The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
Working Paper Series Financing Development: The Role of Information Costs WP 08-08R Jeremy Greenwood University of Pennsylvania Juan M. Sanchez Federal Reserve Bank of Richmond Cheng Wang Iowa State University This paper can be downloaded without charge from: http://www.richmondfed.org/publications/economic_ research/working_papers/index.cfm Version: August, 2009 Financing Development: The Role of Information Costs by Jeremy Greenwood, Juan M. Sanchez and Cheng Wang Working Paper No. 08-08R Abstract To address how technological progress in …nancial intermediation a¤ects the economy, a costlystate veri…cation framework is embedded into the standard growth model. The framework has two novel ingredients. First, …rms di¤er in the risk/return combinations that they o¤er. Second, the e¢ cacy of monitoring depends upon the amount of resources invested in the activity. A …nancial theory of …rm size results. Undeserving …rms are over …nanced, deserving ones under funded. Technological advance in intermediation leads to more capital accumulation and a redirection of funds away from unproductive …rms toward productive ones. Quantitative analysis suggests that …nance is important for growth. Keywords: costly-state veri…cation, economic development, …nancial intermediation, …rm size JEL Nos: E13, O11, O16 A- iations: University of Pennsylvania, Federal Reserve Bank of Richmond, and Iowa State University An abridged version of this paper, containing the theory sections, is forthcoming in the American Economic Review. A variant of the model with credit rationing is developed in NBER Working Paper No. 13104. 1 Introduction Financial development “accelerates economic growth and improves economic performance to the extent that it facilitates the migration of funds to the best user, i.e. to the place in the economic system where the funds will earn the highest social return,” noted Goldsmith (1969, p. 400) some thirty …ve years ago. Information production plays a key role in this process of steering of funds to the highest valued users. If the costs of information production drop, then …nancial intermediation should become more e¢ cient with an associated improvement in economic performance. A theoretical model is developed where reductions in the cost of information processing allow for more e¢ cient capital allocation. Quantitative analysis suggests that cross-country di¤erences in the e¢ ciency of intermediation may account for a substantial, but not the dominant, part of cross-country income di¤erentials. Some stylized facts of the kind illustrated in Figure 1 were o¤ered by Goldsmith (1969) to suggest that …nancial intermediation might be important: First, the ratio of business debt to GDP has risen. In 1952 business debt was 30 percent of GDP.1 Today it is 65 percent. Second, the value of …rms relative to GDP has also moved up. In 1951 …rms were worth 50 percent of GDP, while today they are valued at 176 percent. This may be evidence of improved intermediation; now it is easier for …rms to enter stock and bond markets and raise funds. Direct measures of the impact of improved e¢ ciency in the …nancial system on the economy are hard to come by. Improvements in the e¢ ciency of …nancial intermediation, due to improved information production, are likely to reduce the spread between the internal rate of return on investment in …rms and the rate of return on savings received by savers. The spread between these returns re‡ects the costs of intermediation. This spread will include the costs of ex ante information gathering about investment projects, the ex post information costs of policing investments, and the costs of misappropriation of savers’funds by management, unions, etc., that arise in a world with imperfect information. One may 1 Data sources are provided in Appendix 14.2. 1 2.0 2.0 0.7 Debt Capital 0.5 1.0 0.4 0.5 1.8 Capital / Output Business Debt / GDP 1.5 Market Value of Firms / GDP 0.6 1.6 1.4 Value 0.3 1940 1960 1980 1940 2000 1960 1980 Year Year Figure 1: Trends in …nancial intermediation 2 2000 observe little change in the rate of return earned by savers over time, because aggregate savings will adjust in equilibrium so that this return re‡ects savers’rates of time preference. If the wedge between the internal rate of return earned by …rms and the rate of return received by savers falls, due to more e¤ective intermediation, then the capital stock in the economy should rise. Indeed, there is some evidence that this may be the case. Figure 1 plots the capital-to-GDP ratio for the business sector of the economy, where the capital stock includes intangible investments such as …rms’research and development, following the lead of Corrado et al. (2006). It has increased signi…cantly over the postwar period. Of course, an economy’s capital-to-GDP ratio may rise on other accounts, too. For example, lower taxes on capital income should increase it, as should declines in the price of capital goods due to investment-speci…c technological advance. To the extent that capital stock measures exclude intangibles, or fail to incorporate improvements in quality due to embodied technological progress, more e¤ective intermediation will be picked up as increases in productivity, instead. There is evidence of the above phenomena in the cross-country data, too. Figure 2 plots the relationship between interest-rate spreads and output for a sample of 49 countries. As can be seen, there is a negative association. Additionally, capital-to-output ratios and output are positively related, in a sample of 48 countries. Hence, there is evidence of capital deepening both in the US time-series data and in the cross-country data. While the above facts are stylized, to be sure, it will be noted that empirical researchers have used increasingly sophisticated methods to tease out the relationship between …nancial intermediation and growth. This literature is surveyed masterfully by Levine (2005). The upshot is that …nancial development has a causal e¤ect on economic development; speci…cally, …nancial development leads to higher rates of growth in income and productivity. A general equilibrium model of …rm …nance, with competitive intermediation, is presented to address the impact that …nancial intermediation has on economic development. At the heart of the framework developed here is a costly-state veri…cation paradigm that has its roots in classic work by Robert M. Townsend (1979) and Stephen D. Williamson (1986). The model here has two novel ingredients, though. First, in the standard costly-state veri…- 3 2.4 0.14 0.12 1.6 Interest-rate spread k/o (capital to GDP) 2.0 1.2 0.8 0.4 0.10 0.08 0.06 0.04 0.02 0.00 0.0 0 5000 10000 15000 20000 25000 30000 35000 0 5000 10000 15000 20000 25000 30000 GDP per capita GDP per capita Figure 2: The cross-country relationship between per-capita GDP, on the one hand, and interest-rate spreads and capital-output ratios on the other cation paradigm, the realized return on a …rm’s investment activity is private information. This return can be monitored, but the outcome of this auditing process is deterministic: once monitoring takes place, the true state of the world is revealed with certainty. This is true whether or not a deterministic decision rule for monitoring is employed, as in Townsend (1979) or Williamson (1986, 1987), or a stochastic one is used, as in Ben Bernanke and Mark Gertler (1989) or John H. Boyd and Bruce D. Smith (1994)— see also Townsend (1979, Section 4). By contrast, in the current setup the outcome of monitoring is random. Speci…cally, the probability of detecting malfeasance depends upon the amount of resources devoted to policing the returns on a project and the e¢ ciency of the monitoring technology. In the model a project’s funding and level of monitoring will be jointly determined. The information asymmetry between the …rm and the intermediary gives the …rm an opportunity to exploit its private knowledge about the realization of the investment return. In particular, it can extract rents from the intermediary, and hence savers. Second, a …rm’s production technology is subject to idiosyncratic randomness. This is true in the standard costly-state veri…cation framework as well. Here, though, there is a distribution across …rms over the distribution of these returns. In particular, some …rms 4 may have investment projects that o¤er low expected returns with little variance, while others may have projects that yield high expected returns with a large variance. Two key features in the analysis follow from these ingredients. First, the setup yields a …nance-based theory of the equilibrium size distribution for …rms. This theory derives from the facts that: (a) investment opportunities di¤er in the expected returns and levels of risk that they o¤er and (b) producing information about these returns is costly. A simple threshold rule for funding results. All …rms with an expected return at least as great as the cost of raising capital are funded. Funding is increasing in a …rm’s expected return, and is decreasing in its risk. Loan size is determinate for a given type of project because the costs of …nancing a project are increasing and convex in the level of the monitoring activity. Thus, high mean projects receive limited funding because the costs of monitoring will rise disproportionately with loan size. The riskier a project is, the bigger is the di¤erence between the returns in good and bad states. This increases the incentive for a …rm to lowball its earnings in good states. Hence, more diligent monitoring will be required, which increases its …nancing cost. In an abstract sense, one could think that the diminishing returns in information production modeled here provide a microfoundation for Robert E. Lucas’s (1978) span of control model. Second, the framework provides a link between the e¢ ciency of the …nancial system and the level of economic activity. Such a tie was developed earlier in the models of Valerie R. Bencivenga and Smith (1991), Greenwood and Boyan Jovanovic (1990), Levine (1991), and Albert Marcet and Ramon Marimon (1992). The analysis here provides a crystal clear delineation of the Goldsmith (1969) mechanism, however. It stresses the connection between the state of technological development in the …nancial sector, on the one hand, and capital accumulation, both along the extensive and intensive margins, on the other. If technological improvement in the …nancial sector occurs at a faster pace than in the rest of the economy, then …nancial intermediation becomes more e¢ cient. Loans are monitored more diligently and the rents earned by …rms shrink. Additionally, lending activity will change along both extensive and intensive margins. Projects with high (low) expected returns will now receive 5 more (less) funds. Those investments with the lowest expected returns will be cut. At high levels of e¢ ciency in the …nancial sector the economy approaches the …rst-best equilibrium achieved in a world without informational frictions. This reallocation e¤ect distinguishes the analysis from earlier research by Shankha Chakraborty and Amartya Lahiri (2007) and Aubhik Khan (2001) that also embed the standard costly-state veri…cation framework into a growth model. In these frameworks all …rms receive the same amount of capital. In order to assess the importance of …nancial intermediation for the level of economic activity, the developed model is solved numerically. The model is calibrated to match some stylized facts for the U.S. economy. It is found that the model can replicate the increase in the U.S. capital/output ratios, shown in Figure 1. It does a very good job matching the …rm-size distribution for the U.S. economy for the year 1972. The improvement in …nancial sector productivity required to generate the observed rise in the U.S. capital/output ratio also appears to be reasonable. In the baseline model, improvements in …nancial intermediation account for 1/3 of U.S. growth. The calibrated model is then taken to the cross-country data. Others have tried to quantify the importance of factors such as limited investor protection [Castro et al. (2009)] for economic development. The emphasis here is on taking a micro-foundation …nance model to the data in order to quantify the importance of information production for economic development. Townsend and Ueda (2006) estimate a version of the Greenwood and Jovanovic (1991) model using Thai data. Their model emphasizes the role that …nancial intermediaries play in producing ex ante information about the state of the economy at the aggregate or sectorial levels. Financial intermediaries o¤er savers higher and safer returns. In the current analysis, intermediaries produce ex post information about …rm-level returns. The return earned by savers is …xed (and therefore safe) in the analysis. The gain to the economy from improved intermediation derives from squeezing, via information production, the rents appropriated by …rms on capital investment. This leads to a reduction in borrowing rates and more e¢ cient capital allocation. The model’s imputed measures for the e¢ ciency of …nancial intermediation in var- 6 ious countries match up very well with independent measures. It also does a reasonable job predicting the di¤erences in cross-country interest-rate spreads or capital/output ratios shown in Figure 2. Financial intermediation turns out to be important quantitatively. For example, in the baseline model Sri Lanka would increase its GDP by 70 to 76 percent if it could somehow adopt Switzerland’s …nancial system. World output would rise by 28 to 48 percent if all countries adopted Switzerland’s …nancial practice. Still, the bulk (or 81 to 87 percent) of cross-country variation in GDP cannot be accounted for by variation in …nancial systems. 