Stunted Firms: The Long-Term Impacts of Colonial Taxation∗ Gabriel Natividad Universidad de Piura† 17 November 2017

Abstract Using the geocoded 2008 census of Peruvian firms, I study how the colonial mita forced-labor system (1573–1812) impacts firm investments. Multidimensional regression discontinuity models show lower levels of fixed assets and inventories, less likely use of a commercial name, and less likely registration with the tax authority for firms within mita boundaries. Local population impacts in mita regions — more school attendance, lower insurance, and no different house wealth — are consistent with attitudes towards lower business capital accumulation and more tax avoidance. A centuries-long disadvantageous tax burden in mita regions may explain the deeply rooted causes of firm stunting.

∗

I am very grateful to Lourdes Alvarez and the Ministry of Production of Peru for access to data. I would also like to thank Lu´ıs Cabral, Alex Coad, Sinziana Dorobantu, Felipe Gonzalez, Deepak Hegde, Borja Larrain, Sonia Laszlo, Sudipta Sarangi, Jos´e Tessada, Rodrigo Wagner, Giorgio Zanarone, and seminar participants at CUNEF, IE Business School, the Peruvian Central Bank, Pontificia Universidad Catolica de Chile, Universidad Andres Bello, Universidad de Chile, Universidad de Piura, and Universidad Diego Portales for comments, and to Cristi´ an Figueroa, Efra´ın Natividad, Walter Noel, Leandro Pez´ an, and especially Renzo Severino for excellent research assistance. Errors are my sole responsibility. † Contact: UDEP, Department of Economics, 162 Martir Olaya, Miraflores, Lima 18, Peru. Phone: +51 2139600. E-mail: [email protected]

1

Introduction

Why do some firms thrive and grow while others struggle for subsistence and remain small? The stark contrast of firm sizes in an economy remains a long-standing puzzle, with explanations ranging from microeconomic factors (Lucas (1978), Cabral and Mata (2003), Bloom and Van Reenen (2010)) to macro frictions preventing the reallocation of resources (Hsieh and Klenow (2009), Buera and Shin (2013), Hopenhayn (2014)). While much insight into firms’ choices and constraints has been gained from these inquiries, little is known about the role of historical institutions in determining firm size and investments.1 This paper builds on the growing literature on the long-run impact of institutions (Acemoglu, Johnson, and Robinson (2002), Banerjee and Iyer (2005), Nunn (2008)) to ask whether the sharp contrasts in firm-level investment behavior may reflect a historical imprint, and if so, through what mechanisms. The importance of understanding financial and economic outcomes in a historical institution context has attracted much interest, typically in a cross-country fashion (e.g., La Porta et al. 1997, Djankov, McLiesh and Shleifer 2007, Qian and Strahan 2007). Despite its breadth, the cross-country approach has raised endogeneity concerns (Nunn 2012, Michalopoulos and Papaioannou 2014), thus leading to a recent focus on within-country studies offering methodological advantages. To analyze the institutional roots of firm investment behavior, this paper introduces new data on the population of firms in the 2008 Peruvian business census and extends the multidimensional regression discontinuity setup pioneered by Dell (2010) in the study of Peru’s colonial mita system. The paper also employs granular data from the 2007 census of individuals and households and a centuries-long colonial registry of local treasury transactions to examine the mechanisms for the long-term business impacts of this historical institution. Between 1573 and 1812, the Spanish colonizers of Peru subjected local male populations in specific communities to the mita, a forced-labor regime implying the mandatory deployment of adult males from their hometowns to newly discovered mines in Southern Peru and Bolivia. The 1 Kumar, Rajan, and Zingales (1999) and Laeven and Woodruff (2007) analyze the historical background of legal systems and their connection with firm size using industry-aggregated firm data.

1

colonial authorities determined the geographic boundaries of the mita. Those communities located within the assigned boundaries were subject to the mita regime, whereas communities located just outside the boundaries were exempt. Several features of the colonial mita suggest that it radically altered local economies in ways plausibly connected with their production capabilities. By removing male adults from their hometowns to work at remote mining locations and by triggering the liquidation of physical assets of those tax subjects not willing to serve at the mines, the mita severely distorted the equilibrium levels of capital and labor from a production function standpoint. Combined with the well-documented evidence that the mita assignment was quasirandom, these characteristics suggest that the mita offers a relevant and powerful natural laboratory to assess whether and how contemporary firm investment patterns differ across historical institution regimes. Specifically, a regression design modeling the multidimensional discontinuity in longitude and latitude across mita borders is employed to assess the long-term impacts of the mita on firms in these regions comparing them to counterfactual firms just outside. Taking advantage of the geocoding of firms at the census-block level for the discontinuity setup (e.g., Black 1999), the main tests focus on firms within a close distance (10 kilometers) of the mita boundary and include numerous boundary segment fixed effects and firm-level controls. The models employ alternative polynomial functions of distance, latitude, and longitude that ensure the measured impact of the mita is discontinuous. Importantly, the broad coverage of the business census facilitates inference and the availability of information on physical and intangible investment decisions helps give a more complete assessment of firm outcomes. The results of these models indicate a negative discontinous impact of the mita institution on firm size as reflected in firms’ fixed assets, with a statistical significance that is robust to various specifications. Moreover, the results show a strong negative influence of the mita on firms’ commercial inventories, an important kind of short-term investment. Because the regressions use both firm age fixed effects and industry fixed effects, these findings of smaller firm size and inventory investments may not be attributed to an industry composition or life-cycle argument (Cabral and

