The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
T o b in 's q a n d A s s e t R e tu r n s : I m p lic a tio n s f o r B u s i n e s s C y c le A n a l y s i s Lawrence J. Christiano and Jonas D.M. Fisher Working Papers Series Macroeconomic Issues Research Department Federal Reserve Bank of Chicago October 1995 (W P -95-14) FEDERAL RESERVE BANK OF CHICAGO October 1995 T o b i n ’s q a n d A s s e t R e t u r n s : I m p l i c a t i o n s for B u s i n e s s C y c l e A n a l y s i s Law rence J. Christiano* Federal Reserve Bank of Minneapolis, Northwestern University, National Bureau of Economic Research, and Federal Reserve Bank of Chicago Jonas D . M. Fisher* University of Western Ontario ABSTRACT The marginal cost of plant capacity, measur^1 1 the price of equity, is significantly procyclical. Yet, the price of a major intermediate input into expanding plant capacity, investment goods, is countercyclical. The ratio of these prices is Tobin’s q. Following convention, we interpret the fact that Tobin’s q differs from unity at all, as reflecting that there are diminishing returns to expanding plant capacity by installing investment goods ( “adjustment costs”). However, the phenomenon that interests us is not just that Tobin’s q differs from unity, but also that its numerator and denominator have such different cyclical properties. We interpret the sign switch in their covariation with output as reflecting the interaction of our adjustment cost specification with the operation of two shocks: one which affects the demand for equity and another which shifts the technology for producing investment goods. The adjustment costs cause the two prices to respond differently to these two shocks, and this is why it is possible to choose the shock variances to reproduce the sign switch. These model features are incorporated into a modified version of a model analyzed in Boldrin, Christiano and Fisher (1995). That model incorporates assumptions designed to help account for the observed mean return on risk free and risky assets. We find that the various modifications not only account for the sign switch, but they also continue to account for the salient features of mean asset returns. We turn to the business cycle implications of our model. The model does as well as standard models with respect to conventional business cycle measures of volatility and comovement with output, and on one dimension the model significantly dominates standard models. The factors that help it account for prices and rates of return on assets also help it account for the fact that employment across a broad range of sectors moves together over the cycle. *Prepared for the Economics and Finance Conference, held at Washington University in St. Louis on Septem ber 15-16, 1995. We are grateful to Michele Boldrin, Martin Eichenbaum, and Narayana Kocherlakota for their advice. Christiano thanks the National Science Foundation, and Fisher thanks the SSHRC for financial support. The views expressed herein are those of the authors and not necessarily those of the Federal Reserve Bank of Minneapolis or the Federal Reserve System. 1 In tro d u c tio n The price of a marginal unit of plant capacity, as measured by the price of equity, is sig nificantly procyclical. Yet, the price of a major input into expanding plant capacity, new investment goods, is countercyclical. In this paper we provide a quantitative account for this s ig n s w i t c h p h e n o m e n o n . We do so in a model that does at least as well as previous models in accounting for key features of (a) asset returns and (b) the business cycle. The features of asset returns that we have in mind include the observed low average return on risk free assets and the high average return on equity. The features of business cycles we have in mind include standard measures of comovement and volatility as well as measures of persistence. To capture (a) and (b), we build on the recent model by Boldrin, Christiano and Fisher (1995) (BCF). In particular, we assume consumption and investment goods are produced by distinct production technologies and that there are limitations on the mobility of factors of production across sectors. In addition, we adopt the habit persistence specification of preferences proposed in Constantinides (1990) and Sundaresan (1989). We show that the sign switch phenomenon can be accounted for by a combination of separate shocks to the two production technologies and the “adjustment cost” model of investment analyzed in Lucas and Prescott (1971) and the references they cite. According to this model, the more quickly a firm attempts to incorporate new investment goods into an existing plant, the less effective those goods are on the margin at expanding plant capacity. After verifying that our model can account for the sign switch phenomenon and that it continues to accoimt for mean asset returns as in BCF, we examine its business cycle characteristics. We find that, not only does the model do about as well as standard models on the business cycle statistics usually emphasized, but the model actually makes a step forward on one particularly important business cycle fact. Perhaps the d efi ni ng characteristic of business cycles is comovement: activity across a broad range of sectors moves up and down together over the business cycle (see Lucas (1981, p.217) and Sargent (1979, p.215).) Standard real business cycle models are consistent with this fact in that they imply that the outputs of the consumption sector and the investment goods sector are both procyclical. 1 However, we report evidence suggesting that employment across these sectors is procyclical as well.1 Standard real business cycle models are inconsistent with this evidence. They have the property that consumption is smoothed over the cycle: in a boom, consumption rises relatively little, as the improvement in technology is partially offset by a reallocation of factors of production out of consumption and into investment goods sectors. The opposite occurs in a recession. This is why standard models have the dubious implication that hours worked in the production of consumption goods is countercyclical. Although this is a feature of most real business cycle models, it is not a feature of ours. The following section provides a brief, nontechnical overview of the analysis. After that, we document the empirical properties of equity prices and prices and quantities of investment goods. Then, we formally describe our model and present the quantitative analysis. Finally, we present concluding remarks. 2 O v e rv ie w o f th e A n a ly s is In what follows we first discuss the cyclical properties of investment prices, and then we go on to explain how our model accounts for these properties. We then discuss the sign switch phenomenon. Finally, we discuss the business cycle implications of the model. I n v e s t m e n t G o o ds P r i c e s The time series behavior of the price of an important component of investment goods, producers’ durable equipment, has recently been documented and analyzed by Greenwood, Hercowitz and Krusell (1992). They show that the price deflator of this good, as measured by Gordon (1990), is counter cyclical. In addition, they document that, starting in particular in the 1980s, this price index exhibits a downward trend. These trend and cyclical character istics are a feature of household durable goods too. Together, these two components account for about 65 percent of the value of overall investment activity. The remaining components of investment—investment in structures and residential investment—also exhibit a down ward trend in their price starting in the 1980s, but they do not display the same significant 1See also Benhabib, Rogerson and Wright (1991, ftn.14), Murphy, Schleifer and Vishny (1989). 2 countercyclicality. As a result, the price index of overall investment activity is only slightly countercyclical. We follow Greenwood, Hercowitz and Krusell (1992) in interpreting these features of the price data as reflecting that investment dynamics are driven by both demand and supply shocks. Demand shocks are modeled as arising from a technology shock that is common across investment and consumption goods sectors. A boom triggered by this kind of shock generates a relative shift in demand towards investment goods for consumption-smoothing reasons and so produces a rise in their price. Supply shocks are modeled as arising from a disturbance that is specific to the technology for producing investment goods. Innovations in this shock generate a negative covariance between the price of investment goods and output. We parameterize the variances of our two shocks so that the model reproduces the observed weak countercyclicality of investment goods prices and also reproduces the estimated variance of the aggregate Solow residual reported in Christiano (1988). We accommodate the trend in the relative price of investment goods by the assumption that the investment-specific technology shock is a random walk with a positive drift. The implication that disturbances from this source are permanent is consistent with the notion that they represent shocks in the rate of arrival of innovations. At the same time, we posit that the economy-wide shock is transitory and has no trend. This is the only shock affecting the sector producing consumption goods, and its transitory nature captures the notion that the only disturbances to that technology are shocks to the weather, or natural disasters, or perhaps even labor disputes .2 The drift in the investment-specific technology shock is set to reproduce the observed rate of growth in consumption. The single autoregressive parameter in the stationary economy-wide shock is selected to reproduce the slight negative autocorrelation in the growth rate of the measured Solow residual reported in Christiano (1988). 2In effect, th e m odel tak es th e positio n th a t we know how to bake b read or serve a h e a rty m eal a b o u t as well as we did cen tu ries ago. P erm a n en t shifts in th e technology for producing con su m p tio n goods are viewed as b ein g em bodied in cap ital. 3 The Sign Switch To see how adjustment costs help account for the sign switch, it is useful to first consider the benchmark case in which there are no adjustment costs. Then, the price of equity—we identify this with the marginal cost of new plant capacity—and the marginal cost of new investment goods are identical, i.e., Tobin’s q is identically equal to unity. Thus, in the absence of adjustment costs, the weak countercyclicality of investment goods prices would be shared by equity prices and there would be no sign switch. Under adjustment costs, the price of equity becomes procyclical. This is because adjust ment costs have the effect of (i) reducing the response of equity prices to investment-specific technology shocks (these shocks make equity prices countercyclical), and (ii) enhancing their response to aggregate shocks (a force for procyclicality). The reason for (i) is that with adjustment costs, an investment-specific technology shock triggers two opposing effects on equity prices. On the one hand, the fall in the price of investment goods exerts a down ward pressure on the price of equity. On the other hand, the higher level of investment is associated with a drop in the margin *1 *ffoctiveness with which investment goods enhance new plant capacity, and this exerts upward pressure on the price of equity. The reason for (ii) is that with adjustment costs, an aggregate technology shock triggers two reinforcing effects on equity prices. This type of shock generates rises in both the price and quantity of investment goods. The rise in the price generates upward pressure on equity prices, and under adjustment costs the rise in investment does too. Our adjustment cost formulation is controlled by a single curvature parameter. We set this parameter so that the model reproduces the observed positive correlation between equity prices and output. O th er M odel Im plications We have enough free parameters so that our model can exactly capture the sign switch phenomenon. To test the model, we examine other implications. First, our adjustment cost formulation generates an elasticity of investment to Tobin’s q , an object for which there exist empirical estimates. We compare our model’s implications with these estimates. 