of the following three classes: log, power, exponential, i.e. with c + c > 0, u (c) = with c + c > 0, c exp (c) with c > 0 ln ( c + c)
( c+c)1 1

t u ci t ,

c 0 c 0

(2)

We impose c 0 for log and power utility to allow for a subsistence level for consumption, and we impose c > 0 for the exponential case. Markets and property rights There are spot markets for the nal good (which can be used for both consumption and investment) and complete nancial markets, i.e. there are no constraints on transfers of income across periods. We follow Chatterjee (1994) in assigning the property rights on capital to the rm and the ownership of the rm to the household.1 This is a dierent arrangement of property rights from the one you are used to seeing. Here households own shares of the rm. We will let the initial level of wealth, at date t, dier across agents. Let ai t be the
i i individual wealth of type i at time t. Then, ai t = st At , where st is the share of the rm-

wealth equals the value of the rm. Besides intertemporal trading, there will be no other securities traded among agents in equilibrium, since there is no risk. Technology and rms problem The aggregate production technology is Yt = f (Kt ) with f strictly increasing, strictly concave and dierentiable. The representative
If we had chosen to model the rms problem as static (i.e. the rm rents capital services from households), every argument in this lecture would still hold. You should check this, as well as every other claim I make without proving it!
1

value owned by consumer i at time t. By summing both sides of this equation over i and P i i exploiting the fact that N i=1 st = 1 for every t, we obtain that aggregate household

rm owns physical capital and makes the investment decision by solving the problem X p [f (K ) I ] (3) At = max {I } pt =t s.t. K +1 = (1 ) K + I , where pt is the time t price of the nal good. Lets dene real prots t f (Kt ) It .

Then, it is easy to see that At is the value of the rm, i.e. the present value of future prots discounted at rate (p /pt ), the relative price of consumption between time and time t. Recall that pt /pt+1 = 1 + rt+1 where r is the interest rate. It is useful to compute the rst-order condition (FOC) of the rm problem with respect to Kt+1 by substituting the law of motion for capital into (3) . The problem becomes: pt+1 pt+2 [f (Kt+1 ) Kt+2 + (1 ) Kt+1 ] + ... max f (Kt ) Kt+1 + (1 ) Kt + {Kt+1 } pt pt+1 with FOC: 1= pt+1 0 [f (Kt+1 ) + (1 )] . pt (4)

Households problem Given complete markets, the maximization problem of household i at time t can therefore be stated as: max i
{c } X =t

s.t.
X =t

t u ci (5)

i p ci pt at

i where ai t is the wealth of agent i in term of consumption units at time t. Let t be the

Lagrange multiplier on the individual i time t Arrow-Debreu budget constraint. Solution Consider the log-preferences case. From the FOC of the household problem at time t with respect to consumption at time , we have: i 1 t 0 i t u c = t p = i t p c + ci 4

ci =

t c . i t p

(6)

Substituting this FOC into the budget constraint of (5), we can derive an expression for the multiplier i t:
X =t

i t

1 (1 )

t c = pt ai t i p t X p = pt ai c t
=t

1 i t

= (1 ) pt ai t+c

"

X =t

(7)

Lets now substitute the expression on the last line into equation (6) evaluated at = t, i.e. ci t = 1 i t pt c ,

where

in order to solve explicitly for ci t: " # X 1 ci (1 ) pt ai p c t = t + (1 ) c pt # =t " X p 1 + (1 ) ai = c (1 ) t p t =t = pt , c + (1 ) ai t, # " X t p 1 c (1 ) p ,c pt =t

(8)

(9)

is a function of the subsistence level and of the whole price sequence pt = {pt , pt+1 , ...}. Thus, we have the optimal individual consumption rule t ci + (1 ) ai t = p ,c t,

(10)

which is an ane function of asset holdings at time t for each type i. More in general, when period utility belongs to the families in (2), then preferences share a common property. They are quasi-homothetic, i.e. they have ane Engel curves in wealth: the wealthexpansion path is linear.2
When c = 0, preferences are homothetic because the constant in the consumption function becomes zero and Engel curves start at the origin, i.e. they are linear. However, linearity of the wealth-expansion path is not aected by the constant c .
2

