Government spending and firms dynamics

Government spending and firms dynamics Pedro Brinca Nova SBE Miguel Homem Ferreira Nova SBE December 2nd, 2016 Francesco Franco Nova SBE Abstract Using firm level data and government demand by firm we assess the transmission mechanism of fiscal policy through the firm side. The first contribution of this paper is the matching between Compustat and Federal Procurement data, which has the value and duration of all contracts between the federal government and private firms. With this new database we evaluate the impact of government demand on firm dynamics, namely firms capital and labor decisions. We see that an increase in the value or in the duration of the government demand increases firms investment and has no impact on firms employment. We plan to assess the ability of an heterogeneous firms model with monopolistic competition in matching these stylized facts and better understand the fiscal policy transmission mechanism from the firm side. Keywords: Public expenditure, firms, heterogeneous firms JEL Classification: E22; E62; H32; H72 Nova School of Business and Economics, Universidade Nova de Lisboa, Portugal Nova School of Business and Economics, Universidade Nova de Lisboa, Portugal Nova School of Business and Economics, Universidade Nova de Lisboa, Portugal 1

1 Introduction The recent financial crisis has brought back the interest on fiscal policy multipliers and transmission mechanisms. Before the financial crisis research was more focused on monetary policy, given the belief that fiscal policy shocks took too long to impact the economy. The interest on fiscal policy was limited to understanding which type of models, New-Keynesian or Neoclassical, better replicated movements in key variables, such as real wages or consumption, in response to fiscal shocks. During the financial crisis, with monetary policy reaching the zero lower bound, the interest on fiscal policy resurged. Even though the recent focus on fiscal policy, the transmission mechanisms are not yet clear. At an empirical level the size of the multipliers are still not clear and may depend on the composition of government spending, how it is financed or country specific characteristics. At a theoretical level New-Keynesian models found it difficult to replicate the behavior of firms markups in response to fiscal shocks, while Neoclassical models were not able to achieve big fiscal multipliers so far. Fiscal policy has received greater focus in the recent years. However, the literature evaluating the response of firms dynamics to fiscal shocks is still limited. At a theoretical level the focus has been more on the households problem. Models with heterogeneous households have been explored to understand the transmission mechanisms from government spending to households decisions, such as consumption or hours worked. Although, firms response to fiscal shocks has not yet been evaluated under a framework of heterogeneous firms, to the best of our knowledge. At an empirical level, the transmission mechanisms from government spending to firms decisions has not received much focus as well. Ramey and Nekarda (2011) have addressed this question, evaluating how industries output, wages, hours worked, productivity and markups react to fiscal shocks. Some literature has equally evaluated the transmission mechanism via financial markets. Badoer and James (2016) showed an increase in government debt issuance affects the interest rate in capital markets 2

and limits firms debt issuance. An increase in interest rates in financial markets is found by Denis and Sibilikov (2010) to impact negatively firms investment. Despite the few contributions of literature addressing firms responses to fiscal shocks, there is a vast contributions evaluating the impact of fiscal shocks on aggregate variables. With respect to aggregate investment, contributions have been given by Blanchard and Perotti (2002), Perotti (2004), Ramey (2012). All papers presente evidence of a crowding out effect of government spending on private investment. To fully understand the crowding out effect found, we need to understand firms reactions to fiscal shocks and the transmission mechanism. As Gourio and Kashyap (2007) show, aggregate investment variations derive from firms investment spike. In this paper, we contribute to fill this gap in literature, addressing firms responses to a fiscal shock. Our first contribution is the matching of two different data sets: Compustat and Federal Procurement Data System. On Compustat we have firm level data for all the listed firms. On the Federal Procurement Data System we have data on all the contracts between the Federal Government and private firms. The contracts data that we have is the value, the duration, the signing date, among other observables. By matching these two data sets we can explore how do firms react to an increase in government demand, how it affects firms investment and employment decisions, if there are peer effects and more deeply understand the transmission mechanism of fiscal spending through the firm side. Empirical results indicate that whenever government demand directed to a firm increases, this firm will suffer an increase in revenues and will invest more and the longer the contract the stronger the effects on investment. No effects on employment have been found. The structural model will allow us to better understand the transmission mechanism from government demand into firms dynamics. The model will follow closely the framework by Hopenhayn (1992), whit firms heterogeneity coming from the productivity factor, and a continuum of identical households of size L s, with no 3

