You are on page 1of 106

Economics 7343

Macroeconomic Theory I
Dietrich Vollrath

Fall • 2013


1 Basic Questions 2 Preliminaries 2.1 Gross Domestic Product . . . . . . . . . . . . 2.2 Investment and Accumulation . . . . . . . . . 2.3 Production, Wages, and the Return to Capital 2.4 The Consumption Problem . . . . . . . . . . . 2.5 Government Budgets . . . . . . . . . . . . . . 2.6 Open Economies . . . . . . . . . . . . . . . . 3 The Solow Model 3.1 Firms and Production . . . . . . . 3.2 Accumulation and Dynamics . . . 3.3 Implications of the Solow Model 3.4 Consumption and Welfare . . . . 3.5 Primitive Economic Growth . . . 3.5.1 The AK model . . . . . . . 3.5.2 Multiple Capital Types . . 3.6 The Open Economy . . . . . . . . 3.7 Nominal GDP and Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1 7 . 7 . 8 . 9 . 10 . 10 . 11 . . . . . . . . . . . . . . . . . . . . 13 13 18 21 30 33 33 36 41 44 47 47 50 54 62 64 69 69 77 81 86 86 95

4 Productivity: Growth and Fluctuations 4.1 Technological Progress . . . . . . . . . . 4.2 Dynamic Responses to Shocks . . . . . . 4.3 Productivity and Aggregate Fluctuations 4.4 Productivity and Long-Run Growth . . . 4.5 Cross-country Income Differences . . . . 5 Savings and the Supply of Capital 5.1 The Fisher Model . . . . . . . . . . . 5.2 The Over-lapping Generations Model 5.3 Infinitely-lived Savers . . . . . . . . 5.4 The Ramsey Model . . . . . . . . . . 5.4.1 The Centralized Solution . . . 5.4.2 The Decentralized Solution . . . . . . . . . . . . . i

. . . . . . .5. . . . . . . . .4. Technology. . . . . . . 98 5. . . . . . . . . . . . . . . . . . . . 102 . . . . . . . . . . . .3 Population. . and Growth .5 Fluctuations and Savings . . . .

1 plots real GDP per capita for four countries over a long time frame. as those may well influence the amount of spending on goods and services in the economy.which is what GDP measures. So we focus on GDP because that appears to be a good indicator of real living standards. Why do we care specifically about currently produced goods and services. as opposed to the market for a specific good or service. This means our main focus is not a transaction that simply move money between people (a loan from me to you). what do we see in the data regarding currently produced goods and services per person. So taking that focus as a given. or GDP per capita? Figure 1.C HAPTER 1 Basic Questions In macroeconomics we’re interested in aggregate economic activity. Second. we tended to study those almost exclusively. real GDP per capita in the U.000.000 in 1870. who combined various national statistical sources to produce a consistent series of real GDP per capita for a large sample of countries. This figure is on a log-scale.000 by 2009. so that the increases in vertical space correspond to percentage increases in real GDP per capita. and not financial transactions themselves? To me. That does not mean we will ignore financial transactions completely. inertia is powerful and once Simon Kuznets developed the national income product accounts. you can give me 1. or whatever term you like) is tied to real consumption of goods and services. there are two reasons. we’re particularly interested in transactions in which people buy currently produced goods and services . The data are from Maddison (2009).000 dollars. buy a car). First. This corresponds to an average growth 1 .000. which measures GDP. From about $3. but unless I actually buy goods or services with that money. That is. from 1870-2009. Of all those transactions. What do we see in this figure? First.S. a 10-fold increase. I haven’t changed my real living standard. or well-being. but rather on what you do with that loan (e. and more importantly. that for these four countries at least. Aggregate economic activity is made up of all the transactions that people and firms enter into in any period of time. there has been a steady increase in real GDP per capita over time. welfare (or utility.g. was about $30.

These experiences can be traced similarly across many other countries. Japan is distinctly poorer than the other countries up until about 1940. and the subsequent spike in activity from WWII. Second. Finally.1: Long-run real GDP per capita. Japan. B ASIC Q UESTIONS 30.000 Germany Japan 500 1870 1890 1910 1930 Year 1950 1970 1990 2010 Figure 1. 1870-2009 rate of about 1. 5.000 Real per capita GDP (1990 dollars) 10. In particular. which is why the lines are not perfectly straight.K. They lead to several big questions that we would like to address in macroeconomics.000 U. you can see the gigantic dip that is the Depression in the 1930’s in the UK and US. The US starts out at a similar level of living standards to the UK. Even outside of this dramatic experience. every country has distinct periods where real GDP per capita looks to be below or above trend. At that point. note that the growth of real GDP per capita is not consistent. 2. the US is distinctly better off than the rest in terms of GDP per worker.S. but after the Depression in the 1930’s. note that there are distinct level differences between countries over long periods. • What determines the long-run growth rate of real GDP per capita? • What determines the differences in levels of real GDP per capita between countries? 2 .7% per year.000 U. There are fluctuations year by year. The growth rates of the UK. The data does not run long enough to see any distinct drop from the financial crises starting in 2008.000 20.1. In Germany and Japan there is a sizable drop following WWII as these countries were devastated. and Germany are all similar to this over long stretches of time. in the post-WWII recovery period it then rapidly accelerates up to the level of the UK and Germany before leveling out at a similar living standard.

the acquisition of skills by the workforce. In addition. M V = P Y . meaning how many times that stock of money gets turned over through transactions in the economy. So when we study growth and development we will focus on describing the accumulation of those productive assets. What the QT says is that nominal GDP (P Y ) is proportional to the turnover of money 3 . Differences in levels of real GDP per capita will be due to differences in underlying propensities to save. euro) or by the amount of those units (i. and adopt is a vast literature that we’ll touch on as we go. as they study business cycles and the occurrences of things like the Great Depression and the financial crises of the recent years. the size of the money stock is irrelevant). and adopt new technologies. and P is the price of one of those units. learn.e. There appears to be good evidence that nominal values do drive real GDP over short time frames. indicates that they are highly correlated. V is the velocity of money. Figure 1. A simple plot of the series of real GDP and nominal GDP over time in the U.stocks of productive assets like capital. the skills of the workforce. When nominal GDP falls. and you can see the close co-movement of the two series. The second two questions are what you are more likely to associate with “macroeconomics”. What precisely drives the propensities to save. Y is units of real output.2 shows this plot. acquire skills. The amounts of capital per worker and skills per worker will be important determinants of the differences in levels between countries. M is the stock of money. so does real GDP. That is. That is. then we’ll leave behind the classical dichotomy. we will allow for the possibility that nominal things (money supplies) can influence real things (current production of goods and services). You may recall this from an intermediate macroeconomics class. When we discuss the second two big questions regarding business cycles. so in this course there will be a decided bias in the material towards these questions. This is the area that I am particularly interested in. real GDP per capita is not influenced by the unit of account (dollar.S. and vice versa. We’ll establish that the adoption of technologies is the fundamental driver of this growth rate.that real and nominal things are unrelated.• What causes the small fluctuations around the trend growth of real GDP per capita? • What causes the historically rare massive collapses of real GDP per capita? The first two questions are the general area of study of economic growth and development. Those things conspire to produce the steady trend growth rate of about 1. we will not get into a deep discussion of exactly how to measure this stock at the moment. but not of their long-run growth rates. technology again will be an important factor in determining differences across countries. Studying growth and development we generally accept the classical dichotomy .7% per year for a large number of countries. yen. and the adoption of new technologies. I find that the simplest way to approach this is through what is called the Quantity Theory of Money. and the general technology level. Over long periods of time we think that real living standards only depend on real things .

such as input supplies and productivity. 2 This second extreme is a common one to adopt in intermediate classes when developing the IS/LM. an increase in real output Y .1 We want to think about whether nominal factors (like the money supply or velocity of money) can influence real GDP. assume that Y is fixed in the short run by real factors. That is. or some combination of both? Two extremes are the easiest way to think about the possible answers. For this to be true. then any change in M V must lead to a change in Y only. specifically. 4 . in the short run. there must be some kind of market failures or nominal rigidities present in the economy. it must be that nominal GDP adjusts solely through an increase in P . The second extreme is to assume that the price P is fixed in the short run. as well as perfectly competitive markets for labor and capital will lead to perfect price flexibility and hence no nominal effects on output. But is the increase in P Y achieved through an increase in the price level P . Any fluctuations in Y must arise through real fluctuations in productivity or the supply of input factors. Fluctuations are then driven by nominal factors like M V . Changes in money supplies or velocity have no real effect. perfect competition and free entry by firms. Then if M V rises. In this case. United States 1959-2012 (M V ). B ASIC Q UESTIONS 15 Percent growth versus one year prior 10 5 0 −5 1960q1 Real GDP 1970q1 Nominal GDP 1980q1 1990q1 Quarter 2000q1 2010q1 Figure 1. and not by real factors such as productivity or factor supplies. First. Y . M V = P Y must be true at all times. then nominal GDP must rise.2: Growth rates of Nominal and Real GDP. by the definition of each of the terms. If M V rises for some reason.1. Why do prices take time to adjust? Why don’t firms enter to compete with firms that keep their prices too high?2 1 One note is that the QT is not actually a theory. We’ll go through a simple set of assumptions that lead to this outcome. it is an accounting description.

We will set up a few models that have imperfect markets and constraints on changing prices that lead to nominal demand having an impact on real output. and nominal GDP P Y would remain unchanged. then P should fall. then there is not anything that would necessarily imply that nominal GDP moves identically with real GDP. 5 . and the fluctuations in real GDP are only due to real shocks (changes in labor supply or productivity changes). it seems unlikely that nominal factors have no impact on the economy. However. if Y were to rise. so that P Y co-moves with Y so closely. we see in the figure that nominal and real GDP move very closely together. Note that figure 1. That is. if prices are fixed. If Y is fixed in the short run by real factors. So in the end. but M V remained fixed.2 tells us something about how to distinguish these cases. then this implies that prices P must be relatively fixed. then nominal shocks can influence the economy. However. If real flucatuations are always driving business cycles.


2) 7 . The inclusion of the depreciation term is why Yt is known as “Gross” domestic product.1 Gross Domestic Product Gross domestic product (GDP) is the total value of all goods and services produced within the borders of a given country in a given period (e. δKt . This will be familiar from your intermediate class in macro.1) The first element is compensation. Kt . We assume each person earns wt in wages. Let’s start with the income breakdown. and Kt is the amount of physical capital in the economy. “Net” domestic product would be N DPt = Wt + rt Kt .C HAPTER 2 Preliminaries To begin we need to establish a few accounting identities and definitions that will be used extensively in the course. Here we define M Yt = Ct + It + Gt + Xt − Xt (2. An alternative way of breaking down GDP is the expenditure method. that breaks down in period t. δ represents the fraction of the capital stock. The final term. Other variables will be defined similarly. National product accounts tell us that GDP can be split up into Yt = Wt + rt Kt + δKt . and for us represents the total return to capital. We’ll denote aggregate GDP as Yt . Now. so aggregate wages are Wt = wt Nt . and the various breakdowns that this provides for aggregate output. (2. 2. The basis for the identities are the national income product accounts. then GDP per capita is denoted yt = Yt /Nt . If we have Nt people. with capital letters denoting aggregates and lower-case letters per-capita terms. or wages. The second component.g. there are various ways to break down GDP. is depreciation. rt Kt . rt is the return to each unit of capital. is called “operating surplus” in the national accounts. a year).

we can write ∆Kt+1 = F (Kt . Recalling the production definition of GDP. Nt ) (2. Namely. in your intermediate class.2 Investment and Accumulation One of the first things we’ll study is the Solow model (and an extension called the Ramsey model) which involves the decision about how much to consume (Ct ) and how much to invest (It ). When we study the Solow model. In the Solow model we’ll be attempting to solve this difference equation to find out how big the capital stock is at any given point in time.4) which tells us how capital tomorrow is determined by the capital stock today.6) which is a difference equation in Kt . The are related in the circular flow diagram that you might remember from intermediate macro.Kt and Nt . Nt ). we’ll begin by assuming that Gt = 0 and Xt − Xt = 0.5) which describes the economy-wide accumulation of capital. in various iterations. and M Xt − Xt is net exports.3) which says that GDP is a function (not surprisingly referred to as the “production function”) of Kt and Nt . will form the budget constraints in our optimization problems. Here we will specify the production of GDP as Yt = F (Kt . Nt ) − Ct − δKt (2. the capital stock and the population. GDP definitions. Finally. respectively. Yt = F (Kt . GDP. so we can write down the following ∆Kt+1 = Yt − Ct − δKt (2. 8 . Note that each of these three are accounting identities. we’ll make some simple assumptions about how Ct is determined. Using the expenditure definition of GDP. while the Ramsey model will make the decision regarding Ct the outcome of an explicit utility maximization problem. The only reason you’d spend money on investment goods is because you expect to get something in return.2. So one of the first things we’ll do is add the following dynamic equation to our arsenal ∆Kt+1 ≡ Kt+1 − Kt = It − δKt (2. They are simply three different ways of conceiving of one single object. future GDP that you can consume. Gt is government spending. we will have a production definition of the economy. The important point will be that GDP depends upon the factors of production . They hold by definition at all points in time. P RELIMINARIES where Ct is consumption spending. To M keep our lives simple. It is investment spending. Note that we do not specify individual value-added by firm. as well as our investment decision today. Rather we provide a single aggregate production function that sweeps up all these value-added’s together. for example. This. we know that It = Yt − Ct . might have been referred to as the “value-added” approach. 2.

To go further. Now.13) (2. one of the more important elements involved in our work is the production function. and the Return to Capital As noted.7) or if we scale up both inputs by z . Wages. Owners of capital earn it’s marginal product (similar to workers earning a wage equal to their marginal product).3 Production. Nt ). we know that if firms pay out wages in this manner. We’re going to use several assumptions regarding this function. all we have are a set of different ways of defining GDP Yt = F (Kt .10) (2. so their net return is FK − δ .9) Because of the constant returns to scale. The above. This means that the return to capital is rt = FK − δ . Nt ) = wt Nt + rt Kt + δKt = wt Nt + FK Kt = Ct + It = Ct + ∆Kt+1 + δKt (2.3.14) and recall that all the items on the right-hand side are equal to each other as well.15) where the depreciation term has canceled. Wages. take the derivative of the above with respect to z . for example. Wt = wt Nt = FN Nt .8) which. Nt ) = FK Kt + FN Nt (2.2. as well as assumptions regarding the competitiveness of production. then output scales up by exactly z as well. Essentially. We can divide GDP up into the total product of capital (FK Kt ) and the total product of labor (FN Nt ). though. to figure out more explicitly the terms Wt and rt . Let’s put what we know together. this means that zF (Kt . Production. we’ll assume that the function F is constant returns to scale.11) (2. then firms should be paying each unit of labor a wage equal to it’s marginal product. Nt ) = F (zKt . gives us another way of decomposing GDP. F (Kt . First. This enforces the same budget constraint as does the economy-wide accumulation equation in (2.12) (2. which yields F (Kt . gives you a constraint that 9 . We know. and the Return to Capital 2. (2. that Ct + ∆Kt+1 = wt Nt + rt Kt (2.6). if the economy is perfectly competitive. but their capital is depreciating at the rate δ . Specifically. zNt ) (2. in effect. then it must be that rt Kt + δKt = FK Kt .

we can add an additional constraint that dictates how the governments finances evolve. Their individual budget constraint will be ct + ∆at+1 = wt + rt at (2. If not. focusing on individual choices over consumption and accumulation using (2.18) which looks even more like our aggregate constraint. 2.2. The point is that when we go through our individual problem we’ll be able to use this to inform us about how aggregate consumption moves: provided our assumption about identical individuals holds.4 The Consumption Problem We’ll spend some time focusing on people’s decisions about what consumption should be.19) which is just our aggregate constraint above. focusing on the accumulation equation in (2. which for now consists only of physical capital. consider government spending.17) (2. Gt . The most important point is that whether we are doing the Solow model. at ) can be used for two things: consumption and changing your assets. then in aggregate we’d have Nt ct + Nt ∆at+1 Ct + ∆At+1 = = Nt wt + rt Nt at wt Nt + rt At (2. Tt . plus the amount owed on government debt. or doing the Ramsey model. we are using the same constraint. we’ll start adding back in the other elements of expenditure. P RELIMINARIES highlights how net income (wages plus returns to capital) is allocated to either consumption or an increase in the capital stock. 2. if we assume that the economy is closed. Note that this is really similar to the aggregate constraint above. which b . This leaves us with Ct + ∆Kt+1 = wt Nt + rt Kt (2. b Gt + (1 + rt )Bt = Bt+1 + Tt (2. Bt . For this. and there are Nt individuals who are all identical.6). so At = Kt . If this is the individual constraint.15). then individuals assets have to be equal to the total amount of assets in the economy. Finally. These are either taxes. must be equal to government revenues. we’d have more work to do.16) which says that your income (wages plus the return on your assets.20) This says that government spending. or has a real interest rate of rt 10 . First.5 Government Budgets After we’ve covered the consumption and investment decisions in detail.

This is Ft = At − Kt . For the moment. in an open economy. Once we do this.21) (2. total assets owned by citizens do not have to exactly equal the physical capital stock. Gross National Product is GN Pt = Wt + rt At + δAt (2. net exports and imports. Let’s define a new object. The distinction matters because. then citizens have assets less than total assets in the domestic economy. so they must own foreign assets (i. If Ft < 0. net foreign assets. and increasing bond holdings. let’s keep them identical.e. Your income is wages. and so some of the domestic capital stock must be owned by foreigners.27) which looks a lot like the income definition of GDP.2. Open Economies bonds that are due in the next period. there are three options for your net income: consumption. minus taxes. physical capital located in other countries). Bt+1 . Ft . we’ll need to distinguish GDP from GNP. except that now we have assets (At ) rather than the physical capital stock (Kt ).22) (2. (2.25) = Ct + ∆Kt+1 + δKt + Tt + ∆Bt+1 − = wt Nt + rt Kt + δKt which we can use to write b Bt − T t Ct + ∆Kt+1 + ∆Bt+1 = wt Nt + rt Kt + rt (2.24) (2. Now.6.23) (2. so GDP = GNP). all bonds are only one-period bonds. We know what defines GDP. At = Kt . increasing capital.6 Open Economies We introduced government. For now. returns to bond holdings. One other note is that I’ve implicitly assumed that assets earn an identical return of rt no matter where they are located. (In a closed economy. Going back to our GDP definitions. we have now that Yt = Ct + It + Gt b = Ct + It + Tt + ∆Bt+1 − rt Bt b rt Bt (2. 11 . returns to capital.28) If Ft > 0. so we aren’t going to concern ourselves with things like 10 versus 30 year bonds. so now let’s add the last element of national accounts. You could be more refined and have the return earned on domestic assets be different from that earned on foreign. 2.26) and we have an expanded version of the constraint facing the economy. by definition. then citizens have total assets higher than the total assets in the domestic economy (which is just the physical capital stock).

we get ∆At+1 ∆At+1 = = Yt + (rt + δ )Ft − Ct − Gt − δ (Ft + Kt ) Yt + rt Ft − Ct − Gt − It − δKt + It Yt − Ct − Gt − It + rt Ft Xt − M Xt (2. but actually has some meaning. then they are paying for them with imported goods or using the net payments they get from already owning assets in your country. by definition. ∆At+1 = GN Pt − Ct − Gt − δAt (2. Note.34) (2.29) (2.31) or GNP is just GDP adjusted by the return on net foreign assets. This. 12 . The term on the left is the capital account. subtract off all of the non-investment spending done on domestic goods (Ct and Gt ). If you are acquiring foreign assets (∆Ft+1 > 0).37) ∆At+1 − (It − δKt ) = ∆At+1 − ∆Kt+1 ∆Ft+1 = = + rt Ft M Xt − Xt + rt Ft which is a big mess. The accumulation of physical capital in the domestic economy takes place according to the same mechanics: ∆Kt+1 = It − δKt . now that assets and physical capital don’t need to be identical for our citizens. It tells us that you have to pay for what you get. again. and noting that ∆At+1 = ∆Ft+1 − ∆Kt+1 . We haven’t done any economics yet.2. along with the definitions of asset and capital accumulation.35) (2. or with the earnings on your existing net foreign assets. Using the prior decomposition of GNP. is simply accounting.32) which says that you take your total income (GNP). then you must be paying for them with either exports (in an amount greater than you import). The net result is the change in your asset holdings. the change in net foreign asset holdings. P RELIMINARIES With this definition we get GN Pt = = = Wt + (rt + δ )(Ft + Kt ) Wt + (rt + δ )Kt + (rt + δ )Ft Yt + (rt + δ )Ft (2. If foreign countries are acquiring assets in your country on net (∆Ft+1 < 0). and subtract off the depreciation of your existing assets.30) (2. that this is simply another transformation of the accounting equations for GDP and GNP.33) (2.36) (2. This is net exports plus the payments on net foreign assets (which if Ft < 0 could be negative). However. we need an accumulation equation for assets as well. though. The right-hand side is called the current account.

we are assuming that there is perfect competition among firms in producing output. We will also assume that there is free entry into production. Within the Solow model we can (crudely) discuss the source of fluctuations in GDP around trend. and hence zero profits. and what we’re left with in our model is useful in explaining why GDP moves the way it does. as it turns out. The assumptions aren’t made because we think they are strictly true. By itself. Hopefully our assumptions are relatively unimportant. there will be no incentives to innovate. you can think of it as the most stripped down model of the macroeconomy that we can usefully put down on paper. We will assume that all firms in the economy produce an identical good. The mechanics of the Solow model sit at the heart of nearly every macroeconomic model.1 Firms and Production Let’s begin with the determination of output. but rather because we can’t possibly hope to model the economy precisely. prices will adjust instantly. and hence production.C HAPTER 3 The Solow Model In this chapter we build up a basic Solow model of the economy. This stripping down makes several heroic assumptions that we’ll slowly relax over the course of the class. so that the existing firms cannot establish any kind of cartel to gain market power. Without market power and profits. These are going to be very restrictive assumptions. and so there will be no way for nomianl shocks to generate fluctuations in the economy. The dynamic nature of the economy arises because the amount of savings done today will influence the capital stock. Without market power. That is. In short. there is no specialization or differentiation between firms. and we have to accept some simplifications to proceed. 3. the determination of the trend growth rate of GDP. and hence 13 . There is thus no possibility of an individual firm having market power. and the influence of taxation and government spending. Thus the split of current output between consumption and savings is crucial. in the future. openness to foreign capital flows.

you get less than double the output. of which there are n total. Consider multiplying each factor of production by some factor z . and X1 is equal to capital while X2 is equal to labor. From a general perspective... Y ..Xn ) . zX2 ... The marginal product of a factor of production is simply the derivative of the production function with respect to that factor. is produced by some combination of the factors of production denoted by X . X2 . Increasing returns implies that you can more than double output by doubling inputs. But we need not necessarily be that restrictive. In particular. X2 . The function F (. The first property to think about is returns to scale. and this function is constant returns to scale (see the following boxed section for a definition of returns to scale)..5) Decreasing returns means.. T HE S OLOW M ODEL no deliberate productivity improvements.. Returns to scale refers to how this scaling affects output.Xn ) (3. What combinations of factors of production produce an identical level 14 .2) so that output.. we’ll assume constant returns.3..3) (3. . . This is related to the concept of an isoquant. Ni ) (3.. that if you double inputs used. Write a production function in general as Y = F (X1 .4) (3. ...1) where Ki is the capital stock used by firm i and Ni is the labor used by firm i. Decreasing Returns : F (zX1 . practically. and constant returns means output exactly doubles. ∂Xi (3. there are several properties and terms worth understanding in more detail. zXn ) < zY Increasing Returns : F (zX1 . . we’ll use n = 2.. . (3. M P Xi = ∂F (X1 . zXn ) = zY. we can think about substitution between factors of production. zXn ) > zY Constant Returns : F (zX1 . . zX2 ..6) Finally.. Each firm is assumed to produce output according to the following production function Yi = F (Ki . but this will form a good starting point. zX2 . Production Function Properties One of the primary elements of any macro model is the production function. We’ll relax those eventually. Typically.) is identical for each firm. In almost every case.

