Letting Luck Decide: Government Procurement and the Growth of Small Firms

Abstract I estimate the causal effects of demand shocks, stemming from government procurement, on the growth of small firms in Ecuador. I assemble a unique dataset using several new administrative sources and, as identification strategy, exploit a governmental procurement process that allocates public contracts through a randomised contest. I find a positive and significant effect of demand shocks on firm growth. On average, an increase in demand of 10 per cent will increase wage expenses and fixed assets by approximately 5 per cent during the year of the shock. I also find no evidence of spillover effects from demand shocks on sales to the public or private sector. Finally, as in other studies, I show that demand positively impacts firm growth but, contrary to other findings, this effect is temporary and only observed during the year of the shock.


Introduction
Small and medium-sized firms (hereafter SMEs) contribute up to 45 per cent of total employment and 33 per cent of GDP in developing countries (Kushnir, Mirmulstein, & Ramalho, 2010). Despite this, the majority of small firms never grow beyond a few employees (Nichter & Goldmark, 2009). The importance of firm growth for economic and political reasons is evidenced by the number of public policies that have been created to promote it.
Economic theory provides two different approaches to explain firm growth. On one hand, firms can grow due to intrinsic factors such as managerial ability (Lucas, 1978), increases in productivity (Jovanovic, 1982), and experience (Hopenhayn, 1992). Public policies meant to address intrinsic factors include access to credit, management development programmes, and financial literacy programmes. On the other hand, a set of recent papers suggest that demand factors, such as networking and reputation effects, might be equally important in explaining firm growth (Fishman & Rob, 2003;Syverson, 2004). In such cases, public policies that restrict competition and favour small enterprises might positively impact the development SMEs. Argentina's Ley 25.551 (2001) stipulates that goods provided by small firms receive a price margin of seven per cent; in Brazil, government purchases that are below a minimum threshold are exclusively destined to SMEs (Lei Complementar N.123, 2006). The restriction of government procurement processes to certain, by assumption, less competitive firms implies that such programmes have an opportunity cost. Are these demand-driven programmes effective in promoting the growth of SMEs?
To empirically evaluate the effects of demand, the researcher needs to isolate it from other factors. This is a complicated prospect because the relation between demand and growth is unclear. A firm may experience growth due to a shift of the demand curve induced by, for example, changes in preferences or exogenous price increases of substitute products. Conversely, a firm that grows may benefit from increased market exposure and economies of scale, leading to an increase in demand. To overcome such identification problems, previous studies have relied on firm-level price data that allows to decompose demand from productivity shocks (Foster, Haltiwanger, & Syverson, 2008). When such detailed information is not available, researchers impose a structure on the demand and production functions and obtain estimates of unobserved demand shocks through the regression residuals (Pozzi & Schivardi, 2016). Hebous and Zimmermann (2016) exploit the timing of public government contracts and estimate that a one dollar increase in government purchases increases the capital investment of U.S. firms by 7 to 11 cents. Ferraz, Finan, and Szerman (2015), use a quasiexperimental design based on the public bidding process in Brazil and find that winning a contract increases firm growth by 2.2 per cent during the quarter the contract is awarded.
In this study, I examine the short-and long-term impacts that demand shocks, stemming from government purchases, have on the financial performance of SMEs. For this purpose, I exploit the menor cuantia process, a feature in Ecuador's public procurement law that awards contracts using a lottery. Using this process as a source of variation in firm demand, I assemble a unique dataset that combines firm-level financial information with public records for 1,179 firms that participated in the process for the years 2009-2012. I compare the changes in balance sheet indicators between the winners and the runners-up of the contests, at the extensive and intensive margins.
I find that demand shocks significantly affect firms' short-term growth during the year of the shock. Firms that won a contract report, on average, 22 per cent higher revenues and current assets, and 7 per cent higher fixed assets than firms that did not win. The intensive margin analysis suggests that increasing demand by 10 per cent will increase wage expenses and fixed assets by approximately 5 per cent. The effects of demand shocks are temporary and are only observed during the year of the shock. A year after winning a contract, gross revenues and current assets revert back to pre-shock levels, and there are no differences in wage expenses and fixed assets between winners and runnersup of the contests. Moreover, I find that, outside the menor cuantia process, there are virtually no differences in sales to the government or the private sector between these two groups.