4 Second, the progress that our model makes on the comovement puzzle - the fact that sectoral employment moves together over the business cycle - reflects two assumptions, (i) Following BCF, we assume factors of production must be allocated prior to the realization of the current period shocks. This assumption is intended to capture the various real-world frictions that make it difficult for workers to quickly leave a job in one sector and start work in another. In our model, it is simply not possible to immediately shift factors of production across sectors in the period of a shock, (ii) There is also little incentive to shift labor resources out of the consumption goods sector in the periods after a positive technology shock. Employment in the production of consumption goods rises in the periods after a favorable shock in the investment goods sector because the associated wealth effect makes consumption goods more valuable. Employment also rises in the periods after an aggregate shock. The expansion in the supply of consumption goods in the period of the shock has the effect of raising the value of consumption goods in subsequent periods because of the effects of habit persistence in the utility function. The transient nature of the aggregate shock then ensures that employment must be high to satisfy that desire. Employment in the production of consumption goods is procyclical because it is procyclical relative to each of the shocks in the model.3 Third, we document that our model implies low risk aversion on the part of households. As in BCF, steady state relative risk aversion with respect to a bet on wealth is restricted to unity. In addition, our model can account for the observed equity premium by assuming a very small degree of relative risk aversion with respect to bets on consumption. However, the ability of the model to account for the equity premium with low relative risk aversion with respect to consumption reflects some seemingly counterfactual implications for equilibrium consumption growth. It implies consumption growth is negatively autocorrelated—in the data it is positively autocorrelated—and it overstates the innovation in consumption. To understand how these implications of the model help account for its success in ex plaining the mean equity premium, it is useful to repeat an observation in BCF. They argue 3A n o th e r fe a tu re of o u r m odel environm ent helps account for (ii); nam ely, our assu m p tio n th a t u tility is lin ear in leisure. W ith in a ce rta in class of utility functions, th is assu m p tio n m axim izes th e likelihood of positive com ovem ent of secto ral em ploym ent. We discuss th is issue in d etail below. 5 that the key to getting an equity premium in a model like ours lies in generating an equilib rium process for the capital gains component of the return on equity which has the “right” pattern. In practice this translates into the requirement that (i) households have a strong desire to buy assets when consumption is high and to sell assets when consumption is low; and (ii) the nature of the technology has the effect of frustrating these desires. Habit per sistence in preferences delivers (i) and the limitations on the mobility of resources deliver (ii). Another way to enhance (i) is to construct a model environment in which equilibrium consumption growth is negatively autocorrelated. This feature particularly enhances the motive to smooth consumption and, thus, to buy assets when there is a positive innovation in consumption. 3 P ric e D a ta In this section, we present our analysis of the dynamic properties of postwar U.S. data on share prices and the price of new investment goods. Our results are that the price of aggregate investment goods is slightly countercyclical and displays a downward trend, particularly beginning in the 1980s. The cyclical behavior of the price of equity differs sharply from that of investment. Equity prices are significantly procyclical. 3.1 N e w I n v e s t m e n t G o o d s Prices We study the components of U.S. investment reported in the National Income and Product Accounts (NIPA). We also consider the annual price indexes for consumer durables and for business equipment reported in Gordon (1990) for the period 1947-1983. Investment price indices were divided by the implicit price deflator for household consumption of nondurables and services and then logged prior to analysis. We now consider the trend and business cycle characteristics of these data .4 Trend 4For a re la te d discussion, see F isher (1994a). 6 Figure 1 graphs the price data, together with their Hodrick-Prescott (HP) trend, for the period 1947QI-1995QI. Our broadest measure of investment is the NIPA measure of business fixed investment plus consumer durables. Note from Figure la that the associated implicit price deflator displays roughly no trend until the 1980s, whereupon it takes a sharp turn down. In interpreting this, note the difference between the Gordon price series for household durables and business equipment and the associated implicit price deflators from the NIPA. Gordon’s are the series that fall sharply throughout the sample in Figures lb and lg. He argues that the difference reflects the failure of the NIPA data to properly take into account quality improvement in consumer and producer durables. This suggests that, despite the apparent lack of trend prior to the 1980s in Figiue la, investment good prices probably were falling then. The behavior of the price of residential investment (Figure le) and of structures investment (Figiue If) suggests, though, that the fall in aggregate investment prices probably was slower before the 1980s than after. Figure 2 displays the ratio of the various categories of investment to Gross Domestic Product. In each case, the solid line depicts expenditure shares, that is, the numbers are formed as a ratio of nominal investment to nominal GDP. The dashed line depicts the ratio in real terms. Note in Figure 2 a that the ratio in value terms of our broad measure of investment is roughly stationary, while the ratio in real terms trends up from about 21 percent of GDP in the early 1950s to about 27 percent of GDP now. Thus, the fall in the price of investment goods in the 1980s has been offset by a simultaneous increase in real output. This is also a feature of components of investment, for example consumer durables (Figure 2b) and business equipment in the 1980s (Figure 2g). Investment in structures appears to be an exception, with quantity not rising by enough to offset the reduced price in the 1980s (Figure 2f). Figures 2h and 2i indicate that the sum of private consumption of nondurables and services, and government purchases, expressed as a ratio to total output, is roughly stationary. However, the share of the components does not appear stationary. We infer from Figures 1 and 2 that, to a first approximation, the aggregate data display balanced growth in expenditure share terms, but that the quantity of investment goods grows 7 more rapidly than the quantity of consumption goods.5 This abstracts from other important features of the data, including the significant upward trend in the price index of important components of investment prior to the 1980s. B usiness Cycles Now consider the business cycle properties of the price data. Figure 3 displays the deviations of the (logged) prices from their HP trend (solid line) together with the associated deviations for log GDP (dashes). Casual inspection confirms the Greenwood, Hercowitz and Krusell (1992) finding that the price of equipment is strongly countercyclical. Figure 4 displays the associated cross correlation functions and associated plus and minus two standard deviation error bands. First, note that the contemporaneous correlation is negative, though not significantly so, for our broad measure of investment (see Figure 4a). Durables are countercyclical—though significantly so only when correlated with output one quarter in the past—but the correlation between fixed investment prices and output is not significant. This reflects the very different cyclical behavior of equipment verstis structures and residential investment. Equipment is significantly countercyclical, whereas residential investment is strongly procyclical and structures are acyclical. The data suggest that there are interesting differences in the business cycle properties of the components of investment. Further analysis of these differences is beyond the scope of this paper .6 Our model recognizes only one form of investment, and we calibrate it based on our point estimate for the correlation between the price of aggregate investment and output, which is —0.15. 3.2 S t o c k Prices We consider the cyclical behavior of the S & P 500, Dow Jones and New York Stock Exchange stock price indexes for various industries, as supplied in Citibase .7 In each case, the price 5T h is is co n sisten t w ith th e resu lts of th e form al sta tistic a l analysis presented in E ichenbaum a n d H ansen (1990) an d F ish er (1994b). im p lic a tio n s of th e p rocyclicality in residential investm ent price deflator are explored in F ish er (1995). 7T h ese indexes are b e st th o u g h t of as th e p ro d u ct of price and quantity. W e assum e th a t m ost of th eir business cycle v a ria tio n reflects v ariatio n s in price. 8 index was deflated by the same implicit price deflator for consumption of nondurables and services used to deflate the price indexes of new investment. All data were logged and HP filtered prior to analysis. Table 1 reports the volatilities of the price data, divided by the volatility of output, which is roughly 1.8 percent. In addition, the correlations between the price indexes and the cyclical part of U.S. GDP are also reported. Note that, with two exceptions, these correlations are significantly positive. The two exceptions are the S&P 500 data for the transportation and utilities industries. The dynamic correlations with output are presented in Figure 5. Note that the largest correlations are between the stock price and next quarter’s GDP. These correlations are almost all near 0.5. 4 M odel E conom y This section presents our model economy. There is a single representative household and two production sectors. One produces the consumption good, and the other the investment good. There are two technology shocks: a logarithmic random walk shifts the production function for investment goods, and a stationary first order autoregressive shock shifts both production functions. Households and firms are competitive. In what follows we first present the household problem and a discussion of risk aversion. Then we consider the problem of the firm and equilibrium. We also discuss various features of the equilibrium of the model. 4.1 Household Problem Prior to the realization of the date t random variables, the household evaluates consumption and leisure henceforth according to OO (1 ) [log( C t+j - X t + j ) + 77(1 - H c,t+ j - H ht+j) } , 3=0 where 77 is a positive scalar, H Cit, H l t goods-producing industries, and Xt denote employment in the consumption and investment denotes the habit stock, which is assumed to evolve according to (2 ) X t+1 = h X t + b C t . 9 Here, Ct Et is the expectation operator conditioned on all variables dated and earlier. Also, t denotes consumption, and we specify that utility is linear in leisure following Rogerson (1988) and Hansen (1985). The household budget constraint is (3) c t + s; + sl + Bt < (1 + for x = t = c,i, 0,1,.... Here, and in period t, Bt + (1 + rlt)S U Sf + (1 + denotes date r {_ x) B t- i w xH c<t + w lt Hi't , purchases of shares of equity in industry t x, for denotes purchases of risk-free debt. The rate of return on equity purchased (1 + r®t+1), depends upon the period return on debt, (1 -)- r { ) , depends on the date rate in industry + x t state variables are + 1 state of nature, while the rate of state of nature. Also, t , which is a function of the date The household’s date t t denotes the wage w* state of nature. S 1_x, S \ _ v B t~ \ , X t - In addition, the house hold knows the values of all prices and rates of retiun for each date and state of nature. The household’s problem at time H c, t , H i , t , C t , S f , S l , B t . t is to select values for its time t choice variables, We capture the notion that there is a degree of precommitment in the labor supply decision by imposing a particular information constraint on the variables. In particular, we require that households choose t H c<t, H l t prior to the realization of the date state of nature, while the remaining choice variables are selected afterward. We refer to this restriction on the allocation of work effort as the l i m i t e d l abor m o b i l i t y assumption. The household’s objective is to maximize (1) subject to (2)—(3) and the condition that the future choice variables satisfy the same information constraints. The household’s intratemporal first order necessary conditions for labor are (4) E t - \ W XACtt = rj, x = c , i . Its intertemporal first order conditions are (5) for (6) EtPc,t+i ( 1 + r ex t + 1 ) x = c,i, EtPc,t+i + r{), where Pct+i = = 1= 0A-C,t+1 Act ’ 10 and Ac>t is the derivative of (1) with respect to p c,t+i is the value, in date t Ct, when E t- \ is replaced by E t. The variable consumption units, of a unit of date t + 1 consumption indexed by state of nature and scaled by the conditional probability of that state of nature.8 4.2 Risk Aversion Evaluations of models of asset prices often focus on the implications for risk aversion. One measure of risk aversion is the amount a person is willing to pay to avoid an unanticipated gamble. Two types of gamble are of interest: “gambles on wealth” and “gambles on con sumption.” These are differentiated according to whether agents can use credit markets to mitigate the effects of the outcome of the gamble. With a gamble on wealth, agents have full access to credit markets in the period of the gamble. Constantinides (1990) argues that habit persistence agents have little aversion to gambles like this because they have a relatively painless way of dealing with the state of the world in which they lose. The fall in the present value of consumption that must occur with the loss of a bet on wealth can be accommodated by reducing consumption slowly so that the habit stock has a chance to fall. By specifying (3 to be close to unity and formulating habit persistence in terms of the logarithm, the steady state level of relative risk aversion in wealth is unity in our model (for further discussion, see BCF.) A gamble on consumption has the property that agents have no access to credit markets in the period of the gamble. As a result, the full amount of a loss or gain must be absorbed by current consumption. Agents then have full access to credit markets in the periods after the gamble. We suspect that risk aversion over consumption gambles is harder to measure (or intro spect on) than risk aversion over wealth gambles. Still, it is useful to define a measure of risk aversion over consumption gambles precisely so that we can report on this aspect of the model in the results section. For tractability, we define this concept of risk aversion relative to a slightly simpler environment, in which hours worked is fixed, the rate of return on sav 8In p a rtic u la r, let gt{s') d en o te th e d a te t conditional probability th a t sta te of n a tu re s' will be realized in p erio d t + 1. T h e n gt(s')pc,t+i(V) is th e value— d en o m inated in d a te t consum ption u n its— of a u n it of c o n su m p tio n in d a te t + 1, sa te of n a tu re s'. Here, pCtt+\(s') is th e value of p Cit+ iin s ta te of n a tu re s'. 