Firms and Production of output? This is most simply seen with a two-factor production function: Y = F (X1 . I would have to add a lot of factor 1 to replace the loss of a little of factor 2. Ni ) > 0. In addition. Ni ) > 0 and FKK (Ki . (3. the marginal product actually comes close to zero. K →0 (3. Ni ) = FN K (Ki . Ni ) − RKi − wNi (3. and that firms have no monopsony power. Mathematically.7) which follows from the Implicit Function Theorem. If we add labor. but at a decreasing rate. If the MRTS is small (in absolute value). Without inputs a firm can produce no outputs.10) (3. we are assuming that the production function is concave in Ki . This says that the isoquant is downward sloping (the negative slope) and depends on the relative marginal productivity of the two inputs.8) An increase in Ki increases output.1. Ni ) > 0 and FN N (Ki . Similarly for labor we assume that FN (Ki . 0) = 0. Specifically K →∞ lim FK (Ki . The marginal rate of technical substitution between input 1 and 2 is. They imply that when there is almost no capital in the firm. but as they acquire large amounts of capital. Ni ) = ∞.12) where R is the rental rate that firms have to pay for capital and w is the wage they have to pay for labor. Given all this. for example. our firms will be trying to maximize profits (πi ) πi = F (Ki . we assume that that F (0. A final set of assumptions we make regarding the production function is how it behaves as it approaches extreme values. and we also assume that FKN (Ki .3. 15 .11) These are known as the Inada (1964) conditions. then the marginal product of capital is higher. holding output constant M RT S1. X2 ). Ni ) ≤ 0. Ni ) ≤ 0 (3. The production function is presumed to exhibit diminishing returns to capital: FK (Ki . We are thus assuming that factor markets are perfectly competitive.2 = ∂X2 − M P X1 = ∂X1 M P X2 (3. the marginal gain in output from a small amount of capital is infinite.9) This last assumption says that the marginal product of one factor is raised by the addition of the other. then the marginal productivity of factor 2 is very large relative to 1. So to keep output constant. Ni ) = 0 and lim FK (Ki .

or profits? Start with labor and capital. the share of income left to be paid out as profits is exactly zero (here we are talking about economic profits. (3. Their shares of income can be written as KR Y Nw Y = = K × MPK KαK α−1 N 1−α = =α Y Y N × MPN (1 − α)K α N −α = = 1 − α. Let the aggregate stock of capital be K . We can consider the distribution of income between labor. Ni ) = w.15) (3. As these shares add up to one. That is.13) Given that every firm has an identical production function. and faces an identical R and w. and free entry. while the remaining 1 − α of income will be paid out as wages. firms will have the following first-order conditions FK (Ki . This fits. This should drive economic profits down to zero. There is nothing special about 16 . what fraction of total income is made up of wages. not accounting profits). We have started with an assumption of perfect competition between firms. T HE S OLOW M ODEL To maximize profits. giving us a way of working out exactly what the return to capital and wage will be in an economy with the CD production function in each firm (and perfect competition in output markets and factor markets).16) Firms will set M P K = R. which has the following form: Y = K α N 1−α (3. Exactly how much capital and labor is that? That depends on the supplies of those factors of production.18) What this shows is that exactly α of total income will be in the form of payments to capital. The marginal product of each factor in the CD is MPK MPN = αK α−1 N 1−α = (1 − α)K α N −α . every firm will make an identical choice regarding the amount of Ki and Ni to hire in. and M P N = w. You can confirm easily that the Cobb-Douglas (CD) has constant returns to scale.17) (3. (3. Ni ) = R and FN (Ki . and do not respond at all to w or R. and profits. capital. meaning their supply curves are vertical. The Cobb-Douglas Production Function and Income Distribution The most common form of the production function that is used is the Cobb-Douglas. 1). We assume that both of these are supplied inelastically.3.14) where α ∈ (0. or returns to capital. Y Y (3. and the aggregate stock of labor be N .

This stability in labor’s share is one of Kaldor’s (1957) stylized facts regarding growth. N ). These can obviously be rearranged to show that Ki = K/M and Ni = N/M are the amounts of capital and labor used by each firm. Once we do this and sum over the M firms. Let there be M firms in the economy.S. With free entry.20) Here we can plug into the firm-level production function with what we know about Ki and Ni to reduce this to M M M Y = i=1 F (Ki . Firms and Production the CD that ensures this is true.19) or the total demand equals total supply for both factors of production. Cobb and Douglas developed this production function form precisely to match U. That does not mean labor’s share is identical in every country . (3. we can say something about aggregate output.3. even a single firm will act as a price-taker. As we assumed that each firm produces an indentical good. the M will cancel and we’re left with the aggregate production function Y = F (K. although there is recent evidence from Loukas Karabarbounis and Brent Neiman (2013) that labor’s share is actually declining in several countries. Regardless of the actual level of output. This means that we can easily assume that M = 1 if that is convenient. But there is not tendency for poor countries to have high labor shares (or vice versa). With M firms. Doug Gollin (2002) found that there is no tendency for labor’s share to change as countries get richer. data that showed labor’s share of output to hold steady at around 2/3 for several decades.50 and 0. N/M ) = i=1 F (K. . N ) = F (K. we can write aggregate output as M M Y = i=1 Yi = i=1 F (Ki . exactly α of output is used to pay for capital services.21) In the above series of equations we can pull the 1/M out of the production function due to the constant returns to scale property of F (.1. Ni ) = i=1 F (K/M. it must be that Ki M = K and Ni M = N. (3. N ). What the CD function adds is the fixed shares α and 1 − α for capital and labor.. M ( varies between about 0. M can be any positive finite number. Knowing this. and the rest for labor. Ni ). Across countries.).22) 17 .85. One issue with assuming perfect competition is that we can’t actually pin down the number of firms precisely. (3.

and not an accounting identity. What determines the level of investment? We will assume that it is described as follows It = St . Nt ).25) . and is unavoidable. there is no extra inflow of savings from abroad (or outflow of home savings to other countries).23) where It is the amount of investment done in time t. Recall that all output is identical. So to describe the time path of GDP we need to describe the time paths of Kt and Nt .3. The second term above is depreciation. the only possible source of fluctuations or growth are the stocks of capital and labor. 18 (3. Let’s begin with capital. We will describe it’s dynamics as follows Kt+1 − Kt ≡ ∆Kt+1 = It − δKt (3. If they fluctuate or grow. Having those assumptions gives us a nice aggregate production function to work with. This assumes two major things. perfect competition in output markets and factor markets. that all saved output is costlessly and perfectly translated into investment goods. Some of the homogenous output is “invested”.24) or investment is exactly equal to the amount of output saved (St ). 3. If we know those we can plug through the production function and get Yt . so there is no sense that we are producing machine tools (an investment good) that is differentiated from food (a consumption good). it is very useful to understand exactly what is going on behind that aggregate production function. We assume that it is a constant fraction of output (s) St = sYt . (3. meaning simply that it is set aside. We assume that this rate is unaffected by anything. but as mentioned above eliminates several potentially interesting aspects of the economy. Secondly.2 Accumulation and Dynamics The aggregate production function says that output at time t is Yt = F (Kt . then the financial sector is doing this without charging any fees and without any loss of output. there are a whole series of assumptions built into the aggregate production function. T HE S OLOW M ODEL Often times we’ll jump right to the aggregate production function. First. This economy is closed to capital flows. However. This is simply the amount of output that is set aside for use as capital in period t + 1. In particular. This then leads to another assumption regarding the level of savings. If we want to imagine a financial sector taking in savings and loaning these out to firms. just assuming that Y = F (K. then so will output. Note that this is an assumption. In the baseline Solow model. N ) exists. As you can see. Note that we already have some answers here. A fracton δ of the existing stock of capital Kt just falls apart or breaks down every period.

Accumulation and Dynamics In other words.2. rather than simply aggregate GDP. If we put together all of our various pieces.1 Nt = F (kt . and our “big questions” are all about living standards.3. this leads us to yt = F (Kt . In the final equation I’ve made one more notational adjustment. In the second equation I’ve simply plugged in the production function. . (3. Lower-case yt refers to per-capita GDP to save us from having to write Yt /Nt over and over again. n ∆Nt+1 = nNt . we can find Yt from the production function. the proportion saved is inelastic with respect to the rate of return on savings. which will be part of our motivation to expand on the model. but fails on others. I’ve replaced this with the simpler f (kt ).. We assume that this grows at an exogenous rate. and one that describes how the labor force changes over time. because the function F (. kt . we are generally interested in per capita GDP. Capital letters refer to aggregate values. In this economy it is always the case that a fraction s of output is set aside to be saved. but one that can be relaxed for more realism. Nt ) − δKt . this describes capital growth as a function of capital itself. where we are now assuming that the labor force (Nt ) is identical to the population. In the fourth equation. with that bothersome 1 hanging around. The third equation uses the constant returns to scale of F (.28) Let’s walk though this chain of equations.27) And that’s it. 1) = f (kt ). In particular. Per capita GDP is simply Yt /Nt . The other moving part in this economy is the labor force. Per capita GDP is a good proxy of living standards. The first is just a notational modification. This is a non-linear difference equation. This standard will be maintained throughout the class. We have an equation describing how capital changes over time. This form of the production function is often called the intensive form. . Nt ) Yt = =F Nt Nt Kt . (3. It will turn out that the model does well in describing several economic facts we observe. The Solow model has several implications that we can compare to the data. Knowing Kt and Nt at any point in time.) is non-linear in capital. This is a relatively minor assumption. we are left with the following difference equation for capital ∆Kt+1 = sF (Kt . meaning it is written in per 19 . capital per capita. 1) over and over again.) to divide through by Nt . or anything else. while lower case letters refer to per-capita values. Rather than write F (kt .26) As one can see. I’ve replaced Kt /Nt with it’s per-capita notation. Regardless. (3. To see these implications we will take the Solow model and write it in a more concise form. the level of income..

33) If we simply multiply through by kt . We’re using this discrete-time approximation to keep the notation consistent. closed to foreign capital flows. and constant population growth. s kt (3. as it will be useful as we continue in the course. Nt )/Nt s −δ−n Kt /Nt f (kt ) − δ − n. and it constitutes our complete description of the economy. and the marginal product going to zero as kt goes to infinity. and from that find GDP per capita from the production function. with the marginal product of capital f ′ (.31) (3. then we have ∆kt+1 = sf (kt ) − (δ + n)kt . and we can do that given our equations for capital and labor in (3. The growth rate of kt can be approximated as follows. kt Kt Nt (3.) going to infinity as kt goes to zero. Strictly speaking.30) and we have ∆kt+1 kt = = = sF (Kt . (3. The function f (.3. we can find the level of capital per worker at any period t. . a costless financial sector.27). for an economy with perfect competition among firms producing homogenous output.).27) to plug into (3.. T HE S OLOW M ODEL capita terms. inelastically supplied labor and capital. Hence we need to describe the determinants of capital per worker.26) and (3. Output per capita depends on capital per capita (I’ll often also refer to this as capital per worker to avoid the awkward phrase).30) Note that this is an approximation.) also satisfies the Inada conditions. That is. The function f (.26) and (3. With it.32) (3. so long as the growth rate of population is relatively small (and historically the yearly growth rate less than 2% in all but the most rapidly expanding population) this approximation is essentially an equality. perfect factor markets.34) The above equation is the “Solow equation”. the growth rate of capital per worker is not exactly the growth rate of K minus the growth rate of N . a constant savings rate. It describes how capital per worker evolves over time. but at a decreasing rate. 20 . Specifically f ′ (kt ) > 0 and f ′′ (kt ) < 0. something you’ll confirm for yourself in the homework problems. However. Nt ) −δ−n Kt F (Kt . (3.29) which says that output per capita is rising in capital per capita. Now use equations (3.) inherits all of the properties of F (. ∆Kt+1 ∆Nt+1 ∆kt+1 ≈ − . this equation is all we need to describe how GDP evolves over time.

we can solve for these steady states. sf (k ∗ ). investment exceeds depreciation and the capital stock increases.35) One steady state is where k ∗ = 0. At any capital stock less than k ∗ . once reached. so that over time kt approaches k ∗ . is exactly equal to the amount of capital that is depreciating.3. First we need to establish that there are fixed levels of capital per worker than. and that kt+1 = kt .3 Implications of the Solow Model This model has specific predictions about how the economy will evolve over time. For kt > k ∗ . as well as predictions regarding the relationship of parameters (like the savings rate or the population growth rate) and output per capita.1: The Solow Steady State Note: The steady state is the level of capital. the economy never deviates from. Using (3. This follows from the nature of the Solow equation. the opposite occurs. They occur wherever ∆kt+1 = 0. 3. We can work through these to see how useful the Solow model will be. Because of these opposing effects. as sf (k ∗ ) = (n + δ )k ∗ . (n + δ)k ∗ .34). meaning that capital worker is not growing. If the economy has no capital. at which the amount invested. (3. then it cannot produce any output 21 . k ∗ . Implications of the Solow Model (n + δ )kt sf (kt ) ∆kt+1 > 0 k∗ ∆kt+1 < 0 kt Figure 3. which we denote k ∗ . Implication 1: The economy will eventually end up at a fixed level of capital per worker and stay there. the steady state is stable. We call these “steady states” of the Solow model.3.

The term (n + δ )kt is linear in kt . (3.34) against the value kt .37) Using this intensive form. Figure 3. Solow with the Cobb-Douglas Production Function Recall the Cobb-Douglas production function. (3.39) We can solve the steady state condition for k∗ = s n+δ 1/(1−α) (3. Hence. The term sf (kt ) is a concave function of kt . To see this. then what this figure indicates is that the economy will in fact move towards steady state. we can write the accumulation equation as kα ∆kt+1 =s t −δ−n kt kt (3. The intensive form of the Cobb-Douglas is α yt = f (kt ) = kt . or capital per worker is growing. and δ . note that for any kt < k ∗ . it is easiest to look at a diagram. If the economy is not in steady state. These curves cross where sf (kt ) = (n + δ )kt . The positive steady state is stable. which is precisely our defintion of the steady state. The second thing we need to establish is that the economy will head towards the steady state of its own accord. T HE S OLOW M ODEL to invest in new capital. it is the case that sf (kt ) < (n + δ )kt and so ∆kt+1 < 0. For kt > k ∗ . 1). as it is only a function of the fixed parameters s. and capital per worker is shrinking. no matter where the economy begins.41) From this it can easily be seen that output per person is not growing over time. we will always return to it. To see why. (3.38) and the steady state condition is that sk ∗α = (n + δ )k ∗ . given our assumptions that f ′ (kt ) > 0 and f ′′ (kt ) < 0.1 plots both terms on the right-hand side of equation (3. n. α 1−α Yt = Kt Nt (3. So that intersection tells us where k ∗ is found. it is the case that sf (kt ) > (n + δ )kt and so ∆kt+1 > 0. If we deviate from the steady state. the capital stock tends to move towards steady state. and it will stay at zero capital forever. On the other hand.3.40) and output per person in steady state is therefore y∗ = s n+δ α/(1−α) . there is some value k ∗ > 0 that solves the above equation as well.36) where α ∈ (0. 22 .

lowering the steady state.45) (3. output in period t + 1 is Yt+1 = F (Kt+1 . Implication 2: The steady state level of capital per worker. So in steady state the aggregate economy is growing. shrinking if kt > k ∗ . then the concave savings function shifts upwards. k∗ (3. 23 .3. we know that output per capita will be growing so long as kt < k ∗ . and income per capita.1 one can see that if s rises. This can be seen several ways. in equation (3. indicating a higher steady state value of k ∗ . n+δ f (k ∗ ) (3. Output per capita hits a steady state just the same as capital per worker. aggregate output is growing at the rate of population growth. Once the economy is in steady state. This doesn’t mean that output (in aggregate) is not growing in steady state.43) If s rises. (1 + n)Nt ) = (1 + n)F (Kt . An increase in n rotates the depreciation line up. and other OECD countries there have been remarkably stable growth rates in output per capita over long periods of time. and constant if kt = k ∗ . this derivative is positive if f (k ∗ ) > f ′ (k ∗ ).43). ∂k ∗ f (k ∗ )2 This derivative is positive if f (k ∗ ) > f ′ (k ∗ )k ∗ . n. There’s clearly something missing from the Solow model. and the intersection with the depreciation line shifts to the right. particularly that it is concave. The key step here is to note that Kt+1 = (1 + n)Kt in steady state. then the right-hand side must rise as well. then to balance that equation k ∗ must increase.3. We know that in the U. Nt ) = (1 + n)Yt . and as kt = Kt /Nt . then it must be that k ∗ is lower. Why? We know that kt is constant in steady state. (3. the only way for kt to stay constant is for Kt to grow at exactly the same rate as Nt . but per capita output is stuck.S. Is this true? It is given what we’ve assumed about the production function. From figure 3. the right-hand side is increasing with k ∗ . Nt+1 ) = F ((1 + n)Kt . Re-arranging. Similar logic shows that if n is higher. So our model cannot be precisely right at this point.42) In other words. is positively related to s and negatively to n. Therefore.44) which says that the average product of capital is greater than the marginal product of capital. the steady state condition in the Solow model can be written as s k∗ = . So what must happen to k ∗ in order to make the right-hand side increase? Take the derivative of the RHS with respect to k ∗ and you’ll have f (k ∗ ) − f ′ (k ∗ )k ∗ ∂k ∗ /f (k ∗ ) = . If you examine the steady state for the Cobb-Douglas situation. So if s is higher. Implications of the Solow Model As output per capita is yt = f (kt ). More formally. you’ll see this immediately. This version of the Solow model says that eventually growth in output per capita should have run down to zero.

100000 Real GDP per worker. f (kt ) ∆kt+1 =s − n − δ. To see what this graph should look like.000 20. Countries that have a higher share of output saved (specifically. spent on investment goods) tend to have higher income per capita. Remember that these predictions are about steady states. T HE S OLOW M ODEL These predictions can be evaluated against evidence from across countries.000 NOR SGP BEL NLD USA IRL AUS AUT HKG GBR FRA ITA ISL FIN CAN DNK CHEJPN PRI SWE GRC TWN TTO ISR ESP GNQ NZL KOR BRB CYP PRT TUR MEX IRN CHL MYS ARG CRI GAB URY DOM BWA PANZAF ROM VEN MUS JAM COL BRA SLV GTM PER ECU DZA EGY NAM THA SYR FJI CHN HND LKA MAR PRY IND BOL IDN PHL PAK PNG NGA NIC ZMB COG CMR MRT SEN MLI CIV GMB HTI TCD LSO BEN BGD NPL GHA UGA KEN TZA NER RWA BFA GNB COM GIN MDG TGO MOZ MWI CAF ETH BDI ZAR ZWE 0.000 10.2 and 3. while those with higher population growth rates tend to be poorer. start with the main Solow equation and divide through by kt to put things in terms of growth rates. and so describing differences in living standards between countries requires something more than just the savings rate and population growth rate. s.30 Investment share of GDP. log scale 50. If countries are all at least close to their steady states then the influence of s and n should be apparent. The relationships are not exact. 1988−2008 0.05 0.25 0. We’ll need to upgrade the Solow model to capture more accurately these living standards.46) Now graph the growth rate against the level of kt .2: Relationship of savings s and income per capita From figures 3.20 0. kt kt (3.3 you can see that these rough relationships do hold in the data from the last 20 years.15 0.10 0.000 5. To see this implication. the economy will grow faster the farther away from steady state it starts. 2008. Implication 3: Out of steady state. Another way of saying this is that a model that only has capital and labor is not sufficient to capture differences in output per capita across countries. first 24 .000 1.40 Figure 3.35 0.3.000 2.

03 Population growth rate.48) where the first step in each situation relies on L’Hopital’s rule. This figure identifies the stable steady-state of the Solow model as the point where capital per worker growth equals zero.02 0.46) with respect to kt and you have f ′ (kt )kt − f (kt ) ∂ ∆kt+1 /kt =s .47) (3.00 0. the growth rate of capital per worker shoots up to infinity. the growth rate of capital per worker becomes negative. Figure 4.000 2.000 TUR MEX IRN CHL MYS ARG CRI GAB BWA PAN VEN MUS ZAF DOM JAM COL BRA SLV PER GTM EGY ECU NAM DZA JOR THA SYR FJI CHN HND LKA MAR CPVPRY IDNIND BOL PHL PAK PNG NGA NIC ZMB CMR COG SENMRT GMB HTI MLICIV TCD LSO BGD GHA KEN BEN UGA NPL TZA NER RWA GNB COM BFA GIN MDG TGO MOZ MWI CAF ETH BDI ZAR ZWE −0. n.000 5.3. given that f ′ (kt ) < f (kt )/kt as we established earlier.01 0.000 1. must be negative. As kt goes to infinity. What we see is that as kt goes to zero. log scale 50. More interestingly.49) The sign of this term depends on f ′ (kt )kt − f (kt ). 2 ∂kt kt (3. the figure shows that the farther away from steady state is capital per worker. 1988−2008 0. the higher is the absolute value of the growth rate. So when kt is close to zero. What happens between these extremes? For that we need to know how the growth rate of kt responds to a change in kt .04 Figure 3.000 NOR USA BEL NLD ISL AUS AUT IRL HKG SWE GBR FRA ITA FIN CAN DNK CHE PRI JPN GRC TWN TTO ESP KOR NZL BRB PRT ROM URY SGP ISR GNQ CYP 20. the growth rate approaches infinity. Take the derivative of (3.000 10. and as kt increases the growth rate falls.01 0. 2008.1 plots this relationship. Implications of the Solow Model 100000 Real GDP per worker. until as kt appreoaches infinity the growth rate approaches −n − δ .3. A country with kt very close to zero should grow 25 .3: Relationship of population growth n and income per capita consider how the growth rate acts as kt goes to extreme values of zero and infinity: ∆kt+1 k→0 kt ∆kt+1 lim k→∞ kt lim f ′ (kt ) −δ−n=∞ 1 f ′ (kt ) = s − δ − n = −δ − n 1 = s (3.