This paper contributes to the literature examining how demand shocks affect firm growth and the specific role of government demand on firm dynamics and market outcomes (Czarnitzki, Hünermund, & Moshgbar, 2018;Evenett & Hoekman, 2005;Ferraz et al., 2015;Hoekman & Sanfilippo, 2018;Lee, 2017). The main contribution of this paper is that it highlights that the magnitude, nature, and duration of the shock are important factors to consider when analysing how demand affects firm growth. Shocks that are perceived as temporary or unsustainable only affect short-term measures of growth.
This study also contributes to the literature that examines how firms adjust their factors of production to meet changes in demand and the effect that such adjustments have on firm growth. Seen through the lens of this literature, this study highlights the importance of understanding a firms' capacity constraints and stock and utilisation of factors of production to identify the mechanisms that affect firm growth (Hart, 2017;Kruse, 1993;Kruse, Freeman, & Blasi, 2010;Long & Fang, 2013).
The rest of this paper is divided as follows: Section 2 explains the country context; Section 3 introduces the data; Section 4 discusses the empirical methodology; Section 5 provides the results; and Section 6 concludes.

Public procurement in ecuador
Ecuador is a middle-income country with a population in 2018 of 17 million people, a dollarized economy and a per capita income of approximatively $6,200. Prior to the year 2006, the country experienced political instability and a severe financial crisis. After the 2006 election, the new government vowed to restore public trust. As part of this plan, it enacted a new constitution, transparency laws and, in 2008, the Public Procurement Law (LONSCP, 2008). The Law reformed the procedures for the purchase of public goods and introduced provisions to safeguard the participation of SMEs in public procurement. The National Public Purchases Agency defines SMEs as a firm that has less than 100 employees and has sales lower than 2 million dollars (SERCOP, 2015).
The reform required that all government institutions procure all purchases through the on-line portal Compraspublicas. 1 Before Compraspublicas, government procurement was performed locally, with limited oversight and accountability. The Law stipulated that the process for the procurement of public works under a threshold, precisely 0.0007 per cent of the government's budget (around $150 thousand for the years 2009-2012), had to be done under the menor cuantia ('small amount') process. This process contains two distinct features that are particularly relevant to this study: it is accessible only to SMEs and grants contracts through a randomised lottery.
The menor cuantia process functions through Compraspublicas. The portal connects institutions that procure for services and products (hence projects) with firms, that bid for them. In order for a firm to bid on a project, it must first register in the portal. During this process, firms submit their personal and company information including contact information, professional degrees, and tax data. Once registered, firms are able to browse through the public contracts available and bid for contracts.
From the institution's side, the first step to procure a new public work is to create an entry in Compraspublicas. 2 The project has to include a description of the public work, location, budget, timeline, and project specific requirements. These requirements include experience, qualifications, financial capital, technical abilities, and machinery.
After this step, the project enters into its first phase, acceptance of bids from firms. There are two ways used to notify firms of a new project. First, the system sends automatic notifications to providers. It does so through an algorithm that compares the requirements listed in the project with the competencies listed in the profile of providers. Additionally, the system posts the project to the database of the portal. During this stage, all registered providers are able to search and browse through the available projects and express their interest.
In the second phase of the process, all providers that bid on the project must provide proof that they fulfil the requirements specified. They do this by uploading official documentation to Compraspublicas. For instance, if the project requires specific machinery, then providers must upload the registration and proof of purchase of the equipment. A notable feature of this part of the process is that the requirements for each public work are objective and, in some cases, the system does not allow the provider to complete this phase if they do not meet the minimum cut-offs.
Following this phase, a committee evaluates all the providers that presented a bid. The committee's responsibility is to identify if providers meet the minimum requirements for the project; thus supplementing the system's verification process. To illustrate, suppose that a new construction project requires a minimum of 2 years of previous experience. An interior design firm could, theoretically, qualify for this process. In this case, it is the role of the committee to verify if the experience listed by the firm is relevant. The committee does not rank nor provide a numerical qualification of providers, it only determines if they are qualified to perform the project. The providers that qualify enter into a list. In the final phase of the process, the system automatically and randomly selects one provider from the list. This provider is the winner of the contest and is legally required to perform the project.
