PART III. The Impact of Public Inputs on the Private Economy: An empirical analysis of public capital, unemployment, and employment by sector*

Transcription

1 PART III The Impact of Public Inputs on the Private Economy: An empirical analysis of public capital, unemployment, and employment by sector* January 14, 2003 Christopher N. Annala State University of New York, Geneseo Jones School of Business 1 College Circle Geneseo, NY annala@geneseo.edu Phone: Raymond G. Batina Department of Economics Washington State University Pullman, WA rbatina@pullman.com Abstract: This paper provides three contributions to the existing literature on the impact of public inputs on the private economy. First, this paper fills a gap in the existing research on public capital and the private economy, by examining the effects of public capital on state level rates of unemployment. Second, this paper examines the employment effects of public capital on different sectors of the economy. Third, this paper provides updated estimates of state and local public and private capital stocks by state. The effects of public capital on private sector performance are debated within academic circles and among policymakers. The basic working assumption among proponents is that public capital improves private sector productivity and public capital projects create employment, both of which have positive effects on economic activity and income. The results of this study indicate that under a variety of econometric techniques the stock of public capital has little or no impact on lowering state level unemployment rates or on employment across sectors of the economy. * We are very grateful for the generous support of the Economic and Social Research Institute of Japan s Cabinet Office. We also thank the Kansai Institute for Social and Economic Research for its assistance. We are solely responsible for any errors that remain.

2 1. Introduction This paper provides three contributions to the existing literature on the impact of public inputs on the private economy. First, this paper fills a gap in the existing research on public capital and the private economy by examining the effects of public capital on state level rates of unemployment using state-level US panel data. Second, we study the impact of public capital on sector level employment to determine which, if any, types of employment benefit most from the existence of public capital. Third, this paper provides updated estimates of state and local public and private capital stocks by state. The effects of public capital on private sector performance are debated within academic circles and among policymakers. The basic working assumption among proponents is that public capital improves private sector productivity and public capital projects create employment, both of which have positive effects on economic activity and income. In recent years there has been a substantial amount of research into the effect that public capital has on the private economy and economic growth. Ratner (1983) provides the seminal paper to test for the effect of public capital on private output. Aschauer (1989) found that core infrastructure capital investment provided significant explanatory power for the productivity slowdown witnessed in the United States using national data. Subsequent studies have provided mixed results. Tatom (1991) argues that the relationship found by earlier researchers is spurious due to the nonstationarity of the data. Holtz-Eakin (1993) concludes that when the sample is a panel of the forty-eight contiguous states over the years 1969 to 1986 the elasticity of private output with respect to state and local government capital is essentially zero. Munnell (1990) performs several exercises to determine the impact of public capital on private sector performance. One of these exercises explores the relationship between public capital and employment growth using a sample of individual states. According to 76

3 Munnell, The evidence seems overwhelming that public capital has a positive impact on private sector output, investment, and employment (1990: pp. 26). Demetriades and Mamuneas (2000) examine the impact of public capital on manufacturing employment using a panel of twelve OECD countries; they find that increases in public capital infrastructure increase employment. Using time series data for the United States, Batina (1999) finds that the impact of public capital depends on the proxies used and that the effects are much smaller than those estimated by Aschauer and Munnell. Although inconclusive as to the magnitude of the impact public capital has on economic growth, there appears to be empirical support for the role public capital has in influencing private sector activity. The contribution of this paper is to examine, in greater detail, the effect of public capital on private sector unemployment. The use of public capital projects to reduce unemployment has a long history. For example, as part of Franklin D. Roosevelt s New Deal the Public Works Administration (PWA) was established as part of the National Industrial Recovery Act of The PWA was responsible for the construction of roads, buildings, dams and other projects in an effort to reduce unemployment and increase economic activity. In its 10-year existence, the PWA spent more than $4 billion on public projects. 1 The projects undertaken by the PWA are precisely the types of public capital infrastructure projects which proponents and current researchers argue increase the marginal productivity of capital and labor and increase the rate of growth. Current policymakers still lobby for resources for public capital projects and argue that such projects will enhance economic activity and lower unemployment within their jurisdictions; however the theory is not as clear-cut. Public capital may have a number of different effects on the economy. 1 "Public Works Administration," Microsoft Encarta 98 Encyclopedia, , Microsoft Corporation. 77