There countries do not conform to the predictions of the simple Solow model.6 shows the same kind of plot for all OECD countries starting in 1960. many of them African. Korea (KOR). Do we see this in the data? Yes and no. Going back to our stylized figure 4. it’s quite clear that income per capita grows faster the poorer a country starts out.4: Growth Rates and kt faster than an economy with kt close to the steady state. which would expect them to be growing as fast as Taiwain (TWN). Here.. and it’s failure. Overall.7 when all countries in the world are included the prediction falters.S. Australia. Asian. Given that output per capita is simply a function of kt . There is a mass of poor countries with low growth rates. in figure 3. note that what this implies is that eventually every country 26 . and Central American. it most certainly does. For a subset of currently rich countries. and Botswana (BWA).3. the U. The importance of this relationship.1. Figure 3. However. as the Solow model would predict. Japan. and New Zealand. Figure 3. Again. Canada. An economy with kt higher than the steady state level should actually be shrinking. T HE S OLOW M ODEL ∆kt+1 /kt ∆kt+1 kt >0 k∗ ∆kt+1 kt <0 kt 0 −(δ + n) Figure 3.5 plots growth rates against initial income levels for major European countries. it predicts a negative relationship between growth rates in capital per worker and the level of kt . has to do with what we call convergence. it predicts that growth in output per capita should be declining in the level of yt . the downward slope is obvious.

000 15. 1960-2008.016 0. 1870 3.000 20.000 10. 1870-2008 0.022 GRC FIN NOR SWE ESP ITA 0. 1960−2008 0. Implications of the Solow Model 0.018 0.014 PRT IRL CAN AUT DNK FRA CHE GER USA NLD BEL AUS GBR NZL 0.03 0.04 0.500 Figure 3.500 2.000 GDP per worker.3. 1960 30.000 Figure 3.024 Growth rate.6: Growth rates and income levels.01 0.500 Per capita GDP.000 3.000 5.01 0.000 25.020 0.5: Growth rates and income levels.05 0.01 1. OECD only 27 .04 Growth rate.03 0.04 TUR JPN GRC IRL PRT ESP FIN ITA AUT BEL NOR FRA ISL SWE GBR DNK AUS NLD USA CAN CHE NZL KOR 0.02 CHL ISR 0.000 2.012 500 1.000 1.026 JPN 0. 1870−−2008 0.3.000 40.000 MEX 35.

they both then experience a period of rapid growth (the slopes are much higher relative to the UK and US) until about 1970.. In particular. in that it would have predicted that growth rates fell 28 . Germany and Japan both experienced bombing during World War II that wiped out most of their existing capital stock. predicting just such a catch-up.000 40. and see what happens afterwards. By that point they have caught back up to the UK.02 ZAR 1. both Germany and Japan experience a distinct drop in income per capita immediately after the war. but at this point we just note that our model is incomplete. In this sense. where capital was not destroyed) in the years after the war.3. U. this convergence is not occurring.. An alternative way of seeing the connection of growth rates and income levels is to look at timeseries from different countries. as we saw before. We’ll explore some extensions of the Solow model that may help explain why that is. every economy should end up at k ∗ .S..000 25. 1960-2008.02 NLD USA CAN CHE NZL 0. Figure 3.. we’ll look at countries that had a distinct drop in their kt level. Eventually. It fails. However. Poor countries grow faster than rich countries. 1960−2008 0.00 −0.000 GDP per worker.K. and remain only slightly behind the U.000 20.000 35.000 15. This is convergence in action.S. as they were prior to the war.000 30.06 BWA CHN TWN ROM KOR HKG SGP THA MYS JPN LKA GRC EGY TUR CYP IND IRL IDN MAR PRT ESP FIN ITA AUT LSO TTO PAN NOR PNG FRA BEL MUS PAK ISRPRI DOM CPV ISL MRT SWE GBR DNK GNB COG TZA GAB CHL AUS MOZ MWI BRB COL URY BRA MLI PHL SYR GTM IRN UGA ECU BFA GHA NPL MEX ARG TCD ETH GMB CMR PRY FJI NAM BEN BGD SLV ZAF CRI PER HND JOR JAM BDI CIV NGABOL RWA KEN ZMB COM SEN DZA VEN MDG TGO GIN NER HTI CAF NIC ZWE 0. As can be see.000 10. however. Germany.000 5. According to the Solow model.7: Growth rates and income levels. and Japan from 1870 to the present. This is what we appear to see in the OECD. and so they are catching up.S. all countries should converge to a similar steady state.04 Growth rate. with poor countries catching up. T HE S OLOW M ODEL 0. as their growth rates are too low to catch-up with the rich countries. However. this means that they should have grown more quickly than other countries (in particular the U. the Solow model performs very well.8 plots income per capita for the U. for many of the poorest countries in the world. which means they have identical levels of income per capita. 1960 Figure 3.

29 .51) (3.000 Germany Japan 500 1870 1890 1910 1930 Year 1950 1970 1990 2010 Figure 3.S.52) f ′ (k ) = The second condition shows us that the return on capital in the economy is equal to the marginal product of capital per worker. during that period. This was one of the motivating factors for Solow’s original article. (3. Living standards t t depend on the level. Implications of the Solow Model 30.000 U. Economists had been wondering why the return to capital did get driven down to zero as rich countries accumulated larger and larger stocks of capital. We’ll address that in the next section.K. select countries to zero eventually. 1 This also highlights the importance of distinguishing between growth rates of y and levels of y . N ) − RK − wN = N (f (k ) − Rk − w) . 2. (3. We know already that in steady state k will be constant at k ∗ . let’s modify slightly the firm-level problem. f ′ (k ). To see what happends to the return on capital.1 Implication 4: The return on capital eventually ends up at a fixed level and stays there. The last implication of the Solow model worth emphasizing is how the return on capital acts over time. 5.3.000 20.3. This gives first-order conditions of f (k ) − Rk − w = 0 R.S.8: Income per capita over time. not the growth rate. but it would have been demonstrably better from a material standpoint to have lived in the U. The representative firm (recall that with perfect competition and free entry we can easily just work with one firm) has total profits of π = F (K.000 Real per capita GDP (1990 dollars) 10.000 U. Germany and Japan had high growth from 1950-1970.50) Maximize this alternative way of writing profits over N and k .

the capital stock in t + 1 would be zero. However.4 Consumption and Welfare Up to this point we haven’t actually said anything about welfare. In steady state the influence of capital accumulation on R just offsets the influence of adding labor. However. The key is that FKN > 0. if s goes up. This much follows directly from our prior work. and no government sector to speak of at this point. 3. With our assumption of a constant savings rate. it could even consume all of the existing capital stock at time t. Without an explicit utility function. then it could simply consume all of the output in time t. the marginal product of capital actually goes up. and there would be no output at all. as less output is left aside to consume. then consumption goes down. and does not tend to decline to zero. trying to maximize their utility. consumption must reach a steady state just the same as output per capita. and remain constant thereafter. In steady state. But if the economy did that. no trade. That means that there may be some level of s that delivers the maximum amount of consumption.54) . The Solow model doesn’t have any optimizing individuals in it. If s goes up. kt . But maximum over what time period? If the economy wants to maximize consumption in period t. we don’t know whether that would be worth it or not. and therefore R stays constant even though in aggregate we keep accumulating K . or provided any way of measuring it. (3. By accumulating more capital we drive down the marginal product of capital. it must be that ct = (1 − s)yt . then output per capita rises. But we can think about an alternative.53) That is. We can examine consumption. To see this. though. raising R. meaning higher consumption. s.3. which in our future work will be the item that individuals care about in their utility function. This means that as we add labor. lowering R. So therefore. note that consumption doesn’t share the same striclty positive relationship with the savings rate. consumption per capita is just a fixed fraction of output per capita. as recall that we’ve assumed capital is just unconsumed output. which is to ask what is the maximum level of consumption that the economy can sustain in steady state. In fact. consumption is c∗ = (1 − s)f (k ∗ ) = f (k ∗ ) − sf (k ∗ ) = f (k ∗ ) − (n + δ )k ∗ 30 (3. note that s has two effects on consumption. and no consumption. That means that R is constant in steady state. T HE S OLOW M ODEL and therefore the marginal product of capital per worker will be constant. so there really isn’t any direct way to think about welfare. yt . that output per capita has.

(3.not s times the production function.9: The golden rule capital stock kt where the last equality follows from the Solow equation in steady state. The top curve is the production function . The Cobb-Douglas Golden Rule If we presume that the production function is Cobb-Douglas. Maximize c∗ over k ∗ to find the steady state level of capital per worker that yields the highest consumption in steady state. The difference between these curves is how much we can consume. The maximum amount of consumption comes where the production function is tangent to the depreciation line. The depreciation line shows how much we need to be saving in steady state to keep the capital per worker constant. The first-order condition is f ′ (k ∗ ) = n + δ.9. consider figure 3.4. then how do we find the Golden Rule savings rate and consumption level? 31 . Consumption and Welfare f ′ (k G ) f (kt ) (δ + n)kt cG kG Figure 3.55) This says to set the marginal product of capital equal to population growth plus the depreciation rate.3. To see what is going on.

59) How do we achieve this golden rule capital stock level? We have to set the savings rate so that our steady state capital stock per person is exactly equal to k G .58) which can be solved for the golden rule level of the capital stock kG = α δ+n 1/(1−α) . or “Golden Rule” steady state level of capital per worker as Edmund Phelps termed it. this is α yt = kt (3. (3. The final point to make regarding the Golden Rule is that while it is sustainable in the long run. if the savings rate is set just right.56) and we can write consumption in steady state as c∗ = k ∗α − (δ + n)k ∗ . (3. s n+δ α/(1−α) and maximized this with respect to s we would get exactly There is thus a single k G .57) Maximizing consumption over k ∗ yields the first order condition αk ∗(α−1) = δ + n (3. that provides the maximum level of consumption in steady state.60) Comparing this to the equation for k G shows that we should set the savings rate to exactly s = α in order to reach the Golden Rule. were to destroy part of the capital stock. (3. If a natural disaster. The steady state level of capital per worker is. the economy can achieve the Golden Rule level of consumption. There must be some s such that the steady state k ∗ = k G . this is only true so long as there are no disturbances to the economy. k∗ = s δ+n 1/(1−α) . T HE S OLOW M ODEL α 1−α Let’s begin with a production function of Yt = Kt Nt . So is it possible for the economy to be at this Golden Rule steady state? It is. as we saw previously.3. c∗ = (1 − s) the same answer. and therefore yt < y G . then kt < k G . We’ve already established that k ∗ rises smoothly with s. But once the savings rate is below 32 . Note that if we looked at the steady state value for consumption from the Cobb-Douglas version. By picking the right savings rate. To keep on consuming cG would mean lowering the savings rate below sG . for example. In intensive form.

Primitive Economic Growth sG .5 Primitive Economic Growth As we saw. or as we accumulate kt the marginal product of kt does not decline. The economy is not saving enough to remain at the Golden Rule level of consumption. we will have to abandon this assumption. but that kind of economy is most certainly not a good description of the real world. we need f ′′ (kt ) = 0. but is not something that we expect a real economy to be able to achieve. The Golden Rule consumption level serves as a useful benchmark. with sustained growth in output per capita of roughly 1. there are diminishing returns to accumulating more capital per worker. So if the Solow model is to exhibit sustained growth in steady state. That followed from the fact that f ′ (kt ) < f (kt )/kt . More precisely. which in turn is a consequence of the assumption that f ′′ (kt ) < 0. The A here is simply a scaling parameter. This runs counter to what we see in the data. Nt ) = AKt . The assumption that f ′′ (kt ) < 0 is therefore driving the Solow model to have zero growth in steady state. 3.5. examining the Solow model with these assumptions is useful as a benchmark. is the Solow model capable of reproducing this fact? It is. An economy that has perfect competition. and eventually consumption will have to go to zero as the economy uses up all the existing capital stock trying to maintain consumption at cG . More precisely. inelastic factor supplies. f (kt ) rises more slowly that kt . Specifically. and that never experiences any disturbances can conceivably reach the Golden Rule level of consumption. with f ′ (k ) > 0 but f ′′ (k ) < 0.61) We established that the growth rate of kt fell as kt increased. 3. and so the first term declines towards zero. From the accumulation equation. You can see this more clearly from the Solow accumulation equation f (kt ) ∆kt+1 =s − (n + δ ). What does that mean? It means that f (kt ) does not exhibit diminishing marginal returns. Eventually the additional output we get from a little more capital per worker is just enough to replace the capital per worker that is depreciating. A simple way to ensure that is to assume that the production function is linear. Without having introduced any notions of technology improvement or productivity. in steady state there is no growth in output per capita in the Solow model. then f (kt ) = Akt . although one can 33 . what this means is that as kt rises.2% per year in the OECD nations. the economy no longer has a steady state at k G .1 The AK model Growth runs down to zero in steady state in the Solow model because of our assumptions regarding the production function.5. but only with some very extreme assumptions.3. if F (Kt . kt kt (3. While it seems that these are unlikely to be true in the real world. or that the intensive form of the production function is concave.

62) Here. though.15 0. there is no relationship between the growth rate of capital and the size of kt .11 show growth rates of output per capita plotted against savings and population growth.00 −0.05 0.10 and 3. 0. respectively. 1988−2008 0. then there will be positive growth in capital per worker and hence in output per capita. T HE S OLOW M ODEL also think of it as capturing the productivity of capital. Figures 3.10 Growth Rate GDP per worker. there is a tendency for growth rates to be higher for countries with higher savings rates. So long as sA > n + δ . and falling with n. just as the AK model would predict. s. Is this true in the data? Possibly. The original Solow model would also predict a positive correlation of growth rates and savings rate out of steady state.10: Growth rates and savings rates. The AK model also says that growth rates should remain constant.05 UGA TTO SGP KOR IND THA BWA TWN GNB LKA CHL MYS MUS MOZ IRL TUR EGY DOM NGA GMB TZA HKG IDN TCD URY PAN NOR MLI FIN BGD DNK AUS JAM GBRSWEPHL GRC ROM LSO ZMBSLV USA ETH BEL IRN ISL COL ARG PRT AUT FJI NLD NZL GHA PNG MAR BFA CAN RWA CYP ISR JPN PRI BOL CRI ITA FRA NPLESP GTM SEN PER CHE ZAF MRT PAK HND NAM MEX GAB ECU MWI BRB BRA BEN SYR NER GIN KEN VEN PRY MDG CIV CAF BDI DZA COG CMR NIC TGO HTI COM ZWE 0.10 ZAR 0.25 0. 88−2008 CHN 0. We need to be careful. With that production function. as we don’t know whether countries are in or out of steady state. However. the AK model also says that growth rates are rising with s.3. So this is not definitive proof. kt (3.6 that growth rates demonstrably slow down as countries get richer.5 and 3. 1988-2008 However.35 0. regardless of the level of kt .30 Investment share of GDP.20 0. Growth in capital per worker is constant at sA − (n + δ ). and lower for countries with higher population growth rates. we saw in figures 3. The “AK” model in the title of this section comes from this form of the production function. 34 .40 Figure 3. the accumulation equation becomes ∆kt+1 = sA − (n + δ ). The AK model can replicate the steady growth rates that we see in the data. As you can see.05 0.

11: Growth rates and population growth rates. 1988−2008 0.3. of which there are M .05 0. Each firm is identical. think of it as an index of productivity. Let each perfectly competitive firm in the economy have a production function of α 1−α Yi = BKi Li (3.00 −0. 88−2008 CHN 0. This means that aggregate output is just Y = BK α L1−α and output per worker is y = Bk α . n. Firms take the wage rate and rental rate as given.00 ZAR 0.5. as 35 . Specifically. Primitive Economic Growth 0. That is. capital will have spillover effects on productivity that firms and individuals don’t account for directly.63) where i indexes the firms. The terms B is a scaling term. on average.01 0. so each firm employs an identical amount of capital and labor. M Li = L. Each firm will thus pay out α of output to capital (as we see in the data and our typical model) and 1 − α to labor.05 TTO SGP UGA IND KOR TWN THA GNB BWA LKA CHL MYS MUS IRL MOZ TUR EGY DOM NGA TZA GMB HKG URY IDN TCD NOR MLI FIN BGD PAN DNK AUS GBR GRC SLV LSO SWE ZMB ETH PHL JAM USA BEL ISL COLIRN ARG PRT AUT NLD FJI NZL GHA PNG BFA RWA ITA FRA CYP PRI JPN BOL CRI ESP CAN MAR ISR SEN NPL GTM PER CHE MRT ZAF PAK HND MEX ECU NAM GAB MWI BRB BRA BEN SYR NER GIN KEN VEN PRY MDG CIV COG DZA CAF CMR BDI NIC TGO HTI COM ZWE ROM 0. let B = Ak γ .10 Growth Rate GDP per worker. M Ki = K and total demand for labor is equal to total supply. in the economy. Now. That is. 1988-2008 Externalities and the AK Model One way of introducing AK style production without having to assume that capital’s share in output is α = 1 is externalities. total production is Y = i α 1−α Yi = BM Ki Li = B (M Ki )α (M Li )1−α (3.04 Figure 3. which the firm takes as given.64) and for markets to clear it must be that the total demand for capital is equal to the total supply. At the aggregate level.02 0. what about this term B ? This productivity is presumed to depend on the physical capital per worker in use.03 Population growth rate.

If this is true. and that seems a poor description of reality. they will have to be essentially AK models. and wages would be zero. this looks to hold across countries.3. so the strict AK model seems unlikely to be true. has the implication that capital would earn 100% of output. the production function is θ φ yt = kt ht . but still would have the prediction that growth rates are positively correlated with savings rates. or may demand suppliers meet certain technical standards or logistical requirements. If γ is large enough. Regardless. Let’s use a concrete example to start. In intensive form. Obviously.2 Multiple Capital Types A different approach to embedding sustained growth in the Solow model is to allow for multiple types of capital. although with a more refined notion of what the “K” represents. Labor is unproductive. This setup matches the observation that capital tends to earn a constant share of output. for example. then y = Ak and we have an exact AK model.66) where H is human capital.65) and we have a way of taking a production function with a relatively small α (say 1/3) and making it look more like an AK situation. if the spill-over exists then output per worker is y = Ak γ +α (3.5. Why? For many reasons one could imagine. Firms with high capital to labor ratios may spur competitors to innovate.67) 36 . we do not see that in the data. considering our underlying assumptions regarding firms and how they compenstate factors of production. we would have to believe that wages are simply rents that workers are able to extract from firms. 3. physical and human capital. but not within countries over time. As noted. so there would be no need to compensate it. The AK model also. then we could have sustained growth just through capital accumulation. for these models to generate sustained growth. Let aggregate production be θ φ 1−θ −φ Yt = Kt Ht N t (3. this has a positive productivity effect on all firms. say γ = 1 − α. Even if we relax the assumptions of perfect competition and perfect factor markets. T HE S OLOW M ODEL firms use higher capital per worker. As it turns out. (3.

From this figure. Knowing the ratio. then the ratio will shrink. To solve. Using the accumulation equation for physical capital we find s1−φ sφ ∆kt+1 h = k − (n + δ ). Figure 3.12 plots the growth rates of the two capital stocks against the ratio k/h.70) (kt /ht )θ −φ−θ h1 t The growth rates of both kt and ht depend on the ratio of h/k . which is determined where the two curves cross.68) (3.3. and to −(n + δ ) as k/h goes to infinity. For k . and you have = = sk sh − (n + δ ) − (n + δ ). going to −(n + δ ) as k/h goes to zero. both k and h are growing at the same rate. we can look at the implications of these accumulation equations. plug in for yt . Doing so results in k h ∗ = sk sh (3. 37 (3. For h. and so the ratio remains steady. the growth rate is rising in k/h.71) equal to each other. Similar logic shows that if the ratio is above the steady state. If the economy has kt /ht < (k/h)∗ . the growth rate is declining in k/h. and hence kt /ht will rise over time. treating human capital just like another type of capital ∆kt+1 kt ∆ht+1 ht ∆kt+1 kt ∆ht+1 ht = = yt − (n + δ ) kt yt sh − (n + δ ). what is the growth rate of k and h at the steady state? To solve for this. 1−φ−θ kt kt do we make sense of that? It depends on what we assume about the size of φ and θ. there will be a steady state ratio of k/h. ht sk (ht /kt )φ 1−φ−θ kt (3. (3.71) or the steady state ratio of physical to human capital depends on the relative savings rates. and the rising concavely as k/h increases. we can work out that there is a stable steady state ratio of (k/h)∗ . Primitive Economic Growth Let the two capital stocks accumulate in a Solow manner.72) From this we see that the size of kt potentially influences the growth rate of kt in steady state. which seems intuitive. At that ratio. then we see that k is growing faster than h. How . simply take the steady state ratio of k/h and plug back into one of the accumulation equations. Similar to the original Solow model.69) Take these equations. First. What is this steady state ratio k/h precisely? We know from the diagram that in steady state it must be that k and h grow at the same rate. set the growth rates from (3. Hence (k/h)∗ is a stable steady state. going to infinity as k/h goes to zero. This is easiest to see visually.5.

x Take the case where φ = 1 − θ. What occurs in the case where φ + θ < 1? Then we are back to a typical Solow situation. so must growth in output per capita. If this is true. not that with φ + θ < 1. then the intersection can occur where growth ∗ x in both is positive and the ratio stays constant. then the kt in the denominator θ of the growth rate of physical capital reduces to one. The steady state growth rate is denoted ∆ . then the only possible intersection of these curves is along the axis . where the growth rate of physical capital (and hence of human capital) equals −φ φ s1 sh − (n + δ ). So long as the savings rates are high enough. as kt increases the growth rate of kt continues to drop. with growth dropping to zero. and raw labor (N ) no longer matters. the two curves shift until their intersection occurs exactly on the x-axis. In the figure. To see this. and hence so will output per capita. This will occur to human capital as well. Eventually it must be that the growth rate of kt reaches zero. The only possible way to be at the steady state ratio (k/h)∗ is for both capital stocks to remain constant. This implies that the aggregate production function is Y = 1−φ−θ K H 1−θ . k and h will continue to grow k indefinitely. T HE S OLOW M ODEL ∆x/x ∆ht+1 /ht ∆x ∗ x k/h (k/h) ∗ ∆kt+1 /kt Figure 3. where growth eventually has to wind down to zero.12: Growth with Two Capital Stocks Note: The figure shows the growth rate of k and h relative to the ratio of their stocks. We have an “AK” style result. Because growth in both k and h slows down to zero. If θ + φ = 1.3. 38 .no growth in either. and no longer matters for growth. If θ + φ < 1.