The identification strategy in this study relies on the random allocation of contracts. For a given contract, all providers that qualify to participate in the lottery have, on average, comparable characteristics. The impartiality of the process is ultimately an empirical question, addressed in the empirical section, where it is concluded that menor cuantia projects are, indeed, randomly assigned. Moreover, and regardless of any empirical considerations, there are two major features of the process that suggest that contracts are assigned randomly.
First, no negotiation between institutions and firms takes place at any stage of the process. The price for a given public work is predetermined by the procuring agency and, as a result, no preference is given for one bid being more competitive than another. This is evidenced by comparing the Demand shocks and the growth of SMEs 1265 budgeted and actual costs for a given project. In the menor cuantia process, these values always coincide. In public work projects of higher amounts, which are allocated using different procedures, one can observe considerable variations between the estimated and actual costs. Second, the requirements that are set for each contract are defined in terms of objective criteria and must be verified by legal documents. 3

Data
The data for this study consist of a panel of 1,179 firms that presented bids on a total of 5,475 public works performed under the menor cuantia process between 2009 and 2012. Firm-level data were obtained from the National Bureau of Companies of Ecuador (SUPERCIAS) and include contact information, yearly tax returns, and balance sheet information. 4 Data of public works performed under the menor cuantia process come from the Ecuadorian Procurement Agency (SERCOP) and include contract information for each public work, the unique identification number of each firm that bid on each project, a list of qualified providers, and the winner of the contest. The Supplementary Materials document provides a comprehensive overview of the data gathering process.
The breakdown of qualified firms by year is as follows: 146 in 2009, 543 in 2010, 543 in 2011, and 546 in 2012. Table 1 presents descriptive statistics for the firms in the sample, consisting principally of SMEs in the construction industry. 5 Firms in the sample are young, the average age (years since registration) is 5.1 years. Ninety per cent of firms in the sample are less than 13 years old. For the period 2009-2012, each firm qualified to be part of the random drawing an average of 5.41 times per year, winning a contract, on average, 0.80 times per year. Financially, firms report to have average total assets of $128,589 and average liabilities of $98,202. The average wage expenditure is $25,931 and 90 per cent of firms report less than $60,000 in wage expenditures. Geographically, 55 per cent of the firms in the sample are located in the 10 most populous cities in Ecuador, where approximately 50 per cent of the total population lives. Table 2 provides the description of the public works. The average contract amount is $50,160, approximately 70 per cent of contracts are below $60,000, and 90 per cent of contracts last less than 96 days. The average contract has 6 requirements with 17 qualified providers.

Empirical strategy
The purpose of this study is to estimate the causal effects of demand shocks on firm growth. I use four different measures of growth: gross revenues, wage expenses and fixed and current assets. 6 Assume that the relationship between firm growth and demand can be represented by the following reduced form model: Notes: Descriptive statistics of 1,179 registered firms participating in the menor cuantia process for the years in the sample. Values are arithmetic averages. Income, assets, liabilities, and wage expense are presented in U.S. dollars. Assets (liabilities) include fixed and current assets (liabilities). The source of the data is the balance sheet reports presented by firms to the tax authority. P-values *p < .1; **p < .05; ***p < .01.
where _ y it denotes the growth of firm i during period t, d it is the demand faced by the firm during year t, X it is a matrix of firm-specific covariates, μ i denotes unobserved time-invariant firm characteristics, and it is the error term. I define _ y it to be the difference in logs: _ y it ¼ lnðy it Þ À lnðy itÀ1 Þ " y 2 f gross revenues; wage; and fixed and current assetsg. Estimating this model by ordinary least squares will yield biased results if the demand faced by the firm is correlated with unobserved firm characteristics, μ i , which is likely the case.
To eliminate μ i , one could transform the model by first differencing it. Even though this transformation eliminates μ i , estimating the differenced model by OLS will provide a biased estimate if changes in demand are correlated with time variant unobserved firm characteristics, E½Δ it ; Δd it Þ0.