4 Consider a neoclassical technology, Y = F(K,L,G), where Y represents output, K is private capital, L is labor, and G represents public infrastructure and output is increasing in each of the inputs, that is the derivatives F K, F L, and F G are all positive. An increase in public capital may result in a decrease in K alone, a decrease in L alone, an increase in one input coupled with a decrease in the other input, or a decline in both inputs. For example, if private and public capital are complements and labor and public capital are substitutes, then an increase in public capital will induce an increase in private capital demand and a drop in demand for labor, which may cause the unemployment rate to increase. In addition, there may be other effects as well. First, resources must be withdrawn from the private sector in an economy that is close to full employment and this can cause general equilibrium adjustments that are difficult to predict. Second, the additional taxes used to finance such investments might also create general equilibrium effects that may affect the amount of labor used in the economy. For example, if the additional taxes reduce labor supply and this raises the wage rate, firms may shift from labor to private capital thus raising the unemployment rate. Third, public capital may fuel economic growth attracting more workers to enter the economy. This may lower the wage thus causing firms to shift from private capital toward labor. Further obfuscating the situation is the mobility of factors as it may be possible that a public works project in state A attracts labor from state B. All in all, the impact of public capital investments on the local economy is theoretically ambiguous, and it, therefore, becomes an empirical question. The paper is organized in the following manner. Section two describes the data and estimation technique used in this paper. Section three discusses the results involving unemployment. Section four discusses the results on employment. And section five provides some concluding remarks. 78

5 2. Data and Estimation We use several methods for estimating the impact of public capital on unemployment. First, we employ the conventional least squares dummy variable (LSDV) panel data estimator as described by Hsaio (1986). The model is estimated using fixed effects for both the time dimension and state specific intercepts. Second, we employ the generalized method of moments technique with a lagged dependent variable developed by Arellano and Bond (1991) with lags of the independent variables as instruments for the moment restrictions. Third, we employ the instrumental variables technique developed by Holtz-Eakin, Newey, and Rosen (1988). The basic equation to be estimated is given by: U k i s, t = β i xs, t + µ s + λt + εs, t i= 1, (1) where U represents the unemployment rate in state i in time period t. The parameter µ s is a shift parameter for each state, which is included to capture state specific characteristics, and λ t is a state invariant parameter that captures nationwide time-specific economic factors such as technology, business cycle changes, and federal fiscal and monetary policy. The variables x i s,t represents the i th explanatory variable for state s at time t. The explanatory variables include estimates of private capital, public capital, disaggregated public capital, the growth rate of gross state product, unemployment compensation benefits, and taxes. 2 The sample for our estimation covers the fifty states plus Washington, D.C. for the years 1977 through This specification is similar in spirit to that of Munnell (1990) where the growth rate of output is replaced with the unemployment rate, however as pointed out by Holtz-Eakin the use of state specific intercepts is necessary. The inclusion of state and local taxes in our empirical model accounts for the fact that public capital infrastructure requires the government to raise revenue which may affect the 2 Unionization rates are not available at annual intervals for early years in our study; furthermore, unionization rates are not comparable before and after

6 labor market and hence the unemployment rate if firms react to state and local taxes. Unemployment compensation is included in the estimation to account for the fact that states with more generous unemployment benefits are likely to have higher rates of unemployment as the cost of being unemployed is lower for individuals. Finally, the growth rate of GSP is intended to capture the effect on unemployment of changes in overall economic activity within each state, it is hypothesized that higher growth rates of GSP are associated with lower unemployment rates. The following section describes the calculation of private and public capital stocks for each state. There are no official government estimates of capital stocks at the state and local levels therefore the data come from a variety of sources. Both the private capital and public capital series are constructed using current costs due to data limitations. The final estimates of the private and public capital stocks are then converted into real terms by using the state by state chained-type price indices calculated by the BEA. Private Capital This section describes the calculation of state-level private capital stocks. The private capital stocks at the state-level were calculated based on the procedure outlined by Munnell (1990) and Costa, Ellison, and Martin (1987). The private capital estimates calculated by Munnell cover , therefore updating private capital stock estimates became necessary as there are no official estimates of state and local private capital stocks, or public capital stocks. The procedure involves using proxies from a variety of sources and these proxies are used to apportion national estimates of private capital stocks from the Bureau of Economic Analysis (BEA). Attempts were made to follow the procedure of Munnell (1990) whenever possible for comparison purposes. The procedure resulted in approximately eightypercent of national estimates to be apportioned among the states. The basic equation for the 80

7 calculation of the nonresidential private stock in a given state is based on the following formula: K i = (AGK i /ΣAGK i )AGK + (MFGK i /ΣMFGK i )MFGK + (NFNMFGK i /ΣNFNMFGK i )NFNMFGK (2) where: AGK = BEA current-cost estimate of national value of capital stock in the agricultural sector MFGK = BEA current-cost estimate of national value of capital stock in the manufacturing sector NFNMFGK = BEA current-cost estimate of national value of capital stock in the nonfarm, non-manufacturing sector AGK i = proxy for capital stock in agriculture in state i MFGK i = proxy for capital stock in manufacturing in state i NFNMFGK i = proxy for capital stock in the non-farm, non-manufacturing sector in state i. The BEA stock of agricultural capital was apportioned based on each state's share of the value of land, buildings, and equipment from the Census of Agriculture for 1978, 1982, 1987, 1992, and the Economic Census for Stocks for were based on the 1978 Census. Data from the 1982 Census were used to calculate shares from Data from the 1987 Census were used to calculate shares from , the 1992 Census was used to calculate shares for and the 1997 Census was used to calculate shares for The BEA estimate of manufacturing capital was apportioned based on each state's share of the gross value of depreciable assets from the Census of Manufactures for 1977, 1982, 1987, 1992, and the Economic Census for The procedure for calculating state shares is the same as that for agricultural capital. That is, the estimates of the gross value of 81

