Notes on Inequality in the Neoclassical Model of Growth F. Obiols Barcelona GSE 2 1. Introduction Casual observation suggests that there is substantial heterogeneity across households in dimensions such as human capital and education, age (or position in the life-cycle), income and/or wealth, marital status, labor market status, health, etc. It seems obvious that for a sensible design of policies related to health care, unemployment benefits, and education (amongst others), it would be necessary to take into account these relevant dimensions of heterogeneity. The simple version of the NMG we developed in the previous notes abstracts from any source of heterogeneity, and thus, it would be hard to try to use that plain version of the model to obtain policy prescriptions along these lines. In these notes we develop a preliminary version of the model to incorporate heterogeneity among households, and we study the evolution of inequality over transitions toward the steady state. The model we use is an extended version of the NMG, and although we mainly concentrate on heterogeneity of initial assets, the model could easily encompass heterogeneity in preferences, and in labor income (which could appear as the result of differences in education across workers). The purpose of the model is to shed light on the competitive dynamics of inequality in a particular measure of wealth. The version of the model we develop can also be used to guide policies aiming at reducing inequality in wealth: informally, it is often claimed that more unequal societies tend to “do worse” than more equal ones (i.e., beyond moral/philosophical concerns, informal considerations regarding growth and efficiency seem to advocate for less inequality). The model we develop here will conclude that the trade off between inequality and efficiency is in fact not a very favorable one: less inequality is possible, but at the cost loosing efficiency. 1. Basic model • In the economy there is a large number of agents (consumer-workers) choosing consumption and saving over an infinite horizon. The “name” of the agents is given by i = 1, 2, 3..., N . • Agents are identical in every respect (preferences, endowments, available technologies...) except in their initial amount of assets ai0. • There are competitive firms producing a consumption/investment good. The available technology uses capital and labor, and it displays CRS in these two factors of production. The problem of the firm/s, therefore, is exactly the same one we studied in the notes about the NMG and it wont be repeated here. 2. The problem of one of the agents The problem of a given agent i can be written as: max ∞ X {cit ,ait+1 } t=0 β t log(cit − c̄), β ∈ (0, 1), subject to the budget constraint in every period t: cit + ait+1 = wt + (1 + Rt)ait, non negativity constraints for consumption and assets (ct, at+1 ≥ 0), and taking as given all prices ({Rt, wt}∞ t=0 ), and ai0. There are a few comments that is worth emphasizing before we continue: 1. The log-utility assumption is convenient but not restrictive: the results to be presented below also hold under more general preferences. Furthermore, log-utility is very often used in quantitative studies (e.g., the Real Business Cycle literature). 2. The usual case in the literature is c̄ = 0. If c̄ > 0, then we need to restrict consumption so that cit ≥ c̄ for all t. In this case c̄ > 0 is interpreted as a minimum consumption requirement. It is also possible to consider the case of c̄ < 0, in which the agent derives positive utility even if cit = 0 (a possible interpretation in this case is that c̄ is an endowment of consumption goods, an transfer from the government, which we do not model explicitly). 