The increase in demand caused by winning a menor cuantia contest provides the source of exogenous variation needed to obtain unbiased estimates. Conditionally on qualifying, the random nature of the lottery ensures that the contract allocation is independent of firm-specific characteristics. Firms that did not win the contract serve as an appropriate control group to obtain the effects of demand shocks on growth.
There are two main concerns with using the contracts allocated under menor cuantia as an exogenous source of demand. The first concern is that the lottery may not be random. This would occur if companies or the public institutions were able to manipulate the system. The second concern is participation. Firms can submit bids for multiple projects in a given year. To participate in a lottery, each firm must qualify to enter into the pool. If more productive firms qualify for more contests, then the probability of winning under the process increases. In this case, even if contracts are allocated randomly, they are not exogenous to firm-level characteristics.
The randomness of the contest can be tested empirically. The probability of winning a contest at time t should be orthogonal to any firm-level characteristics observed at time t À 1. Table 3 shows the results of a difference in means for the firms that qualified for the public contests during the years 2009-2012. The results (additional exercises are presented Table A1 and Table A2 of the Appendix A) suggest that there are no significant differences between winners and runners-up, suggesting that the contract allocation is in fact random.
I proceed in two steps. First, I estimate Equation (2) on the extensive margin, by comparing winners of the contests with those that did not win. In this specification _ y it is the measure of growth for company i at time t, d it equals 1 if the firm won a contract during year t and 0 otherwise, and X it represents firm-specific controls. I include as controls the age and location of the firm, geographic characteristics, and regional GDP. All specifications control for time and region fixed effects.
In the second step, I estimate the effect of demand shocks on the intensive margin. I estimate Equation (2) defining d it to be the log of sales from menor cuantia. The coefficient β 1 shows how per cent changes in exogenous demand affect different measures of firm growth. To estimate if demand shocks have an effect beyond the year of the shock, I look at growth at different time intervals, _ y itþi " i 2 f1; 2g. What does _ y it measure? During the year of the shock, _ y it shows the difference in growth between winners and runners-up, with t À 1 being the year of reference. A priori, one would expect to see significant differences in measures of growth between winners and runners-up. This is because winning an additional contract directly impacts balance sheet indicators, such as sales and current asset during the year that the shock occurs. Nonetheless, it is still plausible to observe no differences between participants of the contest during the year of shock. For instance, if firms were capacity constrained and could only perform a limited number of contracts on a given year, then firms that win contracts from menor cuantia will not be able to perform additional work. Analogously, firms that did not win the contest could seek work in the private sector. Under this scenario, firms replace private contracts with public ones, causing no overall changes in the total amount of work performed. Table 4 presents the extensive margin effects of demand shocks on growth during the year of the shock. The independent variable winner takes the value 1 if a firm won a contest at time t and 0 otherwise. Each specification controls for time and region fixed effects and clusters errors at the firm- level. The dependent variable in columns 1 and 2 is revenue growth. Firms that experienced a demand shock report, on average, approximately 22 per cent higher revenues than firms that did not experience a shock. The coefficient of .202 is significant at the one per cent level and is robust to the addition of controls. Columns 3 and 4 present the results for growth of wage expense. The estimated coefficients suggest that firms that win a contract spend, on average, 5 per cent more on wages than non-winners. These results, however, are not robust to the inclusion of additional controls. Columns 4 and 5 report the results on growth of fixed assets. Firms that win a contract report, on average, 7 per cent higher fixed-assets than non-winners. Columns 7 and 8 report the results on current assets. The coefficients are significant at the one per cent level and similar in magnitude to the coefficients estimated for growth of revenues. Overall the results from Table 4 suggest that demand shocks affect firm growth in two distinct manners. For immediate measures of growth, such as revenues or current assets, there is a direct relationship between demand shocks and growth. To illustrate, given that the average yearly revenue of a firm for the sample is $269,230, the estimated coefficient on revenue suggests that winning a contest increases the measure by approximately $ 59,000, which is close to the average value of a menor cuantia contract ($50,160). At the same