Even though we have only derived it for the log-case, it is easy to check that this representation of the consumption function holds also for the other two classes of preferences (power and exponential utility). 1.2.2 Equilibrium aggregate dynamics

Denote aggregate variables with capital letters. From (10), we derive easily that aggregate consumption only depends on aggregate variables (prices and aggregate wealth), but it is independent of the distribution of wealth. By summing over i on the LHS and RHS of (10) with weights i we arrive at: + (1 ) At . Ct = pt , c (11)

Since equilibrium aggregate consumption Ct has the same form as individual optimal consumption choice ci t , it is clear that (11) can be obtained as the solution to the following representative agent problem: max s.t.
P P

{C } =t

t u (C ) (PP)

=t

p C pt At

which is exactly as in (5) but we have replaced small letters with capital letters. Lets make some further progress on the solution. From the FOCs u0 (Ct ) pt . = u0 (Ct+1 ) pt+1 (12)

From (12) and the FOC for the rms problem (4), we obtain the familiar Euler equation of the neoclassical growth model u0 (Ct ) = u0 (Ct+1 ) [f 0 (Kt+1 ) + (1 )] . We can state our rst important result: Result 1.1: If preferences are quasi-homothetic and agents are heterogeneous in initial endowments, the aggregate dynamics of the neoclassical growth model with complete markets admit a single-agent representation. Put dierently, the dynamics of aggregate 6 (13)

quantities and prices are exactly the same as in the standard neoclassical growth model with representative agent. This result depends crucially on the linearity of individual optimal consumption with respect to wealth which, in turn, descends from quasi-homotheticity of preferences. This type of aggregation, where: 1) preferences of the representative agent are the same as the ones of the individual agent, and 2) aggregate dynamics do not depend on the distribution of wealth is called Gorman aggregation, from the seminal article by Groman (1953). It hinges crucially on homothetic utility and heterogeneity only in initial wealth. Two remarks are in order. First, equation (13) governs the dynamics of capital in the representative agent growth model where rms rent capital from households, instead of owning it. Therefore, we have discovered that in complete markets it is irrelevant whether we attribute property rights on capital to rms (and let households own shares of the rms) or to workers (and let rms rent capital from households). Second, equation (13) also governs the dynamics of capital in the social-planner problem. We are still in complete markets, and the Welfare Theorems hold. Steady-state The dynamics of the economy will converge to the steady-state values of capital stock satisfying the modied golden rule f 0 (K ) = 1/ (1 ). Note now that in steady-state pt /pt+1 = 1/ for all t, hence from the denition of (pt , c ) in (9) we conclude that ci = (1 ) ai . In other words, in steady-state, the average propensity to save is , independently of wealth, for every type of household. To conclude, in the neoclassical growth model with complete markets and where agents have heterogeneous wealth endowments, the dynamics of the aggregate variables do not depend on the evolution of the wealth distribution. But is the inverse statement true? Does the evolution of the wealth distribution across households (i.e., wealth inequality) depend on the dynamics of aggregate variables (prices and quantities)? We show below that the answer is: yes, it does.

1.2.3

Equilibrium dynamics of the wealth distribution

From the lifetime budget constraint of agent i at time t pt ci t +

X i p ci = pt at

=t+1

i i pt ci t + pt+1 at+1 = pt at

(14) (15)

pt+1 ai ci t+1 t + = 1 i ai p a t t t

ai t+1 = ai t

pt pt+1

ci t 1 i , at

which expresses the growth rate of wealth for type i as a function of her consumptionwealth ratio. By aggregating over types in equation (14), we can obtain an equivalent equation at the aggregate level: X
i

i ci t

X pt+1 X i i + at+1 = i ai t pt i i pt+1 At+1 = At Ct + pt At+1 Ct pt 1 = At pt+1 At