3.5. Primitive Economic Growth In general, what matters for sustained growth in the Solow model is not the number of capital varieties, but rather the degree of decreasing returns to capital in general. The Lucas Model

A variant on the model including human capital comes from Lucas (1988), who specifies a production function Y = K α (uhL)1−α (3.73) where L is labor, h is human capital per person, and u is the fraction of time spent working. In per capita terms y = k α (uh)1−α . (3.74) Capital accumulates as ∆k = sy − (n + δ )k , as usual. Human capital accumulates as
∆h = φh(1 − u) (3.75)

which says that human capital is accumulated by spending time not working. So we have a tradeoff in output and accumulation. Notice that the growth rate of h is ∆h/h = φ(1 − u), a constant. Therefore human capital grows continuously, and so will output per capita. In addition, as physical capital depends on output (which depends on h), physical capital accumulates continuously as well. Set the growth rates of k and h equal to each other, as in a steady state, and you will find that ∆k/k = φ(1 − u). Therefore the growth rate of output per capita is ∆y = φ(1 − u) (3.76) y given that u is assumed to be a constant. Any change in u has two effects on the economy. First, u affects output immediately (if u goes down, we produce less by working less). Second, u affects output growth (if u goes down, growth goes up). So there is a tradeoff in selecting u. The optimal choice of u will depend on the time preference, as it depends on whether you want output today or in the future. More importantly, the reason there is steady state growth in the Lucas model is not because there is some alternative assumption about human capital accumulation. The source of growth is the assumption that the aggregate production function has constant returns to scale in accumulable factors. To see this, write the total human capital stock as H = hL, and then the production function is Y = K α H 1−α u1−α . With constant returns to generalized capital, there can be sustained growth.
Generally speaking, let the aggregate production function be Y = F (K1 , K2 , ...., KJ , N ) (3.77) 39

3. T HE S OLOW M ODEL where J is the number of types of capital used. If this function has constant returns to scale, then we can divide through by N and write y = f (k1 , k2 , ...., kJ ). Totally differentiate this function dy = f1 dk1 + f2 dk2 + .... + fJ dkJ , (3.79) (3.78)

where f1 is simply the derivative of f () with respect to k1 , and so on. This says that changes in output per worker are the sum of the marginal effect of each capital type times the change in that capital type. Divide this through by y , and multiple and divide each term on the right by the respective capital type dy fJ kJ dkJ f1 k1 dk1 + ..... + . (3.80) = y y k1 y kJ The growth rate of output per capita depends on the growth rate of each individual capital type, multiplied by that capital types share in output. We are still under the assumption that the economy has perfect factor markets, so that each capital type earns exactly its marginal product. fjj For each individual type of capital j , we would assume that the following holds: fj > 0 and < 0. That is adding more kj raises output, but at a decreasing rate. Given this series of

assumptions, it must be that we reach a steady state where all of the capital types grow at the same rate, and output per capita grows at that same rate. If it’s the case that dy/y = dkj /kj for all j types of capital, then equation (3.80) tells us what growth rates are permissible. Rewrite the equation as dy dy = y y f1 k1 fJ kJ + ..... + y y , (3.81)

and consider the term in parentheses. Each individual term is the share of output paid to a type of capital. If those shares add up to one, then any value for dy/y is permissible. That is, if the shares add up to one, growth could possibly be non-zero. The exact growth rate would depend on the savings rates, population growth rates, and depreciation rates. When will the shares add up to one? Only if production function is constant returns to scale in accumulable factors. That is, f (k1 , ..., kJ ) is constant returns to scale. With constant returns, as we accumulate capital of one type, this pushes up the marginal product of all the other types of capital, allowing these marginal products to stay ahead of depreciation and population growth. If f (k1 , ..., kJ ) has decreasing returns to scale, then the shares in the parentheses add up to something less than one. The only possible solution for the growth rate is then for dy/y = 0. In this case, we cannot accumulate capital fast enough to keep the marginal product of each capital type higher than it’s depreciation rate, and growth eventually goes to zero. 40

3.6. The Open Economy Hence, the only way to generate sustained growth in the baseline Solow model is to assume that there are constant returns to accumulable factors in the economy. Perhaps we believe that is true. However, it implies that the share of output paid to raw labor is zero, and we do see unskilled labor earning something. It also implies that higher savings rates should lead to higher permanent growth rates. While across countries nations that save more have some tendency to grow faster, in the time series we see that countries growth rates slow down as they become richer (convergence).

3.6 The Open Economy
It’s possible to expand the Solow model to allow for an economy open to the rest of the world. We have to be careful about what we mean by open, though. Output in the Solow model is homogenous, meaning there is nothing differentiated about what an economy produces. We have no individuals with utility functions, so we don’t know whether there may be demand for the homogenous lumps of output produced in France relative to the ones produces in Japan. So there is no possibility of trade in goods or services. Each economy produces identical lumps of output. If we are interested in international trade, then the Solow model is not worthwhile in its present form. However, we can explore openness to capital flows. That is, we can consider a world in which our economy can send invest in other countries (i.e. buy capital goods in France) or have foreigners invest in our economy (i.e. Japanese firms buy capital here). As we did with domestic savings and investment, we make the assumption that the international financial sector works costlessly and instantly. That is, capital can flow across borders instantaneously and without loss. We do not at this point introduce any friction involved in say, moving a car factory from Nagasaki to Tennessee. In order to understand how our Solow economy works when capital can flow in and out of the country, we need to make several definitions more explicit. First we need to distinguish now between GDP (domestically produced output) and GNP (income earned by citizens of the economy) per capita: GDPt GN Pt = = f (kt ) (R )(kt + ft ) + wt

(3.82) (3.83)

where R∗ is the world rate of return paid by firms to use capital. kt is the stock of capital that is located in the economy, and ft is the net foreign asset position of the economy. Recall that ft can be either negative or positive, depending on whether the economy is a net creditor or net debtor. Total assets are kt + ft for the economy. Because capital can flow costlessly across borders, we presume that it will do so until the rate of return in each economy is exactly equal to R∗ . Otherwise there would be some arbitrage opportunity. Hence the owners of the assets in this economy earn exactly R∗ as a return on those assets. 41

(3.86) This is describing how the current account evolves in a Solow model economy open to foreign capital flows. we have the current account equation being cat = k ∗α + (R∗ − δ )ft − (1 − s)[R∗ (k ∗ + ft ) + (1 − α)k ∗α ] − δk ∗ which after a variety of algebra (and noting that k ∗α = R∗ k ∗ + wt ). So total GNP per capita is the sum of the return on assets and the wage. and the only piece we do not have information for is investment. T HE S OLOW M ODEL In addition. so we’ll assume that s is less than this amount. As before. Hence GDP is yt = k ∗α and we can solve for wages as wt = (1 − α)k ∗α . ∗ From the preliminaries. We’ll ignore population growth to keep the analysis simpler.87) (3. In steady state. gives us cat = sR∗ (k ∗ + ft ) + s(1 − α)k ∗α − δ (k ∗ + ft ). no interstellar capital flows). so that ct = (1 − s)[R∗ (kt + ft ) + wt ]. as far as I know. or the domestic capital stock is held exactly at k ∗ . 42 . Specifically. or that there is sufficient investment each period to just maintain the capital stock at k ∗ . From this. then net foreign assets go to infinity. We know from prior work k∗ = that the total returns to capital plus the total returns to labor must add up to total domestic output.3. then. we can solve for: f∗ = s(1 − α) ∗α k − k∗ . The capital stock evolves as ∆kt+1 = it − δkt . Given that k ∗ is fixed. we can look at how the net asset position will evolve over time. the requirement that capital earns precisely R∗ means α 1/(1−α) (3. (3.84) α Given a production function of f (kt ) = kt .88) Under what conditions will steady state net foreign assets actually be positive? f ∗ > 0 occurs when s > αδ/R∗ . δ − sR∗ (3. so that s → δ/R∗ . the world is one giant closed Solow economy (there are. This implies that it = δk ∗ . Putting it all together. it must be that ∆kt+1 = 0.85) R∗ which follows from setting the marginal product of capital equal to R∗ . what is the return to capital for the world? It 2 One note is that if savings are really big. or when savings are sufficiently large.2 From our perspective. we know that the definition of the current account is cat = yt + (Rt − that δ )ft − ct − it . the workers in the economy earn a wage wt . is there steady state level of net foreign assets in this economy? Do we reach a point where ft+1 = ft or cat = 0? Setting cat = 0 in the prior equation. we will assume that domestic individuals consume a constant fraction of their gross national product.

92) While physical capital is perfectly mobile. sworld (3. we said that we’d have positive net foreign assets only if s > αδ/R∗ . so there is no sense that there can be a real exchange rate (alternatively.89) The world’s rate of return depends on the world savings rate.3. We haven’t introduced any nominal measurement of output. So every time human capital increases. This makes some sense. In the open economy Solow model. The Open Economy would be R = αk ∗ ∗α−1 =α sworld δ α−1/(1−α) = αδ .93) and the physical capital stock is increasing in human capital stocks. Slow accumulation of human capital could explain small capital flows. In other words. sworld . we have that αδ sworld = ∗ . (3. assume that human capital is not. (3. We do not see massive flows of capital between countries. domestic residents will end up with net foreign assets as they save more than the rest of the world does.90) R Going back to our analysis of our domestic economy. but each country maintains the same amount of capital per worker. the real exchange rate is identically equal to one). The marginal product of physical capital is M P K = θk θ−1 hφ and this has to equal R∗ . so there is no such thing as a nominal exchange rate to worry about. Turning this around.91) so that the intensive form is y = k θ hφ . which means that k = ∗ θhφ t R∗ 1/(1−θ ) (3. Y = K θ H φ N 1−θ−φ .6. The only source of variation in current accounts and net foreign asset positions is savings rates. current account differences and net foreign asset positions are the result of differences in fundamental savings rates. Assume that the production function is a Cobb-Douglas like we used in earlier. Each country produces an identical homogenous good. physical capital will flow into a country. 43 . (3. Slow Capital Flows The assumption that capital can flow immediately and costlessly between countries is ex- treme. A modification of the Solow model to include a non-tradable capital stock can help provide an explanation for these “slow” capital flows. we will have positive net foreign assets only if s > sworld .

but those are rare and have non-economic sources. this is different from what was probably presented to you as the aggregate supply curve in an intermediate macroeconomics class. There may be times when a hurricane destroys a significant amount of capital. as we’ve written it. nor is there even convergence as poor countries catch up to rich ones. 44 . and it is the aggregate supply of real output. This makes the Solow model. capital should flow in to poor countries (i. 3 Hoxha. this is an actual aggregation of the supply of firms producing a homogenous output. real output Yt is fixed by the supply of Kt and Nt . perfect competition among firms. the AS curve was simply a relationship between inflation rates and output levels. it will help us to see later on which assumptions are important in allowing for nominal shocks to have real effects. as we have no individuals or households with utility functions that would naturally lead to some kind of demand curve. as we said. This is a supply curve. as that will give us the starting point for expanding. and Vollrath (2013) modify the Gourinchas and Jeanne model and show the conditions under which welfare gains may in fact be quite large from integration. However. Kalemli-Ozcan. This was the direct outcome of several assumptions: a production function that only depends on capital and labor. essentially useless for describing business cycles.3. By lowering the cost of capital flows. There is no growth. and inelastic supplies of both factors. This is one argument for international financial integration. it will be useful to flesh out the Solow model in nominal terms. Thus the supply curve of real output is vertical with respect to the nominal price of real output. There. Here we run into a semantic issue. T HE S OLOW M ODEL Note that in this open-economy Solow model.3 3. those with low values of capital per worker) raising their output per capita.7 Nominal GDP and Fluctuations We saw previously that one implication of the Solow model was that the only source of fluctuations in the economy were the random destruction or creation of capital and/or labor. in some sense. Some recent research by Gourinchas and Jeanne (2006) evaluates the effect of this kind of integration in a version of this model that includes explicit utility functions. This would happen presumably much faster than they would be able to accumulate capital themselves. so what is the demand for the good? At this point we have absolutely no way of describing this. Despite the lack of any obvious source of fluctuations. GDP per capita is precisely f (k ∗ ) at all times.e. So. even if we did know their demand curve. Convergence happens instantly. In particular. perfect factor markets. Here. We have a supply curve for our homogenous good. It is thus an “aggregate supply curve”. There is no alternative good other than real output for households to purchase. as capital flows in costlessly to let a country reach f (k ∗ ). and we can see in the data a distinct downturn in economic activity. and finds limited welfare gains.

we can insist that the transactions in this economy occur using some nominal means of payment. For that to be the case. it gives us a way of pinning down the price of real output in our economy. They pay wages and returns to the owners in dollars. There needs to be an upward-sloping supply curve for real output. if P rises. and there is a fixed supply of K and N in aggregate. Combined with our Solow model aggregate supply curve.3. and vice versa. “Theory” is actually a misnomer. It holds by definition of the terms. then they will increase their demand for Ki and Ni . In other words. Why is output fixed? Because of the assumptions we’ve made about firms and competition. Firms will bid up the price of capital and labor until the point that demand for capital and labor is low enough to clear the market. One thing our simple little model tells us is that nominal shocks have no effect on real output. an increase in the money supply will raise the price. we can determine real output and the nominal price of that output. The Quantity Theory of Money is something you likely have encountered already. every firm is making the same decision. (3. firms are hiring in labor and capital.96) From the firms perspective. money. The Quantity Theory serves as our aggregate demand curve. However. In our economy. all an increase in P will do is raise R and w. it is an identity. variation in the productivity of those 45 . holding money supply constant. (3. so it is the money supply and velocity that determine the price. Ni ) − RKi − wNi .95) meaning that R and w are now nominal returns and wages. Ni ) = R and P FN (Ki . Let’s go back and put the price of real output explicitly in the firms profit function π = P F (Ki . This tells us something about what is necessary to generate fluctuations. and those owners in turn use those dollars to purchase the output of the firms. we do know an accounting identity that relates the supply of money to prices and real output.7. This gives us the first order conditions. Ni ) = w. Rearranging. An increase in Mt does nothing but raise the price level. Regardless. there must be some variation in the amount of inputs used (or as we’ll see in the next chapter. A change in velocity will influence the price level. Now re-do the optimization. as the Quantity Theory is not a hypothesis about how the world works. We know that Yt is fixed by inelastic supplies of factors of production. Holding velocity fixed. Nominal GDP and Fluctuations However. Pt = Mt Vt /Yt . leaving real returns R/P and real wages w/P unchanged. P FK (Ki . as the revenue marginal product of both has increased. It states that Mt Vt = Pt Yt (3. Why? Because output is fixed. While we do not know anything about demand for output.94) of the supply of money (Mt ) times the velocity of money (Vt ) is equal to nominal GDP (the price of real output Pt times real output Yt ).

but a major motivation for studying them is that we see them influencing real output (and employment). meaning they have some market power. In other words. 46 . if when P rises firms are able to hire in more labor or capital. and the entry of firms is costly (either in time or resources). Another is that firms are not perfectly competitive. firms may choose not to hire all of the capital and labor available. so that bidding up the returns or wages pulls in more of that factor. we need to break down some of the baseline assumptions of the Solow model in order to generate meaningful fluctuations in the economy from nominal demand shocks. In that case. What drives nominal shocks may be an interesting question in and of itself. bidding up P would increase their demand for those factors and increase output.3. T HE S OLOW M ODEL inputs). In short. One possibility is capital and labor are supplied elastically to the market. However. then output could indeed rise.

Finally. Having included productivity growth. I’ve jumped directly to the aggregate production 47 . competition and factor supplies. The important insight from the Solow model is that it’s only through productivity improvements that living standards can rise in steady state. we can create an updated model that captures this feature of the data.2% per year.1 Technological Progress We begin by maintaining all of the original assumptions of the Solow model regarding firms. An increase in Et will increase output even without an increase in either capital or labor. as we know.S.1) where Et is a term measuring the productivity of labor. The one change is that production now takes the following form: Yt = F (Kt . Accumulating more factors of production alone is not sufficient. we will attempt to distinguish how much of cross-country income differences are driven by factors of production (capital and labor) versus productivity differences. and other OECD nations averages about 1. Yet. Et Nt ) (4. perhaps as much as two-thirds. trend growth in per capita income in the U. of the difference in living standards across countries is attibutable to productivity differences. We’ll examine whether fluctuations in productivity themselves are capable of explaining fluctuations in output over time around the long trend growth. as factors run into diminishing returns. 4. By incorporating a way for the efficiency of production to increase even without raising inputs of capital or labor. we’ll also have a mechanism for creating fluctuations in output that is not dependent on shocks to capital or labor supplies. The end result will be that a sizable portion.C HAPTER 4 Productivity: Growth and Fluctuations One of the serious faults of the baseline Solow model is that it does not feature any sustained growth in output per capita in steady state.

6) (4. and “depreciates” with both population growth and technological progress. ˜. just capital relative to the that as Et rises. Similar to before. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS function here. of course. this adds to effective labor. we have Kt ˜t ) ≡ F . Growth in capital per efficiency unit is driven by savings. we use the constant returns to scale property to redefine values in terms of efficiency units. ˜t k = (4. so that Et+1 − Et ≡ ∆Et+1 = gEt .2) y ˜t = f (k Et Nt where x ˜ denotes a variable Xt that is scaled by Et Nt . What this means is that when Et goes up.3) As before. Types of Technological Progress In working with technological progress in the Solow model. which is the total labor effort put into production. but it is not the only one available.8) Here again we’ve used the assumption that a perfectly costless financial sector transforms savings into investment. the isoquants of the production function shift inward (we need fewer 48 . It is important to keep in mind ˜t falls.1 (4. Et Nt . k efficiency of labor. Et Nt ) = s −δ−n−g Kt F (Kt . but in the background you can imagine lots of competitive firms using this technology.7) (4. ˜ Kt Et Nt k (4. This form of technological change is known as Harrod-neutral technological change. we have Evaluating the growth rate of k ˜ ∆k ∆Kt+1 ∆Et+1 ∆Nt+1 = − − . Technological progress consists of a fixed growth rate in Et . namely as “labor-augmenting”. meaning that when Et goes up. Dividing the production through by Et Nt .4.4) (4. we can now use our knowledge of the different growth processes to describe this more fully.5) (4. but this does not mean that capital is declining. I have assumed that technology enters the production function in a particular way. Nt )/Et Nt −δ−n−g = s Kt /Et Nt ˜t ) f (k = s − δ − n − g. ∆kt˜ +1 ˜t k It −δ−n−g Kt F (Kt .

in this case with respect to the value of k ˜t+1 = 0. but it also rotates the production function so that the MRTS goes up (meaning that it takes more additional capital to make up for lost labor).9) which is Harrod-neutral. Solow-neutral technical change is written as Yt = F (Et Kt . and therefore everything can be written in Harrod-neutral form. the isoquants shift inward.10) Our Cobb-Douglas production function can be written as Harrod-neutral no matter how we’d like to actually define technological progress. which is called Hicks-neutral technological change. The steady state is defined as the point to a steady state. Nt ). For other production functions. With Hicksneutral changes. Et . Let Hicks Solow Harrod Et . Technological Progress inputs to achieve the same level of output). the EOS between capital and labor is always and exactly equal to one. the MRTS doesn’t change. so long as we define Harrod Hicks 1/(1−α) Solow α/(1−α) Et = Et (Et ) (Et ) . that we are generally not concerned with capital or output per efficiency unit.4.11) Note. In other words. and Et be technological change of the three different types so that we Hicks Solow Harrod have Yt = Et (Et Kt )α (Et Nt )1−α . so the trade-off between capital and labor is not affected. this doesn’t pose much of a problem because we are focusing mainly on Cobb-Douglas production functions. α 1−α To see this. this is not the case and therefore balanced growth may not be possible. sf (k (4. so that where ∆k ˜∗ ) = (δ + n + g )k ˜∗ . (4. is that to have balanced growth in the long-run (meaning that ∆y/y = ∆k/k ). but remain identical in shape. What this means is that Harrod-. Recall that for the Cobb-Douglas. only it is “capital-augmenting”. Nt ) and is similar to Harrod-neutral change. We can simply write a simpler version of this as α Yt = Kt (Et Nt )1−α (4. and Solow-neutral technological change are all simple transforms of each other. we cannot achieve balanced growth unless technological change acts in a labor-augmenting way. A somewhat surprising results. take a Cobb-Douglas production function of the form Yt = Kt Nt . Finally. For our purposes. but 49 . We could alternately write the production function as Yt = Et F (Kt . we must be able to express technological change as Harrod-neutral. As before. so that technology augments all factors equally.1. Hicks-. an implication of this modified Solow model is that the economy will eventually come ˜. due to Uzawa (1961). though. In other words.

Steady state kt is still postively related to s. but in addition the further away from the steady state.13) So the exogenously given growth in labor efficiency provides us with a mechanism by which we can generate trend growth in k and y at the rate g . This means that yt is growing in steady state at the same rate as Et as well. and negatively to n. it must be that FK is constant as well. ˜∗ is stable.4. the more We know that k ˜ falls as k ˜ rises is quickly the economy moves towards it. Et .15) (4.16) . Take the derivative of both sides with respect to K .12) ˜ is in steady state. The logic of why the growth rate of k identical to what we showed in the prior chapter ˜t+1 ∆k ˜t ˜ →0 k k ˜ ∆kt+1 lim ˜t ˜ k k→∞ lim 50 ˜t ) f ′ (k −δ−n−g =∞ 1 ˜t ) f ′ (k = s − δ − n − g = −δ − n − g 1 = s (4. the steady state of the Solow model is stable (i. and examine models that explain the growth rate g endogenously and as the result of explicit economic choices.e. it is negatively related to g . kt is growing over time at the same rate as from which we can infer that even if k Et . which can ultimately be used as one explanation for why the eeconomy cycles around a long-run trend. F (Kt . The other implications of the Solow model are kept intact. the non-zero steady state. To see this. The marginal product of capital FK = f ′ (k ˜ ˜ As kt is constant in steady state. This particular implication will be used in the next section to look at how a Solow economy responds to shocks to the productivity term. We should stress the “mechanism” part. the smaller will be the steady state level of capital per efficiency unit. as this increased efficiency is simply assumed. It is also true that the farther away from steady state an economy is. yt Et y (4. Et Nt ) = Et Nt f (k (4. In this section we consider more about the dynamics of the capital stock away from the steady state and what this can tell us about how the economy responds to shocks. note that ˜t ). In later chapters we will address this point. In addition. The faster that technology grows. and there is no economic reasoning given for why g exists or what value it may take. So what is capital per person? ˜t kt = Et k (4.2 Dynamic Responses to Shocks Whether we include population growth and technological progress or not. just as in the original Solow model. The return to capital will be constant in steady state. 4. to be accurate).14) ˜t ). the faster it will grow. That growth rate is ∗ ˜ ∆yt+1 ∆Et+1 ∆y = + ∗ = g + 0 = g. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS rather with their per capita values.