3. In the budget constraint we are assuming that all agents supply the same amount of labor in efficiency units, and that all agents are equally productive (1 unit of efficient labor). Also, we take Rt = rt − δ, that is, Rt is the return to assets net of depreciation. That is, we have already substituted the law of motion for capital in the budget constraint. All the equilibrium action in the model comes from the consumer’s side, hence it will be convenient to introduce a notion of CE, which we do next. Definition 1: A Competitive Equilibrium in sequence i , ai ∞ for every i = 1, ..., N , a form is a list xh = {c } t t+1 t=0 i P ∞ list xf = {Kt, Ht}∞ t=0 , and a list x = {rt , wt }t=0 , such that: 1) For each i = 1, ..., N , given xp and ai0, xh i solves the utility maximization problem of agent i. 2) Given xp, xf solves the profit maximization problem of the representative firm. In particular, the choices of the firm satisfy: ∂F (Kt, Ht) ∂F (Kt, Ht) = rt, and = wt. ∂Kt ∂Ht 3) All markets clear in all periods: - labor market: N = Ht, P - capital market: i ait = Kt, P - goods market: i cit + Kt+1 = F (Kt, Ht) + (1 − δ)Kt. (notice that as usual by Walras’ Law, one of the above three conditions is redundant). We have in mind that essentially there is only capital in the economy, and so, the supply of capital is given P P by i ait = i kti. However, we could also assume that in addition to capital, there are other assets in zero net supply. One instance of such an asset would be a one period bond: in period t agents are able to buy/sell bit+1 units of bonds at a price qt (in units of period t consumption goods), and each of these bonds will return. • For the “buyers” of bonds, bit+1 > 0, and it represents an additional means of saving. • For the “sellers” of bonds, bit+1 < 0, and it represents an obligation to deliver consumption goods in t + 1. The fact that these bonds are in zero net supply simply means that the initial aggregate amount of bonds is zero (hence the economy starts out with no debts nor P i additional sources of wealth than capital, thus i a0 = P i i k0 ). The market clearing condition for bonds in any period P i = 0, ∀t. Hence, we would is of course that N b i t+1 P P i P P i have i ait+1 = i kt+1 + i bit+1 = i kt+1 . Notice, therefore, that the market clearing condition in the definition above still holds true. Finally, for these bonds to coexist with capital its return must be identical to the return to capital (or agents would only hold the asset with the highest return). This is known as a no arbitrage condition and it requires that 1 − qt = Rt+1, qt (the left hand side is the return on bonds, and the right hand side is the return on capital), which we finally write as qt = 1/(1 + Rt+1). Clearly, the portfolio composition is irrelevant when the no arbitrage condition holds. We conclude that restricting assets to only capital, or to assume a large number of assets including assets in zero net supply, has no economic consequences when the no arbitrage condition holds and there are no frictions. 3. Intermediate results In this section we introduce two intermediate results that will be useful later. 1. Welfare depends on life-time wealth. 2. Perfect aggregation holds. 3.1 Welfare depends on Life-time wealth We will show that consumption in present value is a function of life-time wealth. We will also show that in every period, consumption itself is also a function of life-time wealth. Since welfare depends only on consumption, these facts justify our interest on life-time wealth. Proceeding as in class, take the budget constraint of an agent i in period t + 1 and isolate ait+1 to obtain ait+1 = cit+1 + ait+2 − wt+1 1 + Rt+1 , and insert the resulting expression in the period−t budget constraint, so that after rearranging, we end up with: cit + cit+1 1 + Rt+1 + ait+2 1 + Rt+1 = ait(1 + Rt) + wt + wt+1 . 1 + Rt+1 Repeat the same strategy (get ait+2 from the period t + 2 budget constraint), to get that: cit+1 cit+2 ait+3 i ct + 1+R + (1+R )(1+R ) (1+R )(1+R ) t+1 t+1 t+2 t+1 t+2 wt+1 wt+2 i + (1+R )(1+R ) . = at(1 + Rt) + wt + 1+R t+1 t+1 t+2 If we continue this recursion up to infinity then on the right hand side we get: RHS = (1 + Rt) ait + ∞ X wt+j . Qj j=0 s=0 1 + Rt+s Notice the RHS is, in fact, the definition we introduced in class for life-time wealth: the value of assets (nonhuman assets) today, plus the present value of all future labor income (i.e., of human assets). To be precise, we define life-time wealth as ωti = ait + ∞ X wt+j Qj j=0 s=0 1 + Rt+s Notice that ωti is linear in ait. (1) On the left hand side of the budget constraint we simply have: LHS cit+1 cit+2 i = ct + 1+R + (1+R )(1+R ) t+1 t+1 t+2 i ct+3 + (1+R )(1+R )(1+R ) + ... t+1 t+2 t+3 where we have used a version of the transversality condition: the present value of assets that are arbitrarily away in the future is zero: aT +1 lim Q = 0. T →∞ T t=0 1 + Rt Hence, the LHS is simply the present value of the infinite sequence of consumptions. Hence, we