occupational choice. We extend the model with government sector, with the government buying firms goods, and monopolistic competition, to have a degree of substitutability between firms goods. For now, we are shutting down firms entry and exit. The structural model is still work in progress, but we expect the results of the model will validate empirical findings and shed some light on the fiscal policy transmission mechanisms. The rest of the paper is organized as follows: In section two we present the strategy to match the Federal Procurement and Compustat database. Empirical results are documented in section three. Section 4 contains the theoretical model, which is still work in progress. Section 5 concludes. 2 Compustat and Federal Procurement Data System matching The first contribution of this paper is the matching between the Compustat and the Federal Procurement data system databases. Compustat has information on all publicly listed companies in the US. It has data on investment, number of employees, revenues, debt, interest expenditures, stock market valuation, among others. Federal Procurement Data System dataset has data on the contracts made between the Federal Government and private firms. It has all the information for each contract, such as the name of the vendor company, the value of the contract, the signing date and the completion date. From Compustat we extracted, from 1998 until 2015, data for all the firms in the dataset (22207 firms). From Federal Procurement Data System we extracted all the contracts above 1.000$, since 1998 until 2015. The total number of contracts is 13.890.683. The problem with matching both datasets is that the vendor name in the Federal Procurement Data System does not match exactly the name of the firms in the Compustat dataset. If we are to match the raw data we would only have x firms 4

matched. So we try to understand some of the reasons for the names to be different in the two datasets. This first stage of the matching process consists on finding some common differences between the names of the companies in each database and eliminate these differences. Some of the most common differences between dataset are that in Compustat some company names end in corp or inc while in the federal procurement data they appear as corporation and incorporated. After eliminating this differences we create a loop to find unique matches in the Compustat dataset for the first N letters of the names of the firms on the federal procurement data system. The loop starts with N=4, we then store in a matrix all the federal procurement firms which are uniquely matched with a Compustat firm and we also store the respective Compustat identification code of the matched firm. We then eliminate the matched firms from the initial matrix and we also eliminate the firms that do not find any match at all. Then the loop restarts with N=5 and we repeat this process until we have N equal to the maximum length of a firm s name in the Federal Procurement Data System dataset. Then, we create a loop to try to eliminate wrong matches by seeing, for each previously matched firm, the maximum number of letters for which the firm continues to be matched. After finding the maximum number of letters for which a firm is matched, we set a ratio between the maximum number of letters for which the firm continues to be matched and the total number of letters in the firm s name. Then we define a threshold for this ratio, for which we accept the match. This prevents that a firm that found a unique match with 4 letters, but the fifth is not matched, to be accepted as a correct match. An example of this situation is Solers, inc matched with Solera holdings inc The two companies are different but the name matches until the fifth letter. Although the sixth letter is not matched. We define a threshold so eliminate these situations. We also create another loop to eliminate more wrong matches. This loop consists on seeing the ratio of letters between the matched names. So, it counts the number 5

of letters of the name of the company in the Federal Procurement Data System and compares it with the number of letters in the matched Compustat name. This is to prevent situations like: Nippo corporation matched with Nippon telegraph & telephone corp ntt Here the ratio of the match from the second loop would be 100% given that we dropped the corporation word, all the letters in the name Nippo are matched and from the second loop the match would be accepted. So to eliminate situations like this we create this third loop. After checking the percentage of wrong matches by randomly drawing 500 different matched companies for different thresholds for the second and third loop we decided to establish the second loop threshold at 90% and the third loop threshold at 80%. The number of matched contracts, using the two previous thresholds, is 1.335.162 for 4752 firms. The percentage of wrong matches using these two thresholds is approximately 3%. After matching the contracts, we aggregate the matched contracts per firm and per year, given that we have for several firms more than one contract per year. To decrease even more the wrong matches, we also eliminate the cases in which the government demand is higher than the total revenues of a firm. More 84 observations are dropped. 2.1 Are matched contracts representative of all contracts? Comparing the matched contracts with all the contracts in the dataset we can see that the matched contracts are representative of all the contracts in the Federal Procurement Data System. In figure 1 in appendix it is possible to see that the time series for both the sum of all matched contracts and the sum of all contracts in the dataset. The correlation between both series is 95.75%. And if we look at the growth rate of the sum of all contracts and the growth rate of the sum of the matched contracts both series are very similar (figure 2 in appendix) and the correlation between both series is 75.8%. 6