We want to establish conditions under which an individuals share of total wealth will
i grow over time, i.e. si t+1 > st . First of all, note that:

si t+1 >1 si t

ai At+1 t+1 > i at At

ci Ct t < i at At

(16)

Moreover, from equations (10) and (11), ci ) (pt , c t = + (1 ) , and i i at at and therefore ci Ct t < i at At ) ) (pt , c (pt , c < i at At pt , c ai t At > 0 ) Cti (pt , c = + (1 ) At At

and, thus, summarizing we have the following equivalence (i.e., if and only if) condition: si t+1 >1 si t pt ai t At > 0,

which means that whether consumers i wealth share is increasing or decreasing over time depends on 1) the sign of the constant (equal for everyone) and 2) on her relative position 8

in the distribution. For example, if > 0 then for a consumer whose initial wealth is above average, her share will grow, whereas for a consumer whose initial wealth is below average, her share will fall. And hence the distribution will become more unequal over time. Note that the dynamics of the wealth distribution depend on the entire sequence of prices, hence on the dynamics of aggregate variables in equilibrium. First of all, in absence of subsistence level, c = 0 we have = 0, and therefore the neoclassical growth model with heterogeneous endowments has a sharp prediction for the evolution of inequality. Result 1.2: In the neoclassical growth model with complete markets, homothetic preferences, heterogeneous endowments, but without subsistence level ( c = 0), the wealth distribution remains unchanged along the transition path, i.e. initial conditions in endowments (and inequality) persist forever. The intuition is that if c = 0 then ci = (1 ) ai , so the average propensity to

consume, and save, is the same for every agent. Every agent accumulates wealth at the same rate. In presence of a subsistence level, the dynamics are more interesting. We now determine the sign of , through: Lemma 1.1 (Chatterjee, 1994): The common constant term of the consumption function (pt , c ) is greater (less) than zero if and only if the economy is converging from below (above) to the steady-state, i.e. if K < (>) K . Proof: Suppose the economy grows towards the steady-state, i.e. K < K . From equation (13), the sequence {f 0 (K )} is decreasing and the sequence {p /p +1 } will be decreasing towards 1/. Therefore, the sequence {p +1 /p } is increasing towards , i.e., p +1 /p for all t where the strict inequality holds at least for some . It follows p /pt = (p /p 1 ) (p 1 /p 2 ) ... (pt+2 /pt+1 ) (pt+1 /pt ) < t . ) in (8), use the above equation to obtain From the denition of (pt , c # " " # X X t p p ,c 1 >c (1 ) =c (1 ) t 1 = 0 p t =t =t where the inequality follows from c < 0. QED 9

that

The implications for the evolution of the wealth distribution in an economy growing towards the steady-state (the empirically interesting case) are easy to determine, at this point. In the presence of a subsistence level ( c < 0), > 0. From equation (16) this implies that the average propensity to consume (save) declines (increases) with wealth: poor agents must consume proportionately more out of their wealth to satisfy the subsistence level. In other words: Result 1.3: In the neoclassical growth model with complete markets, homothetic preferences, heterogeneous endowments and subsistence level c < 0, as the economy grows towards the steady-state: (i) the wealth distribution becomes more unequal, as rich agents accumulate more than poor agents along the transition path, and (ii) there is no change in the ranking of households in the wealth distribution, i.e., initial conditions in wealth (and consumption) ranking persist forever. The main conclusion of this lecture is that in this model there is no economic or social mobility. This is not a good model to understand why some individual are born poor and make it in life, while other are born rich and end up poor as rats. This is just a model of castes. Robustness We now discuss how robust this result is to two of the key assumptions made so far in the analysis: 1) all agents have same discount factor , 2) markets are complete. 1. When agents have dierent discount factors, then none of the results hold any longer. Suppose that c = 0 to simplify the analysis. Then, from (10) i i ci t = 1 at ,

therefore the average propensity to save out of wealth is higher the more patient is the individual and from (16), wealth grows faster for the more patient individuals. In the limit, in steady-state, the most patient type holds all the wealth, and the distribution becomes degenerate. 2. In absence of markets and trade (autarky), every consumer has access to her own technology. Each agent i will solve his own maximization problem in