11 ing is constant, and there is no uncertainty. Thus, suppose a household has the following preferences: OO (7 ) x ^ i o g f a - x , ) , t=o where the habit stock evolves as before. At date 0, the household has a given stock of wealth, Wo (= ^ - l + B_i), and habit, X q, and seeks to optimize (7) subject to the following intertemporal budget constraint: The solution to this problem is characterized by ( C , - X , ) = Q 1 ‘, 7 = /3 ( l+ r ) , where (8) A'o) (? -* ) Therefore, the value function for this problem is, apart from an additive constant, (9) = For further details, see BCF. Let Co, C i , ..., be the solution to this problem. Now, suppose the household is confronted with the following gamble: it is given and must give up fiCo [i C q units of consumption goods with probability 1/2 consumption goods also with probability 1/2. We measure relative risk aversion in consumption, RRAC, by the fraction, u, of C q the sacrifice with probability one in order to avoid this gamble. That is, log (C0(l - v ) - X 0) (10) = i {log (C0(l + log (C0(l Here, W\ + u solves 0 v ( W u h X o + b C 0( l - v ) ) - f i ) - X 0) + n ) - X 0) household is willing to + + Pv pv (W , h X o + b C 0( l - m)) (Wl5 h X o + bC0{\ + //))} . = (1 + r) Wo — Co and is unaffected by the outcome of the gamble. We solve this problem on a steady state growth path by factoring Co from (10) and setting and W q/ C q function of X q/ C q to their steady state values. Evidently, the measure of risk aversion we use is a fj,. 12 4.3 F i r m s There are consumption goods producing and investment goods producing firms. Each has a one period planning horizon. Whatever physical capital the firm uses in production in period t + 1 must be put into place by the end of period t. This capital is produced by combining previously installed capital with new investment goods. To finance the purchase of these inputs, the firm issues debt and equity in period t. There are separate equity markets for the two types of firms, and the two types of equity command different, competitively determined, state-contingent rates of return. Since the competitive rate of return on debt is known at the time it is issued, in equilibrium there can be only one rate of return for that financial instrument. When period t + 1 occurs, the firm observes the state of nature and, hence, the prevailing wage rate. It then enters competitive labor markets to hire that amount of labor which maximizes cash flow in that state of nature. The firm’s cash flow is the value of its production, plus its undepreciated stock of capital, net of expenses. The firm’s objective at date is to maximize the date t value of cash flow at t + 1, summed across all possible states t of nature. There are several prices relevant to the firm’s capital decisions. There are the prices of the raw materials used in period t to produce end of period t capital—i.e., the price of new investment goods and of previously installed capital. Also, there is the price (actually, marginal cost) of end of period t capital. This is, in general, different from the date t + 1 price of previously installed capital because the latter reflects the realized state of nature in period t + 1. In sum, these prices are • PfeX)t ~ price of previously installed capital in sector • Pi,t ~ price of new investment goods. • Pk'x,t ~ price of newly produced capital in sector t x = i,c. x = i, c, available for production in + 1. Each of these prices is taken as parametric by the firm. statement of the firm problem in each sector. 13 We now provide a formal The technology for producing consumption goods in t + 1 is (11) CM < where 6 t+ 1 is a covariance stationary shock to technology: (12) 0t +1 = pOt + £ t + 1, 0 < p < 1. The technology for producing new investment goods in t + 1 is (13) I c,t+ i + Ii,t+ i < V t+ iK ° t 4-i(exp(0t+i)-ffi,t+i)1_a where (14) Vt + i = e x p ( p Here, st and pt + p t+ i ) V t . are zero-mean, random variables which are independent of each other and over time and which have standard deviations transformation between I c,t+i and I ltt+i cre and crM, respectively. The linear rate of implicit in (13) guarantees that, in equilibrium, the prices of new investment goods for the consumption and investment goods sectors are equalized. The technology for producing end of period (15) K x<t+ i < Q x ( y, z ) t capital, K x t+ u for industry x is , where (16) for Q x(y,z) x = c,i and ai ip < + i/ip a 2z'1’ 1. In (15)—(16), y denotes previously installed capital and z denotes new investment goods. When ip = 1, (15) corresponds to the conventional linear capital accumulation equation. When ip < 1, then the marginal product of new investment goods is decreasing in the flow of investment. The technology described in (15)-(16) is a special case of the adjustment cost formulation posited in Lucas and Prescott (1971) and in the references they cite. We choose the constants, 0^ > 0 and a2 > 0, to guarantee Qf = Q2 — 1 in nonstochastic steady state. This has the effect of making the nonstochastic steady state properties identical to what they are when ip = 14 1, regardless of the actual value of ip. Also, it has the effect of forcing Tobin’s q to be unity on a steady state growth path. Finally, as discussed further below, it has the effect of normalizing Tobin’s The financing constraint faced by a firm in industry (17) Pkx,ty + and its period (18) 7r?+1 = < Pi,tZ t x q at unity when ip = l.9 is S x>t + B X't , + 1 cash flow constraint is Y x,t+1 + (1 - 6 ) Q X (y , z ) P kx,t+1 - w xt+ l H x,t+l - (1 + r eX it+l) S f - (1 + r{)B t > 0. Here, Y x^+1 is the firm’s gross output, given by (11) or (13), measiued in date t + 1 consump tion units.10 Also, we assume that if period t + Qx is the amount of capital used by the firm dining 1, then (1 —8 ) Q X remains at the end of the period, when it is made available for sale. The firm’s profit function is the value of 7r*+1 denominated in imits of the date consumption good, summed across all possible data variable, p c,t+ i , t + 1 states of nature: firm also makes use of P^t+\- The firm’s objective is to find ,t >-Dx,t EtPc,t + i S Xjt, B x>t, z , y , H x^+1 to solve max 7rf+1, PiX,t -f-1 subject to the relevant production technology and (17)—(18). 9T h e form ulas for a j an d a.2 are ai = [(1 - 5 ) e x p ( - / 2 / ( l - a ) ) ] 1 - ^ , a-2 = [ l - ((1 -6)exp(-fl/(l - a ) ) ) 1^] 10T h u s, for th e co n su m p tio n industry, Yc,t+i = Q° (V,z)a (exT>(9t+ i) H c,t+ i y - a , a n d for th e in v estm en t goods indu stry , Yi,t+i = Piit+\Vt+ i Q c (y, z ) a (exp(9t+i )H itt+ i ) l ~a ■ The is given by (6) in equilibrium and is viewed as parametric by the firm. Other variables viewed as exogenous by the firm are Pfca. t+1,ty^+1,r® f+1, r / for (19) _ max E tp c,t+i^t+i- t 15 x = i,c. The x = i There are a variety of useful ways to write the efficiency conditions associated with this problem. The first order condition for hours worked is (20) = iu£, m p l Xtt+1 where m p l Xit+ 1 denotes the marginal product of labor, denoted in period t + 1 consumption units. Let the marginal value to the firm of an extra unit of K x>t+1 be denoted by (21) Vx,t ~ EtPc,t +1 \pl'Pkx,t+\ where m p k Xyt+ 1 d" (1 ^)-ffcI,t+l] ) denotes the marginal product of capital, denoted in period t + 1 consumption units. The first order conditions associated with (22) VxjQ^t = VXit, z , y , S Xtt, B x>t, P i , t \ V i , t Q i tt = Pkx,t^, E tp c,t+ i(l + r i,t+i) = where A > 0 is the multiplier on the constraint (17), and with respect to its i th argument i — Q*t are E tp c,t+ i(l + r {) = A, is the partial derivative of Qx 1,2. Let the marginal cost of producing K x t +\ by a firm in industry x be denoted by Pk'x,t- It is readily established that TD P P jrkXlt ' Household optimization ensures, via (5) and (22), that A = 1 in equilibrium. This, together k'x,t ~ Q h ~ Q it with (23) implies (24) Vx,t Pk'x,t > b c, i.e., the marginal value of end-of-period t capital is equated to its marginal cost. 4.4 E quilibrium We adopt the normalization that the number of firms of each type and the number of households is one, and we assume that all agents of each type behave identically. A sequenceof-markets equilibrium is then defined in the usual way. Market clearing implies that, in a symmetric equilibrium, the demand for previously installed industry denoted above by z, x capital in period £, equals the supply, (1 —8 ) K Xit■ Similarly, the demand for period t new investment goods by industry x, denoted by y, is 7I t. We proceed now to discuss various features of the equilibrium, including the sign switch, equity premium, Tobin’s q, and comovement of employment. 16 4.4.1 T h e Sign Sw itch We now endeavor to provide insight into how it is that our model can account for the sign switch observations: the fact that the price of equity—which we identify with cyclical, while the price of new investment goods, Pi t P itt, Pk'x,t—is pro is slightly countercyclical. Consider first. Investment productivity shocks alone create a negative covariance between P itt and output, and shocks to aggregate productivity by themselves create a positive covariance be tween P i it for and er^ so that the model generates a slightly countercyclical p f. ae and output. Thus, it should be no surprise that we can select relative magnitudes Now consider costs (i.e., ip Pk'x,t- If Qf.t = 1, as in the conventional formulation without adjustment = 1), then obviously switch. However, when ip < 1, Pk'x,t = Pi t , and there is no way to account for the sign then there is a wedge between these two prices. The wedge has the effect of reducing the impact on of investment-specific technology shocks and of Pk'x,t increasing the impact of aggregate technology shocks. Consider a positive investment shock first. Not surprisingly, in our computational experiments we find that this generates a fall in equilibrium P^t and an increase in two offsetting effects on in I x>t Pk'x,t Pk>x,t- I Xtt. The first relation in (23) indicates that this triggers The fall in P,)t has the effect of driving P ^ t down, but the rise has the opposite effect, by driving falls proportionally less than does Q x2 l Pl t down. In view of this, it is not surprising that after an investment technology shock. Consider now a positive shock to aggregate technology. This triggers an increased demand for capital for consumption-smoothing reasons. Not surprisingly, this results in a rise in rise in I x By reducing more than the rise in the rise in P i <t. Ix t has the effect of driving By reducing the impact on P k> xt costs in effect reduce the source of countercyclicality in Pk'x,t Pi t and also a up proportionally of investment shocks, adjustment Pk > x. This is why the model predicts that this variable is procyclical. In our quantitative analysis, we study an aggregate price index, which we obtain by combining our two equity prices as follows: 1 p _K c,t+ i p r i/c',£5 -nt'.i — ”is — -n - U "f” rs tvf+i J\ t+1 r-j where K t+ 1 = K c>t+1 + K iit+ 1- 17 4.4.2 T o b in ’s q Tobin’s q, the ratio of the marginal value to the firm of K x^+i divided by the marginal cost of a new investment good, is (25) = •* i,t = \ V2, t which is unity when (1 - S ) K al + a2 lx,t a2 if) = 1, * x,t since 0,2 = 1 in that case. The sign switch phenomenon can be stated in terms of the elements of Tobin’s q: the numerator is procyclical, while the denominator is countercyclical. 4.4.3 T h e E q u ity P re m iu m To discuss the equity premium, it is convenient to first obtain an expression for the rate of return on equity. Linear homogeneity guarantees that, in equilibrium, maximized profits (19) are zero. The cash flow constraint (18) then guarantees wf+ l = 0 in each date and state of nature. Using this and (20)-(24), one gets the following equilibrium condition after some algebra: (26) 1 + r | 1+1 = m P k .‘+i + ( 1 ) _ (1 + r ' b f . ,t Here, 7 f = B * /S f Euler equation, denotes the firm’s debt to equity ratio. The household’s intertemporal E tp c,t+ i(l + r (+1) = b implies E tp Ctt+iE t ( l + r^+ l ) = l - C o v t ( p c, t + i , l + r t + i ) or, using (5) and (26) , , E t ( l + r e t+1) (27) -----— —f --------1 = 1+ n ~Covt ( m p k Xtt+1 + ( l - 8 ) P kxtt+1\ p Cyt+1, ----------------- ---------------\ Ek>x,t ) (1 + 7 t ), where the object on the left of the equality is the date t premium on equity in industry is, approximately, E tr l x and <+1 —r { . BCF argue that a key channel by which a change in model specification impacts upon the equity premium operates via its impact on the equilibrium stochastic process for capital gains, P kx,t+i/Pk'x,t- via changes in the stochastic processes for p c,t+i The alternative channels, which operate and m p k x t + i / P k> x t, exert very little direct effect on the conditional covariance. BCF stress that the combination of habit persistence preferences and limited factor mobility are effective in producing the sort of stochastic process for P kxtt+ i / P k ' x,t that results in a sizeable equity premium. 