We also compare the distributions of the matched contacts and of all contracts (figure 6) as well as the correlations between the 5th, 50th and 95th percentiles of the matched contracts and the same percentiles of all the contracts (figures 3, 4 and 5). The distributions are very similar, both with fat left tails, and the correlations between the matched contracts and all contracts in each of the three percentiles is strong (88.25%, 99.48% and 93.96% for the 5th, 50th and 95th percentiles respectively). So the contracts that we match are representative of all the contracts in the dataset. 3 Results We consider only the matched contracts. When aggregating all the contracts per year and per firm we end up with 31664 observations. Although we have a strongly unbalanced panel, given that in the Compustat dataset every year new firms appear and some firms are dropped. To prevent working with an unbalanced dataset we keep only the firms that appear in the Compustat dataset in all years between 1998 and 2015. Given that we want to understand the transmission mechanism of the fiscal multipliers through the firm side we are going to evaluate the impact of variations in government demand on a firm s investment and labor decision, which are the variables in the firms problem in our structural model. We also evaluate the impact of variations of government demand on firms revenues to see if revenues increase or are crowded out. Given this, the regressions that we estimate are I ijt = β 0 Gov ijt + β 1 Gov ijt 1 + β 2 Gov ijt+1 + β 3 I ijt 1 + α i + θ j + σ t (1) Emp ijt = β 0 Gov ijt +β 1 Gov ijt 1 +β 2 Gov ijt+1 +β 3 Emp ijt 1 +α i +θ j +σ t (2) Rev ijt = β 0 Gov ijt + β 1 Gov ijt 1 + β 2 Gov ijt+1 + β 3 Rev ijt 1 + α i + θ j + σ t (3) 7

where I ijt, Emp ijt and rev ijt are the first differences of investment, employment and revenues, of firm i in industry j in year t. Gov ijt is the first difference of government demand for firm s i in year t. α i, θ j and σ t are firm, industry and year fixed effects. All the values of government spending and firms revenues and investment are in million dollars and of employment in thousands of individuals. We include as well a lag and a forward of the first difference of the government demand. The inclusion of government demand in the following period is justified by the possibility of anticipation effects. Ramey (2011) presents evidence that the majority of variations in government spending are anticipated, which may translate into firms adjusting a period before the variations in government demand happen. Including government demand in the previous period tests for the existence of adjustment costs, causing firms to delay their reaction. So, by including the government demand in the previous or in the following period, we are testing for anticipation or adjustment costs. Results are presented in tables 1, 2 and 3. Results indicate that there is a positive impact on revenues. If government demand increases by 1 million dollars, revenues in the same period increase by 0.8 million dollars. Investment results indicate that firms appear to anticipate investment. Whenever the government demand is going to increase by 1 million dollar firms invest more 0.071 million dollars in the previous period. In the period after the increase in government demand firms decrease their investment by 0.12 million dollars. Variations in government appear to almost have no impact on employment, with firms hiring 1 more employee in the period before government increases its demand by 1 million dollars, and firing one employee in the period following the increase in government demand. So firms appear to anticipate the government demand shock one period before it happens. What may also impact the firms decision is the duration of the contract. Shorter or longer contracts may create different firm dynamics. Longer contracts will create more incentives for the firms to either increase their investment or hire more employees than shorter contracts, given that with longer contracts that demand shock will 8

(1) (2) (3) (4) VARIABLES Investment ijt Investment ijt Investment ijt Investment ijt Gov dem ijt -0.003-0.001 (0.029) (0.020) Gov dem ijt 1-0.124*** -0.044 (0.033) (0.027) Gov dem ijt+1 0.071*** 0.029 (0.023) (0.020) Investment ijt 1 0.013 0.040*** 0.040*** 0.013 (0.009) (0.009) (0.009) (0.009) Observations 14,295 15,248 15,248 14,295 R-squared 0.073 0.055 0.056 0.072 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 1: Government demand and investment (1) (2) (3) (4) VARIABLES Employment ijt Employment ijt Employment ijt Employment ijt Gov dem ijt 0.000-0.000 (0.000) (0.000) Gov dem ijt 1-0.001** -0.001** (0.000) (0.000) Gov dem ijt+1 0.001*** 0.000 (0.000) (0.000) Employment ijt 1 0.055*** 0.061*** 0.061*** 0.055*** (0.008) (0.008) (0.008) (0.008) Observations 14,295 15,248 15,248 14,295 R-squared 0.182 0.181 0.181 0.181 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 2: Government demand and employment 9