10

isolation max i

s.t. i i i i k +1 = (1 ) k + f (k ) c i kt given
i with dierent initial conditions kt . It is easy to see that, independently of

{c } =t

t u (ci )

initial conditions, each agent will converge to the same capital stock k , hence in the long-run the distribution of wealth is perfectly equal. Interestingly, we conclude that less developed nancial markets induce less wealth inequality, in the long-run. 1.2.4 Indeterminacy of the wealth distribution in steady-state

One very important implication of the aggregation Result 1.1 is that in steady-state the wealth distribution is indeterminate. From (3), (13) and (10), the set of equations characterizing the steady-state is: f 0 (K ) = 1/ (1 ) , 1 [f (K ) K ] A = 1 N X i ai = A ,
i=1

other words, the multiplicity of the steady-state wealth distributions is of order N 1.3

N We therefore have (N + 3) equations and (2N + 2) unknowns {ci , ai }i=1 , K , A . In However, suppose we start from a given wealth distribution at date t = 0 when the economy has not yet reached its steady-state, then the dynamics of the model are uniquely determined by Results 1.2 and 1.3 and the nal steady-state distribution is determined

ci = (1 ) ai , i = 1, 2, ..., N

as well. So, lets restate our nding in: Result 1.4: In the steady-state of the neoclassical growth model with N agents, heterogeneous initial endowments and homothetic preferences, there is a continuum with dimension (N 1) of steady-state wealth distributions. However, given an initial wealth
This means that, if N = 1 (representative agent), the steady-state is unique. If N = 2, there is a continuum of steady-states of dimension 1, and so on.
3

11

t is uniquely determined, and so is the nal steady-state distribution.

i distribution {ai 0 }i=1 at t = 0, the equilibrium wealth distribution {at }i=1 in every period

Example Consider an economy where N = 2, where the production technology is zf (K ). Then, the set of steady-state equations is zf 0 (K ) = 1/ (1 ) , 1 [zf (K ) K ] , A = 1 1 a1 + 2 a2 = A ci = (1 ) ai , i = 1, 2, So we have 5 equations, but 6 unknowns {a1 , a2 , K , A , c1 , c2 }. We can represent graphlevel of technological progress z , say (zL , zH ) . The gure shows that the model has a continuum of steady-state distributions of wealth of dimension one, all consistent with the uniquely determined aggregate capital stock K . If we pick an initial distribution in the initial steady-state with productivity zL , the equilibrium path to the nal steady-state with productivity level zH is uniquely determined. Finally, in terms of language, this whole section shows that it is important to distinguish steady-state from equilibrium path. In this economy, the equilibrium path is always unique (given initial conditions), but the steady-state is not.

ically all the possible equilibrium paths between two steady states that dier for their

The Negishi Approach

Negishi (1960) suggested a method to calculate the competitive equilibrium (CE) prices and allocations of complete markets economies (in particular, economies for which the rst welfare theorem holds) with heterogeneous households. This method is particularly useful for those economies where aggregation does not go through and, hence we cannot use the representative agent.4 From the rst welfare theorem, we know that any CE is a Pareto optimum (PO), hence it can be found as the solution to a social planner problem with some Pareto weights given to each agent. Suppose we want to compute a particular CE of an economy
In his original paper, Negishi (1960) used this equivalence result to propose a simple way to show existence of competitive equilibria.
4

12

wealth. Can we use the planner problem for this purpose? Negishi showed that the key is to search for the right weights given to each type of agent in the social welfare function

where agents are initially endowed with heterogeneous shares {si 0 }i=1 of the aggregate

of the planner. Each set of weights corresponds to a Pareto ecient allocation, the key is to nd the set of weights which correspond to our desired CE allocation. As an aside, the concept of social welfare was introduced by Samuelson (1956).