18 As is well known, -yf is indeterminate in a model like ours. Equilibrium is consistent with any state-date contingent pattern for 7 f , although the equilibrium quantity allocations in the model are unique. To make the analysis interesting, we must therefore fix 7 * exogenously. We do so by setting 7 f = 7 t = 7 - A numerical value is assigned to 7 in the next section. We define the overall return on equity as r®+1: \^°) re — P ^ K c,t + i e r t+ 1 — r> Kt r c,t+ 1 + 1 PkljK ht+ i r> K ,t -\-1 e r i,t+ U where K t+ 1 = Pfc',tAc,t+ 1 + P k>.,tKi,t+1 -11 4 .4 .4 C om ovem ent To understand our model’s implications for comovement, it is useful to consider the bench mark case where b = h = 0 and the utility of leisure is a power function, separable from consumption. In this case, equilibrium in the labor market associated with the consumption good sector implies via (11), (20), and the appropriate analog of (4) (29) where Ot n .c,t v > 0 and £ > 0. The specification in ( 1 ) corresponds to £ = 0. Because the employment decision is made prior to the realization of the date (30) Hc,t = v{ 1- t shocks H c<t - H itty . It is easily verified that this equation must hold even when the limited labor mobility as sumption is dropped so that the date t labor decision is contingent upon the date t exogenous shocks. Thus, without habit persistence, getting comovement in labor is impossible, with or without limited labor mobility: if £ > 0, then Hl t and Hct must move in opposite directions. The case £ = 1 is also inconsistent with comovement, since employment in the consumption sector is predicted to be constant. Still, relative to preferences based on alternative values of £, £ = 1 appears to be the most favorable to comovement. In our quantitative results below, which are based on £ = 1 , we find that to get comove ment, habit persistence and limited labor mobility are both required. n T h is is ju s t [(1 + r® t+1) 5 c,t + (1 + rf)t+1)5 iit] / ( S c,t + Si,t), after m aking use of th e firm s’ first order co n d itio n s a n d our re stric tio n s on th e d e b t to equity ratio. 19 5 A s s ig n in g V a lu e s to th e P a r a m e te r s In this section, we explain how we assigned values to our model’s parameters. As a prelim inary evaluation of the model, we report on its trend properties and on its implications for Tobin’s q. There are 11 model parameters, three preference parameters, and four each of technology and the exogenous shocks: AM , (31) 7, Pi &ei Pi G\i- We find it convenient to consider 5 parameters, 0 , a , 8 , j , p, and the remaining 6 , h, b, p, cre, p , 07 , separately. Loosely, the first set controls the steady state properties of the model, while the second set controls the business cycle and asset pricing properties. 5.1 P a r a m e t e r s Controlling b t e a d y State We set (32) 0 = 0.99999, a = 0.36, 8 = 0.021,7 = 2/3, p a / { l The indicated value of 0 - a) = 0.004. was selected to maximize the model’s ability to account for the observed low risk free rate. The value of a was chosen so that the model’s implication for the share of GDP earned by capital coincides with an empirical estimate of that quantity based on data for the 1970s and 1980s taken from the NIPA, as reported in Christiano (1988, ftn.3). However, as emphasized there, this is the midpoint of a relatively large range of values for a, determined by the details of how one measures capital income in the NIPA. Also, Christiano and Eichenbaum (1992, p.441) report that the sample average of [1 —( K t+ 1 —I t ) / K t] is 0.021. By setting 8 = 0.021 in the model, this empirical sample average is reproduced along the model’s nonstochastic steady state growth path .12 The value of 7 was selected to match 12C h ristia n o an d E ichenbaum (1992) also re p o rt an e stim ate of a . However, th e ir e stim a te exploits pro p erties of th e s tru c tu re of th e ir m odel, w hich are not shared by our model. 20 the corresponding empirical estimate of the debt to equity ratio reported in Benninga and Protopapadakis (1990). The linear form of preferences for the representative agent was chosen to enhance the model’s implication for the volatility of labor. Finally is the steady state growth rate of consumption in the model, and ft Jia/{ 1 —a ) was selected so that it coincides with the corresponding sample average reported in Christiano and Eichenbaum (1992, p.441). The parameter r] just controls scale, and we set it to 1. The properties of the model along a steady state growth path are reported in Table 2 (see the column marked “Calibrated”). A corresponding set of estimates for the U.S. economy is reported for the entire postwar period and for the period starting the 1980s. We report results for two sample periods because of the evidence described earlier, which suggests the official estimates may underestimate the trend fall in the price of investment goods before the 1980s. Two empirical measures of consumption are reported, C and C, with the latter including government purchases. Given our level of abstraction, it probably makes sense to identify consumption in the model with the sum of household and government consumption.13 With this measure of consumption, the model evidently understates consumption’s expenditure share in output, and correspondingly overstates investment, by about 12 percent of output. To some extent, this mismatch between model and data reflects that our empirical measure of government consumption includes government investment. From the perspective of the model, it makes more sense to include this in our measure of investment. The evidence suggests that this consideration would not entirely close the gap between the model and data. Government gross fixed capital formation (including military) has taken a declining share of GDP. It peaked at about 7.5 percent of output at around the time of the Korean war and has been falling steadily since then. In the decade after 1975, the ratio stabilized at about 3.5 percent of output.14 So, at best, these considerations can account for only a part of the discrepancy between the empirical and model expenditure shares. Given the imprecision 13As is well know n, th is in te rp re ta tio n is form ally rationalized by th e assu m p tio n th a t p riv ate a n d public co n su m p tio n m e perfect su b stitu te s. U nder these circum stances, innovations in governm ent co n su m p tio n w ould be m irro red by eq u al red u ctio n s in p rivate consum ption. Interestingly, th e gross featu res of postw ar U.S. d a ta a p p e a r co n sisten t w ith th is view (com pare F igures 2h an d 2i.) 14T h is is based on an analysis of th e governm ent investm ent d a ta stu d ied in C h ristian o (1988). 21 in the estimated value of a, there is probably room for reducing it in order to improve the model’s implications for expenditure shares. This is consistent with the information in the column marked “a = 0.28” which reports results for a lower value of a that is within the range of estimates reported in Christiano (1988). In the analysis in the next section, we report results for this reduced value of a. However, we do not comment on them because they correspond closely to the results based on a = 0.36. Now consider the growth statistics in the Table 2. Consistent with the model, variables measured in consumption units grow less rapidly than does real investment. However, the difference in growth rates based on the entire postwar period is not as great as the model predicts. For the model to capture this, we would need to introduce growth into our aggregate technology shock too. Note how different, however, the period since the 1980s is. There is a sharp decline in the growth rate of the investment price deflator and a corresponding sharp rise in the growth of real investment (see Figures 1 and 2.) With such a relatively short period there is, of course, a danger of confounding trend and business cycle movements. Still, the reduction in the price trend spans two business cycles (see Figure 1). Our model’s assumption that all growth originates in the investment sector is not a bad approximation to the experience of the past decade. 5.2 P a r a m e t e r s Controlling Business Cycles a n d Asset Pricing The remaining parameters are h, b, xjj, p, er£, <rM. Values for these six parameters were set based on the following six moments of the data: (33) p(Y ,P i) = -0.15, p ( Y , P k0 = 0.30, p(A Solow ) = -0.10, <r(AS o l o w ) = 0.018, and (34) re - r f = Here, p (x , y) 6.63, rf = denotes the correlation between the logged, HP filtered variable ilarly filtered variable, y; variable, x; 1.19. and a(x) p(x) and the sim denotes the first order autocorrelation of the untransformed denotes the corresponding standard deviation. Also, gate GDP, measured in base year prices, and x A Solow 22 Y denotes aggre denotes the logarithmic first difference of the Solow residual, computed using a simple aggregate production function.15 The first two statistics in (33) characterize the sign switch. The two statistics in (34) are taken from Cecchetti, Lam, and Mark (1993) (CLM). Conditional on a set of feasible values for h, b, values for were selected so that ip, p, the model exactly reproduces the four statistics in (33). The model’s implication for these statistics was computed by Monte Carlo simulation. In simulated data sets, time series on the growth rate of the Solow residual, A S o l o w , were computed using the same algorithm used in the data. Thus, an aggregate production function was used for this calculation, even though there does not exist an aggregate production function relationship between aggregate inputs and aggregate outputs in our model. We can define a mapping from feasible h,b, first compute the four parameters, v = [(re —r^),r^]' as follows. For given ip, p , a £, a ^ , as described above. W ith the model h,b now fully parameterized, its implied value of to u was computed by Monte Carlo simulation. In particular, we simulated 500 artificial datasets, each of length 120 observations. In each data set we computed the sample average of the annualized risk free rate and the equity premium on annualized equity returns. The model’s implied value of by the mean of these 500 sample averages. Denote this mapping by the set of feasible h, b u was approximated v = f(h ,b). We define as the set of points in the unit box having the property that < Xt is as close as possible to vt, Ct and AC)t < 0 are never observed in the Monte Carlo simulations used to evaluate /. We chose feasible values of h, the sample estimates of (35) C (b, h) = \vT - Also, the 2 x 2 matrix so that the model’s implied v provided in CLM. Our distance metric is v f(b, b h)]' V Vt f l [ur - f(b, C (b , h ), where h )]. is the CLM estimate of the underlying sampling variance in Op. Let (36) J = C ( b T , h T ), 15 T h e ag g reg ate Solow resid u al is z t = ----- — , w here Y t = Ct + Ic,t + Ii,t, K t = K C)t + K i yt) H t = Hi,t 4- H c,t• In section 6.2 below, we explain our ratio n ale for in te rp re tin g th ese m easures of Yt and Kt as th e “base” year m easures of o u tp u t and capital. 23 where bT , h T values of minimizes £(6, h) over the feasible values of b, h. In practice, we could not find which set b, h J = 0. This procedure for determining the parameters in effect treats the statistics in (33) as though they were known with perfect certainty. Presumably, a procedure which took into account the sampling uncertainty in (33) would “sacrifice” a bit on hitting the elements in (33) that are estimated the least precisely, in exchange for doing better on (34). We have not explored such statistical estimation procedures. We obtained the following results: (37) b = 0.55, h = 0.0, J = 4.23. The corresponding estimates of (38) if) i p , p , a e ,cr^ = 0.40, p = 0.52, a e = 0.017, To see how / and £ vary with b and 0.028. = h, are16 consider Figure 6. It displays the empirical equity premium/risk free rate combination in (33) and 5 and 1 percent confidence intervals about this point based on the estimates of CLM. In addition, there are four lines with stars. Each line corresponds to a particular value of stars correspond to b = 0.4 to b h, as indicated. Starting from the lower left, the = 0.6, in increments of 0.025. For values of b up to 0.6 were feasible. The optimal point, the figure. As the figure makes clear, increases in that is the reason why h h b h = 0.2 and 0.3, not all = 0.55, h = 0.00, is also indicated in sharply increase the risk free rate, and = 0 at the optimum. It is interesting to compare these results to those in BCF. That paper uses the same estimation strategy, but comes up with different estimates for b and h: 0.35 and 0.40 respec tively. These differences reflect differences in the specification of the models: in BCH there is only an aggregate technology shock, which is specified to be a random walk with drift, and leisure enters log-linearly in utility, rather than linearly. In one respect, the two sets of estimates of b and h are similar. Ours imply a steady state ratio of the habit stock to con sumption equal to 0.55, whereas the BCF estimates imply a value of 0.58. Not surprisingly, we report below that the implications for steady state risk aversion in consumption are also quite similar. The implications for steady state risk aversion in wealth are identical. 16T h e im plied values of a\ a n d a 2 are 0.98 and 0.12, respectively. 