1. ˜t is from the steady state. it is always the case that this holds. the growth rate of the capital stock is decreasing.1.17) ˜t )k ˜t − f (k ˜t ). This is negative when f (k ˜t )/k ˜t > f ′ (k ˜t ). ˜t ˜2 ∂k k t (4. the faster it moves towards We asserted before that the farther away k ˜t+1 /k ˜t increases the the steady state.2.1: Growth Rates and k where the first equality in each case is due to L’Hopital’s rule and the second is due to the Inada conditions. Dynamic Responses to Shocks ˜t+1 /k ˜t ∆k ˜t+1 ∆k ˜t k >0 ˜∗ k ˜t+1 ∆k ˜t k <0 ˜t k 0 −(δ + n + g ) ˜t Figure 4. the growth rate of the ˜t . farther away from k 51 . We can see this in figure 4.4. the growth rate is equal to zero and there is a steady state. Between these extremes. where the absolute value of ∆k ˜∗ one goes. Given our assumptions regarding the shape of f (·). Using this information. The sign of this depends on the term f ′ (k or if the average product of capital is greater than the marginal product. Therefore. which can be seen ˜t : by taking the derivative of (??) with respect to k ˜t )k ˜t − f (k ˜t ) ˜t+1 /k ˜t f ′ (k ∂ ∆k =s . we can plot the growth rate of k Where this function crosses the x-axis. capital stock is decreasing in k ˜t against the level of k ˜t in figure 4.

First.19) From this we can evaluate the effect of different shocks to the economy on the growth rate of capital per efficiency unit. consider 52 . The ˜′ . Similarly. increase in Et lowers capital per efficiency unit to k ˜ ˜∗ again. as output per worker can be written as ˜t ) yt = Et y ˜ = Et f (k the growth rate of output per person can be written as ∆yt+1 ∆˜ yt+1 =g+ .18) ˜t kt k ˜ translates directly into a change in the growth rate of so that any change in the growth rate of k capital per worker. yt y ˜t (4.20) (4. Et . Over time k will grow and reach the steady state of k How is growth of capital per person related to this? We know that by the definition of capital per efficiency unit ˜t+1 ∆kt+1 ∆k =g+ (4.4. on an economy initially in steady state. and this increases the growth rate of capital per efficiency unit to x′ . and hence for capital per person and output per person. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS ˜t+1 /k ˜t ∆k x′ 0 ˜′ k −(δ + n + g ) ˜∗ k ˜t k Figure 4.2: A Positive Shock to Et Note: The figure shows the effect of a one-time increase in productivity.

Therefore a one-time shift in Et increases income per person. Does this mean that output per person falls as well? No. capital per efficiency (and hence output per efficiency unit) falls. At the time of the shock.22) (4. leaving the trend growth rate of efficiency still equal to g . Output per worker growth follows exactly the same pattern. as k the growth rate of capital per person falls towards the steady-state level of g . How does the level of output per person respond? Recall that y ˜ has fallen because of the upward shock to Et . You can easily – and in fact probably more easily – write the Solow model in continuous time. Solow in Continuous Time We’ve done the model so far completely in discrete time. As noted before. These immediate and dynamic responses to technological shocks thus form one potential explanation for fluctuations in economic performance over time. Dynamic Responses to Shocks what would happen if the economy began in a steady state and there was a one-time increase in Et . Consider the derivative of output per worker with respect to Et . Notice that these cycles occur without any reference to money or prices (hence the “real” moniker). From (4. at least in terms of growth rates. This drop in capital per efficiency unit implies ˜ must increase. The real business cycle (RBC) literature takes precisely this view.19): ∂yt ˜t ) − Et f ′ (k ˜t ) kt = f (k 2 ∂Et Et which is positive if ˜t ) f (k ˜t ) > f ′ (k ˜ k (4.2. arguing that real technological shocks hitting the economy are capable of explaining the origins of business cycles. given (4.4. Over time.18) we know that this raises the growth rate of that the growth rate of k ˜ accumulates capital per worker as well. This can be seen clearly in figure 4. So a shock to efficiency causes an immediate jump in output per person and increases the growth rate of output per person temporarily while the economy approaches a new steady state. If we reverse this shock.21) or the average product of capital per efficiency unit is larger than the marginal product. we can ask what the growth rate of 53 .2. So this shock to Et generates a boom. even though income per efficiency unit has gone down. this holds by our assumption regarding the concavity of the production function. a negative shock to Et will lower output per person immediately and then lower the growth rate of output per person temporarily until the steady state is reached again. Using a x ˙ to denote the time derivative of a variable x.

26) ˙ ˙ ˙ E N K − − K (t) E (t) N (t) Y (t) −δ−g−n = s K (t) and this is essentially the identical expression to what we had in the discrete time situation. To see this clearly. but again this is not terribly important. with inelastic supplies of inputs. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS capital per efficiency unit is: ˙ ˜ k ˜ k = = = E (t)N (t) K ˙ (t) K (t) E (t)N (t) ˙ E (t)N (t) K K (t) E (t)N (t) (4. It also suggests one way of describing how output fluctuates about it’s trend. would shift the vertical aggregate supply curve.23) − ˙ ˙ K (t)E K (t)N − 2 E (t) N (t) E (t)N (t)2 (4. causing output to go up or down depending on the direction of the shock. or δ = 1. Hence there is no possibility of output varying due to aggregate demand shocks.24) (4. but is not a particularly hard assumption to work around. we need to make one additional assumption. If productivity shocks are driving fluctuations. The instantaneous growth rate of capital per efficiency unit depends on the growth rate of capital (which depends on savings and depreciation). 4. The equation here is exact (as opposed to the approximate discrete time accumulation equation). the growth rate of technology (g ) and the growth rate of population (n). This will highlight the role of productivity.25) (4.4. It will also be helpful to describe efficiency at period t + 1 as Et = ut E0 (1 + g )t 54 (4. The addition of productivity growth has not changed the fundamental assumptions regarding perfect competition among final goods producing firms. however. but for all intents and purposes the discrete and continuous time versions are identical. then this has implications for the time series properties of output. This is that depreciation is complete each period.3 Productivity and Aggregate Fluctuations The Solow model with productivity growth provides a mechanical way of understanding the source of long-run growth in output per capita. We will also assume that n = 0 for simplicity. A change in productivity.27) .

4. lowering output. There 55 . The coefficient in front of ln yt−1 dictates how long these shocks linger. If anything lowers output in t − 1. Our assumptions about perfect competition.32) This says that output per capita is auto-regressive. What the above representation of output per capita suggests is that fluctuations in yt around its trend over time are due to fluctuations in ut . insert our assumption regarding Et and take logs of both sides.4. assume that E0 = 1. then this lowers the total amount saved in t − 1. inelastic factor supplies. (4. (1 + g )t is just the accumulated growth in productivity between period 0 and period t. the idiosyncratic shock to productivity. This is a simple outcome of the Solow model we set up. output per capita in the prior period yt−1 influences output per capita today. (4. We assume that is is log-normally distributed. We can set this to one if we like without changing anything. it must be the case that kt = syt−1 . ln(1 + g ) is roughly equal to g . to match capital’s share of output. That is. Productivity and Aggregate Fluctuations where E0 is productivity in the initial period.30) To work with this. Now we can write output per capita in time t as ln yt = (1 − α)tg + (1 − α) ln ut + α ln s + α ln yt−1 . Hence in expectation ut = 1 and productivity would be at trend. (4.29) (4. First.31) We can make several simplifications here. Plugging this into the expression for output per capita.28) α if we assume a Cobb-Douglas production technology for firms with Yi = Ki (Ei Ni )1−α . With complete depreciation and zero population growth. A negative ut shock today will create a lingering effect on output per capita over the following periods. The term ut is an idiosyncratic shock to Et around it’s trend. However. given our assumptions we’d expect this coefficient to be approximately 0. The autoregressive portion of the expression says that when the economy does get shocked away from trend. we have that 1−α yt = Et (syt−1 )α . with ln(ut ) ∼ N (0. Therefore. Second.3–0. if you actually regress output per capita on it’s lag you’ll find much higher coefficients.3. σ 2 ). and the Cobb-Douglas production function imply that this coefficient is exactly equal to α. We can write output per capita as 1−α α yt = Et kt (4. it doesn’t “snap back” right to trend next period. This leaves us with ln yt = (1 − α)t ln(1 + g ) + (1 − α) ln E0 + (1 − α) ln ut + α ln s + α ln yt−1 . which therefore lowers the capital stock at time t.

With constant returns is has to be that sK = 1 − sN . Solow made an important leap. to obtain our estimate of sN (and hence of α) we simply take total wages paid to workers (wN ) and divide by total revenues. one that is often glossed over in the literature. One of the issues with those time series regressions is that they do not have any way to measure ut . then we should see ut and yt highly correlated. then it is the case that the revenue shares are exactly equal to the elasticity of output with respect to the factor of production. In the Cobb-Douglas case. A common approach starts by writing output as follows: 1−sN Yt = Rest Kt NtsN (4. Now because sN = 1 − α and sK = α under perfect competition and constant returns. One of the arguments in that literature is that the coefficient is actually equal to one. To measure ut we’ll need to measure Et . the time series evidence is not consistent with this basic set-up. meaning that shocks to output in t − 1 permanently affect output in the future. w = FN . So how do we measure Et ? There is no data series on Et . 1−sN Kt NtsN (4. we are using the assumption on constant returns to get that sK = 1 − sN . the residual gives us information about the level of productivity.35) In other words. If productivity shocks are the most important driver of fluctuations in output per capita. the residual reduces to the following: 1−α Rest = Et . With perfect competition. With perfect competition and perfect factor markets. then perhaps the time series regressions are overestimating the coefficient on yt−1 . one can back out the residual directly by simply re-arranging Rest = Yt . 56 . sN is the share of labor in total revenues of firms. (4. 1 To see this. A different approach to evaluating this model is to look specifically at how ut and yt relate to each other.33) where Rest is the “residual”.34) This residual is often called “Solow’s Residual”. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS is an exhaustingly long literature in time series macroeconometrics on exactly what this coefficient is.4. Practically speaking. So to measure Et in this economy all I need to do is calculate the residual. If ut has some auto-correlation of it’s own. Therefore we have to back it out of the data. Et . note that the revenue share for labor is just s N = wN/Y . Regardless. and therefore sN = FN N/Y . or GDP (Y ). To proceed. under our assumption that the economy has perfect markets and constant returns. The right-hand side of this last equality is simply the definition of the elasticity of output with respect to N . To obtain sK . Under the assumption that markets are perfect and production is constant returns to scale. all revenues are divided up into either payments to labor or payments to capital. as he was the first to propose something like this in a separate paper from 1956. this implies that sN = 1 − α. but the principle holds regardless of the exact production technology.1 In practice.

Productivity and Aggregate Fluctuations 10.0 −4.S.00 0.00 −5. 10.0 1959 1955 1984 1951 1973 1966 1962 1965 1976 1972 1964 1983 1950 %Chg in real value−added 1978 5.0 Figure 4.0 2.00 %Chg in real value−added %Chg in residual Year−on−year growth rate 5.0 −2.S.4: Output and Residual Growth Rates in the U.0 1974 1982 1980 2009 −5.0 1977 1999 1968 1997 1998 1953 1994 1963 1988 2000 1996 1985 2004 1971 1986 1992 1987 1993 2005 1989 1979 1995 2006 2003 2010 1969 1952 2012 2011 1981 2007 1957 1961 2002 1956 1967 1960 1990 2001 1970 1991 2008 1954 1975 1958 1949 0.0 6.3. 57 .0 0.3: Output and Residual Co-movement in the U.00 1950 1960 1970 1980 Year 1990 2000 2010 Figure 4.0 %Chg in residual 4.4.

Figure 4.S.3 shows the results. knowing these cost minimization conditions we can proceed to examine 58 . It represents the additional cost of increasing Y incrementally. We can write this as a Lagrangian L = wN + RK + θ(Y − F (K. the share of labor does not equal the elasticity of output with respect to labor. However.36) where θ is the Lagrange multiplier. With imperfect competition.. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS For the moment. Taking first order conditions is straightforward. First. and that gap represents the effect of changes in either Kt or Nt on output. The series are offset somewhat (the output series has a higher average change). Perhaps a better way of seeing the correlation is to just plot the growth rates of output and the residual against each other.4. and similarly for capital. maximize profits by selecting the optimal Y given market conditions. N ) = Y . In a series of papers. What does this imply happens to our calculation of the residual? Cost Minimization and Markups To give us more flexibility in how we deal with firms. it appears that the residual is important in driving fluctuations in output. minimize costs for producing any given level of output Y . From this figure.S. 1990) showed that the residual Rest does not necessarily measure Et if there is imperfect competition among firms or increasing returns to scale. Here you can clearly see that years in which the residual grows significantly. Robert Hall (1988. which then tells us what Et is for the U. output and the residual move very similarly. Now. We can calculate the residual for the U. and yields that the firm will set θFN = w and θFK = R. output also grows significantly. profits. and markups. In other words. It turns out that for our purposes we can get a lot of traction simply by considering the first part of this: cost minimization. and therefore we could conclude that productivity shocks appear to be very important in driving output fluctuations.4. With these assumptions the residual does pick up only fluctuations in Et . The firm’s cost minimization problem is to minimize wN + RK such that F (K. where Y is an arbitrary output amount. This will give us a very simply relationship that we can exploint regarding returns to scale. the multiplier has a very specific interpretation. As you can see. assume that our economy does have perfect competition and constant returns. Second. that is only true under these very strict assumptions. we can break the profit maximization problem up into two parts. θ is the marginal cost of production. N )) (4. In this problem. Does this mean that it is real productivity shocks that drive fluctuations? Only if we assume that we have perfect markets and constant returns. We can plot this log of this Et against the log of Yt to see if in fact output and productivity are highly correlated as our Solow model suggests. as in figure 4.

If we assume the firm has market power. or firms charge exactly marginal cost. Productivity and Aggregate Fluctuations the firm’s profits. Re-arrange the profits of the firm as follows 1 = = θ FN N θ FK K Π + + P F (K. N ) P F (K. these two elasticities add up to the returns to scale in the production function. or they charge a price higher than marginal cost.4. This is equal to θ/P times the elasticity of output with respect to labor FN N/F (K. We will define the following as the markup. N ))/µ. N ) sN + sK + sπ (4. θ. we have increasing returns to scale.39) (4. then we have constant returns to scale. N ) P F (K. Π = = P Y − wN − RK P F (K. N ) (4. Together. 59 . The si terms are simply shorthand for the share of revenue. then it will be the case that µ > 1.and the share of revenue paid out as profits must add up to one. is equal to (FN N/F (K. to complete the relationship. N ) − θFN N − θFK K.40) This says that the share of revenues paid to labor and capital . µ: µ= P . then µ = 1. θ (4. (4. and hence how to maximize profits. Using this defintion of the markup.the first two terms . we do not need to make any specific assumptions at this point to get a useful fact.42) Now. Hence the share of revenues going to labor. or the elasticity of output with respect to labor divided by the markup. If there is perfect competition. we have decreasing returns. In other words.38) where P is the price the firm charges for their output. sN . if those elasticities add up to one. You can confirm this yourself using Cobb-Douglas production functions quite readily. Take the share paid to labor. note what we have on the left-hand side. re-arrange again to get FK K FN N + = µ(1 − sπ ). N ) F (K. but the principle holds for any typical production function. However. This is the elasticity of output with respect to labor plus the elasticity with respect to capital. N ). If those elasticities add up to more than one. and if they are less than one.3.37) (4. F (K. Our assumptions regarding market power would inform us about how P is set (or not) by the firm.41) The markup is the price charged for output relative to the marginal cost of that output.

In other words. and you’ll find details for this in the boxed section. • If production is constant returns to scale (so FN N/F (K. rather than actual productivity shocks. If µ = 1.43) where µ ≥ 1 is the markup over marginal cost that a firm charges. Hall uses the fact. then firms are charging a price that is twice marginal cost. the bigger the effect these factors have on the residual. Et . then the term involving N/K goes to zero and we’re back to Solow’s original version. sN and sK . • If firms have market power (sπ > 0) but do not charge a markup (µ = 1). with imperfect competition it no longer is the case that the Solow residual is a measure of only efficiency. N ) > 1). More importantly. then profits can be positive (sπ > 0) only if the markup is bigger than one. It now picks up any changes in Nt or Kt . or we have perfect competition.34) we are under-weighting labor. if a firm has market power then it will charge a price for its output that is actually higher than the marginal cost of producing it. • If production is increasing returns to scale (so FN N/F (K. then profits can be zero even if the markup is bigger than one. and reduce things we find that 1−α Rest = Et Nt Kt sN (µ−1) . P RODUCTIVITY: G ROWTH AND F LUCTUATIONS This relationship puts some structure on how firms profits. for example. and returns to scale can be related if they are cost-minimizing. That is.4. firms will charge a markup over marginal cost if they have increasing returns to scale. we can see that anything that makes Nt fall (unemployment) will drag down the residual. It’s 60 . What you can see is that sN is therefore less than the elasticity of output with respect to labor. if we plug in the relationship from (4. µ tells us the size of that markup. The larger is the markup µ. If µ = 2. relate to the actual production elasticities.3 may simply reflect the co-movement of employment and output. N )+ FK K/F (K. if there is market power for firms. and over-weighting capital.43) to the residual calculation. In general. then production must be decreasing returns to scale. So the pattern seen in figure 4. that these are related in the following manner sN = 1−α µ (4. even with no market power. N ) + FK K/F (K. markups. So in calculating the residual from (4.44) In other words. First we have to understand how the shares of revenue. From this imperfect competition version of the residual. N ) = 1). (4.

61 .0 Figure 4.00 1950 1960 1970 1980 Year 1990 2000 2010 Figure 4. Productivity and Aggregate Fluctuations 10.0 4. 10.0 8.S.00 −5.0 %Chg in adjusted residual 6.0 1985 1993 1987 1989 1995 2005 2010 1979 2006 1969 2012 1952 2011 2007 1981 1956 1960 1990 1994 1988 1977 1997 1953 1996 2004 1986 1992 1957 1961 1967 2001 1970 2002 0.6: Output and the Adjusted Residual Growth Rates in the U.0 2.S.00 0.0 1974 1982 1991 1980 2008 1949 1954 1958 1975 2009 −5.0 0.3.5: Output and the Adjusted Residual Co-movement in the U.00 %Chg in real value−added %Chg in adjusted residual Year−on−year growth rate 5.0 1984 1978 1959 1951 1955 1950 %Chg in real value−added 1973 1965 1976 1966 1962 1972 1964 1983 1999 1968 1998 1963 2000 1971 2003 5.4.

or some kind of exogenous variation in labor and/or capital supplies. then the points made in the prior chapter hold. Productivity does not appear to be correlated with business cycles.44) and find 1−α Et = Rest Kt Nt sN (µ−1) .4. 62 . Here. 4. but does it also affect our study of long-run growth? With perfect competition and constant returns. Now we have imperfect competition with a markup of 2. Regardless.6. And finally the figures are based on indicies of labor inputs and capital services. It is not the case that years with large increases in productivity tend to be years in which there is positive output growth.5 the growth rate of Et year-over-year is plotted against the growth rate of output. we can rearrange (4. To see whether Et is driving business cycles. So we do not have any definitive evidence that productivity shocks drive the business cycle. One can even see situations. We’ve continued to assume that there are constant returns to scale (meaning that capital’s share of costs is 1 minus labor’s share). And if it is not productivity shocks that drive productivity shocks. As can be seen. For the purposes here. the point is that simply looking at the basic Solow residual does not necessarily provide us with evidence that productivity shocks are responsible for business cycle fluctuations.does not appear to be as closely associated with business cycles. the residual would be a good way of measuring Et . there is essentially no relationship. but the residual will vary with the business cycle. unlike figure 4. which plots the growth rate of output against the growth rate of this adjusted productivity measure. One can argue for larger or smaller estimates.45) 1−α In figure 4. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS possible that Et is unchanging over time. we should back it out of equation (4. There needs to be some upward-sloping aggregate supply curve and nominal shocks. there is a less pronounced co-movement of the two series. (4. which may not accurately measure the size of K or N over time. such as the recessions in 2001 and 2009. we need to make some assumptions about the size of µ.once we more accurately measure it under imperfect competition .3. the residual doesn’t give us any precise information about productivity.44).4. This holds for the particular assumptions that we’ve implemented. Productivity . or growing steadily. A more useful way of seeing this is in figure 4. meaning that prices are twice marginal costs. which could be inaccurate.4 Productivity and Long-Run Growth With imperfect competition. where productivity appears to spike upwards when the economy is shrinking. To do so. but this value illustrates the possible impact of imperfect competition. This may also be incorrect. similar to figure 4. I set it to µ = 2. With µ. That clouds the analysis of business cycles.

Et . and we can reduce the first fraction so The fraction in the parentheses is just the definition of k that we get E sN Rest = s t . 63 .0 1950 1960 1970 1980 Year 1990 2000 2010 Figure 4.4.0 70. there has been steady upward growth in the residual.4. we think that k over time.7: Residual Index in the U. 100. With imperfect competition. and so will be unchanging growth.0 50. even if we are not precisely sure about the exact level of Et at any given point in time.0 60.47) ˜t N (µ−1) k ˜t . or g .S. Productivity and Long-Run Growth and we could track the growth in Et over time. As you can see. scaled by the size of k ˜t will settle down to a steady state value of k ˜∗ .46) ˜t . To see this. This. giving us a measure of trend growth in productivity. (4. From the perspective of long-run Now. Figure 4. write the residual as follows Rest = 1−α Et s (µ−1) Et N Nt Et Kt sN (µ−1) . If that is true.0 Residual index (2005=100) 90. the residual captures productivity Et . is the source of long-run growth in output per capita. despite fluctuations around that trend. the residual can still provide useful information even though it isn’t a pure measure of Et .0 80. From this we can conclude that there has been steady upward growth in productivity.7 plots the Solow residual over time as an index with 2005 set to 100. recall. then the residual can give us a good indicator of the growth of Et . (4.