(1) (2) (3) (4) VARIABLES Revenues ijt Revenues ijt Revenues ijt Revenues ijt Gov dem ijt 0.808*** 0.389*** (0.175) (0.123) Gov dem ijt 1-0.246-0.079 (0.197) (0.168) Gov dem ijt+1 0.437*** 0.151 (0.138) (0.117) Revenues ijt 1-0.062*** -0.002-0.001-0.062*** (0.009) (0.009) (0.009) (0.009) Observations 14,295 15,248 15,248 14,295 R-squared 0.170 0.139 0.139 0.169 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 3: Government demand and revenues be more persistent than with shorter contracts. To account for the effect of the duration and value of the contracts we create an instrument that weights the value of each contract by its duration. The instrument is given by WGD ijt = k contract kijt duration kijt #contracts ijt (4) where WGD kijt is the instrument weighted average government demand for firm i, in industry j in year t. contract kijt is the value of contract k for firm i in industry j in year t, and duration kijt is the duration of the respective contract. #contracts ijt is the number of contracts between firm i and the government in year t. With this instrument, we capture not only the impact of variations in the value of the contract but also the impact of variations in the duration. The regressions we estimate to see the impact on investment, employment and revenues are 10

I ijt = β 0 WGD ijt + β 1 WGD ijt 1 + β 2 WGD ijt+1 + β 3 I ijt 1 + α i + θ j + σ t (5) Emp ijt = β 0 WGD ijt +β 1 WGD ijt 1 +β 2 WGD ijt+1 +β 3 Emp ijt 1 +α i +θ j +σ t (6) Rev ijt = β 0 WGD ijt + β 1 WGD ijt 1 + β 2 WGD ijt+1 + β 3 Rev ijt 1 + α i + θ j + σ t (7) Results of these equations are presented in tables 4, 5 and 6. Results indicate that the higher the value and the longer the contract the more firms will invest, the higher the revenues and no impact on employment decisions. It is curious to notice that now that we weight the government demand by the duration of the contracts firms will take more time to adjust in terms of investment, increasing their investment only in the year after the year in which the contract was signed. Interesting is also the fact that now revenues do not increase only in the year the contract is signed, but increase also in the following year. (1) (2) (3) (4) VARIABLES Investment ijt Investment ijt Investment ijt Investment ijt Gov dem duration ijt -0.469-0.516 (0.411) (0.419) Gov dem duration ijt 1 0.700* 0.717* (0.412) (0.420) Gov dem duration ijt+1 0.089 0.069 (0.410) (0.410) Investment ijt 1 0.013 0.040*** 0.040*** 0.013 (0.009) (0.009) (0.009) (0.009) Observations 14,295 15,248 15,248 14,295 R-squared 0.072 0.056 0.056 0.071 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 4: WGD instrument and investment 11

(1) (2) (3) (4) VARIABLES Employment ijt Employment ijt Employment ijt Employment ijt Gov dem duration ijt -0.003-0.004 (0.006) (0.006) Gov dem duration ijt 1-0.004-0.005 (0.006) (0.006) Gov dem duration ijt+1-0.000 0.000 (0.006) (0.006) Employment ijt 1 0.055*** 0.061*** 0.061*** 0.055*** (0.008) (0.008) (0.008) (0.008) Observations 14,295 15,248 15,248 14,295 R-squared 0.181 0.181 0.181 0.181 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 5: WGD instrument and employment (1) (2) (3) (4) VARIABLES Revenues ijt Revenues ijt Revenues ijt Revenues ijt Gov dem duration ijt 4.255* 4.571* (2.457) (2.561) Gov dem duration ijt 1 4.445* 4.564* (2.464) (2.565) Gov dem duration ijt+1-2.824-3.214 (2.456) (2.451) Revenues ijt 1-0.062*** -0.001-0.002-0.062*** (0.009) (0.009) (0.009) (0.009) Observations 14,295 15,248 15,248 14,295 R-squared 0.169 0.139 0.139 0.169 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 6: WGD instrument and revenues 12