24 Though both sets of estimates imply roughly the same magnitude for the habit stock, their difference lies in how sensitive the habit stock is to recent consumption. In BCF’s estimates the habit stock is relatively insensitive, whereas in our estimates the habit stock is very sensitive to recent consumption. We suspect that this is an important part of the explanation for the difference in estimation results. An important finding in BCF is that the magnitude of the equity premium is decreasing in the autocorrelation of consumption growth. The BCF estimation procedure appears to have exploited this fact by selecting a positive value of h in order to produce negative autocorrelation in equilibrium consumption growth. With a positive value of h, the surge in consumption in the period of a shock leaves the habit stock relatively unaffected in the subsequent period. As a result, the value of consumption in that period is not particularly high and so households cut back their consumption from the high level in the previous period. This reduction is what produces the negative autocorrelation in consumption growth in the BCF model. In our model, there are other sources of negative persistence in consumption growth, and so the estimation strategy has less need to manipulate b and h to accomplish this. Our aggregate technology shock is stationary, and the estimation strategy chooses a low value for its autocorrelation in order to reproduce the negative autocorrelation in the growth rate of the Solow residual. Negative autocorrelation in equilibrium consumption growth is a consequence of this. To help evaluate our parameter estimates, we computed the implied elasticity of invest ment with respect to Tobin’s q . 17 That quantity is 1.66 in our model. For comparison, Abel (1980) reports estimates of this quantity that range from 0.27 to 0.52. Relative to Abel’s estimates, we have understated the degree of adjustment costs (i.e., overstated 17T h e elasticity of in v estm en t in in d u stry x w ith respect to T obin’s q is, using (25) d lo g /x ,t d log qf 1 -i f) 25 ip). 6 Q u a n tita tiv e R e s u lts We quantify the mechanisms in our model that enable it to account for the salient features of asset prices and returns (Table 3). We then go on to examine our model’s implications for business fluctuations and for risk aversion (Tables 4-6). We compare the model’s business cycle implications with the corresponding empirical evidence. We show, for example, that employment across a wide variety of sectors is strongly procyclical. This similarity is partic ularly striking because the trends in these sectors are very different (Figures 7-8). Though the model does not replicate the diversity in trends, it does replicate the procyclicality of employment across sectors. 6.1 Financial M a r k e t s Table 3 presents various statistics which capture the implications of our model for financial variables. The column marked “calibrated” reports results for the model calibrated in the previous section. The columns to the right of that report results based on various perturba tions of the calibrated model, obtained by Monte Carlo simulation. The first column presents the corresponding sample estimates. The Sign Switch Consider the phenomenon that the model was specifically designed to address, the sign switch. That it exactly reproduces the statistics we use to characterize that phenomenon is not surprising—the parameter values were picked in part to accomplish just that. Table 3 is constructed to help assess the role played in accounting for the sign switch by two model features: the assumption of adjustment costs in the installation of investment goods, and the multiple shock assumption. (The intuition about how these factors are supposed to work is reviewed in the overview section above.) The column marked “ip = 0.9” is suggestive of what happens when adjustment costs in the investment function are shut down. In this case, the wedge between the price of equity and the price of investment goods is essentially eliminated. As a result, both have roughly the same correlation with output. That correlation turns out to be nearly zero, because 26 the two shocks in the model have roughly offsetting effects, in terms of their impact on the cyclicality of these prices. To gauge the role played by the multiple shock assumption, consider the column marked “erM = 0” so that the investment-specific technology shock is set to zero. In this case there are only sources of procyclicality in the two prices, and so it is not surprising that there is a strongly positive correlation between equity prices and output, and between the price of investment goods and output. Further insight into our model’s account for the sign switch may be obtained from Figure 9. This figure displays the response of the model variables to a one-standard deviation in novation in the aggregate technology shock (solid line) and in the investment-specific shock (dashed line). Consider the response to the aggregate shock. This response produces a sharp rise in both the investment goods price, P,, and the price of equity, P k'. Consistent with the intuition in the overview and the discussion in the model section, the jump in the equity price exceeds that in the price of investment goods. Now consider the response to an investment-specific technology shock. As anticipated by our earlier discussion, Figure 9h shows that P k> falls relatively little, by comparison with P,. Investment adjustment costs have the effect of muting the response of the price of equity to investment shocks and amplifying their response to aggregate shocks. This is why the price of equity is more procyclical than the price of investment goods, and is at the heart of our model’s account of the sign switch phenomenon. Although the model accounts well for the cyclical comovement with output of investment and equity prices, it does not account well for the magnitude of their cyclical volatility. In the data, the standard deviation of equity prices is a little below 10 percent, while the standard deviation of investment good prices is a little above 1 percent. In the model, these two prices have roughly the same standard deviation, equal to the midpoint between the two empirical standard deviations. Interestingly, the Shiller (1981) “excess volatility” puzzle stands here. Despite its (counterfactually, as we will see) high volatility in interest rates, the model still cannot account for the observed high volatility of stock prices. T h e E q u i t y P r e m i u m a n d R i s k Free R a t e 27 For convenience, our model’s implications for the mean return on assets reported in Figure 6 are reproduced in Table 3. The table also shows that the equity premium is reduced by a factor of 10 by eliminating habit persistence in preferences (see the column, ub — h = 0”), or by dropping the limited labor mobility assumption, that the labor supply decision to each sector is determined prior to the realization of the current period shock (see the column, “Full Labor”). This is consistent with BCF’s conclusion that limitations on factor mobility and habit persistence can produce an equity premium in a business cycle model. By contrast, the model’s ability to account for the risk free rate is, if anything, hurt by habit persistence and limited labor mobility. Other features that are important for the model’s ability to replicate the equity premium are the persistence of the aggregate technology shock and the standard deviation of its innovation. By contrast, the investment specific shock has essentially nothing to do with the equity premium (see the “cr^ = 0” column). This latter is not surprising, in view of (27). Abstracting from the (relatively small) impact of future consumption on the marginal utility of present consumption, the equity premhim would be zero if there were only investment specific shocks. This is because there is no contemporaneous impact on consumption from these shocks. Regarding the aggregate shock, consider the impact of reducing the standard deviation in the innovation in aggregate technology, <re, from its value in the calibrated model to 0.0085 (see the column marked “a e = 0.85%, b = 0.55”). This value of a e is of independent interest because it equates the standard deviation of the innovation in equilibrium consumption, (1 —a)cr£, with an estimate of the corresponding empirical magnitude.18 The drop in the value of cr£ causes the equity premium to fall to 1.45 percent, and it also results in a fall in the risk free rate to 1.93 percent. The equity premium is decreasing in the persistence of the aggregate shock, p. The intuition for this is explored extensively in BCF and is based on the reasoning associated 18A regression of p e r c a p ita co n su m p tion grow th on one lag of itself w ith estim a tio n p erio d 1947:2-1995:1 p ro d u ces a fitte d resid u al w ith sta n d a rd error 0.0054. W hen lagged per c a p ita G D P g ro w th is also included in th e regression, th e s ta n d a rd erro r d rops to 0.0051. T h e d a ta used for these regressions were ta k e n from C itib ase. C o n su m p tio n is m easured as consum ption of nondurables an d services (G C N Q + G C S Q ), a n d th e p o p u la tio n v ariab le is Q P O P . 28 with the permanent income hypothesis. When p is small, the innovation in consumption due to an aggregate shock is temporary, producing a large increase in the demand for new capital for consumption-smoothing reasons, and this in turn generates a big rise in Pk>. The resulting high capital gain is the reason equity is a bad hedge against risk, and this is what underlies the high equity premium. Table 3 shows that the model does less well on the second moment properties of asset returns. In particular, it overstates by a factor of two the volatility of the equity premium and the return on equity. It overstates by a factor of three the volatility of the risk free rate.19 6.2 Business Cycle Implications Table 4 presents the business cycle implications of our model. The format of that table corresponds to that of Table 3. The results in the “Data” column are based on data from the NIPA, which are computed using base year prices. To put our simulated output and investment data on a comparable basis, we measure them in “base” year prices too. The base year in our simulations is the initial observation, when the state of investment-specific technology, Vt , is set to unity, and the model is assumed to be on a steady state growth path, so that the relative price of investment goods is one. Thus, base year output in the model is the simple sum of the physical quantity of consumption and investment goods produced. Similarly, base year investment is the quantity of investment goods. S t a n d a r d B u s i n e s s C y c le S t a t i s t i c s The first 7 rows of Table 4 report the model’s performance in relation to standard business cycle statistics. Its performance is roughly comparable to that of standard models. It shares a problem with standard models in that it understates the volatility of hours worked. 19W e are in v estig atin g w ays of accounting for th e low volatility in th e risk free rate. O ne way, inspired by C am p b ell an d C o ch ran e (1995), is to m ake assum ptions w hich have th e im plication th a t w hen equilibrium c o n su m p tio n d rops close to h a b it, th e u n certain ty in fu tu re consum ption relative to h a b it rises. T hese two p h en o m en a e x e rt opp o sin g effects on th e risk free rate. If they cancel, as in C am pbell an d C ochrane (1995), th e n th e risk free r a te is co n sta n t. 29 The table suggests both shocks play an important role in the model’s business cycle implications. If the investment-specific shock is too small (V M = 0”), then the relative volatility of consumption is too high. If the aggregate shock is too small, then the relative volatility of consumption is too low (see “cre = 0.0085”). Not surprisingly, adjustment costs reduce the volatility of hours worked (see “ip = 0.9”). The limited labor mobility assumption increases the relative volatility of consumption. Again, this is not surprising. Comovement We begin by characterizing the salient characteristics of employment comovement over the cycle. Doing so is complicated to some extent by the fact that some sectors produce both consumption and investment goods, and so the available employment data to not come neatly categorized according to whether they correspond to Hi or H c. To overcome this complication, we report results for 10 different measures of hours worked. Among these there are some that clearly correspond primarily to products”), and others seem clearly related to Hi Hc (for example, “food and kindred ( “construction”). Figure 7 reports the logged data and the associated HP filter trends. W hat is perhaps most striking about these data is the lack of uniformity: some trend up and some trend down and one ( “food and kindred products,” Fig. 7g) even does both. A sharply different picture emerges when one considers the deviations of these data from their HP trend. These are reported in Figure 8. Also reported in Figure 8 is the deviation of logged GDP from its HP trend. The picture that emerges in Figure 8 is one of great similarity among the variables: all comove positively with GDP. They differ in terms of amplitude, but there is little visual evidence of a phase shift. Table 5 reports the correlations with output, and the relative volatility to output, and associated standard errors. In all cases the correlations are large, positive, and statistically significant. Table 4 reports the model’s implications for comovement (see p ( Y , H c) and For ease of comparison, we report the empirical correlation for construction p(Y ,H i).) ( Hi ) and for food and kindred products (H c). Note that the model implies employment in both sectors is procyclical, with the degree of procyclicality being less strong in the consumption sector. 