The R-squared from this regression would be R2 = α2 V ar(ln kt ) . This seems high. then it would be the case that the residual might not give us accurate information on Et . Going back to the original definition of the residual in (4.6–0. if the values of sN or µ are changing over time. you’d find a coefficient on log capital per worker of about 0. given that capital’s share of output is around 0. (4. Remember that µ represents the markup over marginal cost charged by firms.51) 64 .33). and by not including ln Et in the regression. and there is some variance across countries in their ln yt .50) This R-squared tells us precisely the fraction of variation in ln yt that is accounted for by variation in ln kt . If we did that. running that regression will not give us a good estimate of α. That is because it is almost surely true that ln kt is correlated with ln Et . and this would have to reach ridiculous levels to explain the growth in the residual. 4.5 Cross-country Income Differences In thinking about why some countries are rich and some are poor. If we cannot use the simple regression. if you did run this regression. we can re-write this in per-capita terms as 1−sn yt = Rest kt . However. We’d like to know how much of this variance is due to variation in kt (factors). versus productivity (Et ). A simple way to think about doing this would be to simply regress ln yt on ln kt . and this would show up as a rising residual in the data.7.49) We have lots of countries.3–0. then the residual would be rising too. However. Growth in sN is also an unlikely candidate. (4. as it is bounded between zero and one. In practice.4.4. we will get a biased estimate of α. V ar(ln yt ) (4. we should get back a coefficient estimate of α. (4. In per-capita terms. a first step is to understand whether it is factors of production (capital and labor) or productivity that is driving the differences. If µ were climbing over time.48) which in logs is just ln yt = (1 − α) ln Et + α ln kt . it seems unlikely that µ could grow sufficiently and over this long of a period to drive residual growth. if production can be written as 1−α α yt = Et kt . then how do we determine something similar to the Rsquared for the role of capital per worker? One option commonly used involves the residual again. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS One caveat to this.

As before. but it is not systematically related to output per capita. rather than trying to estimate it from the data. then it suggests that capital per worker is sufficient to explain the variation in output per capita observed. then this success ratio is just (1 − sn )2 V ar(ln kt V ar(ln yt ) (4. If this ratio approaches zero. version 8. The PWT include a measure of human capital per worker in their data.52) where we’ve taken the variance of log capital per worker across countries.55) 65 . and others doing similar work. Gollin (2002) finds variation in sn across countries.54) and now the residual will pick up any remaining differnce in output per worker that is not due to physical capital or human capital. simply make an assumption about the size of 1 − sn . Cross-country Income Differences and note that this looks similar to what we have in (4. How is this useful? Consider the following ratio 1−sn V ar(ln kt ) V ar(ln yt ) (4.5. If this ratio approaches one. variation in physical capital per worker only accounts for about 20% of the variation in income per capita across countries.48). The success ratio associated with this kind of analysis would be 1−sn n V ar(ln kt + ln hs t ) . The numerator gives us the amount of variation we’d expect to see in output per capita if the Rest were identical across countries. V ar(ln yt ) (4. researchers have used the data on wages as a fraction of GDP to approximate sn to be about 0. if there is perfect competition 1−α and inelastic factor supplies then sn = α and Rest = Et . α If production is actually yt = kt (ht Et )1−α . Caselli (2005) refers to this as the success ratio. In general.4. We touched briefly on human capital in the prior chapter. on capital kt and the output yt . (4.21. Using data from the Penn World Tables (PWT).6. If we apply some simple rules for logs and variances. and divided it by the variance of log output per capita across countries. One could calculate the success ratio using this equation along with data on capital stocks and output per capita.53) and what Caselli is calculating is equivalent to the R-squared of a regression of log yt on log kt . and uses it as a measure of how important capital is in the variation in living standards across countries. then we could try to back out a residual using 1−sn sn yt = Rest kt ht . This is an average level. The difference with the regression is that Caselli. That is. Here. then it suggests that factors are unimportant in explaining variation in output per capita.0. we can add that additional capital type to the analysis. the success ratio is just 0.

. relative to the U.80 Figure 4. 2011 0. This leaves us with the conclusion that the residual is explaining far more of the gap between countries than accumulation of factors alone. Et .w. The level of h relative to the U. However. That factor is the residual.20 0.60 GDP p.S. indicating that capital stocks predict countries would be far richer relative to the U.S. we see that essentially every country is above the line. Et .S.8: Predicted and Actual Output p.S. than they actually are. we’d expect poor countries to be better off than they are.4. this success ratio is approximately 31%.00 0. making them poorer.20 0.40 0.S. The conclusion is that there must be some other factor that differs with the U. then the dots would line up along the 45 degree line. in 2011.80 0.w.S.60 0. relative to the U. relative to the U. is plotted against the observed level of yt relative to the U. which we think is closely correlated with productivity.8. 2011 0.00 0.. or not quite one-third of the variation in output per worker is due to differences in capital stocks of both kinds.S. k If capital (physical and human) were the only driver of differences in output per worker between countries and the U.. Looking only at capital stocks. 2011 A different way to see the role of physical and human capital stocks is figure 4.w.S. 1−sn sn Measuring Human Capital 66 ..40 0.. 1.00 Predicted GDP p. Thus gaps in living standards are likely due mainly to levels of productivity. P RODUCTIVITY: G ROWTH AND F LUCTUATIONS Using the human capital data.

However the actual empirics don’t tend to change. In practice. If this is the form that human capital takes. and depending on how one precisely models human capital one can get it to vary a lot or a little across countries.57) given our typical production function. you can do more sophisticated versions that allow for different returns for primary. and then variation in human capital by country depends solely on St . The Mincerian name comes from the fact that an equation such as this is what Jacob Mincer proposed as the best fit for the relationship between wages and years of schooling when studying individual workers.58) Log wages depend on an intercept term that involves the capital stock and productivity. It’s not essential that we assume that φ is identical for each year. however. many researchers will assume that the value of φ is identical for each country. we assume that human capital per worker can be written as ht = eφSt (4. There are departures from this. (4. and this is related to wages in a particular way.56) where St is years of schooling for a worker. Specifically. and φ is the percent increase in productivity that a worker gets for each year.4. Take logs of this and you have α 1−α ln wt = [ln(1 − α)kt Et ] + φSt . and then a second term that depends on the years of schooling. 67 . The fact is that we do not have a definitive answer for how to measure human capital in production. or tertiary education. secondary.5. Cross-country Income Differences How do we get a measure of human capital per worker? The norm in the literature at this point is to use a Mincerian setting. then wages can be written (assuming they are equal to marginal products) as α 1−α wt = (1 − α)kt Et ht (4. This means that we assume that human capital depends on years of schooling.


The “goods” that individuals will care about here are distinct in time. rather than in physical characteristics. providing one means through which output may fluctuate or grow over time. This does not appear to be a good description of reality. People will save so that they can enjoy more consumption later in life. as both investment spending (the supply of new capital) and employment (the supply of workers) fluctuate with the business cycle. respectively. we need to specify who is supplying them. 5. Once we’ve described how individuals act to supply these factors. Their budget 69 . To discuss what drives these supplies. Their optimization will result in elastic supplies of capital and labor. we’re going to need to explicitly write down a model of individuals that provide savings and labor to firms. as in a typical intermediate micro problem. and what their motives are. These individuals will have some kind of utility function that they are trying to maximize. people here are going to decide how to allocate their income between buying goods (of whatever kind) today versus buying them in the future. and they’ll take the rental rate for capital and the wage rate as given when they make their decisions. and then incorporate an explicit labor supply decision once we have the basic framework in place. we need to understand that it is fundametally a problem in maximizing utility subject to a budget constraint. In addition.1 The Fisher Model To begin describing savings. we can fold these decisions into the Solow model and see how this changes the predictions of that model.C HAPTER 5 Savings and the Supply of Capital One of the key presumptions we’ve made to this point is that the supply of capital and labor is fixed and inelastic with respect to the rate of return or wage. In other words. and there are long-run trends in these supplies within countries that we would like to understand at some more fundamental level. Rather than choosing how to allocate their income between beer and pizza. investment spending and employment vary across countries. We’ll begin by talking about the savings decision alone.

They have a utility function that looks like this V = U (c1 ) + βU (c2 ) (5. More nuanced models will discard this property (i. then notice that utility actually has a negative value. If σ > 1. we assume that: U ′ (ct ) > 0 and U ′′ (ct ) < 0 (5. However. which posits that people live for only two periods.2) which says that utility goes up as ct goes up. If β > 1 then people are willing to wait. the marginal utility of consumption is falling as ct goes up. Note that we assume that U (·) is identical in both period 1 and period 2. The goods they consume will be identical in all periods of time. The easiest setting in which to describe the savings decision is the Fisher model. but will now also depend on when they earn it. At times it is convenient to flip back and forth in terms of solving problems. Constant Relative Risk Aversion A specific type of utility function that we will use frequently is the constant relative risk aversion (CRRA) one. this is immaterial. If β < 1 then people are impatient.e. Therefore. but the relative price of consumption goods today compared to consumption goods tomorrow will vary with the interest rate. additive separability. The per-period utility function is concave in consumption.1) has a very important property.5. or the utility that one gets from consuming in any particular period. what you consume yesterday has no influence on your utility today. we make assumptions about the form of the period-specific utility function U (·). θ . Another way of saying this is that the marginal utility of consumption in period t is independent of consumption in any other period.1) where V is total lifetime utility. What you consume today has no influence on your utility tomorrow. with durable goods that give utility in multiple periods) but for now we retain it for the simplicity it brings. where β = 1/(1 + θ ). and U (·) is sometimes called the “felicity” function. This means that the marginal rate of substitution between any two periods (say t and t + 1) is independent of any other periods. The term β is the time preference parameter. as 1 It’s also common to see β written in terms of a discount rate. 70 . S AVINGS AND THE S UPPLY OF C APITAL will depend on the income they earn. and would prefer to consume today rather than in period 2.3) where σ is the coefficient of relative risk aversion. In addition to additive separability. This has the form U (c) = c1−σ where σ > 0 1−σ (5. Specifically. If σ = 0 then utility is just linear in consumption. It scale the value of consumption in period 2 by some amount compared to period 1.1 The utility function described in (5.

as well as our risk aversion. We call this the elasticity of inter-temporal substitution (EIS). An implication of the assumption of concavity is that individuals would like smooth consumption.6) which is an indication of the curvature of the utility function. So the parameter σ measures the degree to which we are willing to substitute consumption across time. even though the coin flip has a higher expected value of consumption. More concretely. Specifically.5. A second implication of this concavity is risk aversion. The Fisher Model we’ll see the function still has the proper derivatives. if σ = 1 then the utility function reduces to U (c) = ln c. If σ is large. Taking the derivatives yields U ′ (c) U ′′ (c) = c−σ > 0 = −σc−σ−1 < 0 (5. Individuals will actually give up some consumption to avoid having to face a lottery. 71 . In other words. Finally.1. We’ll see exactly how this fits in with risk later on in this chapter. the more the utility function “curves down” as c increases. How does the utility function reflect consumption smoothing? In other words. what is the elasticity of consumption with respect to a change in the marginal utility of consumption? How much will a person move consumption across time when the marginal utility of this consumption changes. even though the total consumption is identical. In this case − ∂U ′ (c) c U ′′ (c) = −c ′ =σ ′ ∂c U (c) U (c) (5. it tells us how severely average utility goes down relative to the utility of an average.4) (5. an individual will prefer to consume c1 = 15 and c2 = 15 to consuming c1 = 5 and c2 = 25. That is. an individual might prefer to consume c1 = 15 rather than facing a coin flip between c1 = 10 and c1 = 30. Risk aversion measures the elasticity of marginal utility with respect to consumption.5) which matches the assumptions we made regarding utility functions. individuals would rather consume similar amounts in periods 1 and 2 than having a big jump (or drop) in consumption. The larger is σ . and we will not respond much to changes in marginal utility. the EIS is small. It is calculated as U ′ (c) 1 ∂c U ′ (c) = − = (5. The name of the function comes from the definition of risk aversion.7) − ∂U ′ (c) c cU ′′ (c) σ and we can see that it is simply the inverse of the coefficient of relative risk aversion.

which means they have savings of s1 = w1 − c1 . This is now a simple constrained optimization problem. employing one additional assumption.15) = 0 = 0 = 0 (5. Putting those two equations together and re-arranging we have c1 + c2 w2 = w1 + 1+r 1+r (5.5. then in the second period the (1 + r)s1 term is capturing how much they have to pay back to the lender. We assume that they earn a return of r on their savings between periods 1 and 2. but it will be useful to practice using a Lagrangian to solve it. βU ′ (c2 ) 72 (5. their consumption depends not only on their wage. respectively. Form the Lagrangian as L = U (c1 ) + βU (c2 ) + λ w1 + c2 w2 − c1 − 1+r 1+r (5.9) We can combine those two conditions easily.14) which we can then use to solve for the optimal consumption path. If they do borrow. We could simply plug the budget constraint into the utility funtion and solve. the Lagrange multiplier.10) which is sometimes referred to as the lifetime budget constraint of the individual. and λ. S AVINGS AND THE S UPPLY OF C APITAL To go along with the utility function. This yields the following first order conditions: U ′ (c1 ) − λ λ βU ′ (c2 ) − 1+r c2 w2 − c1 − w1 + 1+r 1+r solve them together for the following: U ′ (c1 ) = (1 + r). so that c2 = w2 + (1 + r)s1 . That is that the individual can borrow and lend freely at the rate r. we need some kind of budget constraint.11) and maximize with respect to c1 . The left-hand side is the present discounted value of all consumption.12) (5. while the right-hand side is the present discounted value of all income. c2 . We’ll assume that the individual earns w1 and w2 in wages in the two periods of life. (5. We’ll address what changes if they cannot borrow or lend freely later on. (5.8) In their second period of life. In the first period of life they consume an amount c1 . they can. If they’d like to borrow in the first period (so that s1 < 0).13) (5. but on the savings they set aside in period one. Take the first two conditions and .

The Euler equation yields an answer on how c1 and c2 should be related over time. or the person has falling consumption over time. That is. When will this be true? When either β is very large (and so the individual values the future a lot) or when (1 + r) is large (and so the cost of consuming is very high). c∗ 1 . The budget constraint is fixed by the location of w1 . the budget constraint. then it must be that U ′ (c1 ) > U ′ (c2 ). An Exact Solution with CRRA Utility 73 . consumption should be rising over time if β (1 + r) > 1. then I will not be able to earn interest on it in period 2. This refers to the solution to an optimal dynamic programming model. From the Euler equation. In other words. given our assumptions. In the figure. Likewise. I’d have to give up (1 + r) units of c2 in order to afford an additional unit of c1 . the relative price of c2 in terms of c1 is 1/(1 + r). Figure 5. So what keeps the individual from only buying c2 ? The marginal utility of c1 relative to c2 . and those savings are used to increase consumption in period 2 above their actual wage. c2 . this leaves them with positive net savings in period 1. As c1 drops the marginal utility rises.1 presents the optimal choice problem. but does not describe their exact values. Why? Because you would be foregoing interest payments. If β (1 + r) = 1 exactly. Why? Because if I consume an extra unit in period 1. so at some point it will be worth buying c1 .5. Visually. Note that the relative price of consumption in period c1 is strictly higher than in period c2 . this is no different than a typical intermediate microeconomics problem. It can be helpful to write this as U ′ (c1 ) = β (1 + r). which represent the endowment of the person. note that if β (1 + r) > 1. The Fisher Model This is simply the ratio of marginal utilities equal to a ratio of marginal costs. w2 . The Euler equation describes how the control variable (consumption) is related over time. In addition. U ′ (c2 ) (5. so the individual will be less inclined to want to purchase that good. note that the marginal utility of c2 is scaled down by β . consumption today is more expensive than consumption in the future. That is their optimal choice for consumption. They look for the point at which their indifference ∗ curve is tangent to this budget line. That is.16) and either this form or the prior one are often referred to as the Euler equation.1. The inverse holds as well: if β (1 + r) < 1. which we are doing in a simplified setting. or c1 < c2 . then c1 = c2 and the person chooses to have the same amount of consumption in both periods. I have to give up 1/(1 + r) units of consumption today in order to afford an additional unit of consumption tomorrow. To find this we need to employ the final first-order condition. then it will be the case that c1 > c2 . The interest rate (1 + r) is the relative cost of c1 in terms of c2 . assuming that r > 0.

the amount of c2 relative to c1 rises.19) Note that this preserves the idea that as β (1 + r) rises. The slope of the budget line is −(1 + r ). (5.1: Optimal Consumption Decision Note: The figure shows the optimal decision for an individual with V = U (c1 ) + βU (c2 ) and a budget constraint of c1 + c2 /(1 + r ) = w1 + w2 /(1 + r ). Using this the Euler equation becomes σ c− 1 σ = (1 + r).17) where σ is the coefficient of relative risk aversion. w2 ). Using CRRA utility the felicity function is U (c) = c1−σ where σ > 0 1−σ (5. c2 ). here we can see that an important factor is the intertemporal elasticity of substitution. which dictates where the budget line must go. 74 . However. which occurs at the point (c∗ 1 . βc− 2 (5. The optimal point is where ∗ the indifference curve is tangent to the budget constraint. The individual is thus a net saver.5. or the relative cost of c1 with respect to c2 . The individual has an endowment of (w1 . S AVINGS AND THE S UPPLY OF C APITAL c2 U (c 1 ) Indifference curve with slope = − βU ′ (c ) 2 ′ c∗ 2 (1 + r)s1 w2 s1 c∗ 1 w1 Budget constraint with slope = −(1 + r) c1 Figure 5.18) This solves to c2 = c1 [β (1 + r)]1/σ .

However. which would be the case if the person had log utility.22) This is very ugly. and w2 = 0. 1 + r. regardless of the value of β (1 + r). then c2 will tend to be very similar to c1 . then savings will be positive for sure. The Fisher Model 1/σ . we can think about how savings (and hence the supply of investment in the economy) responds to changes in the return on savings. With this optimal consumption/savings problem in place. Imagine that there is an increase in 1 + r. notice that whether savings are positive or negative depends in part on the relative size of w1 and w2 . The higher it is. The individual is unwilling to substitute between periods even if the interest rate or time preference rates are pointing in that direction. We can simplify somewhat if we consider different extreme cases. and so any small change in β (1 + r) will have a large impact on relative consumption. In this case the fraction reduces to (1 + r)/(2 + r). as you would expect. On the other hand. then they must borrow (s1 < 0) for sure. (1 + r) (5. so the optimal solution is phrased as a proportion of lifetime wealth. then individuals will consume almost exactly half of their lifetime wealth in each period.5. (5. If w1 is very large. the less they consume in period 1.21) There are two parts to this solution. The fraction depends on β and (1 + r). If σ goes to infinity. β . which is very close to 1/2. then the person is willing to substitute consumption between periods. The object in parentheses is the PDV of lifetime wealth. First is a fraction telling us what proportion of the value in parentheses the individual should consume in the first period. U = ln(ct ). Does this increase savings? That depends on three 75 . In other words.20) which can be solved for c∗ 1 = 1 1 + β 1/σ (1 + r)1/σ−1 w1 + w2 (1 + r) . What about savings? They are s1 = w1 − c∗ 1 = β 1/σ (1 + r)1/σ−1 1 + β 1/σ (1 + r)1/σ−1 w1 − 1 1 + β 1/σ (1 + r)1/σ−1 w2 . If w1 = 0 and w2 is large. The fraction of lifetime wealth that should be consumed depends only on the time perference parameter.1. If 1/σ = 1. if they do not like to substitute between periods. then the EIS 1/σ goes to zero. The individual must save something. then the fraction is just 1/(1 + β ). if the EIS is small (σ is large). We can use this relationship in the budget constraint to find c1 + w2 c1 [β (1 + r)]1/σ = w1 + (1 + r) (1 + r) (5. If the EIS gets very large (σ is small).

What this means is that savings does not respond monotonically to the rate of return. now consider someone who has w1 > 0 and w2 = 0. an increase in 1 + r actually lowers their wealth by decreasing the present discounted value of their income. In contrast. if 1 + r goes up. Finally. defined as s1 = w1 − c1 . The only effect of the rate of return on savings is through the wealth effect. would therefore rise. consider someone who has w1 = 0 and w2 > 0. their lifetime wealth is unaffected. As part of the homework you will work out the derivative of savings with respect to 1 + r. the substitution effect dominates. This necessarily means they save less. Here. if EIS > 1. They must borrow to consume in period 1. Hence the income effect dominates and they want to consume more in period 1. It depends on the nature of the utility function (through the EIS) as well as individuals initial position as a saver or borrower. Alternately. then the substitutability of consumption across periods is low. c1 goes up. or all their wealth is in the second period of life. it is worth considering the log form of utility. • The Wealth Effect: In addition to the effect of 1 + r on how individuals allocate their wealth between periods. savings always rise with the rate of return if the individual is a borrower. all else being equal (which it is not).5. Because of this. • The Income Effect: An increase in 1 + r also allows the individual to have higher consumption in the last period for any given amount of savings. and therefore savings tend to go up. individuals do not mind moving consumption between periods. Their savings. and therefore may actually increase their consumption in the initial period. In this case. and the income effect dominates. so that they are always a saver. the wealth effect acts to lower consumption in the initial period. and savings increase in response to 1 + r rising. Whether this income effect outweighs the substitution effect depends on the intertemporal elasticity of substitution (1/σ in the CRRA). but the general pattern that holds once we assume that EIS < 1 can be seen in figure 5. S AVINGS AND THE S UPPLY OF C APITAL separate effects: • The Substitution Effect: The increase in 1 + r is like an increase in the price of consumption in the initial period. the increase in the rate of return means they can save less but still enjoy a large amount of consumption in the future. This effect essentially expands the set of consumption combinations lying within their budget constraint. and the EIS is precisely one. Now. This induces the individual to consume less c1 and more c2 . To gain some intutition for this. Therefore they will consume less in period 1.2. In this case the income and substitution effects exactly offset each other. and if they are doing less consumption in period 1 they need to borrow less (save more). If the rate of return goes up. this means a big decrease in their lifetime wealth (w1 + w2 /(1 + r)). meaning that σ = 1 in the CRRA. savings 76 . the individual is unwilling to lower consumption much in the initial period. and savings actually decline. If EIS < 1. For this individual.

are always increasing with the rate of return. individuals will adjust their savings behavior. 5. with w1 > 0 and w2 = 0. if firms will pay a higher rate of return. and hence to the capital stock. In fact. In other words. The Over-lapping Generations Model s1 s∗ l for σ > 1 and w1 > 0.2. w2 > 0 Saving s∗ 1 for σ = 1 and w1 > 0. individuals who are always savers. Recall that in the Solow model the return on savings was irrelevant to the amount of investment. however. The over-lapping generations (OLG) model is a specific (and not the only) way of setting up ac77 . will save a specifc amount of their wealth. and the savings curve in figure 5.2 never reaches a point where it dips. as the wealth effect dominates the income effect. if σ = 1 and w1 > 0 while w2 = 0. Hence. and this will not change regardless of what the interest rate may be.2 The Over-lapping Generations Model We now have enough knowledge to incorporate endogenous savings into the original Solow model of accumulation we developed. Once savings are already large. the income effect dominates and savings actually decrease with further increases in rates of return.5. then the savings curve is just the horizontal line noted in figure 5. w2 = 0 0 Borrowing 1+r Figure 5.2. Savings rise monotonically at low levels of 1 + r .2: Savings and the Rate of Return Note: The figure shows the optimal savings for an individual with an intertemporal elasticity of substitution of less than one. The Fisher model gives us a way of letting investment spending depend on the return on savings.