8 depreciable assets from the 1977 Census are used to calculate shares for The 1982 Census is used to calculate shares for , the 1987 Census is used to calculate shares for , and the 1992 Census is used to calculate shares for Finally, the 1997 Economic Census is used to calculate shares for the years Non-farm and non-manufacturing capital was apportioned based on proxies in several sectors including: construction, mining, retail and wholesale trade, banking, railroad transportation, trucking, water, and air transportation, electric services, gas services, communications, and services. The method for apportioning the BEA national estimates in the sub-sectors is similar to the procedure above. A state's share of capital in construction is based on the gross book value of depreciable assets from the Census of Construction for 1977, 1982, 1987, 1992, and the Economic Census of The stock of private capital in retail and wholesale trade was apportioned based on each state s share of sales from the Census of Retail Trade for 1977, 1982, 1987, and 1992, the Census of Wholesale Trade for 1977, 1982, 1987, and The Economic Census for 1997 covers both the retail trade and wholesale trade industries. As with manufacturing and agricultural capital the share form the Census year is used to apportion BEA estimates for the previous two years and the following two years. The capital stock in the remaining sub-sectors was apportioned based on a variety of proxies. Assets in the electric services industry were apportioned based on a state s installed capacity or a state s net summer capacity. Estimates of capacity for the Census years 1977, 1982, 1987, 1992, and 1997 are from various years of the U.S. Statistical Abstract, 1979, 1984, 1989, 1994, and 1999 due to lags in data collection and reporting. Assets in the gas services industry are based on a state s share of, end of year miles of pipeline and main for the Census years. Data are from various years of Gas Facts, a publication of the American Gas Association. 82

9 The allocation of assets in banking was based on each state s share of deposits in insured commercial banks. Data on deposits are from the U.S. Statistical Abstract and cover 1977, 1982, 1987, 1993, and Data for 1992 were not reported in the Statistical Abstract, therefore a states share of deposits in 1993 were used to apportion BEA estimates for The estimates of private capital in transportation were apportioned using two procedures. The stock of capital in the railroad industry, which accounts for over forty to sixty percent of BEA estimates of transportation capital between 1977 and 1999, is based on railroad mileage in each state for each of the Census years, 1977, 1982, 1987, 1992, and The data on railroad mileage is taken from Railroad Facts, a publication of the Association of American Railroads. Assets in trucking, air, and water transportation were apportioned based on a state s share of Gross State Product, from the BEA, for each of the sub-sectors. Assets in the three remaining sub-sectors, mining, telecommunications, and services were also apportioned based on each state s share of Gross State Product in those industries. The use of GSP estimates was necessary due to changes in data availability and changes in reporting between 1977 and The share of assets for each sub-sector, in each year, is then multiplied by the BEA national estimates and then summed to arrive at a total for non-farm, non-manufacturing, fixed assets. The sum of all of the sub-sectors described above account for, on average, approximately seventy-four percent of the BEA national estimates of non-farm, nonmanufacturing private capital. This result is almost identical to that reported by Munnell (1990), where the author states that, The sum of asset estimates for all states, for all subsectors, represented nearly three-quarters of the BES national total of non-farm, nonmanufacturing assets. 83

10 Public Capital This section describes the updating of public capital stocks by state for the period Public capital stocks are estimated using the perpetual inventory method according to the equation KG t = IG t + (1 - δ)kg t-1, (3) where KG t is the public capital stock, IG t is gross investment in public capital and δ is the constant rate of depreciation. Estimates of the initial capital stock are taken from Holtz-Eakin (1994) which provides public sector capital estimates from For purposes of consistency and possible comparisons between studies the estimates from 1976 are taken as the initial level of public capital. The level of public sector capital is then estimated according to equation (3), based on estimates of public capital investment from the U.S. Census Bureau series, State and Local Government Finances. The depreciation rates for disaggregated public capital are taken from Fraumeni (1997). The depreciation rate for total capital and education capital is , as this is the depreciation rate for most governmental nonresidential structures. The depreciation rate for highways and sewer systems is Finally, the depreciation rate for utilities is taken as the average of the two rates reported in Fraumeni for private electric light and power capital before and after 1946, for a rate of The six components of public capital above represent on average 82.3 percent of all public capital across all years and all states. Other Variables Data on the unemployment rate by states is taken from the U.S. Statistical Abstract for various years. Additional explanatory variables include the growth rate of Gross State Product from the BEA and state and local taxes and unemployment compensation, both of which are from the U.S. Census Bureau, series, State and Local Government Finances. In the 3 The authors would like to thank Professor Douglas Holtz-Eakin for providing this data. 84