2.1

An Example

Consider our neoclassical growth model of section (1.2) with two types of consumers (N = 2). The agents i problem in the decentralized Arrow-Debreu equilibrium can be written as max t u ci t {ci t } t=0
X

s.t.
X t=0

i pt ci t p0 a0

i where ai 0 = s0 A0 is the initial wealth endowment, given at t = 0. Lets assign the

property rights on capital to the rm, so the rms problem is exactly the one of the previous section. From the FOC of the agent of type i, we obtain i t 0 i F OC ci t u ct = pt ,

where i is the multiplier on the time zero budget constraint. Thus, putting together the FOCs for the two types: 1 u0 (c1 t) = 2. u0 (c2 t) Now, write down the following Negishi planner problem (NP) for our economy max
X t=0

(17)

s.t.

2 {c1 t ,ct ,Kt+1 }

2 2 t 1 u c1 t + u ct

(NP)

2 c1 t + ct + Kt+1 f (Kt ) + (1 ) Kt

K0 given 13

where (1 , 2 ) are the planners weights for each type of household in the social welfare function. The FOCs for this problem are F OC ci i t u0 ci t t = t , i = 1, 2 1 0 0 F OC (Kt ) u0 c1 t = u ct+1 [f (Kt+1 ) + (1 )] (18) (19)

where t is the Lagrange multiplier on the planners resource constraint at time t. Note that putting together the rst-order conditions for consumption for the two agents we arrive at u0 (c1 2 t) = , u0 (c2 1 t) (20)

which tells us that the planner allocates consumption proportionately to the weight it gives to each consumer (with strictly concave utility).5 If we want the NP to deliver the same solution as the CE, we need the PO allocations and the CE allocations to be the same. Given strict concavity of preferences, this implies that, putting together (20) and (17): 2 1 = 2 1 Hence, the relative weights of the planner must correspond to the inverse of the ratio of the Lagrange multipliers on the time-zero Arrow-Debreu budget constraint for the two agents in the CE.
i In particular, for log preferences u (ci t ) = log ct , we derived in equation (7) that 1 1 1 i i = (1 ) p0 a0 = , i (1 ) p0 A0 si 0

therefore we obtain that

1 s2 2 0 = = . 1 s1 2 0

Imposing the natural and innocuous normalization 1 + 2 = 1, we can solve explicitly

2 2 for the two weights: 1 = s1 0 and = s0 , i.e., the weights are exactly equal to the initial i wealth shares. The higher is the initial wealth share si 0 , the lower is the multiplier and

the larger is the Pareto weight i on the Negishi problem: the planner must deliver more
Note also that the ratio of marginal utility across agents is kept constant in every period (a key features of complete markets allocations, also called full insurance ).
5

14

to consumption to the agent who has a large initial share of wealth in the decentralized equilibrium. Result 1.5: Consider an economy with agents heterogeneous in endowments where the First Welfare Theorem holds. Then, the competitive equilibrium allocations can be computed through an appropriate planners problem where the relative weights on each agent in the social welfare function are proportional to the relative individual endowments: those agents who initially have more wealth will get a higher weight in the planners problem. Now, note that using equation (18) we obtain that i t u0 (ci t u0 (ci t t) t) = = . = t +1 t+1 t+1 u0 ci i u0 ci t+1 t+1

Substituting this last expression into (19), we arrive at a relationship between the sequence of capital stocks and the sequence of Lagrange multipliers on the planners resource constraint t = f 0 (Kt+1 ) + (1 ) . t+1 Recall that equation (13) dictating the optimal choice of capital for the rm in the CE stated that pt = f 0 (Kt+1 ) + (1 ) . pt+1 Hence, we have pt t = , t+1 pt+1 in other words, the Arrow-Debreu equilibrium prices can be uncovered as the sequence of Lagrange multipliers in the Pareto problem: intuitively, the multipliers gives us the shadow value of an extra unit of consumption and, in the CE, prices signal exactly this type of scarcity.6 In conclusion, we have uncovered a tight relation between weights of the NP problem and initial endowments in the CE and an equivalence between Lagrange multiplier on the resource constraint of the NP problem and prices in the CE. This strict relationship, that we have uncovered for the log utility case, is true more in general.
Using p0 as the numeraire and imposing the normalizations p0 = 0 = 1, the above relationship implies that pt = t so equilibrium prices are exactly equal to the shadow prices of consumption in the planners problem.
6