S AVINGS AND THE S UPPLY OF C APITAL cumulation and savings based on assuming that each period there is a new generation of Fisher-ian consumers born. The division by 1 + n captures the diluation of the investment across a larger number of young people. This means that capital falls apart completely every period. We assume that only young individuals provide labor.24) (5. Kt+1 = It In per-worker terms. or δ = 1. slopes downward. for whom period t is their period 1.5. Hence every young individual will save something for their old age. Given these assumptions. and will only be able to consume by saving some of their wages as young people. for whom period t is their period 2. it’s worth solving out the simple OLG model. This will work for finding a solution. This includes models of fertility and bubbles involving nominal assets. we get around this by assuming that utility is of the form U (ct ) = ln ct . 1+n (5. And old consumers. and it is their period 2. 78 . We will additionally assume that each generation is 1 + n times larger than the prior one. And here’s where we run into a particularly annoying problem. we have Kt+1 Nt+1 kt+1 = = It Nt Nt Nt+1 it .23) This doesn’t look exactly like the Solow model of accumulation because of the odd way in which we’ve modeled population growth. this means that the income effect outweighs the substitution effect for these individuals. we can describe the accumulation of capital as simply. In terms of our Fisher model. If we assume a general CRRA form for the U (ct ) function. while it is the investment per young person in period t.25) (5. we’re assuming that w1 > 0 but w2 = 0. The supply of savings. Then there is a new generation of young people in period t + 1 for whom t + 1 is their period one. Young consumers. To keep things relatively simple. In period t + 1. as it forms the basis for a number of models that are not particularly interested in savings rates per se. Generally. so to speak. In period t. And given that we are fairly certain that σ > 1. this means that savings do not change with respect to the rate of return. then what will occur given our assumption that w1 > 0 and w2 = 0? It will be the case that there is no wealth effect. but note that we’ve now assumed away the reason for endogenizing savings in the first place .to let it depend on the rate of return. To go further we need to specify the utility function for individuals. then. there are two generations of consumers alive. Along with the assumption that people only have wages in period one. Accounting for capital per young person is done because of a second assumption that we impose to keep the problem manageable. kt+1 refers to the capital per young person in period t + 1. Regardless. we typically assume that depreciation is complete. the young consumers from period t are now old. Old individuals cannot work. and any increase in 1 + r will actually lower savings.

lifetime utility is V = ln c1t + β ln c2.27) So to finish off the model we need to specify how savings and investment are related.3. From equation (5. accumulation of capital is driven by the equation kt+1 = it . that need not necessarily be the case. or savings are costlessly transformed directly into investment goods. 1+n (5. (5. For now. If we use a typical Cobbα Douglas production function. Capital per worker in the future depends on capital per worker today. k = 0 is a steady state as well 79 . then we have that α wt = (1 − α)kt . k ∗ . kt+1 > kt . (5. You can solve this model out for optimal savings of a young person in time t.t+1 .28) kt+1 = 1+n 1+β 1+n To complete the model. This is easy to see visually in figure 5. the economy will reach a steady state with a constant level of capital per worker.2. yt = kt . let’s continue with the presumption that total savings are costlessly translated into new investment. First. Typically.t+1 /(1+ rt+1 ) = wt . The capital per worker continues to grow until it reaches k ∗ .2 . 2 As with the Solow model. we’d just assume that it = s∗ 1t . So we could add some detail on how the financial market works. First. At any level of kt less than k ∗ . only some fraction of it can be actually translated into investment. 1+β 1+n (5. we need to describe how wages are determined. as shown at point A in the figure. the collection of savings and loaning of them to firms to do investment is something that is handled (often) by the financial sector.30) Here we have a difference equation in k . However.5. which are s∗ 1t = β wt . Similar logic explains that if kt > k ∗ then the capital stock will fall towards the steady state. we could propose that for every unit of output that is saved. where I’ve indicated both the time period from the perspective of the individual (1 or 2) and from the perspective of the economy as a whole (t and t +1).29) Putting this into the prior equation yields kt+1 = α β (1 − α)kt . Hence it is stable. We can work through the implications of this simple OLG model similar to what we did with the Solow model. 1+β (5. Alternatively. and the capital stock is increasing. We already noted that their lifetime budget constraint is c1t + c2. The level of kt+1 is mapped against kt . This implies that wt β s1t = . The Over-lapping Generations Model In the simple OLG.26) As noted above.30) you can see that the relationship is concave.

set kt+1 = kt = k ∗ . The OLG assumes total depreciation (δ = 1) and this over-lapping means of population growth.31) This let’s us consider the second set of implications of the Solow model. S AVINGS AND THE S UPPLY OF C APITAL kt+1 kt+1 = kt β 1−α α 1+β 1+n kt A k0 k∗ kt Figure 5. As β/(1 + β ) is the savings rate. From an original endowment of capital of k0 . Recall from the Solow model that I said one could approximate the growth rate of capital 80 .3: Dynamics in the OLG Model Note: The figure shows the evolution of kt+1 given the value of kt .5.30). Solving for the steady state from (5. Where this curve crosses the line of equality. and decreasing in the population growth rate. this is a similar conclusion that we drew from the Solow model. kt+1 = kt . Then we have k∗ = β (1 − α) 1+β 1+n 1/(1−α) . one can follow the evolution of the capital stock period by period as it climbs towards the steady state. (5. But it’s simply an accounting difference. there is a steady state. and those combined end up giving us a slightly different way of tracking what happends to capital. What’s the Difference between Solow and OLG? There isn’t much of one. The steady state is increasing in the time preference rate β .

32) However. kt Kt Nt (5. one where individuals live for an infinite number of periods. if we assume that savings are instantly translated into investment. the fixed savings rate s is just s = 1+β (1 − α). kt (1 + n) Kt (5.36) β as stated before. with the OLG a special case of the Solow. or individuals that are concerned about the future beyond their own lives (for example.3. Infinite lives just expands on the original Fisher model set up. we know that st = 1+ β (1 − α)yt . because we’ve assumed that δ = 1. we can turn to an expanded savings problem. But the two models are essentially identical. The utility function can be written 81 . Essentially. We also know that Kt+1 = It in the OLG. In the OLG. So we have kt+1 = st .34) Multiple both sides of this equation by kt and you have kt+1 − kt = Kt+1 /Nt − kt . Infinitely-lived Savers per worker as ∆kt+1 ∆Kt+1 ∆Nt+1 ≈ − .3 Infinitely-lived Savers With our intuition from the Fisher model.35) You can drop the kt from both sides. the exact relationship is Nt ∆kt+1 = kt Nt+1 ∆Nt+1 ∆Kt+1 − Kt Nt . we will get the same relationship of kt+1 to kt that we have in the OLG. Finally. So from the perspective of the β Solow model. then It = st Nt in the OLG. The infinite number of periods is non-sensical on the face of it. 1+n (5. by Nt we now mean just the young generation of workers.5. Using that and re-arranging slightly the above equation we have Kt+1 1 ∆kt+1 = − 1. (1 + n) (5. First. And we know that Nt+1 = (1 + n)Nt . they care about their children’s consumption).33) If we start from this exact equation in the Solow model and apply the OLG assumptions. but is useful as a reference. (5. 5. we are interested in individuals with long time horizons.

38) and thus β and θ capture the same effect . You will simply need to get comfortable translating back and forth. We have a utility function. (5. rather than the preference parameter β . a1 . this equation describes how our assets evolve based on our consumption decision. and at is our existing stock of assets. We’ll be a little more specific here in writing the budget constraint in two different forms. That tells us how much we have available prior to making any choices. the preference for consumption in period t is β t . S AVINGS as: AND THE S UPPLY OF C APITAL ∞ V0 = t=0 β t U (ct ). (1 + r)t (1 + r)t t=0 ∞ (5. notationally. the present discounted value of assets at infinity must be zero. and for now we will assume that this is constant to simplify the mathematics. to use β .5. One somewhat tedious point is that we often will find it easier to refer to a person’s “time discount rate”.39) where at+1 are our stock of assets tomorrow. Sometimes it will be easier. ct is the amount we consume today. The dynamic budget constraint states that: at+1 + ct = wt + (1 + r)at (5. One cannot accumulate an infinite supply of assets. so that at any period t. For any given periods t and t + 1. Let θ be the rate at which people discount future utility. Note that I’m also running time from period zero forward.40) This lifetime budget is arrived at by applying the dynamic budget constraint over and over again until one has an expression for aT . nor can one accumulate a non-zero amount of debt (aT < 0). You can think of this as the percentage that utility increases by consuming today rather than tomorrow. This dynamic budget constraint is equivalent to the lifetime budget constraint. Very simply β= 1 1+θ (5. the time preference rate for t + 1 is β . 82 . now we require a budget constraint. which is what we worked with in the Fisher model. while other times θ is useful to keep equations clean. but note that now it is to the t power. From the perspective of time zero. The value (1 + r) is the rate of return on our assets. and asserting that aT /(1+r)T goes to zero as T goes to infinity. The only other thing we require given the dynamic budget constraint is an initial asset level. which is going to make accounting for the infinite sums easier. In other words. There is nothing different in thinking of preferences this way.we enjoy consumption today rather than consumption tomorrow. wt is our income today.37) where β is again the time preference rate. For infinitely lived individuals the lifetime budget constraint is ∞ t=0 wt ct = a0 + .

Person A obviously gets to enjoy higher consumption in every period. only the height differs. the budget constraint will tell us the level.4. then the t Euler equation is ct+1 1/σ = [β (1 + r)] . rather than for periods 1 and 2 specifically. Person A has high lifetime wealth.44) (5. If we wanted to solve for the exact path of consumption (e. this makes actually solving this problem somewhat difficult.3. the relative marginal utility must be equal to β (1 + r). (5. Infinitely-lived Savers To maximize utility subject to this lifetime budget constraint we can set up a very large Lagrangian. or equal to one. Note that this condition does not depend at all on the budget constraint. while person B has low wealth. Just as with the Fisher model.41) This is fine. which dictates the slope of the consumption path at every point t. as in ∞ ∞ L= t=0 β t U (ct ) + λ a1 + t=0 wt ct − (1 + r)t t=0 (1 + r)t ∞ . This is because the Euler equation. If this is the case. This can be seen in figure 5. What this says is that for any adjacent periods. However. whether consumption is growing or falling over time depends on whether β (1 + r) is greater than. These are 1 β t U ′ (ct ) − λ = 0 (1 + r)t 1 = 0 β t+1 U ′ (ct+1 ) − λ (1 + r)t+1 and solving these together to eliminate λ we have U ′ (ct ) = β (1 + r) U ′ (ct+1 ) (5. only written with respect to any two adjacent periods.45) 83 .42) (5. Needless to say. A Specific Solution for Infinitely-Lived Consumption −σ Assume that we have CRRA utility.g. The Euler equation essentially gives us a rule for how consumption should evolve from period to period. less than. ct (5.5. the pattern of consumption over time will be identical. t and t + 1. but has an infinite number of first order conditions. In other words. we do not really need all the first-order conditions to understand what the optimal solution should look like.43) which is simply the Euler equation we identified before. it does not matter how large is our total lifetime wealth. which depicts the optimal consumption path for two individuals. While the Euler equation tells us what the pattern of consumption is over time. then we need to incorporate the budget constraint. a specific value for c14 or c1456 ). is independent of the budget itself. but notice that the shape of the the consumption path is identical. Take the first-order condition for time t and time t + 1. so that U (ct ) = c1 /(1 − σ ).

define the following term as lifetime wealth.5. person A has a higher lifetime wealth than person B . and so consumption is higher at every point. S AVINGS AND THE S UPPLY OF C APITAL c Person A cA 0 cB 0 Person B t Figure 5. it must be the case that ct = c0 [β (1 + r)]t/σ (5.46) and we can put this into the budget constraint from (5. However.47) 84 . Therefore the slope of their consumption path through time is identical. W : ∞ W ≡ a0 + t=0 wt (1 + r)t (5.4: Consumption Paths Note: Both person A and person B face identical interest rates and have identical discount rates. Knowing this general Euler equation holds between any two given periods. Before doing so.40) to solve for the actual consumption path.

Can we describe ct in terms of remaining lifetime wealth at time t itself? Note that in any given period t. We can solve the problem the same as before. In contrast. this gives us an expression for ct as a function of W . No matter what period we are in.50) (5. while the Euler equation determines the path over time after c0 .52) 1+r We now have a description of consumption in period zero. (5.51) t=0 c0 (β (1 + r))t/σ (1 + r)t c0 1 1 − (β (1 + r))1/σ−1 c0 = and initial consumption is a fraction of total lifetime wealth. 85 . when β (1 + r) < 1 consumption is falling over time and the fraction of wealth consumed initially will be large. Only when β (1 + r) = 1 will consumption be perfectly smooth over time.3. This determines the level of initial consumption. as in figure 5.54) by using the Euler equation.55) can be referred to as the consumption function.46) that tells us what ct is given c0 .5. total lifetime wealth.55) or consumption in time t should be a fraction of lifetime wealth. However. and the fraction of lifetime wealth consumed at time zero is small. using the relationship in (5. or the rule for determining how much to consume in any given period. we could describe remaining lifetime wealth as ∞ Wt ≡ at + s=0 ws+t (1 + r)s (5. Note that in this last case. finding that ct = (1 + r) − (β (1 + r))1/σ Wt (1 + r) (5.4. Note that if β (1 + r) > 1 then consumption is growing over time. we should always consume the same fraction of our wealth.49) (5.53) and we can describe consumption in all future periods with respect to consumption in period t as cs = ct (β (1 + r))s/σ (5. Infinitely-lived Savers and therefore we can write the budget constraint as ∞ t=0 ∞ ct (1 + r)t = = = W W W (1 + r) − (β (1 + r))1/σ W (1 + r) (5. initial consumption reduces to the simpler function of r c0 = W. Equation (5. can we describe consumption in any given period t? We can.48) (5.

given a savings rate.57) where kt is the capital stock. We’ll see that the steady state solution to this model has many of the properties of the original Solow model. 5. so it must be that at = kt . S AVINGS AND THE S UPPLY OF C APITAL 5. δ is the depreciation of capital. we could imagine that this is an economy populated by a single individual. In the Solow model we presumed that ct = (1 − s)f (kt ). Second. kt .4 The Ramsey Model Recall that the Solow model told us how capital accumulated. Alternatively. the dynamic budget constraint can be written as follows f (kt ) − ct ∆kt+1 = −δ (5. meaning that the maximization is done with full knowledge of the ramifications of one’s decisions. The Ramsey model puts these two concepts together: the capital stock depends on the consumption decision.59) kt kt which is simply the equation of motion from the Solow model written more generally. we’ll consider the centralized problem.1 The Centralized Solution To begin. but predictions about how the economy acts out of the steady state differ. We are assuming in the Solow that the only assets in the economy are capital. In addition. we are assuming that the centralized decision-maker 86 .5. Consumption theory tells us that the amount we save depends on the rate of return to savings. Second. but now ct will be chosen optimally to maximize utility. The important part is that that we are solving for the optimal solution. The amount of capital in the economy determines the rate of return on savings. We can think of a benevolent dictator or social planner attempting to maximize discounted utility for the economy as a whole.4. f (·) is the production function. the dynamic budget constraint here is the same as the dynamic constraint in the consumption problem.58) Two notes about the dynamic budget constraint. (5. and the consumption decision depends on the capital stock. First.56) subject to the following dynamic budget constraint kt+1 = (1 − δ )kt + f (kt ) − ct (5. Maximization The problem is to maximize utility ∞ V0 = t=0 β t U (ct ) (5. a transversality condition will dictate that utility cannot become infinite t→∞ lim β t U ′ (ct )kt = 0.

and so we denote them λt . kt+1 dictates the stock of capital available and output.5. one for each period off into the future. However. recall that the first order conditions were ∂L = β t U ′ (ct ) − λ = ∂ct ∂L λ = β t+1 U ′ (ct+1 ) − = ∂ct+1 1+r 0 0 (5. in the period t + 1. the Lagrangian for this can be written as ∞ L0 = t=0 β t U (ct ) + λt [(1 − δ )kt + f (kt ) − ct − kt+1 ] (5. We’ll need to optimize using the dynamic budget constraint directly. note that now the multipliers are not identical. We do not know the full time path of income.63) The form of the final first-order condition follows because the term kt+1 shows up in two different constraints: in the period t constraint. The Ramsey Model earns all the income in the economy. So for periods t and t + 1. the value of kt is given. Notice that the first two first-order conditions are similar to the typical conditions from the optimal consumption problem.4. But for period t + 1. though.61) (5. However. we can examine only the first-order conditions from two adjacent periods to find the general rules the economy must follow. kt+1 .60) where the Lagrangian has an infinite number of multipliers. 87 . For any given period. which means she has f (kt ) − δkt available to spend. we had a simple way or relating these marginal values. kt+1 is one of the options on how to spend our current income. and the marginal value of wealth in period t + 1 was equal to λ/(1 + r). we can optimize over kt+1 . as we did in the consumption problem. (5.62) (5. This means we cannot roll up some lifetime budget constraint for the planner. so we do not have to maximize with respect to it. Why not? In the optimal consumption problem. In other words. That is because the decisions the central planner makes today will influence income in the future through the capital stock. At time zero. We want to maximize this Lagrangian with respect to ct .64) (5. So the marginal value of wealth was different across periods. and λt for each of the infinite number of periods.65) or the marginal value of wealth in period t was equal to λ. because we have the option about what to set this to. but because the rate of return was given. her wt is equal f (kt ) − δkt . we have the following first order conditions: ∂L = β t U ′ (ct ) − λt ∂ct ∂L = β t+1 U ′ (ct+1 ) − λt+1 ∂ct+1 ∂L = λt+1 [f ′ (kt+1 ) + (1 − δ )] − λt ∂kt+1 = = = 0 0 0.

we end up with U ′ (ct ) = β [f ′ (kt+1 ) + 1 − δ ] (5. the rate of return depends on our consumption decision. while the cost of an extra unit of consumption in the future is still 1/β .71) . Solving the first-order conditions from the Ramsey problem to eliminate the multipliers. It has the same interpretation as before. What the Ramsey model tells us is that the cost to one extra dollar of consumption today is equal to 1 − δ + f ′ (kt+1 ).5. the resource constraints tell us that ct+1 = = f (kt+1 ) − kt+2 + (1 − δ )kt+1 f (f (kt ) − ct + (1 − δ )kt ) − kt+2 + (1 − δ ) [f (kt ) − ct + (1 − δ )kt ] (5. consider just periods t and t + 1. marginal rate of substitution between ct+1 and ct is U ′ (ct ) ∂ct+1 =− ∂ct βU ′ (ct+1 ) which gives us the slope of the indifference curves for this economy. holding constant consumption in every other period of time.68) (5. constant. Visualizing the Solution To understand what the central planner or individual is doing between any two periods. we can examine the maximization in a simple graph. showing all the different combinations of output in the two periods that satisfy the constraints imposed by the fixed values of kt and kt+2 . and so we cannot make a simplification regarding the marginal value of wealth between periods. This equation describes an inter-temporal production possibility frontier (IPPF). in the Ramsey model. holding constant kt and kt+2 .70) (5. Holding V0 .67) which relates consumption in period ct+1 to consumption in period ct . we have to appeal to the third first-order condition that relates λt to λt+1 .66) U ′ (ct+1 ) which is a generalized form of the Euler equation. as well as the capital stock in every other period. S AVINGS AND THE S UPPLY OF C APITAL Now. The Euler equation has to be combined with the resource constraint kt+1 = (1 − δ − n)kt + f (kt ) − ct to solve for the optimal time path of consumption as well as the steady state solution. We can write lifetime utility as V0 = U (ct ) + βU (ct+1 ) + Ω where Ω represents the utility obtained in all other periods in life.69) (5. The slope of the IPPF is ∂ct+1 = − (f ′ (kt+1 ) + (1 − δ )) ∂ct 88 (5. the ratio of marginal utilities must equal the ratio of marginal costs for consumption across periods. First. Instead. and is constant. From the production side.

From this representation. subject to the IPPF. we look for the point at which the indifference curve is tangent to the budget constraint.72) meaning that the IPPF is a concave function. Figure 5. we can also ask ourselves an important question that links this problem to the decentralized problem we’ll explore later. is simply the slope of the IPPF and indifference curve at the optimal solution.5. we have U ′ (ct ) ∂ct+1 ∂ct+1 |Util = − |IP P F = − (f ′ (kt+1 ) + (1 − δ )) = ′ ∂ct βU (ct+1 ) ∂ct which you’ll see is just recovering the Euler equation relating consumption across periods. 89 (5.5: Inter-temporal Optimization and the second derivative of this relationship is ∂ 2 ct+1 = f ′′ (kt+1 ) < 0 ∂c2 t (5.73) . The individual is trying to maximize utility. What is the price of consumption in period t relative to consumption in period t +1 that is consistent with this equilibrium? That price.5 displays this optimal consumption decision.4. The Ramsey Model ct+1 cmax t+1 IP P F c∗ t+1 V0 = U (ct ) + βU (ct+1 ) Slope = 1 + rt+1 c∗ t cmax t ct Figure 5. As in a typical economics problem. As the slopes of the IPPF and indifference curve must be equal at this tangency. which we have called (1 + rt+1 ) in our consumption chapter.