2 The Environment Imagine a world resting in a steady state that is made up of three types of agents: consumer/workers, …rms and …nancial intermediaries. In a nutshell, …rms produce output using capital and labor. The consumer/worker supplies the labor, and intermediaries, the capital. All funding for capital must be raised outside of the …rm. Financial intermediaries raise the funds for capital from consumer/workers. They also use labor in their lending activity. Output is used for consumption by consumer/workers and for investment in capital by intermediaries. The behavior of …rms and intermediaries will now be described in more detail. The consumer/worker plays a more passive role in the analysis, which is relegated into the background by assuming that he supplies one unit of labor and saves at some …xed interest rate, rb.2 3 Firms Firms produce output, o, in line with the production function o = k l1 2 ; Think about a representative consumer with time separable preferences over consumption. The steady state interest rate will then be given by rb = 1= 1, where is his discount factor. 7 where k and l represent the inputs of capital and labor used in production. The variable gives the productivity level of the …rm’s production process. Productivity is a random variable drawn from a two-point vector and Pr( = 1 2( 1 2) 2 2) , = 2 1. =1 respectively.3 ( 1; with 2) 1 The mean and variance of < 2. Let Pr( = are given by 1 1 + 1) = 2 2 1 and Thus, for a given set of probabilities these statistics di¤er in accordance with the values speci…ed for and 1 2. The realized value of is the …rm’s private information. Now, the productivity vector, , di¤ers across …rms. In particular, suppose that …rms in the economy are distributed over productivities in line with the distribution function F : T ! [0; 1], where T 2 R+ and F (x; y) = Pr( 1 x; 2 y). Think of this distribution as somehow specifying a trade-o¤ between the mean and variance of project returns. Due to technological progress in the production sector of the economy, this distribution will evolve over time. Figure 3 plots the density function for F in mean/variance space that is used in the quantitative analysis. The …rm borrows capital, k, from the intermediary before it observes the technology shock, . It does this with both parties knowing its type, . It can employ labor, at the wage rate w, after it sees the realization for . In order to …nance its use of capital the …rm must enter into a contract with a …nancial intermediary. Last, note that a …rm’s production is governed by constant returns to scale. In the absence of …nancial market frictions no rents would be earned on production. Additionally, in a frictionless world only …rms o¤ering the highest expected return would be funded. With …nancial market frictions, deserving projects are underfunded, while undeserving projects are simultaneously over funded. 3 Observe that 1 2 = (1 1 ) 1 1 + (1 2 1 + (1 2 1) 1 2 2 1) 2 2 [ 1 1 + (1 1 2 1 (1 1) 2 1) 2] = (1 8 = 2 1 1 1) 1[ 2 2 1 1 2 2 1+ 2 2 1 2 1+2 2 1 2 ] = (1 2 2 1 2 1+ 2 1 1) 1( 1 2 2 2 1 2 2) . density 0.0006 0.0004 0.0002 0.0000 5 mean, θ 10 15 20 2 4 6 8 10 12 standard deviation, θ Figure 3: The distribution for F in mean/variance space that is used in the baseline quantitative model 3.1 Pro…t Maximization by Firms Consider the problem faced by a …rm that receives a loan in terms of capital in the amount k. The …rm hires labor after it sees the realization of its technology shock, . It will do this in a manner so as to maximize its pro…ts. In other words, the …rm will solve the maximization problem shown below. R( ; w)k maxf k l1 (P1) wlg: l The …rst-order condition associated with this maximization is (1 ) k l = w; which gives l= (1 ) 1= (1) k: w Substituting the solution for l into the maximand and solving yields the unit return function, R( ; w), or r = R( ; w) = (1 )(1 9 )= w (1 )= 1= > 0: (2) Think about ri = R( i ; w) as giving the gross rate of return on a unit of capital invested in the …rm given that state while the variance reads 4 i occurs. The expected gross rate of return will be 1 2 (r1 1 r1 + 2 r2 , r2 )2 . Financial Intermediaries There is a competitive intermediation sector that borrows funds from consumers and lends capital to …rms. While the intermediary knows a …rm’s type it cannot observe the state of a …rm’s business either costlessly or perfectly.4 That is, the intermediary cannot costlessly observe , o and l. The …rm will make a report to the intermediary about its business situation. The intermediary can devote some resources in order to assess the veracity of this report. The payments, p, from a …rm to the intermediary will be conditioned both upon the report made by former, and the outcome of any monitoring activity done by latter. By channelling funds through …nancial intermediaries consumers avoid a costly duplication of monitoring e¤ort that would occur in an equilibrium with direct lending between them and …rms— see Diamond (1984) and Williamson (1986) for more detail. Likewise, in the environment under study, it is optimal for a …rm to borrow from only one intermediary at a time. Suppose a …rm reports that the productivity on its project in a given period is may di¤er from the true state i. j, which The intermediary can devote resources, mj , to verify this claim. The probability of detecting fraud is increasing in the amount of resources devoted to this activity. In particular, let Pij (mj =k) denote the probability that the …rm is caught cheating conditional on the following: (1) the true realization of productivity is …rm makes a report of j; i; (2) the (3) the intermediary spends mj in monitoring; (4) the total 4 Recall that the intermediary knows the …rm’s type, . One could think about this as representing the activity, industry or sector that a …rm operates within. For instance, Castro et al (forthcoming, Figure 3) present data suggesting that the capital goods sector is riskier than the consumption goods one. It would be possible to have a screening stage where the intermediary veri…es the initial type of a …rm. The easiest way to do this would be to have them pay a …xed cost to discover . If the …rm’s type can’t be undercovered perfectly, as in the classic work of Boyd and Prescott (1986), then it may be possible to design the contract to reveal it. 10 amount of borrowing is k (which represents the size of the project). The function Pij (mj =k) is assumed to be monotonically increasing in mj =k. Additionally, let Pij (mj =k) = 0 if the …rm truthfully reports that its type is i. Any lender to the …rm must monitor the whole project to detect cheating, because his claim to pro…ts will depend on the total level of receipts vis à vis the total amount of disbursements paid out to others. Borrowing through a single intermediary then avoids a costly duplication of monitoring e¤ort. A convenient formulation for Pij (mj =k) is 8 > > 1 ( mj1=k) < 1; with 0 < < 1; > > > > < for a report 6= and m =k > 1= ; j i j Pij (mj =k) = > > 0; > > > > : for a report j = i or mj =k 1= : To guarantee that Pij (mj =k) 0, this speci…cation requires that some threshold level of monitoring, mj > k= , must be exceeded to detect cheating. Note that this threshold level of monitoring can be made arbitrarily small by picking a large enough value for ".5 Also, an arbitrarily large value for " can be chosen so that the threshold level of monitoring is very small. Figure 4 makes this clear, while illustrating the function Pij (mj =k). Monitoring is a produced good, measured in units of consumption. The production of monitoring is project speci…c. Monitoring produced for detecting fraud in one project cannot be used in a di¤erent one. Let monitoring be produced in line with the production function 1= m = zlm ; with 0 1= 1, where lm represents the amount of labor employed in this activity. The cost function, C(m=z; w), associated with monitoring is given by C(m=z; w) = w(m=z) : Costs are linear in wages, w. With diminishing returns to scale in production (1= < 1), the cost function is increasing and convex in the amount of monitoring, m, and decreasing and 5 The choice of " can be thought of as normalization relative to the level of productivity in the production of monitoring services–see footnote 7. 11 1.0 0.8 0.8 (m/z)γ, monitoring cost in labor Pij(mj/k), probability of detection 1.0 0.6 0.4 0.2 0.0 0.6 for z = 1.0 0.4 for z = 1.25 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 mj/k 0.4 0.6 0.8 1.0 m/z Figure 4: The functions Pij (mj =k) and C(m=z; w)=w, for " = 100, (the values used in the quantitative analysis) = 0:52, and = 1:86 convex in the state of the monitoring production technology, z. Figure 4 portrays the cost of monitoring in terms of labor, or plots C(m=z; w)=w. Now, exactly which …rms are funded depends on three things: (1) the …rm’s type, ; (2) the state of the monitoring production technology in the …nancial intermediation sector , z; (3) the expense of monitoring e¤ort as re‡ected by the wage, w. As will be seen, when the variance of a …rm’s project becomes larger, the informational problems associated with contracting become more severe. Therefore, high variance projects are less likely to get funded, ceteris paribus. 5 The Financial Contract A contract between a …rm and an intermediary is summarized by the quadruple fk; pj ; pij ; mj g. Here k represents the amount of capital lent by the intermediary to the …rm, pj is the …rm’s payment to the intermediary if it reports bank if the borrower reports j j and is not found cheating, pij is payment to the and monitoring reveals that productivity is 12 i 6= j, and mj is the intermediary’s monitoring e¤ort when j is reported. Denote the value of the …rm’s outside option by v. The intermediary chooses the details of the …nancial contract, fk; pj ; pij ; mj g, to maximize its pro…ts. The contract is designed to have two features: (1) it entices truthful reporting by …rms; (2) it o¤ers …rms an expected return of v. The optimization problem is I( ; v) max p1 ;p2 ;p12 ;p21 ;m1 ;m2 ;k f 1 p1 + rek 2 p2 subject to 1 w(m1 =z) 2 w(m2 =z) g; (P2) p1 r1 k; (3) p2 r2 k; (4) p12 r1 k; (5) p21 r2 k; (6) [1 P12 (m2 =k)](r1 k p2 ) + P12 (m2 =k)(r1 k p12 ) r1 k p1 ; (7) [1 P21 (m1 =k)](r2 k p1 ) + P21 (m1 =k)(r2 k p21 ) r2 k p2 ; (8) and 1 (r1 k p1 ) + 2 (r2 k p2 ) = v: (9) Note that the cost of capital, re, is given by re = rb + ; i.e., the interest paid to investors plus the depreciation on capital. The …rst four constraints just say the intermediary cannot demand more than the …rm earns; that is, the …rm has limited liability. Equations (7) and (8) are the incentive-compatibility constraints. Take (8). This simply states that the expected return to the …rm from reporting state one when it actually is in state two, as given by the left-hand side, must be less than telling the truth, as represented by the right-hand side. Observe that the constraint set is not convex due to the way that m1 enters (8). Therefore, the second-order conditions for the maximization problem are important to consider. The last constraint (9) speci…es that the contract must o¤er the …rm an expected return equal to v, its option value outside. A …rm’s outside option is the expected return that it could earn 13 on a loan from another intermediary. This will be determined in equilibrium. Finally, note the solution for fpj ; pij ; mj ; kg is contingent upon the …rm’s type, = ( 1; 2 ). To conserve on notation, this dependence is generally suppressed. The lemma below characterizes the solution to the above optimization problem. Lemma 1 (Terms of the Contract) The solution to problem (P2) is described by: 1. The size of the loan from the intermediary to the …rm, k, is k= v 2 (r2 r1 )[1 P21 (m1 =k)] (10) : 2. The amount of monitoring per unit of capital when the …rm reports a bad state, m1 =k, solves the problem I( ; v) maxf( 1 r1 + m1 =k 2 r2 where k can be eliminated using (10) above. re)k 1w z k ( m1 ) k vg; (P3) (a) Monitoring in the bad state is simply given by m1 = (m1 =k)k; where m1 =k solves (P3). (b) The intermediary does not monitor when the …rm reports a good state so that m2 = 0: (11) 3. The payment schedule is p1 = r1 k; p 2 = r2 k v= 2 ; (12) (13) p12 = r1 k; (14) p21 = r2 k: (15) Proof. See Appendix 14.1. It is intuitive that there are no bene…ts to the …rm from claiming a better outcome than it actually realizes, since it will only have to pay the intermediary more. The intermediary would like to reduce the …rm’s incentive to report being in the low state. So, suppose the 14 …rm reports a low state. If cheating is not detected, then the …rm pays all of its revenue (minus labor cost) that would be realized in the low state— see (12). If the …rm is caught cheating, then it must surrender all of the revenue (sans labor cost) that it earns in the high state— see (15). Note that due to the incentive-compatibility constraints a false report will never