15

2.2

General application of the Negishi method

In general, without restrictions on preferences, one may not have closed form solutions for the s in the CE, so the algorithm is a little more involved. The objective is to compute the CE allocations for an economy with N types of agents and endowment distribution {ai 0 }i=1 . We can describe the algorithm in four steps:
N

1. In the social planner problem (NP), guess a vector of weights = 1 , 2 , ..., N . P i The normalization N i=1 = 1 means that belongs to the N-dimensional simplex N = { RN + :
N X i=1

i = 1}

and the simplex traces out the entire set of Pareto-optimal allocations. o n N } , K and the implied sequence of 2. Compute the sequence of allocations {ci t t i=1
t=0

t, one needs to solve the N + 2 equations

multipliers {t } t=0 on the resource constraint in each period t. In practice, at every i t u0 ci = t , i = 1, .., N t

the Negishi method simplies enormously the computation of the equilibrium: the Negishi solution requires solving, for every time t, a small system of equations. Recall that, instead, to solve for the CE allocations, at every time t one must set the excess demand function to zero and the excess demand function depends on the entire price sequencean innitely dimensional object. To understand, take another look at the consumption allocation (10) where () depends on the entire price sequence from t onward. Instead of guessing (and iterating over) an innite sequence of prices, one guesses and iterates over a nite set of weights. With a caveat: even though K0 is given, 0 is not. So, one has also to guess a value for 0 . The reason is that, unless you have the right value for 0 , the system will not be on the saddle-path and capital 16

t = f 0 (Kt+1 ) + (1 ) t+1 i N in N + 2 unknowns {ct }i=1 , Kt+1 , t+1 . At every t, (Kt , t ) are given, therefore

N X i=1

i ci t + Kt+1 = f (Kt ) + (1 ) Kt

will diverge. In other words, there is another condition that we need to satisfy in the growth model, the transversality condition. 3. Exploit the equivalence between prices pt and multipliers t to verify whether the time-zero Arrow-Debreu budget constraint of each agent holds exactly at the guessed vector of weight . Specically, for each agent, compute the implicit transfer function i () associated with the assumed vector of weights () =
i X t=0

i i t ci t 0 a0 , for every i = 1, 2, ..., N

(21)

and if (21) holds for agent i with a greater than sign, it means that the planner is giving too much weight to agent i. So, in the next iteration reduce the weight i given to agent i. Note one useful property of the transfer functions: " # N N X X X X i i i i () = t ct 0 a0 = pt Ct p0 A0 = 0
i=1 i=1 t=0 t=0

since the discounted present value of resources of the economy cannot be greater than its current wealth. Put dierently, recall that from the representative rm problem A0 = X pt
t=0

p0

[f (Kt ) It ] =

which establishes the same result. To conclude, the transfer functions must sum to zero. 4. Iterate over until you nd the vector of weights that sets every individual transfer function i ( ) to zero. This vector corresponds to the PO allocations that are aordable by each agent in the CE, given their initial endowment, without the need for any transfer across-agents. Thus, we are computing exactly the CE associated to initial conditions {ai 0 }i=1 . See also Ljungqvist-Sargent, section 8.5.3, for a discussion of the Negishi algorithm.
N

X pt
t=0

p0

Ct

2.3

Non-Gorman aggregation

Maliar and Maliar (2001, 2003) make use of the Negishi approach to prove a more general aggregation result. In their model, agents have non-homothetic preferences in con17

sumption and leisure, and are subject to idiosyncratic, but insurable, shocks to labor endowment. They prove that the strong Gorman aggregation fails, but one can obtain a weaker aggregation result. The aggregate dynamics of the model can still be described by a representative agent (RA). However, the RAs preferences are dierent from preferences of the individual consumer. There exists a new preference shifter for the RA that depends on the distribution of individual productivity shocks. Therefore, the dynamics of aggregate variables do depend on the distribution of the shocks. 2.3.1 The Model