time, the results show that for other measures of growth, such as wages and fixed assets, this relationship, while positive, has a lower a magnitude and statistical significance. Table 5 presents the effects of demand shocks on growth at the intensive margin, where the independent variable is the log of total yearly revenue received from menor cuantia. Columns 1 and 2 show the results for revenue growth, suggesting that an increase of 10 per cent in sales will increase declared revenue by approximately the same amount. While ostensibly trivial, this result provides a good indication that the financial statements used in this study are a reliable source to measure the financial performance of firms. Columns 3 and 4 present the results for the growth of wage expense. The estimated coefficient of 0.05 is significant at the one per cent level and does not change with the addition of controls. This suggests that an increase of 10 per cent in the demand will increase wage expenses by 5 per cent. Columns 5 and 6 present the results on growth of fixed assets, the coefficients suggest that an increase of 10 per cent in the demand will Notes: Least squares estimation of the effects of winning a procurement contract on firm growth. The dependent variables are growth (log differences) of revenue (columns 1 and 2), wage expense (columns 3 and 4), fixed assets (columns 5 and 6), and current assets (columns 7 and 8). The variable winner is a dummy variable taking the value 1 if a firm won a contest at time t and 0 otherwise. Standard errors (in parentheses) are clustered at the firm-level. Age of a firm is reported in years. Contest participated refers to the numbers of contests that a firm qualified for during a given year. The size of a firm is a set of dummies that control for the size (as defined by the National Bureau of Companies of Ecuador) of the firm. The regional controls include local GDP and construction permits issued during the year. P-values *p < .1; **p < .05; ***p < .01.

Results
increase wage expenses by 4 per cent. Columns 7 and 8 report the results of current assets and suggest that an increase of 10 per cent in the demand will increase current assets by 22 per cent. Overall, the results from the intensive margin analysis are similar in magnitude and significance to the ones presented in Table 4. Next, I examine the duration of the effects. This is of particular relevance given that the changes observed could be due to short-term reasons such as hiring more labour to fulfil the contract or renting machinery required for a project. Figure 1 shows the differences in growth rates between firms that won a menor cuantia contract and those that did not win. The differences are shown for the first 2 years after the contest. The figure shows the coefficient for growth estimated using Equation 2 with the 95 per cent confidence interval. The dependent variable is the growth rate in gross revenues, wage expense, and fixed and current assets. The figure reveals two significant insights. First, the year after the shock, winners of the menor cuantia contest experience a decrease in gross revenues and current assets. The decrease the year after the shock is similar in magnitude than the increase experienced the year of the shock. No effect is observed the year after the shock for labour costs and fixed assets. Second, no statistically significant effects in any measure of growth are observed two years after the shock.
One non-pecuniary benefit of winning a contract is that it gives firms experience, reputation, and contacts in the public sector. In this case, it is possible for winning firms to increase their sales to the government outside of the menor cuantia process. Table 6 provides the results of testing the difference in means of the sales to the government between the winners and runners-up. There are virtually no differences in sales to the government after the year of the shock. Moreover, I find that, outside the menor cuantia process, there are virtually no differences in sales to the government or the private sector between these two groups.
I perform several robustness checks to examine the sensitivity of the results. First, I estimate the results looking at each year individually. Second, I use an alternative definition of growth. Third, I estimate the results defining the dependent variable in levels instead of growth. The results are not affected by the use of these alternative specifications. Notes: Least squares estimation of the effects of winning a procurement contract on firm growth. The dependent variables are the growth (log differences) of revenue (columns 1 and 2), wage expense (columns 3 and 4), fixed assets (columns 5 and 6), and current assets (columns 7 and 8). The variable revenue from menor cuantia is the log of revenues obtained from the menor cuantia contest. Standard errors (in parentheses) are clustered at the firm-level. Age of a firm is reported in years. Contest participated refers to the numbers of contests that a firm qualified for during the year. The size of a firm is a set of dummies that control for the size (as defined by the bureau of companies of Ecuador) of the firm. The regional controls include local GDP and construction permits issued during the year. P-values *p < .1; **p < .05; ***p < .01.