occur so that the payments shown by (14) and (15) do not occur in equilibrium. The contract speci…es that the intermediary should only monitor the …rm when it reports a bad outcome (state 1) on its project— see (11). Monitoring in the low state is done to maximize the intermediary’s pro…ts, subject to the incentive-compatibility cum promisekeeping constraint (10), as problem (P3) dictates. Note that the higher is the value of the …rm, v, the bigger must be the loan, k, to satisfy the incentive-compatibility cum promisekeeping constraint (10). This constraint (10) ensures that the contract provides the …rm an expected return equal to what it would earn if it misrepresented the outcome in the good state, 2 (r2 r1 )[1 P21 (m1 =k)]k. Furthermore, this expected return is set equal to the …rm’s outside option, v. The size of the loan, k, is increasing in the amount of monitoring that occurs in the low state, m1 =k. This happens because the probability of the …rm not getting caught from misrepresenting its revenues, 1 P21 (m1 =k), is decreasing in the intermediary’s monitoring activity. The theory of intermediation presented here is at an abstract level, as in Boyd and Prescott (1986). It is not intended to explain the presence of real world forms of intermediation, such as the existence of banks, bond markets, or stock markets. Levine (2005) notes that, empirically speaking, it is of secondary importance whether the source of the development in …nancial systems arises from improvements in banks, stock markets or bond markets. Now, a …nancial contract will be o¤ered by an intermediary to a …rm only if it yields the former nonnegative pro…ts, I( ; v) 0. Suppose that r1 < re. A necessary condition for a contract to yield nonnegative pro…ts is for the intermediary to devote more than the minimal level of resources per unit of funds lent, 1= , to monitoring a report of a bad state. If this is not done, the …rm will always claim that it is in the low state, and the intermediary 15 can only earn a loss on the contract. Brie‡y consider the case where r1 re. Here the …rm’s return on capital in its worst state of nature is at least as large as the cost of capital, re. Firms would desire to borrow an in…nite amount of capital. An equilibrium will not exist. Assumption: r1 = R( 1 ; w) < re = rb + , for all …rm types. Later, a lower bound on the level of productivity in the …nancial sector, , will be imposed that guarantees that this assumption holds whenever z > . This lower bound ensures that the equilibrium wage, w, is high enough so that the assumption will always hold–see (2). Lemma 2 (Interior solution for monitoring) m1 =k > 1= , for all v > 0. Proof. The argument is presented in Appendix 14.1. 6 Competitive Financial Intermediation In the economy there is perfect competition in the …nancial sector. Consequently, an intermediary must o¤er a contract that maximizes a …rm’s value, subject to the restriction that the former does not incur a loss. If an intermediary failed to do so it would be undercut by others. The upshot is that intermediaries will make zero pro…ts on each type of loan. Furthermore, for a …rm to produce, it must make non-negative pro…ts too. It is intuitive that for all this to happen a project must o¤er some potential for a surplus or that 1 r1 + 2 r2 re; i.e., the expected return on capital must exceed its cost. Assume this. The value of a …rm, v, is then determined by the condition v =V( ) arg maxfx : I( ; x) = 0g: x (16) The implications of perfect competition will now be analyzed. Key questions are: (ii) What will be the loan size? (ii) Which …rms will get funded? The size of a loan for a project, k, can now be determined. To do this, substitute (10) in Problem (P3) and solve for the optimal level of monitoring, m1 =k. Plug this solution for 16 monitoring in the objective function for (P3) to obtain a formula for I( ; v).6 Next, solve for v using condition (16). Plugging the obtained formulae for m1 =k and v into (10) yields k = ( + ) ( 1 r1 + 2 r2 =( ) re)( 1 ( ) =( + )=( ) ) ( 1 )( + ( 1 w) =( + )=( ) ) (17) ( z) ] r1 ) 2 (r2 [ =( ) : Equation (17) gives a determinate loan size for each type of funded project. Furthermore, funding is increasing in a project’s expected return and is decreasing in its volatility.7 Lemma 3 (Loan size) The level of investment in a …rm, k, is increasing in its expected net return, 1 r1 + 2 r2 re, and the state of technology in the …nancial sector, z, and is decreasing in the variance of the return, r2 r1 (holding the wage rate, w, …xed). Proof. See Appendix 14.1. Attention will now be directed toward determining which projects will be funded. Consider the set of …rms, A(w), de…ned by A(w) f : 1 r1 + 2 r2 re > 0g: (18) Intuitively, one might expect that this set of projects will be funded in equilibrium because they o¤er an expected return on capital, 1 r1 + 2 r2 , that is greater than its user cost, re. This turns out to be true. The contracting problem (P2) implies that the …rm will never make negative pro…ts, given (3) to (6). The construction of equation (17) suggests that the intermediary will be able to make a loan in this situation, and not incur a loss. 6 The solution obtained for I( ; v) is I( ; v) = ( 1 )( ) + ( 1 w) =( + =( + 1 )( + )=( + + ( z) ) [ ] =( + r1 ) 2 (r2 ) ) ( ) v ( =( 1 r1 + + 2 r2 ) v: re)( + )=( + ) This solution presumes that r1 < re, so there is an interior solution for monitoring, and that 1 r1 + 2 r2 re. It is easy to see that the intermediary’s pro…t function, I( ; v), is \ shaped with the following properties: (i) I( ; 0) = 0; (ii) limv!0 @I( ; v)=@v = 1; (iii) @ 2 I( ; v)=@v 2 < 0; (iv) limv!1 I( ; v) = 1. Therefore, there is only one v > 0 that solves (16). 7 Observe that loan size is a function of z. In this sense, the choice of the constant detection function can be thought of as normalization relative to z. 17 in the odds of Lemma 4 (The set of funded …rms) A necessary and su¢ cient condition for a type- …rm to be active or funded, or for k( ) > 0; I( ; V ( )) = 0 and V ( ) > 0, is that 2 A(w). Proof. Those interested should go to Appendix 14.1. Therefore, as was mentioned in the introduction, a simple threshold rule exists for funding, as characterized by (18). Call A(w) the set of active …rms. From (2) it is easy to see that 1 r1 + 2 r2 )(1 re = (1 )= w (1 )= [ 1= 1 1 + 1= 2 2 ] re 0: Observe that the …rm’s pro…ts are decreasing in wages. A type- …rm will operate when w W ( ), and will not otherwise, where the cuto¤ wage, W ( ), is speci…ed by W( ) =(1 ) (1 )( 1= 1 1 + re 1= 2 2 ) =(1 ) : (19) So, the set of active projects A(w) can be expressed equivalently as A(w) = f : w < W ( )g. (20) The active set depends on the wage because ri = R( i ; w). It contracts (expands) with a rise (decrease) in the real wage, since R( i ; w) is decreasing in w. In equilibrium the wage rate, w, turns out to be increasing function of the state of technology in the …nancial sector, z. Hence, the active set will shrink with technological improvement in the …nancial sector, or a rise in z. This will become clearer in Section 8. Figure 5 summarizes the discussion on funding. As the expected return on a project rises more capital is allocated to it, as is illustrated in the …rst panel. (Note that the direction of the x and y axes is speci…c to each panel.) The increase in funding (or the scale of the …rm) is associated with higher monitoring costs given the increasing, convex form of the cost function (fourth panel). The amount of monitoring done per unit of capital is then economized on (third panel). As a consequence, the odds of detecting fraud drop (second panel). In response the expected rents earned by the …rm rise (…fth panel). As the risk associated with a project rises its funding is slashed (…rst panel). When the di¤erence between the good and bad state widens there is more incentive for the …rm to 18 falsify its earning. The intermediary therefore monitors more per unit of capital lent (third panel). Total monitoring costs fall with risk, because the size of the loan is smaller (fourth panel). The probability of detecting malfeasance therefore moves up since monitoring per unit of capital is now higher (second panel). A …rm’s rents will fall with a rise in risk (…fth panel), because it receives a smaller loan and faces more vigilant policing. 7 Stationary Equilibrium The focus of the analysis is on stationary equilibria. First, the labor-market-clearing condition for the model will be presented. Second, a de…nition for a stationary equilibrium will be given. Third, it will be demonstrated that a stationary equilibrium for the model exists. On the demand side for labor, only …rms with 2 A(w) will be producing output. On the supply side, recall that the economy has one unit of labor in aggregate. The labormarket-clearing condition will then appear as Z [ 1 l1 ( 1 ; 2 ) + 2 l2 ( 1 ; 2 ) + 1 lm1 ( 1 ; 2 )]dF ( 1 ; 2 ) = 1. (21) A(w) It is now time to take stock of the situation so far by presenting a de…nition of the equilibrium under study. It will be assumed that the economy rests in a stationary state where the cost of capital is re = rb + . De…nition 1 Set the steady-state cost of capital at re. A stationary competitive equilibrium is described by a set of labor allocations, l and lm , a …nancial contract, fp1 ; p2 ; p12 ; p21 ; k; m1 ; m2 g, a set of active monitored …rms, A(w), …rm values v, and a wage rate, w, such that: 1. The …nancial intermediary o¤ers a contract, fp1 ; p2 ; p12 ; p21 ; k; m1 ; m2 g, which maximizes its pro…ts, I, in accordance with (P2), given the cost of capital and wages, re and w, and the value of …rms, v. The intermediary hires labor for monitoring in the amount lm = (m=z) . 2. The …nancial contract o¤ered by the intermediary maximizes the value of a …rm, v, in line with (16), given the prices re and w. 3. A …rm is o¤ered a contract if and only if it lies in the active set, A(w), as de…ned by (18), given re and w. It hires labor, l, so as to maximize its pro…ts in accordance with (P1), given, w, and the size of the loan, k, o¤ered by the intermediary. 19 (i) k P12(m1/k) (ii) 0.015 0.025 0.030 0.020 0.040 0.035 Mean, r 0.030 0.025 Mean, r 0.035 Std dev, r 0.020 0.040 0.015 Std dev, r (iv) m1/k w(m1/z)γ/k (iii) 0.015 0.030 0.025 0.035 Mean, r 0.040 0.015 0.020 0.040 Std dev, r Mean, r (v) 0.040 Mean, r 0.030 0.035 0.030 0.025 Std dev, r v v/k (vi) 0.015 0.015 0.020 0.035 0.020 0.040 0.025 Std dev, r 0.020 0.035 Mean, r 0.030 0.025 Std dev, r Figure 5: The determination of …rm size as a function of the mean and standard deviation of r = R( ; w) 20 4. The wage rate, w, is determined so that the labor market clears, in accordance with (21). When will an equilibrium exist for the economy under study? To address this question, let 1 maxf 1 : ( 1; 2) 2 T , over all 2 g. Next, de…ne the constant ! by the equation R( 1 ; !) = re: (22) The constant ! speci…es a lower bound on the feasible equilibrium wage rate.8 When w = ! for a type-( 1 ; 2) project it will happen that r1 = re. In this situation the intermediary could simply ask for a payment of r1 k in both states of the world and engage in no monitoring. It would earn zero pro…ts. The …rm’s pro…ts would be 2 (r2 r1 )k. It would desire a loan of in…nite size. Therefore, as the equilibrium wage, w, approaches ! from above the equilibrium de…ned above will eventually become tenuous. The situation is portrayed in Figure 6, which graphs the demand and supply for labor. The demand for labor is portrayed by the solid line labeled L. The properties of this demand schedule are established during the course of the proof for Lemma 5. Demand is downward sloping in w. The question is whether or not it will cross the vertical supply schedule for labor. Now, at a given wage rate, loan size increases with more e¢ cient intermediation. This leads to more labor being demanded when intermediation improves. In other words, it can be shown that the demand for labor schedule shifts rightward with an upward movement in z. Now, de…ne as the level of z such that the demand curve intersects the supply curve at the point (1; !). Lemma 5 (Existence of an equilibrium) There is a constant exists a stationary equilibrium for the economy. such that for all z > there Proof. See Appendix 14.1. 8 The function R( ; w) is continuous and strictly decreasing in w, with limw!0 R( ; w) = 1, and limw!1 R( ; w) = 0; hence, ! is well de…ned. 21 Wages wˆ≡ max W (τ ) τ L, for z > ζ w ω L, for z = ζ 1 Demand, Supply Figure 6: Existence, demand and supply for labor 22 8 The Impact of Technological Progress on the Economy The primary goal of the analysis is to understand how technological advance in the …nancial sector a¤ects the economy. To this end, the impact that technological progress, in either the …nancial or production sector, has on the portfolio of funded projects will be characterized. To develop some intuition for the economy under study, some special cases will be examined. 8.1 Balanced Growth In the …rst special case, technological progress in the …nancial sector proceeds in balance with the rest of the economy. Speci…cally, assume that the economy is moving along a balanced growth path where the Therefore, Ft+1 ( 1 ; 2) 1= i ’s grow at the common rate g 1= and z grows at rate g 1=(1 = Ft ( 1 =g; 2 =g). ) 9 . The salient features of this case are summarized by the proposition below. 