11 estimation the unemployment rate and the growth rate of GSP enter as percentage rates, all other variables enter as natural logs. Descriptive statistics for each of the variables appear in Table 1. In order to test for the possibility of unit roots in the data we perform two panel unit root tests. The results of the modified Levin and Lin test (1993) which is distributed N[0,1] and the Im, Pesaran, and Shin ( 1997) panel unit root tests appear in Table 2. The results are dependent upon the inclusion of a linear time trend and on the number of AR lags included in the ADF test. The advantage of the IPS test over the LL test is that the LL test statistic requires all cross-sections to have the same AR order, the IPS test however allows for the AR lags to vary between cross-sections. Taking all of the evidence from the various tests and depending upon the number of AR lags and the inclusion of a linear time trend or not, we can say for the most part that all of the variables are stationary except for perhaps the stock of education capital where the IPS test statistic is and the ten-percent critical value is

12 3. Results: Unemployment To better understand the relationship between public capital and the unemployment rate several equations are estimated including: 1. An equation in current levels with total public capital using the fixed effects LSDV estimator. 2. An equation with one-period lags of public and private capital as suggested by Demetriades and Mamuneas (2000) to account for the fact that the current unemployment rate may depend on previous levels of capital. 3. An equation with disaggregated public capital as suggested by Batina (1999), among others. We also estimate an equation with one-period lags of disaggregated public capital. 4. We also estimate an equation in first differences for both total public capital and disaggregated public capital, as suggested by Holtz-Eakin (1994). When estimated in first differences the state specific effects are eliminated. 5. We then estimate the equation with lagged unemployment to capture the possibility of inertia. In this case the appropriate estimator is given by Arrelano and Bond (1991) where we employ GMM estimation with lags of the independent variables serving as instruments. 6. Finally we report the results of estimations based on the model proposed by Holtz- Eakin, Newey, and Rosen (1988) to account for possibility simultaneity. As noted by Holtz-Eakin the small sample size may of a concern. The results of the estimation in levels appear in the column (1) of Table 3, estimates of time and state specific dummy variables are not reported. The results indicate that both higher levels of private capital and the growth rate of GSP result in lower unemployment rates, as indicated by the positive signs and both coefficients are statistically different from zero at 86

13 the ten-percent level. The estimation also indicates that higher unemployment compensation benefits and higher taxes both result in higher rates of unemployment. The coefficient on public capital is positive and statistically different from zero indicating that a larger stock of public capital results in a higher unemployment rate, this may be a result of either laborsaving investments or an indirect effect through changes in factor prices. To test for the stationarity of the estimated regression we employ the LL test and the IPS test on the residuals based on the procedure developed by Kao and Chiang (1997). The modified LL test statistic is and is significant at the 1-percent level. The IPS test statistic on the residuals is and is also significant at the 1-percent level; these results indicate that the estimated regression equation is stationary. It follows that the variables in that equation are cointegrated. Column (2) of Table 3 provides the results when the equation is estimated with lags of both private and public capital. The qualitative results are very similar when one-period lags of private and public capital replace current levels. The results of the estimation with the current levels of disaggregated public capital appear in column (3) of Table 3. The coefficient values on the other explanatory variables change only slightly when the equation is estimated with disaggregated public capital. The results indicate that the stock of education capital has a negative sign, implying that a larger stock of capital devoted to education lowers the unemployment rate. All other forms of public capital have a positive sign, which again indicates that a larger public capital stock tends to increase the unemployment rate. The LL and IPS panel unit root tests are again employed to test for the stationarity of the regression equation. The modified LL test statistic is and the IPS test statistic is , both of which are significant at the 1-percent level indicating that the estimated equation is stationary and the variables are thus cointegrated. The results of the estimation with lagged disaggregated public capital appear in column (4) of Table 3. Again, the results of this estimation are qualitatively similar to the estimation with 87

14 current levels, where the coefficient on public capital in education is negative and all other public capital coefficients are positive. Table 5 provides estimates of the elasticities, measured at the mean, for the two basic equations with total public capital and disaggregated public capital, i.e. columns (1) and (3) from Table 3. Highway capital tends to have the largest effect in magnitude (0.311), followed by education capital ( ), sewer and sanitation capital (0.109), and utilities capital (0.069). The results of the estimation in first differences with total public capital appear in column (1) of Table 4. Estimating the equation in differences has the appeal that it shows how the growth rate of the independent variables is related to a change in unemployment. Only the coefficients on public capital and unemployment benefits are significant at conventional levels when estimated in first differences. As with the previous estimates the coefficients on unemployment compensation benefits and total public capital are both positive. Column (2) of Table 4 presents the results of the first difference estimation with disaggregated public capital. The coefficients on highway capital, education capital, and sanitation and sewerage capital are all significantly different from zero and positive, as is the coefficient on the first difference of unemployment compensation benefits. These results indicate that the growth rate of public capital actually leads to higher rates of unemployment. Table 6 presents the results of the GMM estimation with lagged unemployment included as an independent variable, both state and time dummies are included in the estimation. Column (1) provides the results when the equation is estimated with total public capital. Column (2) provides the estimates of the impact of disaggregated capital. With the inclusion of lagged unemployment the results differ for several variables. First, lagged unemployment is positive and significant indicating that there is an inertia effect of unemployment. Second, we again find that unemployment benefits have a positive effect on 88