30 To gain insight into how the model manages to deliver this comovement, consider the response of variables to the aggregate shock, reported in Figure 9. Note that consumption rises by nearly 1.1 percent in the period of the shock. In a standard model, the impact of an aggregate shock on consumption would be much lower, as workers and capital are predicted to instantaneously switch out of the consumption sector and into the investment sector.20 Thus, the specification of technology in our model prevents a countercyclical response of hours worked in the consumption sector in the period of the shock. The transient nature of the technology shock, together with habit persistence, ensures that the employment response in the consumption sector remains strong and positive in subsequent periods. Habit persis tence implies that because consumption was high in the period of the shock, the value of consumption is high in the periods after. Our estimate, h = 0, plays an important role here. The transience of the shock implies that to supply consumption in those periods requires high labor effort. And this is exactly what happens according to Figure 9f. Now consider the response to an investment shock. The response of employment in the consumption sector to this shock is positive—presumably reflecting a wealth effect—but very close to zero. The fact that hours wuik^J in the consumption sector is procyclical relative to both shocks guarantees that our model is able to account for the comovement phenomenon. To see what happens in our model when these restrictions on intersectoral factor mobility are relaxed, consider the “Full Labor” column in Table 4. That column reports the dynamic implications of a model parameterized just like our calibrated model, with the only excep tion that employment responds flexibly to the shocks. Note that now employment in the consumption sector is countercyclical. Evidently, to get comovement, habit persistence is not enough, and the limited labor mobility assumption is needed too. In section 3 we showed that with the limited mobility assumption only, and not habit persistence, comovement is not possible either. That is, to get comovement in our framework, both habit persistence and the limit labor mobility assumption are required. Several features of our framework have played an important role in delivering our comove ment result. These include the transitory nature of the aggregate technology shock, our small 20To b e concrete, b akery equ ip m en t an d bakers in th e “food and kindred p ro d u cts” sector are p red icted to tra n sfo rm in sta n ta n e o u sly into bulldozers and bulldozer drivers in th e “co n stru ctio n secto r.” 31 estimated value of h, and, as explained in the previous section, the linear specification of utility in leisure. These considerations help explain why BCF do not find comovement. Their aggregate technology shock is a random walk, their estimated value of h is positive, and they adopt a log-linear specification of utility in leisure. Persistence One indicator that our calibrated model introduces persistence is that its implied growth rate of the Solow residual has autocorrelation —0.1, while equilibrium output growth has autocorrelation 0.02 (see Table 4). Christiano (1988) shows that a standard business cycle which reproduces a pattern like this in the Solow residual implies first order autocorrelation in output growth equal to roughly —0.1. Another indicator of endogenous persistence can be seen in the parameterization, ae = 0.0085, b = 0.65. In this case, the growth rate of the Solow residual is essentially uncorrelated over time. Yet, the autocorrelation in output growth is 0.11 (see Table 4). 6.3 I m p l i c a t i o n s for R i s k A v e r s i o n It is useful, for purposes of assessing the plausibility of our model, to document its impli cations for RRAC, as defined in (10). This is done in Table 6, which reports the values of (xlOO) associated with various values of /i and the various values of b u we consider, along a steady state growth path. We report risk aversion for our calibrated model and the various perturbations on it studied above. In addition, for comparison, we report consumption risk aversion implied by the parameterization considered in Constantinides (1990) and for the production and exchange models studied in BCF. Note that for our calibrated model (b = 0.55), a household would be willing to pay 2.5 percent of one period’s consumption in order to avoid a fair bet on 10 percent of that consumption. This is a low level of risk aversion, particularly by comparison with the levels of risk aversion required in other studies that seek to account for the equity premium. Moreover, our model’s implication for the risk free rate and equity premium is very close to the empirical values of these variables (recall Figure 6.) So, it is natural to investigate how 32 it is that our model manages to do this, w ith so little risk aversion. The explanation lies in two apparently counterfactual im plications of the model. W h y is R R A C s o L o w in th e C a lib r a te d M o d e l? F irs t, note from Table 4 that our model implies the autocorrelation of consumption growth is —0.14, whereas the corresponding em pirical estimate is 0.19. Th is im plication of the model reflects the importance of the aggregate shock, the transitory nature of which enhances the model’s ability to account for the equity premium for a given level of R R A C. The transitory nature of the aggregate shock to technology implies that households have a strong smoothing motive when there is a positive innovation. To investigate the quantitative importance of these considerations, we examined a version of the model in which p — 0.99, so that the aggregate technology shock is almost a random w alk. Now, consumption growth is also virtu a lly a random walk. Predictably, we found that this change reduces the equity premium to 3.46 percent (see Table 3, b = 0.55, p = 0.99.) To offset this, we raised b to 0.60. According to Table 6, w ith this specification of u tility, households are w illing to give up 2.9 percent of a period’s consumption to avoid a 10 percent gamble, up only a little from 2.5 when b = 0.55. Although this is a higher level of risk aversion, it is perhaps not beyond the realm of em pirical plausibility. Interestingly, this version of our model preserves the basic features of the calibrated model: its ability to account for the sign switch, comovement, and the basic features of the business cycle. The second reason our calibrated model can account for the observed equity premium w ith so little risk aversion is that it overpredicts the innovations in consumption. In the model, the standard deviation in the innovation in consumption is ju st (1 — a )c re ~ .011, for reasons discussed above. B u t, in the data this quantity is about one-half of one percent, 0.005. Not surprisingly, this feature of the model also enhances its ability to account for the equity premium. To investigate how important it is, we studied a version of the model w ith (1 —c t ) a e = .0054 (see the indicated column in Tables 3 and 4). Not surprisingly, the equity premium is reduced substantially w ith this change, down to 1.45 percent. We then increased b to 0.65, and this returned the equity premium up to 4.10 percent. Interestingly, we still can account qualitatively for the sign switch and comovement observations. T h is value of b 33 im plies higher risk aversion in consumption (see Table 4 ), but presumably not a level that economists w ill find im plausibly high. Note that this change also raises the autocorrelation of consumption growth to nearly zero. T h is reflects that increasing b enhances the motive to smooth consumption. W h at L evel o f R R A C D o es it Take ? Thus, relatively sm all increases in the model’s im plication for risk aversion in consumption can move it in the direction of being more consistent w ith the consumption data. B u t we have not moved the model all the way. W hat sort of risk aversion in consumption would that im ply? The answer is in B C F . They study a pure exchange economy in which equilibrium consumption growth is modelled based on U .S. consumption data. They accoimt for the equity premium w ith b = 0.58 and h = 0.30. According to Table 6, w ith these parameters, a household is w illing to give up 6.7 percent of consumption to avoid a fa ir, 10 percent gamble. Some w ill perhaps view this as a high degree of risk aversion. Does this mean that, necessarily, to accotxnt for the observed equity premium, high R R A C is required? The answer may be yes. B u t, there are at least three reasons to think that the answer might actually be no. A ll of these reasons build, in different ways, on the notion that the inform ation observed by the economic analyst and that observed by households differ in some way. F irs t, from the analysis above, it is clear that the details of the consumption process m atter a lot for determining how much R R A C is required to account for the equity premium. Y et, there is little confidence in the quality of this data (see W ilcox (1992).) Gibbons (1989) cites this low quality as a reason for ignoring consumption altogether in evaluating asset pricing models. The range of uncertainty about the consumption data when these quality considerations are integrated w ith the usual sampling uncertainty may include parameterizations of consumption which permit accounting for the observed equity premium w ith low R R A C.21 21One indicator of data uncertainty is the fact that the first order autocorrelation in the growth rate of BCF’s consumption data is 0.34. This reflects that their measure of consumption and their sample period dif fer from ours. This high level of consumption autocorrelation underlies their estimate of RRAC. Presumably, they would have reported a lower RRAC, had they used our data set. 34 Second, suppose all the features of the univariate stochastic process underlying the con sumption data were known accurately. Quah (1990) has shown that, even a process in which the univariate representation is a first order autoregression in growth rates w ith posi tive A R (1 ) parameter is consistent w ith an unobserved components representation in which transitory shocks play a very large role. The analysis in this paper has exhibited various em pirical considerations that make such representations plausible, although the calibrated model fails to reproduce crucial features of the type of statistical environment contemplated in Quah (1990). T h is is because the univariate representation of equilibrium consumption exhibits negative persistence in its growth rate. The parameterization, a e = 0.0085, b = 0.65, does exhibit features of Quah’s environment: in this case, the univariate representation of consumption resembles a random walk because its growth rate is nearly uncorrelated over time; yet the innovation in consumption entirely reflects its transitory component (see Table 4). In the statistical environment like the one studied by Quah, as long as agents observe the two underlying components driving consumption, their demand for equity may be driven in an im portant way by the transitory component, possibly leading to a large premium on equity. We are currently exploring this possibility further. T h ird , as is well known, various transformations are applied to the data, which are likely to have the im plication that measured consumption displays more persistence than the actual consumption choices made by agents. The fact that the data are aggregated over time is perhaps the prime example of this possibility.22 Thus, agents could be living in an environment w ith relatively little persistence in consumption, which could be reflected in a high equity premium (like in our a £ = 0.0085, b = 0.65 model), even though published data exhibit substantial persistence due to time aggregation. For a quantitative investigation of this idea in a closely related context, a discussion of the “Deaton paradox” for consumption, see Christiano (1989). 22 Another possibility is seasonal adjustment, which is thought by some to have the effect of smoothing the data. However, some preliminary analysis suggests this may not be an important part of the explanation of the gap between our model and the data. We generated 1,000 observations from the calibrated model. The first order autocorrelation of consumption growth in this data is —0.0773. We then applied the version of the Census X -ll seasonal adjustment procedure implemented in RATS’ EZ-X11 program to seasonally adjust the data (we used the “multiplicative adjustment” option). The first order autocorrelation of the growth rate of the seasonally adjusted artificial consumption data is —0.0382. We concluded that the smoothing implicit in seasonal adjustment does very little to increase first order autocorrelation in growth rates. 35 To summarize this discussion of risk aversion, our model can account for the mean equitypremium and risk free rate w ith low risk aversion. On the one hand, we argued that the model’s apparent counterfactual im plications for consumption played an im portant role in this result. On the other hand, the model’s im plications are not v e r y far off. Moreover, we discussed a variety of considerations that could in principle reconcile the model w ith the data. These considerations make us optim istic that a model can be found that accounts for the key first moments of asset prices without having counterfactual im plications for consumption. 