This slope is 1 + rt+1 = f ′ (kt+1 ) + (1 − δ ). (5.74)

If we were to save one extra dollar, what would we receive? We’d get the marginal return on this new capital, but some of this dollar would be lost to depreciation, so the net return on the savings is f ′ (kt+1 ) − δ . Steady State Solution As in the Solow model, we can look for steady states in which consumption, capital, and output are all constant. What this steady state implies is that ∆kt+1 = 0 and ∆ct = 0. Let c∗ and k ∗ be the long-run steady state equilibrium levels of consumption and capital. From the Euler equation, we have in steady state that U ′ (c∗ ) = β [f ′ (k ∗ ) + 1 − δ ] U ′ (c∗ ) which implies that f ′ (k ∗ ) = 1 +δ−1 β (5.75)


and if we have an explicit form for the intensive production function we could solve this for an exact value of k ∗ . An explicit form of the utility function would yield exact solutions for steady state consumption. Recall from the previous section that we determined that the return to savings, 1 + r, is equal to f ′ (k ) + 1 − δ . In steady state, then, the return to savings is 1 + r∗ = 1 + f ′ (k ∗ ) − δ = 1 + 1 1 +δ−1−δ = β β (5.77)

which tells us that the steady state return to savings is a function only of the time discount rate. This has to hold, because the only way for a steady state to even exist is if consumption is constant over time. Consumption is constant over time only when (1 + r) = 1/β . This is one point where the usefulness of talking about discount rates (θ) is cleaner than talking about β . Recall that β = 1/(1 + θ). Given that, the steady state return to capital is r∗ = θ (5.78)

which tells us that the steady state interest rate is equivalent to the rate at which individuals discount the future. This provides a little intuition for how patience influences the return to capital and hence the capital stock. Impatient people have a high discount rate, and to get them to save enough to replenish the capital stock every period we must offer them a large return to their savings. The only way for the return to savings to be large is to have a small capital stock. Hence impatient people lead to small steady state capital stocks. 90

5.4. The Ramsey Model In addition, we can compare this steady state solution to the Golden Rule steady state of the Solow model. Recall that the Golden Rule steady state was achieved when we set f ′ (k GR ) = δ (ignoring population growth). In contrast, the Ramsey model tells us that the steady state marginal product of capital is f ′ (k ∗ ) = 1/β + δ − 1. As 1/β − 1 > 0, it must be the case that the marginal product of capital is higher in the Ramsey steady state than under the Golden Rule. This implies that k ∗ < k GR . In other words, the Ramsey model delivers a steady state capital stock below the Golden Rule level. Why does this arise? In the Ramsey model, we have incorporated a time discount rate, β , into the utility function, and therefore the short-run costs of achieving the Golden Rule (high savings and low consumption) are factored against the discounted future gains (high consumption). Because of the discounting, it becomes optimal to forego the Golden Rule consumption level in the long run in return for some higher consumption in the present. An Explicit Ramsey Model

Let us use a CRRA function for per-period utility, so that
U (ct ) =
−σ c1 t 1−σ


and the production function is Cobb-Douglas
α yt = kt .


In steady state, we determined that it must be that f ′ (k ∗ ) = θ + δ , which given the production function implies that 1/(1−α) α k∗ = (5.81) δ+θ and note the similarities to the Solow model steady state with the Cobb-Douglas function. In place of the savings rate, we have α, which you’ll recall is actually the Golden rule savings rate. However, in addition to dividing by the depreciation rate, we also must divide by a term related to the time discount rate. In steady state, the dynamic budget constraint is
k ∗ = k ∗α + (1 − δ )k ∗ − c∗ (5.82)

which can be solved for, using (5.81),
c∗ = 1 δ+θ

(1 − α)δ + θ αα








ct ∆ct+1 > 0

∆ct+1 = 0 ∆ct+1 < 0

k∗ Figure 5.6: Consumption Dynamics


Dynamics To understand what is happening out of the steady state, we can construct a phase diagram of the Ramsey model. This is built off of the two difference equations that the optimization delivers: the Euler equation and the dynamic budget constraint. They are: U ′ (ct ) = β [f ′ (kt+1 ) + 1 − δ ] U ′ (ct+1 ) ∆kt+1 = f (kt ) − ct − δkt (5.84) (5.85)

and both are non-linear difference equations. The phase diagram will look at how each variable, consumption and the capital stock, changes based on given values of ct and kt . Taking consumption first, note that the Euler equation implies a growth rate in consumption, ∆ct+1 , between periods t and t + 1. If the capital stock is at steady state, and f ′ (kt+1 ) = 1/β − 1 + δ , then it must be that ct = ct+1 and ∆ct+1 = 0. Graphing this, we have a vertical line at k ∗ in figure 5.6. Regardless of the level of consumption, if kt = k ∗ consumption is not changing. Now what if capital is not at the steady state? Consider kt < k ∗ . If the capital stock is smaller than k ∗ , the marginal product of capital is relatively large, and so the term β [f ′ (kt+1 ) + 1 − δ ] is 92

Turning to the capital stock. A similar analysis reveals that for any values of kt > k ∗ . ∂ct = f ′ (kt ) − δ ∂kt ∂ 2 ct ′′ 2 = f (kt ) < 0 ∂kt (5. and so U ′ (ct ) > U ′ (ct+1 ). it must be the case that ∆ct+1 < 0. must be hump-shaped. Let’s begin at steady state. we see that consumption (the difference between these two curves). The Ramsey Model ct ∆kt+1 < 0 ∆kt+1 = 0 ∆kt+1 > 0 k∗ Figure 5. Thus ct < ct+1 and ∆ct+1 > 0. we can perform a similar analysis. Mathematically. This tells us that ct = f (kt ) − δkt . So for any level of kt < k ∗ it must be that consumption is growing over time. with a maximum at the Golden Rule level of f ′ (kt ) = δ .5. then the present is “expensive” relative to the future. We cannot say exactly what ∆ct+1 is. as the present is “cheap” relative to the future.4.86) (5. but we do know that it is negative in this region.7: Capital Dynamics greater than one. when ∆kt+1 = 0. this follows from examining the derivatives of ct with respect to kt . If this ratio is greater than one.87) which indicates that the relationship between consumption and capital is convex with a maximum 93 . Graphing f (kt ) and δkt together as in the Solow diagram.

respectively. Off of this line. The intersection of the curves at point X is the steady state.7. Graphing this in figure 5. A and B violate the transversality condition and are non-optimal. indicated by the dashed lines. If the economy is consuming less than the steady state amount. ct < f (kt ) − δkt .6 and 5.8. This shows the phase diagram of the Ramsey model. Now. In those regions. Given an initial stock of k0 . there must be extra savings available to increase the capital stock.5. In only the southwest and northeast regions are the dynamics pointing towards the steady state. at the Golden Rule.8: The Ramsey Diagram kt Note: The Ramsey diagram combines the dynamics of consumption and capital. if we combine diagrams 5. there is only one path of consumption and capital. This should make intuitive sense. If the economy finds itself with a combination of consumption and capital that put it on this line. indicating the direction of change in both consumption and the capital stock in each of the four distinct regions defined by the two steady state curves. the solid line represents all the points at which ∆kt+1 = 0. what is happening? Below the curve. we have figure 5. ∆kt+1 < 0. The arrows indicate the dynamics governing consumption and capital in each of the four regions. that leads to the steady state in the long run. the point at which neither consump94 . the capital stock will not be changing. The point X is the steady state. only the consumption choice corresponding to C puts the economy on the path to the steady state. or the capital stock is growing. and so ∆kt+1 > 0.7. S AVINGS AND THE S UPPLY OF C APITAL ct ∆ct+1 = 0 X A C B ∆kt+1 = 0 k0 k∗ Figure 5. A similar analysis suggests that everywhere that ct lies above the curve.

The economy cannot simply jump to k ∗ . But we are not necessarily at the steady state. There is one point. internalizing the effect of consumption decisions on the accumulation of the capital stock. The decision facing the economy (and built in implicitly in our optimization problem) is to determine what the right level of consumption is in period zero that will put the economy on the trajectory towards the steady state. we will not leave this point. We can use this phase diagram to understand how the economy responds to shocks and changes in technology. and a similar analysis yields a similar path for the northeast quadrant in which we begin with more capital than in the steady state. 5. The point of the stable arms is that there is a single optimal consumption level for any given capital stock. will allow for the capital stock to grow as well as consumption. Consider an economy with the capital stock of k0 .4. it will put itself in the northwest quadrant. that if we select this initial level of consumption. This cannot be optimal. This solution is unsustainable because it implies infinite consumption and a zero capital stock in the long run. in that once we reach it. The Ramsey Model tion nor capital changes. the economy will approach the steady state point of X and will reach a consumption level that can be sustained indefinitely. it will begin to acquire capital and consumption will rise.2 The Decentralized Solution In the previous section we assumed that a single optimizing agent made the decisions for the economy. On the other hand. who are trying to optimize their profits by employing factors of production. but eventually the economy will spill over into the southeast quadrant of the diagram and consumption begins to fall while the capital stock accumulates to enormous proportions. Over time. Only one consumption level leaves the economy in a position to reach the steady state and sustain consumption into infinity. If the economy selects a very large consumption amount (point A). This path is called the “stable arm”. C. who are trying to optimize their lifetime utility by making savings decisions and b) firms. Now we separate the decision-making and consider a) consumers.4. because consumption goes to zero. if the economy chooses a very low level of consumption (so that savings are large). 95 . This point is stable. they must save over time to acquire the steady state capital stock. violating the transversality condition. and the phase diagram allows us to analyze how the economy will act outside of the steady state. less than k ∗ . but prior to addressing those issues we will establish that the same solutions we derived here can be obtained by a decentralized economy in which individuals and firms are acting independently.5. and the dynamics in this region tell us that the capital stock will start falling and consumption will start rising.

94) . Firms require both capital and labor to operate. xt is additional income (dividends. To obtain capital for use in period t.89) where at are the assets of individuals. Nt ) = FK (Kt . paying a rate Rt for those funds. wt . ∞ V0 = t=0 β t U (ct ) (5. and xt as given. we assume that all firms exist for only one period. we consider a perfectly competitive economy. and ct is their consumption. For simplicity. The value rt is the interest rate that individuals can earn in the financial market in period t. The individual takes the time path of rt .90) U ′ (ct+1 ) This gives us our first piece of the solution. which says that U ′ (ct ) = β (1 + rt+1 ). Profits in period t will be Πt = F (Kt .92) (5. a firm borrows money from the financial market. and the return Rt is their net cost of capital.93) To cast these results in per capita terms so that they align with our individual problem.). wt is labor income. consider writing the firms profits as Πt = Nt [f (kt ) − wt − Rt kt ] 96 (5. Nt ) − wt Nt − Rt Kt (5. etc.5. Nt ) = wt Rt .88) subject to the dynamic budget constraint at+1 = (1 + rt )at + wt + xt − ct (5. (5. Maximization yields that FN (Kt . Individuals maximize utility. so they do not have long-run financing considerations. Firms As discussed in the chapter on the Solow model. We know that the solution for the individuals is to have their consumption be dictated by the Euler equation. The firm’s demand for these funds depends on it’s profit maximization decision. (5. S AVINGS AND THE S UPPLY OF C APITAL To fit this all together.91) where wt is the wage rate the firm takes as given. we will need to introduce a financial market that links the savings of individuals with the capital requirements of firms. Individuals Here we have nothing more than the infinitely-lived consumption problem of the previous chapter.

95) (5. and from them it receives an amount (1 + Rt − δ )Kt .98) = 0 0 f (kt ) − f (kt )kt ′ (5. (5.101) (5. To clear the market it must be that these amounts are equal Nt (1 + rt )at = (1 + Rt − δ )Kt (5. What this implies is that (1 + rt ) = rt = (1 + Rt − δ ) Rt − δ. Nt )Kt (5. it must be that F (Kt . (5. This market received deposits of at from Nt individuals. Market Clearing and Equilibrium The financial market we have in mind here works with zero frictions.104) 97 (5. The Ramsey Model and maximizing over Nt yields [f (kt ) − wt − Rt kt ] + Nt 1 −f ′ (kt )kt = + Rt kt Nt Nt [f (kt ) − wt − Rt kt ] + [−f ′ (kt )kt + Rt kt ] = wt and the maximization over capital yields the simple f ′ (kt ) = Rt .102) From the firm maximization problem.99) and given the firms profit-maximizing decisions.97) As firms are perfectly competitive and the production function is constant returns to scale. Nt ) = FN (Kt . Nt )Nt + FK (Kt .100) and as this is a closed economy. we know that they will employ capital until Rt = f ′ (kt ). and must repay those individuals an amount Nt (1 + rt )at . The financial market loaned an amount Kt to firms to use in production.5. the total assets in the economy must be equal to the total capital stock. Using this last result in the Euler equation for individuals yields U ′ (ct ) = β (1 + f ′ (kt+1 ) − δ ) U ′ (ct+1 ) (5. where note that we have incorporated the fact that some of the capital depreciates in use. or Nt at = Kt . and so the total amount returned to the financial market is smaller by this amount.103) . it must be the case that profits (xt ) are equal to zero.96) (5.4. so that in the end we have rt = f ′ (kt ) − δ.

At .106) (5. and we know from (5. we know from (5. Given the dynamic budget constraint. and so the market solution is Pareto Optimal.108) . they must have the exact same solution for the steady state and the dynamics. Putting all this information together in the dynamic budget constraint. but as all of output is paid to either labor or capital (i. As the equations describing the motion of consumption and capital over time are identical in the de-centralized and centralized economy. we have kt+1 = (1 + f ′ (kt ) − δ )kt + f (kt ) − f ′ (kt )kt − ct which reduces to kt+1 = f (kt ) + (1 − δ )kt − ct (5. and xt = 0. What we’ve seen here is that a decentralized solution delivers exactly the same result. We have complete markets in all goods (consumption in every period). meaning that all costs and benefits were internalized by the optimizing agent. We can write the growth rate of assets per capita as ∆at+1 ∆At+1 ∆Nt+1 = − at At Nt (5.103) what the return on assets is. Note that the centralized solution was an optimal solution.105) and this is identical to the original accumulation function for capital we described in the centralized model. we assume that the total population (or alternately.107) which we can evaluate by determining the growth rate of aggregate assets. it must be that At+1 = (1 + rt )At + wt Nt + xt Nt − ct Nt 98 (5.97) how to describe the wage rate. all of the households) are growing at the rate n per period. we can consider how the model changes when we incorporate growth in the number of people and growth in productivity.e. The value xt represents other income to the individual. We know from market clearing that at+1 = kt+1 and at = kt .5. Technology. S AVINGS AND THE S UPPLY OF C APITAL which is identical to the Euler equation derived for the centralized economy. and Growth As we did previously. there is no other income to pay out. Population Growth To add in population growth.4.3 Population. at+1 = (1 + rt )at + wt + xt − ct . Finally. knowing that the market can (under our assumptions about how firms and markets work) deliver the same results. profits are zero). This is an example of First Welfare Theorem in action. 5. Now consider the individuals dynamic budget constraint. This is useful because we can concentrate on working with the centralized problem.

and xt = 0 which leads to kt+1 = f (kt ) + (1 − δ − n)kt − ct and this is exactly what we derived in the Solow model.110) (5. Therefore.112) (5.115) or the growth rate of population is smaller than 1/β − 1.109) Given this budget constraint. which you’ll recall is the steady state value of r∗ .4. utility is the discounted sum of per capita consumption. 99 . it has to be that β (1 + n) < 1 (5. That is. So if we have utility over total utility of each generation. The Ramsey Model so that the growth rate of assets per capita can be written as ∆at+1 wt xt ct = rt + + − −n at at at at which can be rearranged into the following constraint at+1 = (1 + rt − n)at + wt + xt − ct . For this formulation of the problem to make sense. as increasing the number of individuals each period makes maintaining a large capital stock per person more difficult. the steady state level of capital is lower the larger population growth is. As we did before. If we did not. ∞ V0 = t=0 β t N0 (1 + n)t U (ct ) (5. This follows somewhat mechanically.5. Note that we did not alter the utility function in this analysis. wt = f (kt ) − f ′ (kt )kt .111) and notice that the addition of population growth has increased the marginal product of capital in steady state. rt = f ′ (kt ) − δ . Solving with this dynamic constraint. An alternate version of utility would be the following. (5. then utility (which is based on assets per person) could grow to infinity. we have an Euler equation U ′ (ct ) = β [f ′ (kt+1 ) + 1 − δ − n] U ′ (ct+1 ) which then leads to a steady state value of k ∗ that solves f ′ (k ∗ ) = 1 +δ+n−1 β (5. the financial market sets assets equal to the capital stock. we need the population to grow more slowly than assets.113) (5.114) and utility is the discounted sum of total utility in the economy. the Ramsey model solution is not functionally different.

We can evaluate the growth rate of assets per efficiency unit using this equation.116) where the ˜ · indicates a variable written in per-efficiency unit terms.119) (5. and therefore ˜ ˜t ) − f ′ (k ˜t )k ˜t .123) (5. Et . Utility maximization takes place as before. we kt = a ˜t . (5. Applying this. which is simply consumption per person. (5.121) and the first order conditions for two adjacent periods t and t + 1 are ∂L = β t U ′ (˜ ct Et )Et − λt ∂c ˜t ∂L = β t+1 U ′ (˜ ct+1 Et+1 )Et+1 − λt+1 ∂c ˜t+1 ∂L ˜t+1 ) + (1 − δ − n − g ) − λt = λt+1 f ′ (k ˜ ∂ kt+1 100 = = = 0 0 0.117) (5.5. S AVINGS AND THE S UPPLY OF C APITAL Technological Change Here we can again adapt the model as we did with the Solow version. How does this alter the problem? Given the dynamic budget constraint.122) (5. but note that people only care about consumption per person. ∞ L= t=0 ˜t + f (k ˜t ) − c ˜t+1 ˜t − k β t U (˜ ct Et ) + λt (1 − δ − n − g )k (5. Therefore. as well as noting that w ˜t = f (k ˜t = 0.120) which is identical to the accumulation equation for the Solow model with technological change. knowing that Et+1 = (1 + g )Et and Nt+1 = (1 + n)Nt . Yt = F (Kt . allowing the production function that incorporates technological change. grows exogenously at the rate g . you’ll see in the following Lagrangian that we have modified the term inside the felicity function to be c ˜t Et . x ˜t ) − δ . Et Nt ) where the efficiency term. it must be that At+1 = (1 + rt )At + w ˜t Et Nt + x ˜t Et Nt − c ˜t Et Nt (5.124) . and rt = f ′ (k have that ˜t+1 = (1 − δ − n − g )k ˜t + f (k ˜t ) − c k ˜t (5.118) The financial sector again ensures that capital and assets are equal to each other. ∆˜ at+1 w ˜t x ˜t c ˜t = rt + + + −n−g a ˜t a ˜t a ˜t a ˜t and this gives us a dynamic constraint for assets per efficiency unit of a ˜t+1 = (1 + rt − n − g )˜ at + w ˜t + x ˜t − c ˜t . not consumption per efficiency unit.

We do know that consumption is growing. just as in the Solow model.128) which is complex. Without functional forms for utility and production. The Euler equation can be written as ct+1 ct σ ˜α−1 + (1 − δ − n − g ) = β (1 + g ) αk t+1 (5. but also on the growth rate of technology rate. So it is not the case that U ′ (ct ) = U ′ (ct+1 ). The Ramsey Model Solving these all together to eliminate the multipliers. so this can be solved for k = ˜∗ α (1+g)σ−1 β 1/(1−α) −1+δ+n+g (5. U (ct ) = ct /(1 − σ ). An Explicit Solution with Technological Change ˜α . This implies that in steady state it must be the case that ˜∗ ) > f ′ (k 1 −1+δ+n+g β (1 + g ) (5. This tells us that consumption and capital per person must grow at the rate g in steady state. Technological growth gives the economy a built-in increase in consumption over time. The difference is that the future is discounted now by not only β . and noting that Et+1 = (1 + g )Et we have an Euler equation of U ′ (ct ) ˜t+1 ) + (1 − δ − n − g ) . So with technological change.125) This tells us that the optimal growth rate of consumption now depends not only on the discount ˜t+1 ) + (1 − δ − n − g ). and the interest rate f ′ (k itself. and a CRRA utility Using a standard Cobb-Douglas production function. and this lowers the marginal utility of consumption in the future. it will have to be k ˜ that are constant. so that y ˜t = k t 1−σ function. as this involves the ratio of marginal utilities of consumption per person.5. β . not per efficiency unit. but also by the growth rate of 101 .4. This means that k product of capital will be relatively high.126) or the marginal product of capital has to be larger than the value that would make consumption ˜∗ has to be lower. = β (1 + g ) f ′ (k U ′ (ct+1 ) (5. though.127) and in steady state we know that ct+1 = (1 + g )ct . ˜ and c In the steady state. so that U ′ (ct ) > U ′ (ct+1 ). we can provide more details on the solution to the Ramsey model. the marginal constant over time. but similar in form to the original solution without technological progress. We cannot even draw much information from the Euler equation. it isn’t possible to be more explicit about the solution.

Yes. Why? Because the growth of technology allows the economy to enjoy large consumption in the future without having to save for it. the response of the economy is more nuanced in the sense that savings will respond endogenously. 5. but not real things like output or employment.5 Fluctuations and Savings Having developed a whole structure of the Ramsey model to endogenize savings. the more the economy ignores the future and consumes today. Exactly how fast differs by model. however. there is still perfect competition and perfect price flexibility. and what that means for how savings rates respond to economic growth. In terms of fluctuations. there is nothing about having endogenous savings that changes the conclusions we made before regarding the role of nominal demand in fluctuations. such as prices. There are some homework problems asking you to think about how the savings rate responds to 1 + r. S AVINGS AND THE S UPPLY OF C APITAL technology. Even though the supply of capital is endogenous. No. does this model suggest anything different than what we’ve already seen in the Solow model? Yes and no. We will still need to introduce some kind of market imperfection or nominal rigidity to have any impact of nominal demand on output. following a positive productivity shock both models predict that economic growth will increase temporarily until the economy returns to steady state.5. This means that savings rates will change dynamically as the economy accumulates capital. 102 . as opposed to s. that fundamentally savings are still exogenous . what do we find? First. However. so the economy will always be operating at full capacity. in response to a productivity shock. So a change in nominal demand will change nominal quantities. in explaining that savings rates (specifically st /yt ) do respond to the rate of return on savings. in the sense that the general pattern of response to a productivity shock will look awfully similar in the two types of they depend on β . That is. as in the Solow model. The Ramsey model is more nuanced. The larger this growth rate.