Demographics The economy is inhabited by a continuum of innitely lived agents, indexed by i I [0, 1]. Denote by i the measure of agents i in the set I and normalize R the total number of agents to one, I di = 1, so that averages and aggregates are the same. Initial heterogeneity is in the dimension of initial wealth endowments. Uncertainty Agents are subject to idiosyncratic productivity shocks to skills. Let i t be the shock of agent i, and suppose shocks are iid, with mean 1, and dened over the set E . This is not necessary, but it simplies the notation. Preferences Preferences are time separable, dened over strieams of consumption, given by U = E0 where period utility is given by u (ct , 1 nt ) =
X t=0

i t u ci t , 1 nt .

1 (1 nt )1 1 c1 t +A 1 1

(22)

and note that preferences are not quasi-homothetic, unless = . Markets and property rights There are spot markets for the nal good (which can be used for both consumption and investment) whose price is normalized to one, and complete nancial markets, i.e. agents can trade a full set of state-contingent claims. The agents portfolio is composed by Arrow securities of the type ai t+1 () which pay one unit of consumption at time t + 1 if the individuals shock is and zero otherwise. Let pt () R the price of this security and pt () ai t+1 () d the value of such portfolio for agent i. 18

Technology and rms problem The aggregate production technology is Yt = Zt f (Kt , Ht ) with f strictly increasing, strictly concave and dierentiable. The representative rm rents capital from households. Ht is aggregate labor input in eciency units, R i i i.e. Ht = I i t nt d . Household problem For agent i: max E0
X t=0

ci t

i kt +1

i {ci t ,kt+1 ,at+1 ()} s.t.

i t u ci t , 1 nt

(23)

i i i i i pt () ai t+1 () d = (1 ) kt + wt t nt + at t
i k0 , ai 0 given

Equilibrium This is a CM economy. The First Welfare Theorem tells us that the equilibrium is PO, so we can use a social planner problem to characterize the equilibrium by applying the Negishi method. The key, as usual, is to nd the right weights that guarantee that allocations are aordable for each agent, given their initial endowments. Aggregation? Given that preferences are not homothetic, we know that Gormans strong aggregation concept will not hold. But can we, nevertheless, obtain a RA whose choices describe the evolution of the aggregate economy? And how the preferences of the RA look like? Letting t be the multiplier on the aggregate feasibility constraint, from the FOC with respect to individual i in the Negishi planner problem: i ci = t t = t wt i i A 1 ni t t i i t
1

(24)

Rearranging gives

ci t = i 1 ni t t =

Ai t wt

and note that consumption of individual i is proportional to its weight in the social welfare function. Leisure is directly proportional to its weight (a wealth eect) and inversely 19

i 11/ t

proprtional to individual productivity: eciency arguments induce the planner to make high-productivity individuals work harder. Integrating the two FOCs across agents gives Ct = Z
i ci t d

1 Ht = 1 Now, note that

Z
I

i t

di Z
I

(25) Ai t wt
1

i i i t nt d

=1

i 11/ i d t

ci t

i 1 ni t t =

Now, consider the social welfare function for the planner Z i 1 1 (1 ni (ct ) 1 1 i t) +A d 1 1 I

1/ (i ) (i t) i 1 nt = R (1 Ht ) 1 i ) (i )11/ di ( t I

1 R i

1 i

di

Ct

ci t

Ai t wt

i 11/ 1 ni t t =
1

=R

(i )
1

(i ) di I

Ct
1

(26) i 1/ t (27)

Ai t wt

and substitute the two expressions in (26) into the social welfare function: " " #1 #1 1 1 1/ (i ) (i (i ) ) t Ct 1 (1 Ht ) 1 R 1 1 R 11/ Z i ) i i ) di ( di ( ( ) t I I +A di i 1 1 I =
R
I

i (i )
1 (i )

di

1 1 Ct

di

1 +A

i (i )
1 (i )

(i t)

11/

( )

11/ i di t

di 1

(1 Ht )1 1

which yields the utility for the RA (1 Ht )1 1 Ct1 1 + AXt 1 1 where hR i 1 i i 11/ i ( ) (t ) d I hR i = 1 (i ) di I 20

Xt

Therefore the RA problem which describes the aggregate allocations for this economy becomes: max E0
X t=0

{Ct ,Ht ,Kt+1 }

s.t.