1= Proposition 1 (Balanced growth) Let the i ’s grow at rate g 1= and z increases at rate g 1=(1 ) . There exists a balanced growth path where the capital stock, k, wages, w, and rents, v, will all grow at rate g 1=(1 ) . The amount of resources devoted to monitoring per unit of capital, m1 =k, remains constant. Proof. Refer to Appendix 14.1. In this situation the …nancial sector is not becoming more e¢ cient over time, relative to the rest of the economy. The amount of monitoring done per unit of capital invested remains constant over time. Thus, the probability of a …rm getting caught by misrepresenting a high level of earnings, P21 (m1 =k), is constant over time too. For any particular project type, the spread between the return on capital (net of labor costs) and the interest earned by investors, 1 r1 + 2 r2 re, is …xed over time. One could easily allow the production of monitoring services 1= to require capital as well labor, say as given by m = zkm lm with 0 < 1= ; ; 1= + Now, for a balanced growth path to obtain, z would have to grow at rate g 9 =(1 ) < 1. . The In order to get a …xed interest rate assume that the consumer/worker has isoelastic preferences over consumption. Then, in standard fashion along a balanced growth path, rb = g =(1 ) = 1, where is the coe¢ cient of relative risk aversion. 23 existence of a balanced growth path results from the fact that the probability of detection, P21 (m1 =k), depends on the employment of monitoring services relative to the size of the loan. If loan size did not enter this function, then technological advance in the non-…nancial sector would lead to a drop in the cost of monitoring as the economy’s capital stock rose. That is, there would be a positive feedback loop from the state of development in the non-…nancial sector to the state of development in the …nancial one, as is modeled in three di¤erent ways in de la Fuente and Marin (1996), Greenwood and Jovanovic (1990), and Harrison, Sussman and Zeira (1999). 8.2 Unbalanced Growth The above case suggests that for technological progress in the …nancial sector to have an impact it must outpace advance in the rest of the economy. Suppose that this is the case. Then, one would expect that as monitoring becomes more e¢ cient those projects o¤ering the lowest expected return will be cut. Proposition 2 (Technological progress in …nancial intermediation) Consider z and z 0 with z < z 0 . Let w and w0 be the wage rates associated with z and z 0 , respectively. Then, A(w0 ) A(w). Additionally, if = ( 1 ; 2 ) 2 A(w) A(w0 ) and 0 = ( 01 ; 02 ) 2 A(w0 ) then 1= + 2 ( 2 )1= < 1 ( 01 )1= + 2 ( 02 )1= . 1( 1) Proof. See Appendix 14.1. An increase in z makes …nancial intermediation more e¢ cient. For any given wage rate, w, the aggregate demand for labor will increase for two reasons. First, more capital will be lent to each funded project. Second, more labor will also be hired by the intermediary to monitor the project. Since the demand for labor rises, the wage rate must move up to clear the labor market. This increase in wages causes the set of active projects, A(w), to shrink, with the projects o¤ering the lowest expected return being culled. Alternatively, technological advance could occur in the production sector and not the …nancial one. Here, the lack of development in the …nancial sector will hinder growth in the rest of the economy. Speci…cally, technological advance in the production sector of the economy will drive up wages. This leads to the costs of monitoring rising. Therefore, less 24 is done. This lack of scrutiny by intermediaries now allows …rms with marginal projects o¤ering low expected returns to receive funding. 1= Proposition 3 (Technological progress in production) Suppose all the i ’s increase by the factor g 1= , holding z …xed. Then, the set of active projects, A(w), expands with the new projects o¤ering lower expected returns than the old ones. Proof. See Appendix 14.1. 8.3 E¢ cient Finance An extreme example of Proposition 2 would be to assume that z grows forever. Then, the …nancial sector will become in…nitely e¢ cient relative to the rest of the economy. This leads to the fourth special case. 2 , with Proposition 4 (E¢ cient …nance) Suppose T is a compact and countable subset of R+ a positive measure of projects for each type, = ( 1 ; 2 ). Then, 1. lim A(w) = A 1= 1 (x 1 ) arg max[ z!1 =( 2. lim m1 =z = 0; for z!1 1= 2 (x 2 ) + 2A , 3. lim m1 =k = 1 and lim P12 (m1 =k) = 1, for z!1 z!1 4. lim p2 = r2 k; for z!1 5. lim v = 0; for ]; 1 ; 2 )2T 2A , 2A , 2A , z!1 6. lim w = w =(1 z!1 ) (1 )f max[ =( 1= 1( 1) + 1= 2( 2) 1 ; 2 )2T ]=e rg =(1 ) =(1 ) ; (23) 7. lim z!1 Z A(w) kdF = k ( )1=(1 re ) f max[ =( 25 1 ; 2 )2T 1= 1( 1) + 1= 2( 2) ]g : (24) Proof. Refer to Appendix 14.1. As the cost of monitoring borrowers drops, the intermediation sector becomes increasingly e¢ cient. The …nancial intermediary can then perfectly police loan payments without devoting a signi…cant amount of resources in terms of labor to this activity, as points (2) and (3) in the proposition make clear. Since …rms are operating constant-returns-to-scale production technologies, no rents will accrue on their activity— see point (5). Firms must pay the full marginal product of capital to the intermediary— point (4). That is, the spread between a …rm’s internal rate of return (before depreciation), 1 r1 + 2 r2 , and the user cost of capital, re = rb + , vanishes, where the latter is made of the interest paid to savors, rb, and rate of depreciation, . In this world only projects with the highest return are …nanced, as point (1) states, even though they may be the most risky. In the aggregate any idiosyncratic project risk washes out. Therefore, in the absence of a contracting problem, only the mean return on investment matters. And, with constant-returns-to-scale technologies everything should be directed to the most pro…table opportunity. The wage rate, w , and aggregate capital stock, k , in the e¢ cient economy are determined in standard fashion by the conditions that the marginal product of capital for the most pro…table projects must equal the user cost of capital, re, and the fact that the labor market must clear. These two conditions yield (24) and (23). (By comparison, consider the standard deterministic growth model with the production technology o = k l1 w =(1 ) (1 )[ 1= =e r] =(1 ) and k and one unit of aggregate labor. Here [ =e r]1=(1 ) ( 1= ) =(1 ) . The di¤erences in the formulae are due to two facts that pertain to the current setting: (i) the best projects from a portfolio T are chosen; (ii) there is uncertainty in .) 9 9.1 Fitting the Model to the U.S. Economy Procedure The quantitative analysis will now begin. To start with, assume that …rms now produce output in line with the production function 26 o = x k l1 ; where x represents a known level of aggregate productivity. To simulate the model, values must be assigned to its parameters. This will be done by calibrating the framework to match some stylized facts for the U.S. economy. Some parameters are standard. They are given conventional values. Capital’s share of income, , is chosen to be 0:33, a very standard number. Likewise, the depreciation rate, , is set to 0:07, again a very common number. The chosen value for the discount factor, = 1=(1 + rb), implies that the interest rate earned by savers is 6:3 percent. This is in conformity with Cooley and Prescott’s (1995, p. 19) estimate of 6:9 percent for the real return to capital over the postwar period. The concept used for the capital stock is much narrower here, though; i.e., it is just the stock of business capital. Therefore, matching the low observed capital-output ratio will be harder. Nothing is known about the appropriate choice for parameters governing the intermediary’s monitoring technology, or ; ; and . The selection of a value for a normalization (relative to some baseline level of z). Therefore, set little is known about the distribution of returns facing …rms. Let R …rms of expected total factor productivity (TFP); i.e., m = ( wise, v = R v m amounts to = 100. Similarly, be the mean across 1 1 + 2 2 )dF . Like- will denote the mean over …rms of the logarithm of the volatility of TFP; i.e., ln[ 1 2( 2 2 1 ) ]dF . In a similar vein, will represent the correlation between the means and (ln) volatilities of …rm-level TFP, while 2 m and 2 v will denote the variance of these …rm-level variables. Assume that these means and (ln) volatilities of …rm-level TFP are distributed according to a bivariate truncated normal, N ( need to be selected for the parameters ; ; m; v; 2 m; 2 v; m; v; 2 m; 2 v; ). Thus, values and . Let Targetsj represent the j-th component of a n-vector of observations that the model should match. Similarly, Output(param) denotes the model’s prediction for this vector. The model’s solution will be a function of the list of calibrated parameters, param. These parameters are picked to minimize a weighted sum of the squared deviations between the 27 data targets and the model’s output: min param n X weightsj [Targetsj -Outputj (param)]2 ; (25) j=1 where weightsj represents the weight attached to j-th target. The data targets will now be discussed. The distribution of returns across …rms will be integrally related to the distribution of employment across them. Firms with high returns will have high employment, other things equal. Therefore, the size distribution of …rms for the year 1972 is chosen as a data target. Eight points on this distribution are picked. As was mentioned in the introduction, to the extent that …nancial intermediation has become more e¢ cient over time, relative to the non…nancial sector, one would expect that the capital/output ratio would rise. If a nation’s productivities in the …nancial and non-…nancial sectors increase in a balanced way then just output will increase. Denote an aggregate quantity in bold. The model provides a mapping between the aggregate level of output (per person), o, and the capital/output ratio, k=o, on the one hand, and the state of technology in its production and …nancial sectors, x and z, on the other. Represent this mapping by (o; k=o) = M (x; z)–the dependence of this mapping on the model’s parameters has been suppressed for notational convenience. Now, while the states of the U.S.’s …nancial and non-…nancial technologies are unobservable directly, this mapping can be used to make an inference about (x; z), given an observation on (o; k=o), by using the relationship (x; z) = M 1 (26) (o; k=o): Therefore, TFP’s in the …nancial and non-…nancial sectors are picked to match GDP and the capital/output ratio for the years 1972 and 2000. This determines x1972 , z1972 ; x2000 , and z2000 , as functions of the other estimated parameter values and, of course, the target values for o and k=o. The calibrated parameter vector is param (; ; ; m; v; 2 m; 2 v; ; x1972 ; z1972 ; x2000 ; z2000 ). Thus, the calibration procedure is picking 12 parameters to match 12 observations, as best as can be done subject to the restriction (26). Table 1 presents the parameter values for the model. 28 Payroll, Cumulative Share 1.0 0.8 0.6 0.4 Data, 1972 Model, 1972 0.2 0.0 0.6 0.7 0.8 0.9 1.0 Establishments, Percentile Figure 7: Firm-size distribution, 1972 9.2 Results The model’s prediction for the 1972 …rm-size distribution is shown in Figure 7. It …ts remarkably well. The …rm-size distribution shifts somewhat between 1972 and 2000. Now, monitoring and the provision of …nancial services are abstract goods, so it hard to know what a reasonable change in z should be. One could think about measuring productivity in the …nancial sector, as is often done, by k=lm , where k is the aggregate amount of credit extended by …nancial sector and lm is the aggregate labor that it employs. By this traditional measure, productivity in the …nancial sector rose by 3.8 percent between 1973 and 2000. Berger (2003, Table 5) estimates that productivity in the commercial banking sector increased by 2.2 percent a year over this same period (which includes the troublesome productivity slowdown) and by 3.2 percent from 1982 to 2000. The model’s ability to match cross-country di¤erences in the e¢ ciency of …nancial intermediation and di¤erences in the interest-rate spreads between borrowers and lenders will be addressed in Section 10. 