15 the unemployment rate and that the growth rate of GSP has a negative effect on the unemployment rate. Third, the main qualitative difference between the GMM estimation and the LSDV estimation is that the stock of private capital has a positive impact on the unemployment rate. Fourth, we still find that total public capital is positively related to unemployment. However, under this estimation method both education and sewer and sanitation capital are negatively related to the unemployment rate. The results of the HNR estimator appear in Table 7. These results indicate that public capital is either positively related to unemployment or not statistically different from zero and essentially agree with the results of Table 4. 89

16 4. Results: Employment Based on the conflicting results for the impact of public capital on unemployment presented in this paper versus the results on employment from previous studies, we also examine the impact of public capital on state level employment and employment by sector. The goal of this line of research is to determine which sectors, if any, are most affected by the stock of private capital. The basic equation to be estimated is similar in spirit to existing location models where the primary determinants of a firm s location decision are generally wages and taxes. The equation to be estimated is similar to that of Munnell (1990), where she finds a positive and statistically significant effect of public capital on employment, however since we employ panel data techniques the additional explanatory variables are unnecessary. According to Holtz-Eakin (1994) results are similar with state specific effects or the use of additional explanatory variables. Table 8 provides summary statistics on employment by sector (total private employment, agricultural services, mining, construction, manufacturing, transportation and public utilities, wholesale trade, retail trade, finance, insurance, real estate, and services), average wage and salary disbursements by sector, public capital, current expenditures, and total taxes over all states and time periods. The data on employment and wage and salary disbursements are available from the Bureau of Economic Analysis and the government finance data are available from the Census Bureau, the data on the public capital stock are described above. Average real wage and salary disbursements by sector are calculated by taking total wage and salary disbursements by sector and dividing by employment in that sector, then deflating by the state-by-state chain level price index. Table 9 presents the results of the IPS panel unit root test for each of the variables in our study. The results indicate that almost all of the variables are stationary, the exceptions being employment in agricultural services, wholesale trade, and transportation and utilities, and public capital in education. 90

17 We begin with the fixed effects LSDV model to estimate the following equation: EMP k j i s, t = β i xs, t + µ s + λt + εs, t i= 1, (4) Where EMP j i s,t represents employment in sector j, in state s, in time period t, and x s,t represents the i th explanatory variable in state s including the average wage and salary disbursements by sector, public capital, current expenditures, and total taxes. Our estimation assumes that public capital is a pure public good, that is, the entire public capital stock enters the estimated equation for each sector. Results of the fixed effects LSDV estimates are presented in Table 10. The results indicate that total public capital has a positive and statistically positive effect on total employment, manufacturing, and transportation and utilities. For all other sectors the impact of public capital is either statistically insignificant or negative. The results are in-line with expectations in that a priori we would anticipate both manufacturing and transportation to benefit more from public capital than services, etc. Based on IPS panel unit root tests of the residuals we find that several of the regressions may not be stationary. To account for the potential nonstationarity we estimate the same equation using the HNR estimator based on the first differences which eliminates the state specific effects, thereby eliminating the inconsistency stemming from the presence of correlated state effects. The HNR estimator also has the advantage of addressing the possibility of simultaneity bias. Again, we estimate the equation in first differences with the twice lagged independent variables serving as instruments. The results of this estimation appear in Table 11. The results indicate that public capital has a negative and statistically significant effect on employment for almost all sectors. Second, current expenditures tend to have a positive effect on employment across sectors. Third, we also find that taxes are insignificant for all but the manufacturing sector. These results may not come as a surprise to those familiar with the literature on the effect of state and local taxes on economic growth. Several researches have 91

18 found very small or insignificant effects of taxes on economic growth. Following a survey of the literature Wasylenko (1997) concludes that state and local fiscal policies have become increasingly similar which has negated some of the impact of taxes and spending. Annala and Perez (2001) provide evidence that state and local public capital has converged over time, implying that public capital may not be a determining factor in firm location and therefore employment growth. Annala (forthcoming) also finds that state and local fiscal policies have become more alike over the past twenty-five years. The results of the Arellano and Bond (1991) dynamic fixed effects estimator appear in Table 12. The Arellano and Bond estimator is similar to that of HNR in that it is based on an instrumental variable approach which accounts for possible simultaneity. In terms of the focus of this paper, public capital, the results are qualitatively similar between the two approaches in that public capital has a negative effect on total employment and employment for each sector with two exceptions, wholesale trade and finance, where the impact is positive. In general, it appears that wages have a negative effect on employment. And we also find that current expenditures under this method have a positive effect on total employment, a negative effect for most sectors except manufacturing, retail trade, and services, where it is positive. 92