Discussion and conclusions
In this paper, I estimate the causal effects of demand shocks on firm growth using as a source of exogenous variation the shocks from the menor cuantia process. I find that, in the short term, demand shocks significantly affect firm growth. Firms that win the contest report higher revenues and fixed assets and spend more on wages and short-term assets than those that did not win.  The following table presents the results from a Student t-test difference in means. The term 'Sales-Winners' and 'Sales-Runners-Up' refer to all government sales outside of the menor cuantia process for firms that won and lost in the menor cuantia process, respectively. The column 'Difference' denotes the differences in sales between winners and runners-up. P-values *p < .1; **p < .05; ***p < .01.
Demand shocks and the growth of SMEs 1271 What can explain the short-term results? Due to data limitations, I cannot empirically test alternative hypotheses, however, the findings are consistent with three potential explanations. The first is that government contracts are more attractive (more profitable, less risky) than private contracts (Duggan, Starc, & Vabson, 2016). In this case, firms have incentives to substitute private with public contracts to increase their profitability. Ecuadorian law requires firms to share 15 per cent of profits with employees, a fact that explains the increase in wages (Ministerio del Trabajo del Ecuador, 2017). The profit sharing scheme may also contribute to firms investing in more capital.
A complementary explanation is that winning a government contract relaxes the binding liquidity or credit constraints that firms face. For instance, if firms can use a government contract as collateral to qualify for a loan; then, they could use those funds to purchase new capital or train their existing employees.
A final explanation is that firms face capacity constraints (no economic slack) and must adjust their factors of production to meet the additional demand at the intensive margin or extensive margin (Hart, 2017;Weitzman, 1984). This may entail increasing the utilisation of workers and capital or hiring more workers and purchasing more capital. These adjustments may increase expenses due to the costs associated with procedures such as acquiring new machinery, payment of overtime wages, or training employees to use new tools.
While the short-term results are consistent with Hebous and Zimmermann (2016), Lee (2017) and Ferraz et al. (2015) and in this study I find no evidence of an increase in growth in the years following the shock. The reason for this may be a direct consequence of the menor cuantia process which, in contrast to the aforementioned studies, provides small monetary contracts and focuses on SMEs. As discussed in Evenett and Hoekman (2005), procurement discrimination may have no effect on long-run outcomes in markets that are competitive or have low barriers to entry such as menor cuantia. Moreover, the random nature of the menor cuantia may create a demand that is perceived as unsustainable and reduce the incentives to invest in long-term assets or permanent workers.
There are important limitations concerning the generalisation of these results to other industries and countries. During the sample period, 2009-2012, the construction sector in Ecuador experienced strong growth due primarily to heavy government investment (Instituto Nacional de Estadística y Censos (INEC), 2017). In addition, Ecuador's employee protections for short-term employment are flexible; fixed-term contracts are used for permanent tasks; there are no limits on length and renewal of fixed-term contracts; and there are no notice periods for redundancy dismissal (World Bank, 2019). For these reasons, further studies are needed to understand how the nature, magnitude, and duration of the demand shocks impact the long-term growth of SMEs. Although this study contributes to the nascent evidence that demand shocks can have a positive impact on firms, it also cautions that certain policies aimed at fostering firm growth might instead sustain firms characterised by 'businesses with low-growth intentions and a lack of interest in innovating' (Acs, Åstebro, Audretsch, & Robinson, 2016). Notes 1. www.compraspublicas.gob.ec. 2. Each project must be approved in the government budgetary process. This process is done during the previous fiscal year. 3. A potential concern is the committee's discretion to qualify providers. A committee might try to provide preferential treatment to a firm by being stringent in their review of other firms and thus limiting the number of qualified providers. To overcome this potential limitation, I exclude from the sample a firm if, during any contest, it was the only one qualified into the pool. 4. All values are obtained from firms' balance sheet documents, as reported to the tax authority (Servicio de Rentas Internas). 5. According to SERCOP, A micro firm has between 1 and 9 employees and gross sales and assets of less than $100,000.
A small firm has between 10 and 49 employees and sales and assets between $100,000 and 1 million dollars. A medium firm has between 50 and 99 employees and sales between 1 and 2 million dollars. 6. For revenues I use total sales; for wages, I use the total expenditure on salaries, wages, and commissions; for fixed and current assets I use the definition as stated in the International Financial Reporting Standards (IFRS).