7 C o n c lu d in g R e m a r k s A number of researchers have argued that asset pricing data contain useful inform ation for macroeconomists. For example, the early work of H all (1978) and Hansen and Singleton (1982,1983) showed how to use asset pricing data to test im plications of equilibrium models and estimate their parameters. Data on asset prices are not only useful for evaluating models, but also for providing guidance about how to further develop them. In view of these considerations, it is surprising that b^Cuess cycle researchers have made relatively little use of asset pricing data. In this and a previous paper w ith Michele Boldrin, we took a few steps in this direction. In B C F , we explored modifications in a standard business cycle model that could ac count for the observed high equity premium and low risk free rate. In this paper, we were interested in understanding an observation that we in itia lly found puzzling: equity prices are procyclical, while investment prices are (weakly) countercyclical. Although the literature on Tobin’s q prepares one for the possibility that these two prices are not identical, we were nevertheless surprised find that their business cycle dynamics are so very different. In this paper we incorporated the features proposed in B C F to account for key aspects of the first moment properties of asset returns, together w ith additional features designed to account for the business cycle properties of asset prices. After establishing that there is a param eterization of our model that can account for the price and rates of return on assets, we tinned to see what this parameterization implies for business cycles and for risk aversion. We find that the model does at least as well as standard business cycle models in ac 36 counting for conventional business cycles facts. On two dimensions, we find that the model actually represents a step forward relative to the standard business cycle model. F irs t, as in B C F , we find that the modifications designed to account for mean asset returns help confer an internal propagation mechanism to the model. Second, in our model environment, the modifications also allow the model to be consistent w ith the fact that employment is procyclical across a broad range of sectors. The basic features that we use to account for the asset pricing phenomena are habit persistence preferences and lim itations on the ability to quickly move factors of production both cross-sectionally and intertem porally. These same lim itations, by slowing the economy’s ability to respond to shocks, have the effect of introducing persistence. A t the same tim e, lim itations on intersectoral m obility, coupled w ith habit persistence, have the effect of making employment across sectors move up and down together over the cycle. The results in this paper and in B C F support the notion that the same frictions needed to account for the salient features of asset prices and returns are also useful in understanding the salient features of business cycles. 37 A T h e P l a n n e r ’s P r o b l e m The quantities in a competitive equilibrium of our model can be computed by solving the following planner problem: maximize OO £ o £ / ? {h i(G - X t ) + r j ( T - H c>t - H i>t) t =0 + A x>t [ X t+1 - h X t - b C t} + A c,t [ ^ ( e x p ^ O ^ t ) 1-" - + A m \V t K * t (e x p (^ )^ ,i)1_Q ~ h ,t ~ I ii.t, + a ,« [O ' ((1 - 6 ) K c,„ [ , , ) - K CJ+,\ + «i,i [0‘ ((1 - i ) K , , u h ,t) - K k w ] } subject to 9 t = V t = exp(/2 + p Q t-x + e t , £ t P t ) V t- ~ i N ( , 0, c r 2 ) P t ~ -A(0, c r 2 ) Here, Ay t, y = x , c , i and f2y f, y = c , i are Lagrange m ultipliers. The planner is assumed to choose H c<t and H t t prior to the realization of the date t state of nature, while the remaining choice variables (including the m ultipliers) are selected afterward. To solve the planner problem we first have to transform it into its stationary form. The variables in the transformed problem are c t , k Ctt + i , k ^ t+ i , x t+ i , i c^ ,i i,t , AC)f, A,)t, AI)t,u>Cit and dJi't- These are defined as follows (not all variables are transformed in the same w ay): - Ct Ct ~ ya/(l-a) ’ _ Ic,t *Clt “ yV(l-a) > L- - ,*+i Kc’t+l ~ yl/(l-a) ’ ■ i W _ _ K i ,t+ 1 1 ~ yl/(l-a) ’ _ Ii,t - yV(l-a) ’ 38 X t+i Xt+1 ~ ya/il-a) \ = _T/a/ Ac,i V t (1_a^AA c>t, \ _t/1 — V*/(1_q )aA t^ , \ — T/a^ 1“a^A AX|t — Vt l\Xjt In what follows it w ill also be helpful to use the following: Y itt Y<* ya/(l-a) ’ y^1 y*'1 t/ 1 /(1 - q ) Vf where Y c>t - K “t ( e x p ( 9 t ) H c<ty “ and Y i>t = VtK ° t ( e x p ( 9 t ) H itt ) 1 The first order conditions for an interior solution to the stationary version of the planner problem can be rearranged to form the following system of equations: (39) \ x,t - ( 3 E Ai , t 1 c t+i - e x p ( -(J L t+ i)x t+ i 1 -P E Q2,l (40) exp(—/}t+ i)(A Iif+1/H- Qt = 0 V c ,t+ l ,Q +1 —6xp( —£ l t + l ) x + e xp (—/2f+i ) ( l - 6 ) — ^— Q ch t+ 1 t+ l b^x,t+ i ) a '~ ^C,t-f1 = 0 V 2 ,i+ 1 (41) Ai,t+i V2,f+1 1 (42) £7 \ c t - e x p ( - f tt)x t (43) E r2 / = + exp (—M«+i)(l \ ^t,t+l V2 ,t - b \ x<t) - M i - ®)jr- - v n: (1 - a ) j f ~ ~ 9 n: 0 / =o c,t = o (44) 0 (45) i c ( “I- %x,t V i,t In these expressions 0 denotes the information set that includes realizations of the tech nology shocks up to and including time t , and f2* denotes the information set identical to Qf except that the time t realizations of the technology shocks are excluded. Also, Q Jt, j = 1,2 denote partial derivatives of Q * w ith respect to its first and second argu ments, respectively, evaluated at the time t values of the arguments, for x = c , i . Fin ally, At = 1 - a ) and /2t+ 1 = j j t / a . 39 Making use of the resource constraints implicit in the formulation of the planner problem, we can collapse the model to the following seven equations E [ Vj(kCtt,kc,£+l> kCjt+27 ^ x , t , ^x,t+i) ^ i , t , ki,t+1> 2> xti%t+lj %t+2jH ct^H ct+\, Hi,t, Hi ,*+l> i @t>@ t + 1 , f^t, (J't+i ')| = 0, E \ V j { k c,ti k c , t +li k c , t + 2 , k i , t , k i , t + 1) ^-4,4+2) 2+, j = 1,2, 3, 6,7; 2-4+2) H Ctt , -He,4+1> -Hf.t) -Hf^+l, '^i ,4>AIi4+ i,Aijt,Ai]4+x,^4,04+x,//4,^4+i)|f2(] =0, j = 4, 5. Here v j ( • ) , j = 1 , 2 , 3 , 6 , 7 are equations (39), (40), (41), (44), and (45) respectively, and Vj (•), j = 4 , 5 are equations (42) and (43), respectively. These equations can be solved using the methods described in Christiano, Fisher and Valdivia (1995). We can use the multipliers from a solution to the planner problem to compute the relative prices studied in the main text. First, the relative price of the new investment good is given by p — -n,4 = •'V* Ai,t ^c,4 AC)t V t 1 Second, the prices for K c, t + \ and K i t + i are Pk'c,t = a c,4 A c,4 a 4,4 u’ct 1 Pk[,t = AC)t Vt’ ^ ’c A c,t ^4,4 1 A ctVt’ respectively. Third, the prices for installed capital are P _ Q l,4^c,4 _ LOct Q l,t fcc,t = Act ~ p M _ ^?1,4^*,4 _ ^4,4^1,4 = A ct ~ A~t~ K ' Notice that each of these prices will trend downward if V t has a positive trend. To use these formulas we require expressions for \ c , t , u c t and u h t . These can be computed from the first order conditions from the planner problem. The expressions derived in this way are given by Ac.t y c,t A j 4 ^c,4 = exp(jlt)xt b \ x ,ti , •, and Q h A,; '4 ,4 Wi,t" Qit' 40 R eferences Abel, Andrew, 1980, “Em pirical investment equations: An integrative fram ework,” C a m e g ie R o c h e s t e r C o n f e r e n c e S e r i e s o n P u b lic P o lic y , vol. 12, pp. 39-91. Benhabib, Jess, Richard Rogerson, and Randall W right, 1991, “Homework in macroeco nomics: Household production and aggregate fluctuations,” J o u r n a l o f P o l i ti c a l E c o n om y, vol. 99, pp. 1166-1187. Benninga, Simon and A ris Protopapadakis, 1990, “Leverage, time preference and the ‘equity premium puzzle,’ ” J o u r n a l o f M o n e ta r y E c o n o m ic s , vol. 25, pp. 49-58. Boldrin, Michele, Lawrence J . Christiano, and Jonas D.M . Fisher, 1995, “Asset pricing lessons for modeling business cycles,” National Bureau of Economic Research working paper 5262. Cam pbell, John Y ., and John H. Cochrane, 1995, “B y force of habit: A consumptionbased explanation of aggregate stock market behavior,” National Bureau of Economic Research working paper 4995. Cecchetti, Stephen G ., Pok-sang Lam , and Nelson C . M ark, 1993, “The equity premium and the risk-free rate: Matching the moments,” J o u r n a l o f M o n e ta r y E c o n o m ic s , vol. 31, pp. 21-45. Christiano, Lawrence J ., 1988, “W hy does inventory investment fluctuate so much?” J o u r n a l o f M o n e t a r y E c o n o m ic s , vol. 21, pp. 247-280. Christiano, Lawrence, J ., 1989, “Comment on ‘consumption, income, and interest rates: Reinterpreting the time series evidence,” ’ in O livier Jean Blanchard and Stanley F is cher, editors, N B E R M a c r o e c o n o m ic s A n n u a l 1 9 8 9 , pp. 216-233. Christiano, Lawrence J ., and M artin Eichenbaum, 1992, “Current real-business-cycle the ories and aggregate labor-market fluctuations,” A m e r ic a n E c o n o m ic R e v ie w , vol. 82, pp. 430-450. 41 Christiano, Lawrence J., Jonas Fisher, and Victor Valdivia, 1995, “A linearization method for solving dynamic, general equilibrium models,” unpublished manuscript. Constantinides, George, 1990, “Habit formation: A resolution of the equity premium puz zle,” J o u r n a l o f P o l i t i c a l E c o n o m y , vol. 98, pp. 519-542. Eichenbaum, Martin, and Lars Peter Hansen, 1990, “Estimating models with intertemporal substitution using aggregate time series data,” J o u r n a l o f B u s in e s s a n d E c o n o m ic S t a t i s t i c s , vol. 8, pp. 53-69. Fisher, Jonas, 1994a, “A dynamic, general equilibrium empirical analysis of real and mon etary aspects of the business cycle,” unpublished Northwestern University Ph.D. dis sertation. Fisher, Jonas, 1994b, “Relative prices, complementarities, and co-movement among com ponents of aggregate expenditures,” University of Western Ontario research report 9405. Fisher, Jonas, 1995, “W hy does residential investment lead business investment over the business cycle?” manuscript, University of Western Ontario. Gibbons, Michael R., 1989, “On the volatility of bond prices,” C a r n e g ie - R o c h e s te r C o n f e r e n c e S e r i e s o n P u b lic P o l i c y , vol. 31, pp. 139-175. Gordon, Robert J., 1990, T h e m e a s u r e m e n t o f d u ra b le g o o d s p r ic e s , University of Chicago Press. Greenwood, Jeremy, Zvi Hercowitz, and Per Krusell, 1992, “Macroeconomic implications of investment-specific technological change,” Institute for Empirical Macroeconomics discussion paper 76, Federal Reserve Bank of Minneapolis. Hall, Robert E., 1978, “Stochastic implications of the life cycle permanent-income hypoth esis: Theory and evidence,” J o u r n a l o f P o litic a l E c o n o m y , vol. 86, pp. 971-987. Hansen, Gary, 1985, “Indivisible labor and the business cycle,” J o u r n a l o f M o n e t a r y E c o n o m i c s , vol. 16, pp. 309-327. 42 Hansen, Lars Peter, and Kenneth Singleton, 1982, “Generalized instrumental variables estimation of nonlinear rational expectations models,” E c o n o m e tr ic a , vol. 50, pp. 1269-1286. Hansen, Lars Peter, and Singleton, Kenneth J., 1983, “Stochastic consumption, risk aver sion, and the temporal behavior of asset returns,” J o u r n a l o f P o l i ti c a l E c o n o m y vol. 91, pp. 249-265. Lucas, Robert E., Jr., and Edward Prescott, 1971, “Investment under uncertainty,” E c o n o m e t r i c a , vol. 39, pp. 659-681. Lucas, Robert E., Jr., 1981, “Understanding business cycles,” reprinted in S tu d ie s in B u s i n e s s - C y c l e T h e o r y , M IT Press. Murphy, Kevin M., Andrei Shleifer, and Robert W. Vishny, 1989, “Building blocks of market clearing business cycle models,” N B E R M a c r o e c o n o m ic s A n n u a l 1 9 8 9 . Quah, Danny, 1990, “Permanent and transitory movements in labor income: An explana tion for ‘excess smoothness’ in consumption,” J o u r n a l o f P o litic a l E c o n o m y , vol. 98, pp. 449-475. Rogerson, Richard, 1988, “Indivisible labor, lotteries and equilibrium,” J o u r n a l o f M o n e ta r y E c o n o m ic s , vol. 21, pp. 3-16. Sargent, Thomas J., 1979, M a c r o e c o n o m ic th e o r y , Academic Press. Shiller, Robert, 1981, “Do stock prices move too much to be justified by subsequent changes in dividends?” A m e r i c a n E c o n o m ic R e v ie w , vol. 71, pp. 421-436. Sundaresan, Suresh M., 1989, “Intertemporally dependent preferences and the volatility of consumption and wealth,” R e v ie w o f F in a n c ia l S tu d ie s , vol. 2, pp. 73-89. Wilcox, David W., 1992, “The construction of U.S. consumption data: Some facts and their implications for empirical work,” A m e r ic a n E c o n o m ic R e v ie w , vol. 82, pp. 922-941. 43 Table 1: Business Cycle Properties of Stock Prices Second Moment Statistics Dow Jones Industry S&P 500 ------ corr(p, y ) corr(p,y) . °JL . 