19 5. Concluding Remarks The ultimate effect of public capital on unemployment is a fundamentally empirical question since there are a variety of conflicting effects that may occur when the government undertakes an investment in public capital. This paper has provided several important contributions to the existing body of knowledge on the empirical effects of public capital. More specifically, we examined the effects of public capital on unemployment and presented evidence that indicates that different types of public capital have different effects on statelevel rates of unemployment. Second, we studied the impact of public capital on employment in a variety of sectors including agriculture, manufacturing, mining, construction, transportation, wholesale trade, retail trade, finance, and services. In general, we found that public capital either has a negative effect on employment, or no effect at all under the preferred method of estimation that accounted for stationarity problems and simultaneity bias. Finally, we provided updated estimates of state and local public and private capital. Our results indicate that there is no overwhelming evidence that supports the contention that public capital projects reduce the unemployment rate or increase employment for most of the sectors we studied using state-level panel data for the US. For example, we found that public capital projects either raise the unemployment rate in most cases (Tables 3, 4, 6, and 7) or have no effect at all. There is some mild evidence that the stock of education capital is negatively related to the unemployment rate (Tables 3 and 6). However, it is possible that education capital is proxying for the stock of human capital. It would not be surprising to discover a negative correlation between human capital and the unemployment rate. We also find that public capital lowers total employment and employment in most of the sectors we considered, agriculture, construction, transportation, retail trade, and services, under both GMM estimation and HNR's method. And employment in mining and 93

20 manufacturing was unaffected or negatively related to public capital under both methods. Only employment in wholesale trade and finance were positively related to public capital, but only under GMM estimation. Why might public capital raise the unemployment rate or lower employment in a variety of industries? One possibility is that public capital and labor are substitutes. A public capital project might cause private firms to shift away from labor toward private capital thus possibly reducing employment and raising the unemployment rate. Second, 49 states in the US must balance their budgets with Vermont being the only exception. Public capital projects may force states to raise taxes to obtain funding, ceteris paribus. In that case, higher taxes may reduce labor supply thus raising the wage causing firms to shift from labor toward private capital. More research could usefully be done in this area. Our results call into question some of the results of earlier researchers using different data and methods. At this point, we feel one should be skeptical of claims that public capital projects will create jobs and reduce unemployment. 94

21 References American Gas Association, various years, Gas Facts, Arlington Virginia, American Gas Association. Annala, C., Forthcoming, Have state and local fiscal policies become more alike? Evidence of Beta convergence among fiscal policy variables. Public Finance Review. Annala, C. and S. Perez, 2001, Convergence of public capital investment among the United States, Public Finance and Management, Volume 1, Number 2. Aschauer, D., 1989, Is public expenditure productive? Journal of Monetary Economics 23, Arellano, M. and S. Bond, 1991, Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations, The Review of Economic Studies, 58, Association of American Railroads, various years, Railroad Facts, Washington, D.C., Association of American Railroads. Baltagi, B., 2001, Econometric Analysis of Panel Data, second edition, John Wiley and Sons, Ltd. Batina, R., 1999, On the long run effect of public capital on aggregate output: Estimation and sensitivity analysis, Empirical Economics 24, Besley, T. and A. Case, 1995, Incumbent Behavior: Vote-Seeking, Tax Setting, and Yardstick Competition, American Economic Review 85, Bureau of Economic Analysis, Fixed Assets Tables, available at: Bureau of Economic Analysis, Regional Accounts Data, available at: Case, A., H. Rosen, J. Hines, 1993, Budget Spillovers and Fiscal Policy Interdependence: Evidence from the States, Journal of Public Economics 52, Costa, J., R. Ellson, and R. Martin, 1987, Public capital, regional output, and development: Some empirical evidence, Journal of Regional Science, 27, Demetriades, P. and T. Mamuneas, 2000, Intertemporal output and employment effects of public infrastructure capital: Evidence from 12 OECD economies, The Economic Journal, 110, Fraumeni, Barbara, 1997, The Measurement of depreciation in the U.S. national income and product accounts, Survey of Current Business, July,