29 Table 1: Parameter Values Parameter De…nition Basis = 0:33 Capital’s share of income Standard value = 0:07 Depreciation rate Standard value = 1=(1 + rb) = 0:94 Discount factor Cooley and Prescott (1995) = 100 Pr of detection, constant Normalization = 0:52 Pr of detection, exponent Calibrated to …t targets = 1:86 Monitoring cost function Calibrated to …t targets means Calibrated to …t targets variances and correlation Calibrated to …t targets TFP’s Calibrated to …t targets m = 1:54; v = 0:80 2 m = 0:49; 2 v = 0:66; = 0:88 x1972 = 60:364; z1972 = 1:083e5; x2000 = 88:549; z2000 = 6:785e5 Table 2 presents results for some other variables of interest. Here, the aggregate value R for a variable is again indicated in bold, so that x = A(w) xdF for x = m1 , w 1 (m1 =z) , k, R etc. The expected value for x is given by x dF . Monitoring becomes less expensive A(w) as z rises (relative to x). This results in the amount of monitoring per unit of capital rising, as re‡ected in the larger values for m1 k for 2000 versus 1972. As a consequence, the likelihood of intermediaries detecting fraud increases. The fraction of a …rm’s output dissipated in pure rents, v o, declines. This can be seen another way. The internal rate of return, i, earned by a …rm on its investment is given by i = 1 r1 + 2 r2 . The average internal return earned by …rms, weighted by their level of investment, will then be de…ned R R by i kidF [k dF ]. Likewise, denote the average rate of return earned by A(w) A(w) R R the intermediary on its lending activity by ei ( 1 p1 + 2 p2 k)dF [k A(w) dF ]. A(w) The gap between these two returns, e i ei, measures the average excess return earned by …rms due to rents. This excess return is squeezed as rents shrink. In similar fashion, the average spread between the rates of return that intermediaries and savers earn, s ei rb, re‡ects the costs of intermediation incurred by the necessity to monitor borrowers. This interest rate spread declines as the costs of intermediation fall due to technological progress 30 in information production. Rousseau (1998, Figure 4) presents evidence suggesting that …nancial innovation reduced loan-deposit spreads in the U.S. between 1872 to 1929. Li and Sarte (2003, Table 3), using a structural VAR, present evidence suggesting that drops in the cost of …nancial intermediation account for a signi…cant part of long-run ‡uctuations in U.S. manufacturing output. A rise in the probability of detecting fraud relaxes the incentive constraint (8), and makes it easier to lend more capital to …rms. This has two e¤ects. First, as borrowing rates decline renting more capital becomes pro…table. This results in a higher aggregate amount of capital being invested per unit of output produced, k o. Second, for a given amount of lending, funds are redirected toward those …rms o¤ering the highest rate of return. Along with the …rst e¤ect, this increases GDP, o. Denote the levels of capital and output that would obtain in the …rst-best economy by k and o . As can be seen, capital and output steadily rise, relative to their …rst-best outcome, as z moves up (relative to x). Model TFP is quite volatile across plants, as can be seen from Table 2. Hsieh and Klenow (2008, Tables 1 and 2) report (weighted) standard deviations of 0.45 and 0.85 for 1977. The (weighted) number found here lies in the middle of their range. For the U.S., they show little or no increase in this number over time. This is similar to what is found here. Hence, the volatility in plant-level TFP required to match the 1972 U.S. …rm size-distribution appears to be reasonable. Finance is important in the model. This can be gauged by undertaking the following counterfactual question: By how much would GDP have risen between 1972 and 2000 if there had been no technological progress in the …nancial sector? As can be seen from the third column of Table 2, output would have risen from $22,097 to $34,590 or by about 1.6 percent a year (when continuously compounded). This compares with the increase of 2.4 percent ($22,097 to $43,268) that occurs when x rises to its 2000 level. Thus, about one third of the increase in growth is due to innovation in the …nancial sector. Likewise, the model predicts that about 15 percent of TFP growth was due to improvement in …nancial intermediation. The …nancial system actually becomes a drag on development when z is not 31 allowed to increase. Wages rise as the rest of the economy develops. This makes monitoring more expensive. Therefore, less will be done. As a consequence, interest rates rise and the economy’s capital/output ratio drops. Without an improvement in the …nancial system, the …rm-size distribution actually moves slightly in the wrong direction. Table 2: Impact of Technological Progress in the Financial Sector 1972 2000 Counterfactual Production Sector, x 60.364 88.549 88.549 Financial Sector, z 1.083e5 6.785e5 1.083e5 0.1018 0.1738 0.0809 0.9651 0.9734 0.9608 1.029e6 3.004e6 1.431e6 0.1010 0.0768 0.1132 530.43 807.30 760.86 0.6953 0.6763 0.7131 0.1443 0.1174 0.1611 0.0787 0.0737 0.0818 0.0656 0.0436 0.0783 0.0149 0.0099 0.0180 1 0.0638 0.0638 0.0638 o 1.5400 1.7613 1.4279 Monitoring-to-capital, m1 k R Pr of detecting fraud, A(w) P12 dF Financial sector productivity, k lm R A(w) dF Rents to output, v o R R TFP x A(w) ( 1 1 + 2 2 )dF dF A(w) R p Std ln(TFP), 1 2 A(w) k( ln 2 ln 1 )dF Internal return (weighted), i Lending rate, ei Excess return, e = i ei Interest rate spread, s = ei Return to savers, rb = 1= Capital-to-output ratio, k rb R A(w) dF Capital relative to …rst best, k k 0.4452 0.5628 0.3903 Output relative to …rst best, o o 0.7128 0.7879 0.6740 $22,097 $43,268 $34,590 GDP, o 32 10 Fitting the Model to the World Economy Ever since Goldsmith (1969), economists have been interested in the cross-country relationship between …nancial structure and economic development. An implication of the current model is that as the state of technology in the intermediation sector advances, the spread between borrowing and lending rates in an economy will shrink, while its capital-to-output ratio and level of aggregate output increases. The cross-country data is suggestive of such a relationship, as Figure 2 shows. For the cross-country analysis assume that production in a nation is undertaken in the manner described earlier, but let x and z now represent country-speci…c productivity factors for the non-…nancial and …nancial sectors. While the state of a country’s productivities in the non-…nancial and …nancial sectors is unobservable directly, once again the mapping (26) can be used to make an inference about (x; z), given an observation on (o; k=o). This is done for a sample of 40 countries, using the parameter values listed in Table 1. This implies that the distribution of potential projects di¤ers across countries by the factor of proportionality, x, an assumption needed both for discipline and tractability. The results are reported in Table 7 in Appendix 14.2. By construction the model explains all the variation in output and capital/output ratios across countries.10 Still, one could ask how well the measure of the state of technology in the …nancial sector that is backed out using the model correlates with independent measures of …nancial intermediation. Here, take the ratio of private credit by deposit banks and other …nancial institutions to GDP as a measure of …nancial intermediation, as reported by Beck et al. (2001). (Other measures produce similar results but reduce the sample size too much.) Additionally, one could examine how well the model explains cross-country di¤erences in interest-rate spreads, s. 10 The model predicts a positive association between a country’s rate of investment and its GDP. Castro et al. (2009, Figure 1) show that this is true. It is stronger when investment spending is measured at international prices, as opposed to domestic ones. They resolve this di¤erence within the context of a twosector model where the relative price of capital goods is endogenously determined. In their framework, capital goods are more expensive to produce in poor countries. This happens because this sector is risky, implying that the costs of …nance are high in countries with poor investor protection. Thus, this puzzle could be resolved here by adopting aspects of their two-sector analysis. 33 Table 3 reports the …ndings. The correlation between the imputed state of technology in the …nancial sector and the independent measure of …nancial intermediation is quite high. Thus, it appears reasonable to use the constructed values of z for investigating the relationship between output and …nancial development. Interestingly, Finland and Peru both have a capital-to-output ratio of about 1.6. The model predicts Finland’s z is about 160 percent (continuously compounded) higher than Peru’s–the former’s ln(z) is 12.99, compared with 11.40 for the latter; again, see Table 7 in Appendix 14.2. But, recall that the units for ln(z) are meaningless, since monitoring is abstract good. If one measures productivity in the …nancial sector by the amount of credit extended relative to the amount of labor employed in the …nancial sector, as was done earlier, then the analysis suggests that intermediation in Finland is about 145 percent (continuously compounded) more e¢ cient than in Peru. Why? Finland has a much higher level of income per worker and hence TFP than does Peru ($40,603 versus $10,200). Therefore, given the higher wages, monitoring will be more expensive in Finland. To give the same capital/output ratio, e¢ ciency in Finland’s …nancial sector must be higher. As can be seen, the interest-rate spreads predicted by the model are positively associated with those in the data. The correlation is reasonably large. That these two correlations aren’t perfect, should be expected. There are other factors, such as the big di¤erences in public policies discussed in Parente and Prescott (2000), which may explain a large part of the cross-country di¤erences in capital/output ratios. Di¤erences in monetary policies across nations may in‡uence cross-country interest rate spreads. Additionally, there is noise in these numbers given the manner of their construction— see Appendix 14.2. Table 3: Cross-Country Evidence interest-rate spread …nancial intermediation Correlation(model, data) 0.37 34 0.72 11 The Importance of Financial Development for Economic Development It is now possible to gauge how important e¢ ciency in the …nancial sector is for economic development, at least in the model. To this end, note that the best …nancial and industrial practices in the world are given by x = maxfxi g and z = maxfzi g, respectively. Represent country i’s output, as a function of the e¢ ciency in its industrial and …nancial sectors, by oi = O(xi ; zi )–this is really just the …rst component of the mapping M (x; z). If country i could somehow adopt the best …nancial practice in the world it would produce O(xi ; z). Similarly, if country i used the best practice in both sectors it would attain the output level O(x; z). The shortfall in output from the inability to attain best practice is O(x; z) O(xi ; zi ). Luxembourg turns out to have the highest value for x, and Switzerland for z. The percentage gain in output for country i by moving to best …nancial practice is given by 100 [ln O(xi ; z) ln O(xi ; zi )]. The results for this experiment are plotted in Figure 8. As can be seen, the gains are quite sizeable. On average a country could increase its GDP by 31 percent, and TFP by 10 percent. The country with the worst …nancial system, Sri Lanka, would experience a 76 percent rise in output. Its TFP would increase by 26 percent. While sizeable, these gains in GDP are small relative to the increase that is needed to move a country onto the frontier for income, O(x; z). The percentage of the gap that is closed by a movement to best …nancial practice is measured by 100 O(xi ; zi )] 100 the sample.11 [O(xi ; z) O(xi ; zi )]=[O(x; z) G(xi ; zi ). Figure 9 plots the reduction in this gap for the countries in The average reduction is this gap is only 17 percent. For most countries the shortfall in output is accounted for by a low level of total factor productivity in the non-…nancial sector. Therefore, the importance of …nancial intermediation for economic development depends on how you look at it. World output would rise by 28 percent by moving all countries to the best …nancial practice— see Table 4. This is a sizeable gain. Still, it would only close 12 11 Luxembourg has been deleted from the graph. The reduction in its gap is 100 percent. 35 percent of the gap between actual and potential world output. Dispersion in cross-country output would fall by about 17 percentage points from 77 percent to 58 percent.12 Financial development explains about 28 percent of cross-country dispersion in output by this metric. Table 4: World-Wide move to financial best practice, z Increase in world output (per worker) 28% Reduction in gap between actual and potential world output 12% Fall in dispersion of ln(output) across countries 17% ( ' 77% - 59%) Fall in (pop-wghtd) mean of (cap-wghtd) distortion 15% ( ' 17% - 2%) Fall in (pop-wghtd) mean dispersion of (cap-wghtd) distortion 4.65% ( ' 5% - 0.35%) The presence of informational frictions causes the expected marginal product of capital, 1 r1 d= + 2 r2 , 1 r1 + to deviate from its user cost, re. De…ne the induced distortion in investment by 2 r2 re. For a country such as Sri Lanka these deviations are fairly large. The (capital-weighted) mean level of this distortion is 29 percentage points. It varies across plants a lot, as indicated by a coe¢ cient of variation of 32 percent. This is the type of resource misallocation e¤ect emphasized by Restuccia and Rogerson (2008). Here, the distortion is modelled endogenously. If Sri Lanka adopted the Swiss …nancial practices the average size of this distortion would drop to 1.5 percentage points. Its standard deviation across plants collapses from 9 percentage points to just 0.3 percentage points. The elimination of this distortion results in capital deepening among the active plants. Average TFP would rise by 26 percent in the model, as ine¢ cient plants are culled. For the world at large, the average size of the distortion is 17 percentage points, with an average coe¢ cient of variation of 28 percent. The mean distortion drops to 2 percentage points with a world-wide movement to …nancial best practice. The average standard deviation across plants falls from 5 percentage 12 The impact of …nancial intermediation on income will be larger if the former is allowed to a¤ect TFP in the production sector more directly. Erosa and Hidalgo-Cabrillana (2008) undertake a theoretical analysis where entrepreneurs produce an intermediate good that is important for the production of …nal output. They employ a Lucas (1978) style span of control model. A limited ability to enforce …nancial contracts leads to a poor selection of entrepreneurs in the economy. This channel of e¤ect may be important because Levine (2005) documents that …nancial development has a causal impact of productivity. A related quantitative model is in Amaral and Quintin (2005). They emphasize the capital accumulation channel. 