Ct1 1 (1 Ht )1 1 + AXt 1 1

(28)

Ct + Kt+1 = (1 ) Kt + Zt f (Kt , Ht ) K0 given Some remarks are in order: 1. We have found a RA, but its preferences are not the ones of the individual agent. Note that we have Ht instead of Nt , and note that we have a new preference shifter Xt . First reason why this is not Gorman aggregation. 2. The preference shifter, in general, depends on the distribution of shocks and endowments, therefore aggregate dynamics do depend on the distribution. Second reason why this is not Gormans aggregation. Note, however, that the distribution is exogenous: its a simple problem. 3. Suppose = . Then utility is quasi-homothetic. If there are no skill shocks, but only dierences in endowments, then Xt = 1 and Ht = Nt . We are back to Gormans aggregation. If there are idiosyncratic shocks, then Xt 6= 1 and Gorman aggregation aggregation to hold. 4. Suppose 6= . Then utility is not quasi-homothetic. Even if there are no skill shocks, but only dierences in endowments, then Xt depends on the distribution of endowments and Gormans aggregation fails. Finally, note that the assumption that agents can trade a full set of claims contingent on all possible realizations of idiosyncratic labor productivity shocks is not very realistic, as we will argue later in the course. It is mainly a useful theoretical benchmark. fails, which establishes that you need heterogeneity to be only in wealth for Gorman

21

2.4

Notes

Gorman (1953) developed the theory of aggregation of individual preferences. His main result is that when preferences are homothetic and households dier in initial wealth levels only, social preferences do not depend on the distribution of individual wealth. Stiglitz (1969) discusses the income and wealth distribution dynamics in the context of a Solow growth model, where individual savings are assumed to be linear in capital. Our discussion in sections 1.2 is based on Chatterjee (1994). Caselli and Ventura (2000) extend the Gorman result to economies where agents also dier in their endowments of eciency units of labor. They work in continuous time and apply their results to the transitional dynamics of the neoclassical growth model and to an economy with Arrow-style externalities. Maliar and Maliar (2001, 2003) are examples of how the aggregation theorems apply to economies with both idiosyncratic and aggregate uncertainty when markets are complete. In particular, they show that even when preferences are non-homothetic, in certain cases one can obtain closed-form aggregation, although the aggregate preferences depend on the distribution of individual shocks. See also chapter 4.d in Mas Colell for a more abstract discussion of the existence of a representative consumer.

References
[1] Caselli, Francesco and Jaume Ventura (2000), A representative Consumer Theory of Distribution, American Economic Review. [2] Chatterjee, Satyajit (1994) "Transitional dynamics and the distribution of wealth in a neoclassical growth model", Journal of Public Economics [3] Gorman, Community preference eld, Econometrica [4] Maliar, Lilia and Serguei Maliar (2001) "Heterogeneity in capital and skills in a neoclassical stochastic growth model", Journal of Economic Dynamics and Control [5] Maliar Lilia and Serguei Maliar (2003) "The representative consumer in the neoclassical growth model with idiosyncratic shocks", Review of Economic Dynamics [6] Negishi Takashi (1960), "Welfare Economics and Existence of an Equilibrium for a Competitive Economy", Metroeconomica 22

[7] Samuelson, Paul (1956), Social Indierence Curves, Quarterly Journal of Economics [8] Stiglitz, Joseph (1969), Distribution of Income and Wealth among Individuals, Econometrica.

23