write the following inter-temporal budget constraint: cit + ∞ X cit+j = (1 + Rt)ωti. Qj j=1 s=1 (1 + Rt+s ) (2) This is our first result: present value of consumption is a function of life-time wealth. We will now use the FOC of the consumer problem to get rid of the consumptions dated t + 1, t + 2,... and show that in every period t, ct is a linear function of life-time wealth (that is, Engel curves (also called expansion paths) are linear in life-time wealth). With log-utility the Euler Equation in period t reads: (cit − c̄)−1 = β(1 + Rt+1)(cit+1 − c̄)−1, and after rearranging: cit+1 = β(1 + Rt+1)cit + c̄(1 − β(1 + Rt+1)), or cit+1 1 − β(1 + Rt+1) i = βct + c̄ . 1 + Rt+1 1 + Rt+1 This expression can be substituted in the LHS of the inter-temporal budget constraint we obtained above. In fact, we can repeat the same strategy in t + 1 to get cit+2 1 + Rt+2 = βcit+1 + c̄ 1 − (β(1 + Rt+2) , 1 + Rt+2 and combining with the previous expression, we get: cit+2 (1+Rt+1 )(1+Rt+2 ) 1−β(1+Rt+1 ) 1+Rt+1 1−β(1+Rt+2 ) +c̄ (1+R )(1+R . t+1 t+2 ) = β 2cit + βc̄ Repeating this procedure, and substituting the resulting expressions in the equation for LHS, we finally write: ∞ 1 − β(1 + R X cit c̄ t+1+j ) LHS = + . Qj 1−β 1 − β j=0 1 + Rt+1+s s=0 Writing down together the LHS and the RHS, we finally obtain cit = (1 − β)(1 + Rt)ωti − c̄Pt, (3) with ∞ 1 − β(1 + R X t+1+j ) Pt = . Qj j=0 s=0 1 + Rt+1+s Notice that Pt is the same for all i (hence it is independent of i). Equation (3) above states our second result and shows that consumption in period t is a function of life-time wealth in period t. Hence, it is justified to focus on the evolution of inequality in life-time wealth. To follow the evolution of ωti over time and over agents in general is a cumbersome task. The reason is that ωti depends on all future prices as well as on the asset level brought from the precedent period, which in turn depends on consumption on that period, which in turn depends on all future prices from that period onwards. That is, prices depend on aggregate assets, and aggregate assets in t + 1 may depend in a complicated way of the period t distribution of assets among agents. That is, in general it is true that in order to know aggregate assets we need to know equilibrium prices, and in order to know equilibrium prices we need to know aggregate assets and its distribution. Fortunately, however, in our model there is a couple of simple facts that help us to find a way out of this perverse loop: 1. Equilibrium prices depend only on aggregate assets, and not on it’s distribution over agents (remember that in the FOC of the firm -which must be satisfied- all that matters is to know aggregate capital and aggregate labor (hence the way they are actually distributed is in fact irrelevant). We conclude that, at least to compute prices, the distribution of wealth is irrelevant.∗ ∗ This conclusion would not be true if in addition to aggregate assets, prices were a function of the standard deviation of assets (the first moment of the distribution of assets over agents), or other higher order moments of that distribution. 2. The second interesting observation follows from Eq. (3) above: consumption per capita is a linear function of life-time wealth per capita: X ci X ωi t = (1 − β)(1 + R ) t − c̄P . t t N N i i P i Furthermore, ωti is linear in assets, hence i ωt /N is linear in per capita assets. The crucial implication of this fact is that consumption per capita in period t is independent of the distribution of assets in period t: again, it only depends on the amount of assets per capita in that period. In particular, if in two economies the term Pt is the same, per capita assets are also the same, and the only difference is in the way these assets are distributed among the agents, then consumption per capita in both economies will necessarily be the same. It follows from the statement above that all we need to be able to follow consumption and assets in per capita terms is the equilibrium sequence of prices, which in turn depends only on aggregate assets and not in its distribution. In oder words, if we were given the equilibrium allocation in per capita terms (say {ct, at+1}∞ t=0 ), then we would be able to compute prices, and hence, we would be also able to compute consumption, assets, and lifetime wealth for every agent in the economy and over all time periods. 