We also evaluate if the number of contracts signed between a firm and the Federal government affects the firm s dynamics. This measure may translate the government preference for a firm s products. The more contracts signed the higher the government preference. To evaluate if government preferences affect firms dynamics we consider the following specifications I ijt = β 0 #contracts ijt + β 1 I ijt 1 + α i + θ j + σ t (8) Emp ijt = β 0 #contracts ijt + β 1 Emp ijt 1 + α i + θ j + σ t (9) Rev ijt = β 0 #contracts ijt + β 1 Rev ijt 1 + α i + θ j + σ t (10) Results are presented in tables 7, 8 and 9. Results indicate that an increase by 1 in contracts signed between a firm and the federal government increases investment in 0.20 million dollars and revenues in 1.47 million dollars. Employment again is not affected. (1) VARIABLES Investment ijt Number of contracts ijt 0.195*** (0.062) Investment ijt 1 0.039*** (0.009) Observations 15,248 R-squared 0.056 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 7: Number of contracts and investment 13

(1) VARIABLES Employment ijt Number of contracts ijt 0.001 (0.001) Employment ijt 1 0.061*** (0.008) Observations 15,248 R-squared 0.181 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 8: Number of contracts and employment (1) VARIABLES Revenues ijt Number of contracts ijt 1.469*** (0.379) Revenues ijt 1-0.002 (0.009) Observations 15,248 R-squared 0.140 Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 Table 9: Number of contracts and revenues 14

4 Structural model The stylized facts that we have from this empirical exercise are that government demand has a positive impact on firm s revenues and investment, with longer contracts having a stronger impact on both variables. With the theoretical model, we want to validate the empirical results and more deeply understand the transmission mechanism from government demand to firms dynamics. We use an incompletes market with heterogeneous firms model to capture this transmission mechanism. 4.1 Perfect competition We start with a simple model with perfect competition. We then introduce monopolistic competition in order to distinguish the firms products so that the government has different preferences for each product in the market. This way, by changing the government preferences we change the demand of the government for a firm s products. 4.1.1 Households The economy is populated with a continuum of identical households of size L s. Each household maximizes is expected lifetime utility t=i β i c1 γ t+i 1 1 γ, (11) where 0< β <1 is the discount factor and c t is consumption at time t subject to the following budget constraint a t+1 = w t + a t (1 + r t ) + Π t c t (12) where a t is riskless bond, w t is the cost of labor, r t the return on the bond and Π t are total profits by firms. We assume that each household holds the market portfolio and that they supply labor inelastically. 15

4.1.2 Firms We assume there exists a large number of firms. Each firm is characterized by a stochastic productivity term z. We assume that z is a Markov chain z [z 1,..., z N ], where P r (z t+1 = z j z t = z i ) π ij function is then given by 0, and N j=1 π ij = 1. Firms production y t = z t (k α t l 1 α t ) θ (13) with 0< θ <1 and 0< α <1, with l t and k t being the labor and capital hired by the firm. Given the idiosyncratic productivity z t and the capital in place k t, the employment choice is the solution for π t = max z t (kt α lt 1 α ) θ w t l t (14) l t The value of a firm with productivity z i at time t is given by { v(z it, k t ) = max π t x t + R t+1 x t,k t+1 N } π ij v(z jt+1, k t+1 ) j=1 (15) s.t. k t+1 = k t (1 δ) + x t 4.1.3 Recursive Competitive equilibrium Households first order conditions, given no aggregate risk, is c γ t = β(1 + r t+1 )c γ t+1 (16) which yields a discount factor of R t+1 = with private assets in zero net supply 1 = β c γ 1 + r t+1 t+1 c γ t (17) a t = 0 (18) 16

For a given wage level w and a given interest rate R the capital and labor choices of firm i with productivity z i in period t are given by [ N k it = π ij z j=1 ] 1 θ(1 α) 1 1 θ(1 α) it 1 θ ( 1 Rt (1 δ) αθr t ) 1 θ(1 α) θ 1 [ w t (1 α)θ ] θ(1 α) θ 1 (19) l it = k which we can simplify as αθ 1 θ(1 α) it [ ] 1 w θ(1 α) 1 1 t z 1 θ(1 α) it (20) (1 α)θ k it = A it 1 κ(r t, w t ) (21) l lt = B lt 1 λ(r t, w t ) (22) Let p i be the time-invariant distribution of firms over idiosyncratic productivity values. The aggregate capital and labor demand in the economy will be given by K t+1 = κ(r t+1, w t+1 ) p i A it (23) with the market clearing conditions being L d t = λ(r t, w t ) p i B it 1 (24) L s t = L d t (25) and the overall production of the economy being C t = Y t K t+1 + K t (1 δ) (26) Y t = i p i y it = i p i z it (A it 1 κ(r t, w t )) αθ (B it 1 λ(r t, w t )) (1 α)θ (27) 17