36 points to a mere 0.35. Last, it will be noted that the model could be used to make an inference about productivities in the production and …nancial sectors, x and z, by using interest rate spreads, s, instead f 1 (o; s). of the capital output ratio, k=o; i.e., by using the mapping of the form (x; z) = M Erosa (2001) uses interest-rate spreads to quantify the e¤ects of …nancial intermediation on occupational choice. The correlation between the model’s prediction between the state of technology in the …nancial sector and the independent measure of …nancial development is again high–Table 5. The model does a reasonable job predicting cross-country capital-tooutput ratios. Financial development is now more important, but it still does not explain the bulk of cross-country variation in output–see Table 6. The di¤erence in the quantitative signi…cance obtains from the fact that the percentage variation in cross-country interest rate spreads is much larger than in capital/output ratios. The mean distortion in the world is now 21 percentage points, with an average coe¢ cient of variation of 30 percent. Table 5: Cross-Country Evidence capital-output ratio …nancial intermediation Correlation(model, data) 0.49 0.70 Table 6: World-wide move to best fin. prac., z f 1 (o; s)] [alternative results obtained when using interest-rate spread match, (x; z) = M Increase in world output (per worker) 48% Drop in shortfall between actual and potential world output 21% Fall in dispersion of ln(output) across countries 12% (' 75% - 62%) Fall in (pop-wghtd) mean of (cap-wghtd) distortion 18% ( ' 21% - 3%) Fall in (pop-wghtd) mean dispersion of (cap-wghtd) distortion 5.44% ( ' 6% - 0.56%) 37 Sri Lanka India Bolivia Morocco Mauritius Nicaragua Colombia Philippines Costa Rica Uruguay Honduras Brazil Turkey Mexico Argentina Ireland UK Panama US Portugal Italy Spain Iceland Peru Finland Netherlands France Denmark Canada New Zeland Australia Belgium Luxembourg Austria Japan Israel Thailand Norway Switzerland GDP per worker, % change 120 100 80 60 40 20 0 Figure 8: The impact of a move to …nancial best practice on GDP per worker 38 Sri Lanka India Bolivia Morocco Mauritius Nicaragua Colombia Philippines Costa Rica Uruguay Honduras Brazil Turkey Mexico Argentina Ireland UK Panama US Portugal Italy Spain Iceland Peru Finland Netherlands France Denmark Canada New Zeland Australia Belgium Austria Japan Israel Thailand Norway Switzerland Reduction in gap to the frontier, % 40 20 0 Figure 9: The impact of a move to …nancial best practice on the gap in GDP per worker 39 12 12.1 Robustness Analysis Intangible Investments and Capital’s Share of Income Suppose part of investment spending is undertaken in the form of intangible capital. As a result, measured investment may lie below true investment. This will lead to measured income, GDP, falling short of true output, o. This injects an upward (a downward) bias in the measurement of labor’s (capital’s) share of income. Speci…cally, in context of the standard neoclassical model, with a Cobb-Douglas production function, measured labor’s share of income, LSI, will appear as LSI = o GDP (1 ) > (1 ): Corrado, Hulten, and Sichel (2007) estimate the amount of intangible investment that was excluded from measured GDP from 1950 to 2003. They show that when output is adjusted to include these unrecognized intangibles, true output, o, is 12 percent higher than measured output, GDP, for the period 2000-2003. As a consequence, it is easy to calculate that =1 GDP o LSI = 1 1 (1 1:12 0:33) = 0:41: How does this larger estimate for capital’s share of income a¤ect the analysis? The calibration procedure described by (25) is redone for the case where = 0:41. The results are in accord with those obtained earlier. The model again …ts the U.S. data well. In particular, it matches the …rm-size distribution for 1972 very well. With no …nancial innovation, GDP would have risen by about 1.77 percent a year, compared with its actual rise of 2.4 percent. Hence, …nancial development accounts for about one quarter of the growth in GDP. About 18 percent of measured TFP growth is due to improvements in …nancial intermediation. Financial intermediation is now a little more important for economic development, at least when the model is used to match up GDP and capital/output ratios across countries. World output would increase by 33 percent, as opposed to the 28 percent found earlier, 40 if all countries moved to the best …nancial practice. When interest-rate spreads are targeted instead of capital/output ratios, …nancial development is slightly less important than before–world output would rise by 47 percent. Therefore, all in all, the results obtained earlier are quite robust to a change in capital’s share of income (when the model is suitably recalibrated). 12.2 Varying the Degree of Substitutability between Capital and Labor Let output be produced according to a CES production function of the form o = [ k + (1 1 )(x l) ] ; with 1. This production function will have have implications for how labor’s share of income, LSI, will vary across countries. To see this, think about the one sector growth model. Here labor’s share of income can be written as LSI= (w=l)=(w=l + rk) = 1=[1 + (r=w)(k=l)]: Therefore, labor’s share will rise whenever (r=w)(k=l) falls. With the above production function, 1=(1 = ) represents the elasticity of substitution between capital and labor. Hence, in response to a shock in some exogenous variable, z, it will happen that d ln(r=w)=dz = (1= )d ln(k=l)=dz. If the shock induces capital deepening [d ln(k=l)=dz > 0] then labor’s share will rise or fall depending on whether the elasticity of substitution is smaller or bigger than one. In the cross-country data, labor’s share either rises slightly or remains constant with per-capita income.13 restricted so that 1=(1 This suggests that for the quantitative analysis ) < 1, which implies should be < 0; i.e., capital and labor are less substitutable than Cobb-Douglas. Let = 0:38, roughly in line with Pessoa, Pessoa and Robb (2005). The calibration procedure described above is redone for this value for . The CES framework does not …t the …rm-size distribution for 1972 as well as the Cobb-Douglas case. It does worse predicting the shift in the 2000 distribution, although the movement is still in the right direction. In 13 That is, r=w will decrease by more (less) than k=l rises when the elasticity of substitution is smaller (greater) than one. 41 fact, if one allowed for 0 to be freely chosen in the calibration procedure then a value close to zero (Cobb-Douglas) would be picked. For the U.S. economy, the CES speci…cation predicts a rise in labor’s share from 0.73 to 0.75 as the capital stock deepens. The model with a CES production function has a di¢ cult time matching the observed variation in crosscountry capital/output ratios. Labor’s share varies from 0.70 to 0.75. All in all, both the U.S. and cross-country data prefer the Cobb-Douglas speci…cation. With a CES production structure world output would increase by 18 percent, if all countries move to the best …nancial practice. This is lower than the Cobb-Douglas case. This occurs because the potential for capital deepening is more limited the higher the degree of complementarity between capital, which is reproducible, and labor, which is …xed, in production. 13 Conclusions What is the link between the state of …nancial intermediation and economic development? This question is explored here by embedding a costly-state veri…cation framework into the standard neoclassical growth paradigm. The model has two novel ingredients. As in the standard costly-state veri…cation paradigm, the ex post return on a project is private information and an intermediary can audit the reported return. The …rst ingredient is that likelihood of a successful audit is increasing and concave in the amount of resources devoted to monitoring. The cost of auditing is increasing and convex in the amount of resources spent on this activity. Second, there is a distribution over …rm type, each type o¤ering a di¤erent combination of risk and return. Two key features follow from these ingredients. First, a …nancial theory of …rm size results. All …rms are funded that earn an expected return greater than the cost of raising capital from savers. Funding is increasing in a project’s expected return and decreasing in its variance. The size of a …rm is limited by diminishing returns in information production. Second, a Goldsmithian (1969) link is created between the state of …nancial development and economic development. The presence of informational frictions leads to a distortion between the expected marginal product of capital and its user cost, the interest paid to 42 savers plus capital consumption. This distortion is modelled endogenously here. As the e¢ cacy of auditing increases, due to technological progress in the …nancial sector, the size of this distortion shrinks. The upshot is an increase in the economy’s income. Intuitively, the rise in income derives from three e¤ects: (a) as the spread shrinks there is more overall capital accumulation in the economy; (b) capital is redirected toward the most productive investment opportunities in the economy; (c) less labor is required to monitor loans, which frees up resources for the economy. The developed model is taken to both U.S. and cross-country data. It is calibrated to …t the U.S. …rm size distribution for 1972 and the rise in the U.S. capital-to-output ratio between 1972 and 2000. It captures these features of the data well. The model’s predictions for the e¢ ciency of …nancial intermediation in a cross-section of 40 countries matches up well with independent measures. It does a reasonable job mimicking cross-country capitaloutput ratios and interest-rate spreads. The mechanism outlined above has quantitative signi…cance. The average measured distortion in the world between the expected marginal product of capital and its user cost falls somewhere between 17 and 21 percentage points. The average coe¢ cient of variation in the distortion within a country is 28 to 29 percent. World output could increase by 28 to 48 percent if all countries adopted the best …nancial practice in the world. Still, this only accounts for 13 to 19 percent of the gap between actual and potential world output. This happens because the bulk of the di¤erences in cross-country GDP are explained by the huge di¤erences in the productivity of the non-…nancial sector. There are two natural extensions of the above framework. The …rst would be to allow for long-term contracts. On this, Smith and Wang (2006) embed a long-term contracting framework into a model of …nancial intermediation. Clementi and Hopenhayn (2006) and Quadrini (2004) have also examined the properties of dynamic contracting for …rm …nance in worlds with private information. The use of dynamic contracts could mitigate the informational problems discussed here. How much is an open question. In a competitive world, such contracts may be severely limited by the ability of each party to leave the relationship at any point in time and seek a better partner. Second, …rms often use internal funds to 43 …nance investment. Incorporating internal …nance may a¤ect the importance of …nancial development. Perhaps accessing external funds involves a …xed cost, and hence is economized on. Then, internal funds may be hoarded in case good investment opportunities come along. An extension along these lines calls for a dynamic theory of the …rm–Cooley and Quadrini (2001) or Gomes (2001). In situations where the …nancial stakeholders of …rms delegate its operation to others, however, the returns on any investment activity (whether internally or externally …nanced) will need to be monitored. Ideally, any predictions from such a model should be matched up with a cross-country data set on the sources of …rm …nance. References Amaral, Pedro, S. and Erwan Quintin. “Financial Intermediation and Economic Development: A Quantitative Assessment.”Mimeo, Department of Economics, Southern Methodist University, 2005. Beck, Thorsten, Demirgürç-Kunt, Asli and Ross Levine. “A New Database on Financial Development and Structure.”World Bank Economic Review, 14, 3 (September 2000): 597-605. Beck, Thorsten, Demirgürç-Kunt Asli and Ross Levine. “The …nancial structure database.” In: Asli Demirgürç-Kunt and Ross Levine (eds) Financial Structure and Economic Growth: A Cross-Country Comparison of Banks Markets, and Development. Cambridge: MIT Press 2001: 17-80. Bencivenga, Valerie R. and Bruce D. Smith. “Financial Intermediation and Endogenous Growth.”Review of Economic Studies, 58, 2 (April 1991): 195-209. Berger, Allen N. “The Economic E¤ects of Technological Progress: Evidence from the Banking Industry.” Journal of Money, Credit, and Banking, 35, 2 (April 2003): 141-176. Bernanke, Ben and Mark Gertler. “Agency Costs, Net Worth and Business Fluctuations.”American Economic Review, 79, 1 (March 1989): 14-31. Boyd, John H. and Bruce D. Smith. “How Good are Standard Debt Contracts: Stochastic versus Nonstochastic Monitoring in a Costly State Veri…cation Environment.” Journal of Business, 67, 4 (October 1994): 539-561. Boyd, John H. and Edward C. Prescott. “Financial Intermediary-Coalitions.”Journal of Economic Theory, 38, 2 (April 1986): 211-32. 