22 Holtz-Eakin, D., 1993, Solow and the states: Capital accumulation, productivity, and economic growth, The National Tax Journal, Holtz-Eakin, D., 1994, Public-sector capital and the productivity puzzle, The Review of Economics and Statistics, Holtz-Eakin, D., W. Newey, and H. Rosen, 1988, Estimating vector autoregressions with panel data, Econometrica, 56, Hsaio, C., 1986, Analysis of Panel Data, Cambridge University Press. Im, K., M. Pesaran, and Y. Shin, 1997, Testing for unit roots in heterogeneous panels, manuscript (Department of Applied Economics, University of Cambridge). Kao, C. and M. Chiang, 1997, On the estimation and inference of a cointegrated regression in panel data, (Department of Economics, Syracuse University) Levin, A. and C. F. Lin, 1993, Unit root tests in panel data: New results, discussion paper #93-56 (University of California at San Diego). Munnell, A. 1990, How does public infrastructure affect regional economic performance? New England Economic Review, September/October, Ratner, J., 1983, Government capital and the production function for U.S. private output, Economic Letters, Tatom, J., 1991, Public capital and private sector performance, Economic Review, Federal Reserve Bank of St. Louis, May/June, U.S. Bureau of the Census, Census of Agriculture, 1978, 1982, 1987, 1992; Census of Construction, 1977, 1982, 1987, 1992; Census of Manufactures, 1977, 1982, 1987, 1992, Census of Retail Trade, 1977, 1982, 1987, 1992, Census of Wholesale Trade, 1977, 1982, 1987, U.S. Bureau of the Census, Economic Census, U.S. Bureau of the Census, State and Local Government Finances, available at: Wasylenko, M., 1997, Taxation and economic development: The state of the economic literature, Federal Reserve Bank of Boston, New England Economic Review, March/April,

23 Table 1: Summary Statistics for State-Level Data, 50 States and Washington, D.C., Mean Maximum Minimum Standard Deviation Unemployment Rate 6.23% 18% 2.2% 2.13% Growth Rate of GSP 3.80% 21.50% % 2.81% Log Private Capital Log Public Capital Log Education Capital Log Highway Capital Log Sewer and Sanitation Capital Log Utilities Capital Log Taxes Log Unemployment Compensation

24 Table 2a: Results of Im, Pesaran, and Shin panel unit root test: Critical values for the test statistic with an intercept only are, 1%: -1.82, 5%: -1.73, 10%: Critical values for the test statistic with an intercept and linear time trend, 1%: -2.46, 5%: -2.38, 10%: Intercept Intercept and Trend AR 1 AR 2 AR 1 AR 2 Annual unemployment rate Growth rate of real GSP Unemployment compensation Total taxes Total non-residential private capital Total public capital Total education capital Total highway capital Total sewer and sanitation capital Total utilities capital Table 2b: Results of the modified Levin and Lin panel unit root test ~ N[0,1] Intercept Intercept and Trend AR 1 AR 2 AR 1 AR 2 Annual unemployment rate Growth rate of real GSP Unemployment compensation Total taxes Total non-residential private capital Total public capital Total education capital Total highway capital Total sewer and sanitation capital Total utilities capital

25 Table 3: LSDV estimation results, 50 states plus Washington, D.C. Dependent variable is the unemployment rate. (Standard errors in parentheses) (1) Current Capital Stocks Log Total Public Capital (0.355) (2) Capital Stocks lagged one period (t-1) (0.329) (3) Disaggregated Public Capital (4) Disaggregated Public Capital lagged oneperiod (t-1) Log Total Private Capital (0.484) (0.461) (0.494) (0.467) Log Highway Capital (0.427) Log Education Capital (0.370) (0.429) (0.372) Log Sewer and Sanitation Capital (0.253) (0.253) Log Utilities Capital (0.189) (0.180) Growth rate of GSP (1.945) (1.919) (1.962) (0.946) Log Total Unemployment Compensation Benefits (0.108) (0.103) (0.107) (0.102) Log Total Taxes (0.381) (0.377) (0.391) (0.385) Standard Error of Regression

26 Table 4: LSDV estimation results, 50 states plus Washington, D.C. Dependent variable is the first difference of the unemployment rate, no state specific effects (Standard errors in parentheses) (1) Total Capital Stocks Log Total Public Capital (0.844) (2) Disaggregated Public Capital Log Total Private Capital (0.612) (0.610) Log Highway Capital (1.283) Log Education Capital (1.182) Log Sewer and Sanitation Capital (0.606) Log Utilities Capital (0.526) Growth rate of GSP (1.575) Log Total Unemployment Compensation Benefits (0.119) Log Total Taxes (0.456) (1.571) (0.119) (0.459) Standard Error of Regression

27 Table 5: Elasticities, measured at the mean, from the LSDV estimation. Total Capital Disaggregated Public Capital Stocks Private Capital Public Capital Education Capital Highway Capital Sewer and Sanitation Capital Utilities Capital Growth Rate of GSP Taxes Unemployment Compensation Table 6: GMM dynamic fixed effects estimation results, 50 states plus Washington, D.C. Dependent variable is the unemployment rate, both state and time period dummies are included in the estimation. (Standard errors in parentheses) (1) Total Capital Stocks Lagged unemployment (U t-1 ) (0.030) Log Total Public Capital (0.777) (2) Disaggregated Public Capital (0.029) Log Total Private Capital (0.624) (0.610) Log Highway Capital (1.092) Log Education Capital (0.630) Log Sewer and Sanitation Capital (0.512) Log Utilities Capital (0.463) Growth rate of GSP (0.020) Log Total Unemployment Compensation Benefits (0.111) Log Total Taxes (0.471) (0.019) (0.108) (0.569) 101