4.2 Monopolistic competition To introduce government demand directed to a specific firm, we extend the previous model with monopolistic competition. This way we can have a government preference matrix that will dictate the government demand directed to each firm. 4.2.1 Households The households problem is similar to the model with perfect competition. The only difference is that now total consumption will be given by ( 1 C = 0 ) 1 c t (s) 1 ɛ 1 ɛ ds which is equivalent to having the demand for each product given by (28) ( ) ɛ pt (s) c t (s) = C t (29) Given this, the households budget constraint is now given by P t a t+1 = w t + a t (1 + r t ) + Π t p t (s)c t (s)ds T t (30) where T t are lump sum taxes. Given that we now have prices in our model, the only difference between the households new first order condition and the one with perfect competition is the incorporation of inflation c γ t = β(1 + r t+1 )c γ t+1 P t P t+1 (31) 4.2.2 Government The government will run a budget balance. All the government revenues come from the lump sum taxes on households. The government then uses all the revenues to buy firms products. Total government demand is given by ( 1 G = 0 ) 1 (ω t (s)g t (s)) 1 ɛ 1 ɛ ds (32) 18

and the government demand by good is given by ( ) ɛ pt (s) g t (s)ω t (s) = G t (33) with ω t (s) being the weight the government gives to good s. ω t (s) represents the preferences of the government towards good s and we use this matrix to change the government demand directed to a firm s product. P t 4.3 Firms The demand the firms face now is the sum of households and government demand. So D = C + G and with monopolistic competition the firms problem now becomes max p t y t w t l t l t ( ) ɛ pt (s) s.t y t (s) = D t P t (34) y t (s) = z t (kt α lt 1 α ) θ Given the new profit function, the value function of the firm will stay the same. 5 Simulations 6 Conclusions 19

References Badoer, D. C. and James, C. M., 2016. The determinants of long-term corporate debt issuances, The Journal of Finance, volume 71, issue 1, 457-492. Beck, T., Demirgüç-Kunt, A. and Maksimovic, V., 2005. Financial and Legal Constraints to Growth: Does Firm Size Matter?, The Journal of Finance Volume LX, No. 1, 137-177. Blanchard, O. and Perotti, R., 2002. An empirical characterization of the dynamic effects of changes in government spending and taxes on output, Quarterly Journal of Economics, volume 117, issue 4, 1329-1368. Denis, J. D. and Sibilikov, V., 2010. Financial constraints, investment, and the value of cash holdings, Review of Financial Studies, volume 23, issue 1, 247-269. Gourio, F. and Kashyap, A. K., 2007. Investment spikes: New facts and a general equilibrium exploration, Journal of Monetary Economics, volume 54, supplement 1, 1-22. Graham, J. R., Leary, M. T., and Roberts, M. R., 2014. How does government borrowing affect corporate financing and investment? NBER working paper. Hopenhayn, H. A., 1992. Entry, exit and firm dynamics in long run equilibrium, Econometrica, volume 60, No.5, 1127-1150. Pastor, L. and Veronesi, P., 2003. Stock valuation and learning about profitability, Journal of Finance 58, 1749-1789. Perotti, R., 2004. Estimating the effects of fiscal policy in OECD countries, IGIER. Ramey, V. A. and Nekarda, C., 2011. Industry evidence on the effects of government spending, American Economic Journal: Macroeconomics 3, 36-59. Ramey, V. A., 2011. Identifying government spending shocks: It s all in the 20

timing, Quarterly Journal of Economics volume 126, issue 1, 1-50. Ramey, V. A., 2012. Government spending and private activity, NBER working paper No.17787. 21

7 Appendix Figure 1: Time series of both the sum of all matched contracts and of the sum of all contracts Figure 2: Growth rate of all contracts and growth rate of matched contracts 22

Figure 3: Time series of the 5th percentile of matched contracts and of the same percentile of all contracts Figure 4: Time series of the 50th percentile of matched contracts and of the same percentile of all contracts 23

Figure 5: Time series of the 95th percentile of matched contracts and of the same percentile of all contracts Figure 6: Distribution of all contracts 24