44 Castro, Rui, Clementi, Gian Luca and Glenn M. MacDonald. 2009. “Legal Institutions, Sectoral Heterogeneity, and Economic Development,” Review of Economics Studies, 76(2): 529-561. Chakraborty, Shankha and Amartya Lahiri. “Costly Intermediation and the Poverty of Nations.”International Economic Review, 48, 1 (February 2007): 155-183. Clementi, Gian Luca and Hugo A. Hopenhayn. “A Theory of Financing Constraints and Firm Dynamics.” Quarterly Journal of Economics, 121, 1 (February 2006): 229– 265. Cooley, Thomas F. and Edward C. Prescott. “Economic Growth and Business Cycles.” In: Thomas F. Cooley (ed) Frontiers of Business Cycle Research. New Jersey: Princeton University Press, 1995: 1-38. Cooley, Thomas F. and Vincenzo Quadrini. “Financial Markets and Firm Dynamics.” American Economic Review, 91, 5 (December 2001): 1286-1310. Corrado, Carol A., Hulten, Charles R. and Daniel E. Sichel. “Intangible Capital and Economic Growth.”National Bureau of Economic Research (Cambridge, MA) Working Paper No. 11948, 2006. Diamond, Douglas W. “Financial Intermediation and Delegated Monitoring.”The Review of Economic Studies, 51, 3. (July 1984): 393-414. Erosa, Andres. “Financial Intermediation and Occupational Choice in Development.” Review of Economic Dynamics, 4, 2 (April 2001): 303-334. Erosa, Andres and Ana Hidalgo Cabrillana. “On Capital Market Imperfections as a Source of Low TFP and Economic Rents.”International Economic Review, 49, 2 (May 2008): 437-473. Foster, Lucia, Haltiwanger, John and Chad Syverson. “Reallocation, Firm Turnover, and E¢ ciency: Selection on Productivity or Pro…tability?” National Bureau of Economic Research (Cambridge, MA) Working Paper No. 11555, 2005. de la Fuente, Angel and Jose Maria Marin. “Innovation, Bank Monitoring and Endogenous Financial Development." Journal of Monetary Economics, 38, 2 (October 1996): 269-301. Goldsmith, Raymond W. Financial Structure and Development. New Haven: Yale University Press, 1969. Gomes, Joao. “Financing Investment.” American Economic Review, 91, 5 (December 2001): 1261-1285. 45 Greenwood, Jeremy and Boyan Jovanovic. “Financial Development, Growth, and the Distribution of Income.” Journal of Political Economy, 98, 5 (Part 1, October 1990): 1076-1107. Harrison, Paul, Sussman, Oren and Joseph Zeira. “Finance and Growth: Theory and New Evidence.”Mimeo, Board of Governors of the Federal Reserve, 1999. Heston, Alan, Summers, Robert and Bettina Aten, Penn World Table, Version 6.1, Center for International Comparisons at the University of Pennsylvania (CICUP), October 2002. Hobijn, Bart and Boyan Jovanovic. “The Information-Technology Revolution and the Stock Market: Evidence.”American Economic Review, 91, 5 (December 2001): 12031220. Hopenhayn, Hugo A. and Richard Rogerson. “Job Turnover and Policy Evaluation: A General Equilibrium Analysis.”Journal of Political Economy, 101, 5, (October 1993): 915-38. Hsieh, Chang-Tai and Peter J. Klenow. “Misallocation and Manufacturing TFP in China and India.”Mimeo, Department of Economics, Stanford University, 2008. Khan, Aubhik. “Financial Development and Economic Growth,” Macroeconomic Dynamics, 5, 4 (June 2001): 413-433. Levine, Ross. “Stock Markets, Growth, and Tax Policy.” Journal of Finance, 46, 4 (September 1991): 1445-65. Levine, Ross. “Finance and Growth: Theory and Evidence.” In: Philippe Aghion and Steven N. Durlauf (eds) Handbook of Economic Growth, Volume 1A. Amsterdam: Elsevier B.V., 2005: 865-934. Li, Wenli and Pierre-Daniel G. Sarte. “Credit market frictions and their Direct E¤ects on U.S. Manufacturing Fluctuations.” Journal of Economic Dynamics and Control, 28, 3 (December 2003): 419-443. Lucas, Robert J., Jr. “On the Size Distribution of Business Firms,” Bell Journal of Economics, (1978): 508-23. Marcet, Albert and Ramon Marimon. “Communication, Commitment and Growth.” Journal of Economic Theory, 58, 2 (December 1992): 219-249. McGrattan, Ellen R. and Edward C. Prescott. “Taxes, Regulations, and the Value of U.S. and U.K. Corporations.”Review of Economic Studies, 72, 3 (July 2005): 767–796. Parente, Stephen L. and Edward C. Prescott. Barriers to Riches. Cambridge, MA: The MIT Press, 2000. 46 Pessoa, Samuel de Abreu, Pessoa, Silvia Matos and Rafael Rob. “Elasticity of Substitution Between Capital and Labor and its applications to growth and Development.” Pier Working Paper 05-012, 2005. Quadrini, Vincenzo. “Investment and Liquidation in Renegotiation-Proof Contracts with Moral Hazard.”Journal of Monetary Economics, 51, 4 (May 2004): 713-751. Restuccia, Diego and Richard Rogerson. “Policy Distortions and Aggregate Productivity with Heterogeneous Plants.”Review of Economic Dynamics, 11, 4 (October 2008): 707-720. Rousseau, Peter L. “The Permanent E¤ects of Innovation on Financial Depth: Theory and U.S. Historical Evidence from Unobservable Components Models.” Journal of Monetary Economics, 42, 2 (October 1998): 387-425. Smith, Anthony A. and Cheng Wang. “Dynamic Credit Relationships in General Equilibrium.”Journal of Monetary Economics 53, 4 (May 2006): 847-877. Townsend, Robert M. “Optimal Contracts and Competitive Markets with Costly State Veri…cation.”Journal of Economic Theory, 21, 2 (October 1979): 256-293. Townsend, Robert M. and Kenichi Ueda. “Financial Deepening, Inequality, and Growth: A Model-Based Quantitative Evaluation.” Review of Economic Studies, 73, 1 (January 2006): 251–293. Wang, Cheng. “Dynamic Costly State Veri…cation.” Economic Theory 25, 4 (June 2005): 887-916. Williamson, Stephen D. “Costly Monitoring, Financial Intermediation, and Equilibrium Credit Rationing.”Journal of Monetary Economics, 18, 2 (September 1986): 15979. Williamson, Stephen D. “Financial Intermediation, Business Failures, and Real Business Cycles.”Journal of Political Economy, 95, 6 (December 1987): 1196-1216. 14 14.1 Appendix Theory Proof for Lemma 1. First, substitute the promise-keeping constraint (9) into the objective function to rewrite it as ( 1 r1 + 2 r2 re)k 1 w(m1 =z) 47 2 w(m2 =z) v: Next, it is almost trivial to see that optimality will dictate that p12 = r1 k and p21 = r2 k, since this costlessly relaxes the incentive constraints (7) and (8). Next, drop the incentive constraint (7) from problem (P2) to obtain the auxiliary problem now displayed: subject to e ; v) I( max f( 1 r1 + p1 ;p2 ;m1 ;k [1 re)k 2 r2 1 w(m1 =z) vg; (P4) p1 r1 k; (27) p2 r2 k; (28) P21 (m1 =k)](r2 k p1 ) r2 k (29) p2 ; and 1 (r1 k p1 ) + 2 (r2 k (30) p2 ) = v: The strategy will be to solve problem (P4) …rst. Then, it will be shown that (P2) and (P4) are equivalent. Problem (P4) will now be solved. To this end, note the following points: 1. The incentive constraint (29) is binding. To see why, suppose not. Then, reduce m1 to increase the objective. 2. The constraint (28) is not binding. Assume, to the contrary, it is. Then, (29) is violated. This happens because the right-hand side is zero. Yet, the left-hand side is positive, given that p1 r1 k < r2 k, so that [1 P21 (m1 =k)](r2 k p1 ) > 0. 3. The constraint (27) is binding. Again, suppose not, so that p1 < r1 k. It will be shown that exists a pro…table feasible deviation from any contract where this constraint is slack. Speci…cally, consider increasing k very slightly by dk > 0 while adjusting p1 and p2 in the following manner so that (29) and (30) still hold. Also, hold m1 …xed. The implied perturbations for p1 and p2 are given by 2 3 2 3 1 dp [1 P21 (m1 =k)] 1 4 1 5 = 4 5 dp2 1 2 2 3 2 r P (m =k) (r2 k p1 )(m1 =k )dP21 =d(m1 =k) 4 2 21 1 5 dk: 1 r1 + 2 r2 48 Note that such an increase in k will raise the objective function. It will now be demonstrated that the optimization problems (P2) and (P4) are equivalent. e ; v) First, note that I( I( ; v), because problem (P4) does not impose the constraint (7). e ; v) It will now be established that I( I( ; v). Consider a solution to problem (P4). It will be shown that this solution is feasible for (P2). On this, note that point 1 implies that r2 k p2 = [1 P21 (m1 =k)](r2 k p1 ) r2 k p1 ; so that p1 p2 : Now, set m2 = 0 in (P2), which is feasible but not necessarily optimal. Then, constraint (7) becomes p1 e ; v) p2 , which is satis…ed by the solution to (P4). Therefore, I( I( ; v). Last, with the above facts in hand, recast the optimization problem as I( ; v) max f( 1 r1 + p2 ;m1 ;k re)k 2 r2 subject to r2 k p2 = (r2 r1 )k[1 1 w(m1 =z) P21 (m1 =k)]; vg; cf (29); and r2 k p2 = v= 2 , cf (30): The above two constraints collapse in the single constraint (10), by eliminating r2 k p2 , that involves just m1 and k. The problem then appears as (P3). Proof for Lemma 2. Suppose that the solution dictates that m1 =k 1= . Then, from (P3) it is clear that the optimal solution will dictate that m1 =k = 0. This happens because P21 (m1 =k) = 0 for all m1 =k 1= , yet monitoring costs are positive for all m1 =k > 0. Next, by substituting (10) into (P3) it is easy to deduce that the intermediary’s pro…t function can be written as + 2 r2 re r1 ) 2 (r2 Q 0 as r1 Q re. I( ; v) = [ 1 r1 49 1]v = r1 re v r1 ) 2 (r2 (31) Therefore pro…ts are negative if r1 < re and v > 0. Hence, a contract will not be o¤ered when m1 =k 1= . Proof for Lemma 3. Clear from equation (17). Proof for Lemma 4. Necessity: From problem (P3) it is clear that the intermediary will incur a loss when 1 r1 + 2 r2 by Lemma 2. Su¢ ciency: Suppose that 1 r1 re + 0 and v > 0, for any 2 A(w), because m1 =k > 1= re > 0 for some 2 r2 2 T . By equation (17) it is immediate that the intermediary will issue a loan k > 0. By construction it will earn zero pro…ts on this loan. Recall that the derivation of (17), discussed in the text, used a solution for v. This solution is v =V( )=( )( + + )=( ) 1 ( ) =( ) ( 1 r1 + 2 r2 ( 1 )( + re) Therefore, the …rm will earn positive rents too. + + )=( [ 2 (r2 ) (32) r1 )] ( 1 w) (z= ) Proof for Lemma 5. To begin with let k = K(w; ; z) represent capital stock that will be employed by a type- …rm when productivity in the …nancial sector is z and the wage rate is w > !. Similarly, let m1 =z = M (w; ; z) denote the amount of monitoring services, relative to z, that will be devoted to this project. Given this notation, the expected demand for labor by both the …rm and intermediary for a project of type L(w; ; z) ( 1= 1 1 + 1= 2 2 )[ (1 ) w Note that this demand is only speci…ed for 1 ] K(w; ; z) + is 1 M (w; (33) ; z) : 2 A(w) and w > !, where ! is de…ned by (22). In order to characterize L(w; ; z); the properties of its components K(w; ; z) and M (w; ; z) must be developed when 2 A(w) and w > !. First, take K(w; ; z). On this, rewrite equation (17) as (34) k = K(w; ; z) = ( 1 1 + 2 2 rew(1 )= )( + )=( ) 50 [ 2( 2 1) ] =( ) z =( 1) w 1=[ ( 1)] ; : where i (1 )(1 )= 1= i and ( + ) ) 1 =( =( ( ) ) ( 1 + )( + )=( ) . Note the following things about this solution for k: (i) The level of investment in a …rm, k , is continuous and strictly decreasing in w; (ii) k ! 0 as w ! W ( ) = [( 1 1+ 2 r] 2 )=e =(1 ) (when z is …nite); (iii) k is continuous and strictly increasing in z. Now, switch attention to the second term in L(w; ; z). A formula for M (w; ; z) can be derived in the same manner as the one for K(w; ; z). It is m1 =z = M (w; ; z) = where [( + ( 1 )=( 1 + 2 2 1= + )] rew(1 )= )( f =[ 2 ( 2 +1)=( 1 )]g ) w 1=( 1=[ ( ) 1)] 1=( z 1) ; (35) . Note the following things about this solution for m1 =z: (i) m1 =z is continuous and strictly decreasing in w; (ii) m1 =z ! 0 as w ! W ( ) [( 1 1 + r] 2 )=e 2 =(1 ) (when z is …nite); (iii) m1 =z is continuous and strictly increasing in z. Thus, the demand for labor by a type- project has the following properties: (i) L(w; ; z) is continuous and strictly decreasing in w; (ii) limw!W ( ) L(w; ; z) = 0; (iii) L(w; ; z) is continuous and strictly increasing in z. De…ne the function 8 < L(w; ; z); for w W ( ), e L(w; ; z) = : 0; for w > W ( ), (36) R for w > !. The aggregate demand for labor can be expressed as A(w) L(w; ; z)dF ( ) = R R e e L(w; ; z)dF ( ). Now, determine the constant by limw#! T L(w; ; )dF ( ) = 1;where T again the lower bound on wages ! is given by (22). From the simple closed-form solutions for (34) and (35) it is easy to deduce that such a must exist. To summarize, the aggregate R e demand for labor, T L(w; ; z)dF ( ), has the following properties: 1. 2. 3. 4. R R R R T T T T e L(w; ; z)dF ( ) is continuous and strictly decreasing in w for w 2 (!; 1); e L(w; ; z)dF ( ) is continuous and strictly increasing in z for z 2 ( ; 1); e w; L( b ; z)dF ( ) = 0, where w b = max e L(!; ; )dF ( ) = 1. 