5.22 0.30 Composite 5.14 0.36 (0.55) (0.10) (0.53) (0.09) Capital Goods 6.06 0.39 (0.56) (0.09) 0.21 Utilities 4.66 (0.64) (0.11) 6.42 0.30 Finance (1.03) (0.12) Industrial 5.33 0.35 (0.55) (0.09) Transportation 5.82 0.19 (0.79) (0.15) NYSE corr(p,y) 5.20 (0.53) na 0.37 (0.09) na 6.82 (1.40) 6.73 (0.93) 5.62 (0.75) 7.45 (1.08) 0.30 (0.09) 0.38 (0.10) 0.35 (0.12) 0.36 (0.10) Notes: (i) Data source— CITIBASE. The sample period is 1947I-1995I for NYSE Composite, Dow Jones, and 1966I-1995I for the other NYSE variables. The sample period is 19471—19951 for all but two of the S&;P 500 data series. It is 19701—19951 for S&P 500 finance and transportation. Data on gross domestic product (GDP) cover the period 19471—19951. Stock price data are deflated by implicit price deflator for consumption of nondurables and services. (ii) Statistics— All data were logged, and then Hodrick-Prescott filtered prior to analysis. a p denotes the standard deviation of the (detrended) stock price, a y denotes the standard deviation of output, and c o r r ( p , y ) denotes the correlation between p and y . Numbers in parentheses denote the standard errors of <j p / ( tv and corr(p,y), computed as in Christiano and Eichenbaum (1992). For estimation of the relevant zero-frequency spectral density a Bartlett window, truncated at lag four, was used. 44 Table 2: Sample First Moments, Macroeconomic Variables Data 1947:1-1995:1 1983:1-1995:1 ~ C ---- 0.56 0.59 Y* C 0.76 0.75 Y* hi 0.24 0.24 Y* Ph'K 10.6 y* 1.46 AY* 1.85 1.84 1.68 AC 1.58 1.86 AC 3.75 AI 2.35 1.11 APJ 1.81 -2.54 A Pi -0.57 Model Calibrated a = 0.28 0.64 0.72 0.36 11.3 1.60 1.60 0.28 9.1 1.60 1.60 4.44 1.60 -2.84 4.44 1.60 -2.84 Notes: (i) Data from “Data” column are taken from Citibase, except the capital-output ratio, which was taken from Christiano (1988) and covers the period 1956:111-1984:1. Data in “Calibrated Model” column are properties of the nonstochastic steady state. A x ~ first difference of log of otherwise untransformed variable. (ii) U.S. data— C ~ consumption of nondurables (GCNQ) and services (GCSQ ); C ~ consumption of nondurables and services, plus government purchases (GGEQ ); / ~ business fixed investment (G IFQ ) plus household durable purchases (GCDQ ). The preceding variables are measured in 1987 dollars. Y * ~ current dollar Gross Domestic Product (GDP) divided by implicit price deflator for consumption of nondurables and services { { G C N + G C S ) / ( G C N Q + G C S Q ) ) ; Pi I ~ current dollar business fixed investment (G IF) plus current dollar durable purchases (G CD ), divided by implicit price deflator for consumption of nondurables and services; Pi ~ business fixed investment plus household durable goods deflator ((G C D + G IF)/(G C D Q -i-G IFQ )), divided by implicit price deflator for consumption of nondurables and services. Growth rate results are for per capita variables, obtained by scaling by the propulation between 16 and 64 years old (PAN17-PAN19). The growth rate results extend only to the period 1995:1, reflecting the availability of the population data in our version of Citibase. A ll mnemonics correspond to variable names in Citibase. (iii) Calibrated model data—Y * ~ C + P {I. 45 Table 3: Prices and Rates of Return <7e = 0.85% Statistic Data Calibrated Full Labor b = h = 0 b = 0.55 b = 0.65 2.24 3.35 6.66 1.96 7.82 re 8.56 (0.02) (0.01) (0.03) (0.06) (0.08) rJ 1 . 6 4 1 . 9 3 2.56 1 . 6 5 2.78 1.19 (0.05) (0.07) (0.03) (0.03) (0.09) (0.81) _ yf 0 . 3 2 1 . 4 5 4.10 0 . 5 8 5.78 6.63 (0.12) (0.07) (0.04) (0.03) (0.15) (1.78) 34.5 8.67 13.8 41.2 11.3 CT 19.53 (0.04) (0.1) (0.1) (0.03) (0.2) 9.02 2.42 16.0 18.2 3.58 5.27 a rS (4e-2) (0.03) (0.01) (0.01) (0.74) (0.1) 17.4 30.1 10.7 8.28 36.4 19.02 (Tj.e —y/ (0.02) (0.1) (0.1) (0.2) (0.03) (1.73) 3.73 4.85 3.22 3.11 1.11 5.16 ° p, (0.02) (0.02) (0.02) (0.02) (0.02) 4.07 1 . 3 7 2 . 5 0 1 . 6 9 4.77 9 . 3 1 a P k> (0.01) (0.01) (4e-3) (0.01) (0.01) -0.64 -0.42 -0.78 -0.75 -0.15 P(Y,Pi ) -0.15 (4e-4) (4e-3) (3e-3) (3e-3) (0.01) 0.04 -0.08 -0.08 -0.06 0.30 0.30 p(X,Pk>) (4e-4) (4e-3) (0.01) (0.01) (4e-3) 46 0 8.46 (0.08) 2.77 (0.08) 5.69 (0.15) 40.9 (0.2) 18.2 (0.1) 36.0 (0.1) 4.11 (0.01) 4.69 (0.01) 0.77 (3e-4) 0.82 (2e-4) CTit = ip = 0.9 P = 0 8.95 (0.08) 2.80 (0.09) 6.15 (0.15) 42.7 (0.2) 18.2 (0.1) 38.1 (0.2) 5.25 (0.02) 5.13 (0.02) -0.09 (0.01) 0.00 (0.01) 11.1 (0.1) 3.76 (0.10) 7.31 (0.17) 49.4 (0.3) 23.9 (0.1) 42.3 (0.2) 5.65 (0.02) 5.39 (0.02) -0.07 (0.01) 0.33 (4e-3) p = 0.99 b = 0.55 b = 0.60 a = 0.28 5.47 (0.06) 2.01 (0.06) 3.46 (0.11) 29.7 (0.1) 11.8 (0.04) 27.0 (0.1) 4.31 (0.02) 3.53 (0.01) -0.38 (0.01) 0.18 (4e-3) 8.18 (0.08) 2.43 (0.07) 5.75 (0.15) 39.8 (0.2) 16.3 (0.1) 35.9 (0.1) 5.10 (0.02) 4.52 (0.01) -0.28 (0.01) 0.18 (4e-3) 7.98 (0.08) 2.26 (0.08) 5.72 (0.15) 40.9 (0.2) 18.1 (0.1) 36.1 (0.1) 5.18 (0.02) 4.77 (0.01) -0.08 (0.01) 0.37 (4e-3) Notes: (i) The “Data” column contains estimates of the mean and standard deviation of the risk-free return and the equity premium, with standard errors in parentheses over the period 1892-1987 for U.S. data. These estimates are taken from Cecchetti, Lam and Mark (1993). These authors do not report the analogous values for the return to equity; (ii) p ( x , y ) denote the correlation between variables x and y; (iii) Except for the correlations, all statistics are in percent terms; (iv) Rates of return are annualized; (v) Results for the models are based on 500 replications of sample size 120, and Monte Carlo standard errors are reported in parentheses. The latter are the standard deviation, across replications, of the associated statistics, divided by \/500; (vi) Prices and output in the model and the data are logged and HP filtered prior to analysis, rates of return are not filtered. Pi is measured by the ratio of 1987 dollar business fixed investment plus consumption of durables to the implicit price deflator for consumption of nondurables and services. o p k, p ( Y , P k>) are taken from Table 1, and the fact that a y — 1.79 percent. 47 Table 4: Business Cycle Statistics a e = 0.0085 Statistic Data aY 1.79 o c/o y 0.47 (T i/ay 2.91 (Th / (TY 0.82 p (y ,c ) 0.78 p(Y ,I) 0.70 P {Y ,H ) 0.81 p 0.52 (y , h c ) p (Y ,H i) 0.79 p(A Y ) 0.36 p(A C ) 1 0.19 p(AS) -0.10 a{A S) 1.80 Calibrated Full Labor b = h = 0 p = 0.99 b = 0.55 b = 0.65 a fl = 0 1.90 (0.01) 1.19 (4e-3) 0.99 (2e-4) ^ = 0.9 2.65 (0.02) P = o b = 0.55 b = 0.60 2.32 (0.02) 2.32 (0.02) 0.61 (4e-3) 1.80 (0.01) 0.14 (4e-4) 0.61 (4e-3) 1.01 (2e-4) 0.26 (2e-4) 0.46 (4e-3) 1.95 (0.01) 0.44 (le-3) 2.09 (0.01) 0.54 (4e-3) 1.92 (0.01) 0.29 (le-3) 1.80 (0.01) 0.14 (4e-4) 0.63 (4e-3) 2.14 (0.01) 0.20 (le-3) a = 0.28 1.92 (0.01) 2.16 (0.01) 0.56 (4e-3) 1.87 (0.01) 0.20 (le-3) 2.09 (0.01) 0.27 (2e-3) 2.05 (0.01) 0.29 (2e-3) 2.14 (0.01) 0.53 (3e-3) 1.89 (0.01) 0.15 (4e-4) 2.06 (0.01) 0.18 (le-3) 1.90 (0.01) 0.34 (3e-3) 2.06 (0.01) 0.19 (le-3) 0.55 (4e-3) 0.95 (le-3) 0.63 (3e-3) 0.43 (0.01) 0.98 (3e-4) 0.14 (0.01) 0.53 (4e-3) 0.95 (le-3) 0.69 (3e-3) 0.31 (0.01) 0.97 (4e-4) 0.72 (3e-3) 0.32 (0.01) 0.97 (4e-4) 0.68 (3e-3) 1.00 (3e-6) 1.00 (4e-6) 0.48 (3e-3) 0.48 (0.01) 0.96 (le-3) 0.80 (2e-3) 0.49 (4e-3) 0.95 (le-3) 0.38 (4e-3) 0.62 (0.01) 0.94 (le-3) 0.75 (2e-3) 0.62 (0.01) 0.95 (le-3) 0.75 (3e-3) 0.63 (4e-3) 0.91 (le-3) 0.60 (3e-3) 0.23 (0.01) 0.70 (3e-3) -0.38 (4e-3) 0.93 (2e-3) NA (NA) 0.13 (0.01) 0.69 (3e-3) 0.13 (0.01) 0.75 (3e-3) 0.75 (3e-3) 0.43 (3e-3) 0.53 (3e-3) 0.31 (0.01) 0.81 (2e-3) 0.07 (0.01) 0.67 (3e-3) 0.16 (0.01) 0.75 (2e-3) 0.13 (0.01) 0.75 (3e-3) 0.27 (0.01) 0.68 (3e-3) 0.02 (4e-3) -0.14 (4e-3) -0.05 (4e-3) 0.16 (4e-3) -0.003 (4e-3) -0.23 (4e-3) 0.10 (4e-3) -0.12 (4e-3) 0.11 (4e-3) -0.09 (4e-3) -0.12 (4e-3) -0.15 (4e-3) 0.26 (4e-3) -0.14 (4e-3) -0.13 (4e-3) -0.25 (4e-3) 0.12 (4e-3) 0.01 (4e-3) 0.12 (4e-3) 0.01 (4e-3) 0.01 (4e-3) -0.14 (4e-3) -0.10 (4e-3) 1.80 (0.01) -0.13 (4e-3) 1.92 (0.01) -0.10 (4e-3) 1.80 (0.01) -0.03 (4e-3) 1.45 (0.01) -0.03 (4e-3) 1.45 (0.01) -0.23 (4e-3) 1.24 (4e-3) -0.04 (4e-3) -0.29 (4e-3) 2.04 (0.01) 0.01 (4e-3) 1.69 (0.01) 0.01 (4e-3) 0.33 (le-3) 48 1.80 (0.01) 1.70 (0.01) -0.12 (4e-3) 1.64 (0.01) Notes: (i) Figures in the “Data” column are based on U.S. data, covering the period 1947:1-1995:1, taken from Citibase. Consumption is measured by consumption of nondurables and services, GCNQ+GCSQ, measured in 1987 dollars, divided by GPO P, a not-seasonallyadjusted measure of population (including armed forces overseas). Investment is business fixed investment, G IFQ , plus consumption of durable goods, GCDQ, measured in 1987 dollars, and scaled by GPOP. H c is measured by LWH20X, which is employment in the industry in Table 5 that looks most (to us) like a consumption goods industry. Similarly, H i is measured by LW HCX in Table 5; (ii) With the exception of the correlations and the relative volatilities, all the statistics are reported in percentage terms; (iii) Results for the model are based on 500 replications of sample size 120, and Monte Carlo standard errors are reported in parentheses; (iv) p ( x ) means the first order autocorrelation of the variable x , p ( x , y ) means the correlation between variables x and y, and cr(x). means the standard deviation of x . A x means the first difference of the log of (otherwise untransformed) x . p ( A S ) denotes the first order autocorrelation of the growth rate of the model implied Solow residual, and a ( A s ) denotes the standard deviation of the growth rate of the model implied Solow residual, (v) Variables without A have been logged and HP filtered prior to analysis; (vi) the entry NA (Not Applicable) signifies that the indicated number is not defined. The discussion in Section 4.4.4 indicates that when b = h = 0, then H c is a constant. 49 Table 5: Business Cycle Properties of Hours Worked Hours worked, Ht 1. Household (lhours) 2.Total private (lwhx) 2.a. Goods producing (lwhgx) Construction (lwhcx) Manufacturing: Durable goods (lwhdx) Non durable goods (lwhnx) Food and kindred products (lwh20x) Apparel and textiles (lwh23x) 2.b. Service producing (lwhpx) Retail trade (lwhrx) T — —3 r = -2 corr(Ht,Yt-T) T = •- 1 r = 0 T = 1 0.18 0.09 0.40 0.08 0.46 0.09 0.49 0.08 0.45 0.09 0.65 0.05 0.71 0.05 0.73 0.05 0.64 0.07 0.48 0.09 0.55 0.08 0.25 0.72 0.06 0.75 0.05 0.41 0.13 0.69 0.07 0.61 0.07 0.79 0.06 (Th /(TY 0.82 0.06 1.18 0.05 2.26 0.22 0.11 0.26 0.10 0.11 3.18 0.23 0.25 2.96 0.14 1.43 0.18 0.69 0.08 1.94 0.25 0.66 0.05 0.82 0.05 0.11 0.26 0.11 0.35 0.10 0.10 0.12 0.12 0.40 0.08 0.15 0.56 0.08 0.38 0.11 0.10 0.51 0.65 0.09 0.11 0.81 0.03 0.90 0.03 0.92 0.02 0.79 0.05 0.91 0.02 0.86 0.03 0.52 0.13 0.73 0.06 0.80 0.05 0.88 0.03 r = 2 T= 3 0.51 0.09 0.70 0.81 0.05 0.93 0.05 0.93 0.05 0.83 0.07 0.69 0.07 0.84 0.08 0.81 0.08 0.77 0.69 0.10 0.12 0.92 0.04 0.77 0.05 0.52 0.82 0.08 0.56 0.09 0.46 0.12 0.12 0.60 0.07 0.85 0.06 0.81 0.06 0.35 0.39 0.14 0.08 0.10 0.11 0.82 0.08 0.11 0.11 0.65 0.11 0.64 0.10 0.33 0.11 0.73 0.66 0.47 0.09 0.11 Notes: (i) H ~ measure of hours worked, taken from Citibase, with mnemonic indicated in parentheses in column 1. Y t ~ GDP in 1987 dollars, from Citibase. Sample period— all data are quarterly, and with the exception of lhours, they cover the period 1964:1-1995:1. Lhours covers the period 1947:I-1993:IV. Numbers in the second row of each block are standard deviations, computed using the procedure described in note (ii) to Table 1. (ii) Sources and definitions— data from 1 ~ household survey, manhours employed per week; data from 2 ~ establishment survey, indexes of aggregate weekly hours of production or nonsupervisory workers on private nonagricultural payrolls by industry. 50 Table 6. Measures of Risk Aversion in Consumption. Model Calibrated B C F Calibrated Perturbed, Calibrated Perturbed, Calibrated Perturbed, Calibrated B C F, Exchange Economy b h X 0.55 0.35 0.60 0.65 0.80 0.58 0.0 0.40 0.0 0.0 0.0 0.30 0.55 0.58 0.60 0.65 0.80 0.82 1% 0.02 0.03 0.03 0.04 0.12 0.16 10% 2.5 2.7 2.9 3.6 6.7 7.3 20% 9.8 9.7 10.3 11.9 18.6 12.4 Notes: Entries in the last three columns are the value of 100 x u associated with the indicated row value of b and the column value of fi. The value of u solves (4.10). The model column indicates what motivated the particular parameterization considered. 51 0 065 - Fig . 2f: Stru ctu res Investm ent on 0060 - oio - 0055 - ooe - 0 050 0 045 0 040 0 035 0030 0 025 -* ooe oo/ - ooe 005 0 04 - Fig . 2g: Equipm ent Investm ent Fig . 2i: Governm ent P u rch a se s 0 300 - I I I " " 1 0»0 -i I' l' , I I, l ^ I ' , \ . F ig u re 4: C o rre latio n B e tw e e n Q u arte rly P ric e of In ve stm en t at t and Output at In ve stm en t p rice an d output d ata logged and d etren d ed b y H P filter Fig. 4b: Household durable investment Fig. 4c: Fixed investment Fig. 4d: Nonresidential investment Fig. 4e: Residential investment Fig. 4f: Structures investment Correlation Correlation Correlation Fig. 4a: Fixed investment and durable goods Correlation Fig. 4g: NIPA equipment investment t-k Figure 5: Correlation Between Stock Price M easure at t and Output at t-k Correlation Price and output data logged and detrended by HP filter S&P500: Finance vs output Correlation NYSE: Finance vs output k S&P 500: Utilities vs output Correlation NYSE: Utilities vs output k k k S&P 500: Capital good vs output Correlation Dow Jones vs output k S&P 500: Industrial vs output Correlation NYSE: Industrial vs output k k S&P 500: Transportation vs output Correlation NYSE: Transportation vs output k k k Figure 6: Assigning Values to Preference Parameters, b and h 14 1 2 b = 0.475 1 0 8 b = 0.450 6 b = 0.425 4 h = 0.3 b = 0.400 /j 0 . 2 2 • Corresponds to a particular h, ---1 0 2 - 1 3 Risk free rate (A R % ) 4 i 1 5 i 6 b combination i 8 Fig. 7a: Household hours vsrsu s trend Fig. 7d: Construction hours versus trend Fig. 7b: Total Private hours versus trend no 41) - Fig. 7e: Durable goods hours versus trend 4*0 - *n 4>0466 “ 4tO4)) 4JO- Fig. 7J: Retail trade versus trend Fig. 71: Service-producing hours versus trend 41 47 46 46 44 41 4. M 1/ IV 0050 Fig. la : Household hours vartut output Figure 9: Impulse Response Function for Calibrated Model % Deviation Fig. 9a: Response of C % Deviation Fig. 9b: Response of I Fig. 9h: Response of P^ % Deviation Fig. 9c: Response of h % Deviation Fig. 9d: Response of \Q % Deviation Fig.9e: Response of lj ------ Aggregated Shock - - - • Investment Shock 0