3.2 Perfect aggregation holds The previous reasoning boils down to the fact that, in per capita terms, it is observationally equivalent to study an economy with lots of heterogeneous agents or an economy with a single representative agent whose initial endowment of assets is the per capita assets of the multi agent economy. That is, a perfect aggregation result holds such that the heterogeneous agents economy admits a representative agent representation. Of course, the problem of the representative agent is identical to the one described by the NMG and of which we already have a number of results. In particular, the CE is efficient, which means that we can solve the much simpler planner’s problem, obtain the efficient allocation (which is in fact the competitive allocation), and use the FOC of the firm to obtain the sequence of equilibrium prices corresponding to the transition starting from the given k0. Since we derive all these results using the artificially constructed representative agent, we are entitled to use all the results on convergence we obtained in the NMG. In particular, it should be clear, that 1. In per capita terms, the steady state of the economy is independent of the distribution of initial assets. That is, given initial assets per capita, we converge to the same steady state irrespectively of the initial distribution of assets. 2. The sequence of equilibrium prices is independent of the distribution of assets. 4. Transitional dynamics of inequality in ωti The measure of inequality we will use is the Coefficient of Variation in ωti. This measure is defined as CV (ωti) = sd(ωti)/ωt, where sd(ωti) stands for the standard deviation across agents, and where ωt stands for the average over agents.† † For a given random variable (rv) x, the standard deviation (sd) is a measure of disparity (or heterogeneity). Since the sd is affected by the level of the rv, then we simply normalize the measure by its mean. This way we can compare distributions even if they have very different means. Beyond this, it is possible to derive our results using other measures of inequality such as Lorenz domination, or the Gini Index (at a higher analytic cost, though). Remember that ωti = ait + ∞ X wt+j ∀t, Qj j=0 s=0 1 + Rt+s and so in t + 1 we have that: i ωt+1 = ait+1 + ∞ X wt+1+j Qj j=0 s=0 1 + Rt+1+s (4) holds. The budget constraint in period t reads cit + ait+1 = wt + (1 + Rt)ait, which implies that ait+1 = wt + (1 + Rt)ait − cit, and since cit satisfies Eq. (3), then we finally obtain ait+1 = wt + (1 + Rt)ait − (1 − β)(1 + Rt)ωti + c̄Pt. Insert the expression above in Equation (4) and obtain i = wt + (1 + Rt)ait − (1 − β)(1 + Rt)ωti + c̄Pt ωt+1 P∞ w + j=0 Qj t+1+j , s=0 1+Rt+1+s which can be rewritten as: i = (1+Rt) ait + ωt+1 ∞ X wt+j Qj j=0 s=0 1 + Rt+s −(1−β)(1+Rt)ω i+c̄Pt, t which we finally write as i ωt+1 = β(1 + Rt)ωti + c̄Pt. (5) Applying the sd operator on both sides of the previous expression we get i sd(ωt+1 ) = β(1 + Rt)sd(ωti). P i Dividing by the mean (ωt+1 = (1/N ) i ωt+1) then we get: i ) sd(ωt+1 ωt+1 sd(ωti) ωt = β(1 + Rt) , ωt ωt+1 or that ωt i i CV (ωt+1) = CV (ωt )β(1 + Rt) . ωt+1 It follows from Eq. (5) that β(1 + Rt) ωt ωt+1 =1− c̄Pt . ωt+1 We are now ready to state the main results of this section. i R1. If k0 < k∗ and c̄ = 0, then CV (ωt+1 ) = CV (ωti) for all t. i ) ≥ CV (ωti) R2. If k0 < k∗ and c̄ > 0, then CV (ωt+1 for all t. The reason is that over such a transition Pt is always negative, and so β(1 + Rt)ωt/ωt+1 is always larger than 1. We conclude that in the NMG there are non trivial dynamics for inequality in life-time wealth, which are either invariant over time, or monotone. There is an interesting second twist to the results above, which has to do with the desirability of policies aiming at reducing inequality. First, the CE with the representative agent is efficient (as is the CE with heterogeneous agents), and both the steady state and the transition toward it are independent of the initial distribution of wealth. This means that in a competitive equilibrium no faster/slower growth is possible by implementing a redistribution of assets, and that no redistribution of assets will ever have any impact on the steady state level of capital (hence output), consumption, etc. in per capita terms. A first implication of this fact is