28 Table 7: HNR estimation results, 50 states plus Washington, D.C. Dependent variable is the first difference of the unemployment rate, no state specific effects (Standard errors in parentheses) (1) Total Capital Stocks Log Total Public Capital (0.844) (2) Disaggregated Public Capital Log Total Private Capital (0.612) (0.610) Log Highway Capital (1.283) Log Education Capital (1.182) Log Sewer and Sanitation Capital (0.606) Log Utilities Capital (0.526) Growth rate of GSP (1.575) Log Total Unemployment Compensation Benefits (0.119) Log Total Taxes (0.456) (1.571) (0.119) (0.459) Standard Error of Regression

29 Table 8: Summary Statistics for State-Level Data, 50 States and Washington, D.C., (Natural Logs) Mean Maximum Minimum Standard Deviation EMPLOYMENT (Number) Private wage and salary employment Ag. services, forestry, fishing & other Mining Construction Manufacturing Transportation and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services AVERAGE REAL WAGE (Chained 1996 $) Private wage and salary disbursements Ag. services, forestry, fishing & other Mining Construction Manufacturing Transportation and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services PUBLIC CAPITAL (Thous. Chained 1996 $) Total Public Capital Total Education-Capital Total Highways-Capital Sewer + Sanitation Capital Total Utilities-Capital Total Current Expend (Thous. Chained 1996 $) Total Taxes (Thous. Chained 1996 $)

30 Table 9: Results of Im, Pesaran, and Shin panel unit root test: Critical values for the test statistic with an intercept only are, 1%: -1.82, 5%: -1.73, 10%: Critical values for the test statistic with an intercept and linear time trend, 1%: -2.46, 5%: -2.38, 10%: (Natural logs) Intercept Intercept and Trend EMPLOYMENT (Number) Private wage and salary employment Ag. services, forestry, fishing & other Mining Construction Manufacturing Transportation and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services AVERAGE REAL WAGE (Chained 1996 $) Private wage and salary disbursements Ag. services, forestry, fishing & other Mining Construction Manufacturing Transportation and public utilities Wholesale trade Retail trade Finance, insurance, and real estate Services PUBLIC CAPITAL (Thous. Chained 1996 $) Total Public Capital Total Education-Capital Total Highways-Capital Sewer + Sanitation Capital Total Utilities-Capital Total Current Expend (Thous. Chained 1996 $) Total Taxes (Thous. Chained 1996 $)

31 Table 10: LSDV estimation results, 50 states plus Washington, D.C. Dependent variable is the log of employment by industry (Standard errors in parentheses) Dependent variable employment in industry j. Total Private Employment Ag. Services, forestry, fishing Mining Construction Manufacturing Transportation and Utilities Wholesale Trade Retail Trade Finance, Insurance, Real Estate Services Average wage and salary in industry j (0.018) (0.027) (0.068) (0.047) (0.037) (0.021) (0.030) (0.021) (0.024) (0.017) Total Public Capital (0.024) (0.051) (0.136) (0.070) (0.048) (0.031) (0.037) (0.025) (0.040) (0.024) Total Current Expenditures (0.033) (0.069) (0.175) (0.096) (0.065) (0.042) (0.051) (0.034) (0.054) (0.032) Total Taxes (0.021) (0.044) (0.112) (0.061) (0.042) (0.027) (0.033) (0.021) (0.035) (0.021) Std. error of regression IPS test on residuals * * * * * * Table 11: HNR estimation results, 50 states plus Washington, D.C. Dependent variable is the first difference of the log of employment by industry, no state specific effects (Standard errors in parentheses) Dependent variable employment in industry j. Total Private Employment Ag. Services, forestry, fishing Mining Construction Manufacturing Transportation and Utilities Wholesale Trade Retail Trade Finance, Insurance, Real Estate Services Average wage and salary in industry j (0.109) (0.558) (1.112) (0.802) (0.308) (0.159) (0.204) (0.199) (0.217) (0.098) Total Public Capital (0.098) (0.205) (0.841) (0.507) (0.142) (0.125) (0.122) (0.100) (0.136) (0.109) Total Current Expenditures (0.104) (0.332) (1.292) (0.931) (0.185) (0.122) (0.146) (0.116) (0.140) (0.114) Total Taxes (0.183) (0.466) (1.025) (0.922) (0.205) (0.191) (0.233) (0.204) (0.277) (0.201) 105