2T 51 W ( ); Therefore by the intermediate theorem for all z > there will exist a single value of w that sets labor demand equal to labor supply (or 1)— see Figure 6. Proof for Proposition 1. Express the labor-market-clearing condition as Z (1 ) 1= 1= 1= f( 1 1 + 2 2 )[ ] K(w; ; z) + 1 M (w; ; z) gdF = 1: w A(w) Let the 1= i ’s grow at the common rate g 1= and z grow at rate g 1=(1 ) (37) . Recall that exists a solution to the model without growth, as demonstrated by Lemma 5. It is easy to construct a balanced growth path using this solution. The solution implies that there will be a wage rate that solves (37). Conjecture that along a balanced growth path wages, w, will grow at rate g 1=(1 ) . From (34) it can be deduced that K(w; ; z) will grow at rate g 1=(1 ) . Therefore, the …rst term in braces in (37) will be constant. Equation (35) implies that M (w; ; z), or the second term, will be constant too. The active set A(w) will not change— equation (20). Therefore, labor demand remains constant. Hence, the conjectured solution for the rate of growth in wages is true. Using (32) is easy to calculate that v will grow at rate g 1=(1 ) . Last, since M (w; ; z) is constant it must be the case that m1 is growing at the same rate as z, or g 1=(1 ) . Therefore, m1 =k will remain unchanged along a balanced growth path. Proof for Proposition 2. First, point 2 in the proof of Lemma 5 established that the aggregate demand for labor is continuous and strictly increasing in z. Therefore, at a given wage rate the demand for labor rises as z moves up. In order for equilibrium in the labor market to be restored, wages must increase, since the demand for labor is decreasing in wages— point 1. Last, recall from (19) that a type- project will only be funded when w < W( ) = set of =(1 ) (1 )[( 1= 1 1 + 1= 2 2 )=e r] =(1 ) . It’s trivial to see that as w rises the 2 T satisfying this restriction, or A(w), shrinks; if = ( 1; 2) ful…lls the restriction for some wage it will meet it for all lower ones too, yet there will exist a higher wage that will not satisfy it. Furthermore, observe that W ( ) is strictly increasing in 1= 1 1 + 1= 2 2 . Therefore, those ’s o¤ering the lowest expected return will be cut …rst as w rises, because they have the lowest threshold wage. 52 Proof for Proposition 3. Let the 1= i ’s increase by the common factor g 1= > 1. Suppose that wages increase in response by the proportion g 1=(1 ) . Will the labor-market- clearing condition (37) still hold? The answer is no, because the demand for labor will fall. Take the …rst term behind the integral, which gives the demand for labor by a …rm. From (34) it is clear that K(w; ; z) will rise by a factor less than g 1=(1 ) , when z is held …xed. Therefore, the …rst term in braces in (37) will decline. Turn to the second term. From (35) it is easy to see that M (w; ; z) will drop under the conjecture solution. Therefore, wages must rise by less than g 1=(1 ) , since the demand for labor is decreasing in w (as was established in the proof of Lemma 5). The active set, A(w), will therefore expand, because 1 r1 + 2 r2 increases–see (18). Proof for Proposition 4. The set of projects in T o¤ering the highest expected return is given by A = arg max[ By assumption 2 A then R =( A 1; 1= 1( 1) + 1= 2( 2) ]: 2 )2T dF > 0. Take any equilibrium wage w. From (20) it is immediate that if 2 A(w), since (b r + ) is increasing in 1= 1 1 1 r1 + + 2 r2 1= 2 2 re = (1 . Hence, A )(1 )= w (1 )= ( 1= 1 2 + 1= 2 2 ) A(w) for all w. In equilibrium the wage will be a function of z, so denote this dependence by w = W (z). Now, let z ! 1. It will be shown that w = W (z) ! w , where w =(1 ) (1 )f 1= 1( 1) max[ =( 1 ; 2 )2T + 1= 2( 2) ]=e rg =(1 ) : (38) To see why, suppose alternatively that w ! w e 6= w . First, presume that w e < w . Then, by (19) all projects of type 2 A will be funded since their cuto¤ wage is W ( ) = w > w. e ed (W (z); ; z) = 1, for 2 A . From equations (33), (34) and (36) it is clear that limz!1 L R R d e (W (z); ; z)dF = 1. Therefore, such an Since, A dF > 0, this implies that limz!1 T L equilibrium cannot exist because the demand for labor will exceed its supply. Second, no …rm R d e (W (z); ; z)dF = 0. can survive at a wage rate bigger than w , by (19). Here, limz!1 T L This establishes (23). Last, note that A(w ) = A . 53 It is immediate that A limw"w A(w), because 2 A is viable for all wages w w = W ( ) by (19). It is also true that limw"w A(w) A , since from (19) any project 2 = A requires an upper bound on wages W ( ) < w to survive; that is, for any 2 = A there will exist some high enough wage w such that W ( ) < w < w . Therefore, limw"w A(w) = A = A(w ). This establishes point 1 of the Proposition. To have an equilibrium it must be the case that m1 =z < 1 for 2 A , otherwise the demand for labor would be in…nitely large. From equation (35) this can only happen when rew(1 )= ! 1 1 + 2 2, or equivalently when 1 r1 + 2 r2 ! re. Solve problem (P3) for the optimal level of monitoring, m1 =k, and then use (32) to solve out for v to obtain m1 =k = ( + + ) 1= It is apparent that lim m1 =k = 1; because z!1 [ ( 1 r1 + 2 (r2 1 r1 + 2 r2 2 r2 r1 ) re) ] 1= : ! re. Consequently, a false report by a …rm will be caught with certainty, or lim P12 (m1 =k) = 1. The contracting problem z!1 (P2) then requires lim p2 = r2 k and lim v = 0. A comparison of (32) and (35) leads to z!1 z!1 the conclusion that in fact lim m1 =z = 0, when lim v = 0. Using this result and (38), in z!1 z!1 R e conjunction with the labor-market-clearing condition, T L(w; ; z)dF = 1, then gives (24). 14.2 Data Figure 1: (Left panel) The numbers represent total debt outstanding for businesses (excluding …nancial ones) relative to gross domestic business value added (excluding gross farm value added). This data derives from the Flow of Funds Accounts of the United States (Table D.3: Debt Outstanding by Sector). The source of the data for the value of …rms relative to GDP is Hobijn and Jovanovic (2001, Figure 1, p. 1204). It refers to the total market capitalization of all securities contained in the CRSP data set. (Right Panel) A series for the intangible stock of capital is constructed by backing out the implied data series on investment in intangibles that is reported in Corrado, 54 Hulten and Sichel (2006, Figure 1). Speci…cally, a capital stock series for intangibles is constructed by iterating on the law of motion ki0 = (1 i )ki +ii , stock of intangible capital, ii is investment in intangibles, and i where ki is the current is the depreciation rate on intangibles. Two issues arise with this procedure. First, what is the depreciation rate on intangible capital? A weighted average of the rates reported in Corrado, Hulten and Sichel (2006, p. 23) suggests that it should be 33 percent. McGrattan and Prescott (2006, p. 782) feel that an upper bound of 11 percent is appropriate. Taking a simple average of these two numbers gives 22 percent, the value used here. Second, what starting value for the intangible stock of capital should be used? Along a balanced growth path the stock of intangible capital is given by ki = ii =(g + i ), where g is the growth rate of GDP. This formula is used to construct an initial capital stock for 1947, where g is assigned a value of 0:015. The stock of intangible capital is then simply added to private nonresidential nonfarm …xed assets. The resulting series is divided through by nonfarm business GDP. Figure 7: The data is for establishments. The horizontal axis orders establishments (from the smallest to highest) by the percentile that they lie in for employment. The vertical axis shows the cumulative contribution of this size of establishment to the total payroll in the U.S. economy. The use of payroll data controls for worker heterogeneity; i.e., measures employment in e¢ ciency units. Note that …rms without paid employees (sole-ownership …rms and joint proprietorships) are included in 1972 data. The source for the raw data used is Statistics for U.S. Businesses, U.S. Census Bureau. Figure 2 and Section 10: The data for the interest-rate spread is taken from Beck, Demirguc-Kunt and Levine (2001). It is de…ned as the accounting value of banks’net interest as a share of their interest-bearing (total earning) assets averaged over 1990 to 1995. The numbers for the ratio of private credit to GDP (the measure used for …nancial intermediation) are reported in Beck, Demirguc-Kunt, and Levine (2000) as the sum of deposit money in banks and related institutions, stock market capitalization, 55 and private bond market capitalization. These are averaged over the period 1990 to 1999. The other numbers derive from the Penn World Tables, Version 6.1— see Heston, Summers and Aten (2002). The capital stock for a country, k, is computed for the 1990-2000 sample period using the formula k = i=(g + ), where i is gross investment, g is the growth rate in investment, and is rate of depreciation. This formula heroically assumes that an economy is on a balanced growth path. The depreciation is taken to be 0.06. The growth rate for investment is calculated from the investment data reported in the tables for the period 1950 to 2000. Investment is recovered by using data on investment’s share of GDP and GDP. A country’s total factor productivity, T F P , was computed using the formula T F P = (y=l)=(k=l) , where y is GDP, l is aggregate labor, and is capital’s share of income. A value of 0.30 was picked for . Aggregate labor is backed out using data on per-capita GDP, GDP per worker, and population. The numbers in the analysis are reported in Table 7; the reported values are generally the mean for 1990-2000. Most of the headings in the table are obvious. Here ‘Spread’refers to the Beck, Demirguc-Kunt and Levine (2001) interestrate spread discussed above, while ‘Fin. Dev.’ represents the measure of …nancial development from Beck, Demirguc-Kunt and Levine (2000). Recall that s denotes the average spread in the model between the rates of return that intermediaries and savers. Likewise, o(x; z) [O(x; z) O(x; z)] is the increase in a country’s income if it moved to …nancial best practice, while G(x; z) [O(x; z) O(x; z)]=[O(x; z) O(x; z)] refers to the fraction of the gap between a country’s actual and potential output that would be closed. Last, the mean (capital-weighted) distortion for a country is de…ned by R R d(x; z) ( r + r r e )kdF= dF . The standard deviation of this distortion 1 1 2 2 A(w) A(w) qR R ( 1 r1 + 2 r2 re d)2 kdF= A(w) dF . is then given by A(w) 56 Table 7: Cross-country numbers, data and baseline model (see Section 14.2 for an explanation of the data and headings) Country Sri Lanka India Bolivia Morocco Mauritius Nicaragua Colombia Philippines Costa Rica Uruguay Honduras Brazil Turkey Mexico Argentina Ireland UK Panama US Portugal Italy Spain Iceland Peru Finland Netherlands France Denmark Canada New Zealand Australia Belgium Luxembourg Austria Japan Israel Thailand Norway Switzerland GDP p.w. 7013 5121 6779 11419 23705 5923 12332 7864 13913 20251 6823 18001 14340 22100 25056 46945 39908 15255 57151 30350 50569 40138 39834 10200 40603 46929 45317 44024 45933 36422 45907 50839 80702 45560 37061 40777 11632 47845 45706 Data K/GDP Spread 0.774 0.787 0.839 0.884 0.892 0.915 0.935 1.054 1.085 1.099 1.120 1.170 1.179 1.257 1.291 1.355 1.416 1.532 1.578 1.596 1.610 1.639 1.639 1.655 1.695 1.758 1.758 1.759 1.781 1.794 1.824 1.862 1.881 1.884 1.917 1.933 1.995 2.112 2.122 0.051 0.030 0.035 0.036 0.032 0.064 0.042 0.052 0.056 0.069 0.120 0.094 0.053 0.082 0.016 0.020 0.020 0.039 0.035 0.036 0.038 0.072 0.016 0.015 0.035 0.049 0.018 0.025 0.019 0.023 0.007 0.019 0.018 0.033 0.030 0.031 0.016 Fin. Dev. 0.339 0.512 0.441 0.576 0.683 0.234 0.402 0.890 0.158 0.266 0.282 0.538 0.336 0.556 0.335 0.977 2.481 0.692 3.297 1.046 1.104 1.279 0.940 0.312 1.551 2.430 1.592 1.741 1.706 1.363 1.470 1.487 2.064 1.143 3.043 0.979 1.786 1.403 3.464 ln(z) 7.60 6.91 7.60 8.29 9.10 7.60 8.52 8.52 9.21 9.68 8.70 9.80 9.62 10.34 10.62 11.51 11.63 11.18 12.72 12.18 12.76 12.68 12.67 11.40 12.99 13.49 13.45 13.43 13.60 13.45 13.86 14.21 14.80 14.24 14.26 14.48 13.69 16.16 16.22 57 Model [ = 0:33; (x; z) = M 1 (o; k=o)] s x o(x; z) G(x; z) d(x; z) 0.054 0.053 0.048 0.044 0.044 0.042 0.041 0.033 0.032 0.031 0.030 0.027 0.027 0.024 0.023 0.020 0.018 0.015 0.014 0.014 0.013 0.013 0.013 0.012 0.011 0.010 0.010 0.010 0.010 0.009 0.009 0.008 0.008 0.008 0.007 0.007 0.006 0.004 0.004 38.77 31.17 36.54 50.59 82.19 32.08 51.91 36.32 52.50 67.11 32.09 60.20 51.50 66.72 71.64 106.6 93.40 47.22 112.7 73.35 102.8 87.29 86.80 34.69 86.50 93.60 91.43 89.65 91.66 78.18 90.55 95.96 130.1 88.65 76.53 81.26 34.52 86.58 83.77 0.756 0.752 0.703 0.656 0.629 0.648 0.616 0.546 0.512 0.493 0.507 0.453 0.454 0.397 0.375 0.320 0.293 0.266 0.205 0.217 0.194 0.189 0.190 0.219 0.165 0.134 0..135 0.135 0.125 0.127 0.108 0.090 0.066 0.085 0.079 0.070 0.080 0.002 0.000 0.100 0.071 0.087 0.141 0.332 0.067 0.142 0.073 0.129 0.196 0.057 0.151 0.115 0.168 0.186 0.452 0.293 0.065 0.447 0.132 0.305 0.182 0.179 0.033 0.159 0.171 0.160 0.151 0.152 0.099 0.130 0.135 1.000 0.099 0.062 0.065 0.013 0.002 0.000 0.293 0.286 0.260 0.240 0.236 0.227 0.219 0.179 0.170 0.167 0.161 0.148 0.146 0.129 0.122 0.110 0.099 0.082 0.075 0.073 0.071 0.068 0.068 0.066 0.061 0.054 0.054 0.054 0.051 0.050 0.047 0.043 0.042 0.041 0.038 0.037 0.033 0.022 0.022 Std Dev d(x; z) 0.093 0.090 0.080 0.073 0.072 0.068 0.065 0.052 0.049 0.047 0.045 0.041 0.041 0.035 0.033 0.029 0.026 0.020 0.019 0.018 0.017 0.016 0.016 0.016 0.014 0.012 0.012 0.012 0.012 0.011 0.011 0.010 0.009 0.009 0.008 0.008 0.007 0.004 0.004