that any policy intervention will, at the very least, create inefficiencies. Hence, the model captures a non favorable trade off between equality and efficiency: less inequality is possible, but at the cost of reducing economic efficiency. A second implication of this fact is that there are many wealth distributions that are compatible with steady state. In oder words, the NMG has no prescription regarding inequality at the steady state. In the current version of the model the prediction is that either inequality will remain constant over time, or it will change monotonically (think of what would happen if the economy was decreasing toward the steady state). This essentially means that, according to the model, current levels of inequality are a simple projection of initial levels of inequality, but nothing else. We complement the preceding results with two additional implications of the model. R4. Consider two economies A and B identical in all respects, except that CVA(ω0i ) > CVB (ω0i ). If c̄ > 0 then CVA(ωti) > CVB (ωti) for all t. R5. Consider two economies A and B identical in all respects, except that K0A < K0B < K ∗. If c̄ > 0 then CVA(ωti) > CVB (ωti) for all t. 5. Final thoughts The results we have introduced so far look directly at life-time wealth. Even if the results are interesting, it is difficult to empirically observe ω, so it is difficult to test our theory of competitive inequality dynamics. What we would need is to be able to develop a theory but with the focus on observable measures of wealth and/or income. Suppose we concentrate on assets ait (this is still difficult to measure, but a way simpler than ω!). Following the reasoning along the lines above, we would be able to show that ait+1 = β(1 + Rt)ait + Dt, ∀t ≥ 0, where Dt is again an amalgam of future wages and interest rates that are independent of i. Continuing with the same sort of reasoning we developed before, it is possible to show that Kt i i CV (at+1) = CV (at)β(1 + Rt) . Kt+1 Interestingly enough, one is also able to show that in the NMG, if K0 < K ∗, and it is not too far away from the steady state, then β(1 + Rt) KKt < 1 for all c̄ > 0 t+1 and c̄ ≤ 0. Hence, the conclusion is that close to the steady state, inequality in assets is decreasing over time during a transition from below. Finally, it is also interesting to think of the “agents” in the previous model as countries, or perhaps as “regions” in a country. An important conclusion under this interpretation is that, if there is a well integrated capital market, then the regions (or countries) will converge to its own steady state, and in the long run, there will be differences in wealth that will persist forever. That is, in the long run there will be poor and rich regions/countries. Think for a moment of the possibility of closing down the asset market, so that each region is left in isolation. Then the model looks like a collection of N different regions. The initial differences in assets are simply different initial conditions for capital, but the NMG predicts that there is a unique steady state. This means that in the long run, all economies (regions/countries) would be identical! simply by closing down the capital market. The intriguing issue is, therefore, Why don’t we observe something like this in actual economies? References Here are a few references that may of your interest. 1. Alvarez, M. J., Dı́az, A.: Minimum consumption and transitional dynamics in wealth distribution. Journal of Monetary Economics Vol. 52, (2005). 2. Caselli, F.,Ventura, J.: A Representative consumer theory of distribution. The American Economic Review 90, p. 909 - 926, (2000). 3. Chatterjee, S.: Transitional dynamics and the distribution of wealth in a neoclassical growth model. Journal of Public Economics 54, p. 97 - 119, (1994). 4. Chatterjee, S., Ravikumar, B.: Minimum consumption requirements: theoretical and quantitative implications for growth and distribution. Macroeconomic Dynamics 3, p. 482 - 505, (1999). 5. Sorger, G.: Income and wealth distribution in a simple model of growth. Economic Theory 16, pp. 23-42, (2000). The starting paper is (3), and it was further developed by (2) (continuous time) and (4). The reference in (1) uses similar methods to ours, and conducts a quantitative investigation of the theory. The reference in (5) introduces endogenous leisure.