2

Mata 2003). When analyzing the intangible investment of firms in the form of their decision to hold a commercial name (Tadelis 1999), I also find a strong negative impact of the mita. Holding a commercial name appears to be a nontrivial matter in the region of interest, as fewer than one third of firms have one. The results show that firms in mita regions very close to the boundary are between 20% and 33% less likely to invest in this kind of intangible investment across various specifications with strong statistical significance. By influencing firms’ investment in an intangible instrument for reputation, the mita thus exerts a negative impact on their potential growth. These findings complement recent work on the challenges of reputation building in developing countries (McMillan and Woodruff 1999, Fafchamps and Minten 2002, Macchiavello and Morjaria 2015). A key economic choice of firms in this context is whether to register their business operations with the tax authority, a useful step towards broader economic activities that comes with the cost of paying income taxes (de Soto 1989, Loayza 1996) This registration decision is recorded in the business census. When analyzing the impact of the mita on this registration outcome, I find that firms in mita regions are significantly less likely to hold a business tax identification number, an essential precursor to broader economic activities. The point estimates, ranging between -11% and -23% depending on the specification, indicate a substantial effect. Could the lower investment impacts of the mita be explained by a negative impact of the mita on human development? The findings of children stunting in mita districts documented by Dell (2010) suggest that firm stunting and human development might be directly connected. Yet when I extend the regression analysis explaining business investments to include the ratio of children stunting for the districts of interest as an additional explanatory variable, I find that the independent impact of the mita on firm investments remains strongly negative and significant as reported above. Having found distinct long-run impacts of the mita on firm investments and registration for tax purposes, I turn to analyzing three broad classes of mechanisms for these effects by introducing data with different levels of granularity and time coverage. First, using geocoded information from

3

the 2007 population census, I find that key individual characteristics are discontinuously different across the mita boundary. Specifically, using regression discontinuity models at the level of each individual covered by the population census, and controlling for individual age, I find that local populations in mita regions very close to the boundary show a higher likelihood to attend school and a lower frequency of insurance coverage than populations just outside the boundary. There is also some evidence that individuals are more likely to remain single in mita regions. These patterns suggest that individuals care to cultivate their human capital while facing more risk and avoiding some long-term commitments. Supplementing this analysis with household data from the housing census, I find that the levels of home ownership are not lower for households in mita regions and that their wealth, proxied by a principal components index of the observable characteristics of the house were they reside, is also statistically similar to that of non-mita households. I interpret these local population outcomes of the mita as consistent with deeply rooted preferences, rather than external constraints, leading to lower business capital accumulation and possibly more tax avoidance (e.g., Kirchler 2007). A second mechanism I explore to understand why firms within mita boundaries might be smaller than those outside boundaries is the financial and economic environment of their localities. If firms in mita regions were operating precisely at locations with generally lower levels of financial development or overall economic activity (with the usual caveat of the circular flow between firms and their surroundings), their external environment would be a prime candidate to explain the firm stunting effects. Yet when analyzing the full grid of one-square-kilometer areas around mita boundaries with the usual regression discontinuity design, I fail to find any statistical difference in the presence of banks or in the nighttime light intensity of areas across mita boundaries. These findings again suggest that a story of constraints is not likely explaining the smaller investments and less frequent business registration among mita firms. Third, I employ data tracing further back in time to analyze the economic persistence of the mita. As documented by the pioneering work of TePaske and Klein (1982), during colonial times the Spanish crown organized its treasury activities in a decentralized network of regional

4

treasury offices. Using the geographic detail of these treasuries’ registries for communities on both sides of the mita boundary for three centuries, I examine the yearly individual taxes the treasuries collected in an analogous multidimensional discontinuity framework at the district level including year fixed effects. I find evidence that the mita caused a positive discontinuous impact on the taxation of areas within the mita boundary. Further, I find that this increased taxation of mita areas was persistent over the three centuries of the transactions in the registry, offering a plausible channel for its long-term effects. These results are striking because the mita, a forced-labor system of extraction, subsequently caused higher taxation at mita regions for centuries during the colonial regime, suggesting a complementarity in fiscal policies at the microgeographic level. The results also give a historical perspective to the discussion of the impact of taxation on investment as a driver of economic development (Engen and Skinner 1996, Wasylenko 1999, Chirinko and Wilson 2008, Besley and Persson 2013), with the simplifying yet consequential feature of colonial (non-democratic) taxation that voting was not an available option (Alesina and Rodrik 1994). A rising literature has explored the barriers to firm growth in the form of regulatory obstacles, tax compliance, and size thresholds enacted by government policies (Kleven and Waseem 2013, Gourio and Roys 2014, Garicano, Lelarge, and Van Reenen 2016). By now it is typically accepted that, following their own private incentives, firms may choose to remain small, thus reducing welfare (e.g. Leal-Ord´ on ˜ez 2014), and a direct policy implication would recommend removing such obstacles to make way for economic growth. The findings of this paper offer a new perspective: the local roots of businesses may account for severe cases of firm stunting. The geographic thresholds of institutional boundaries are thus brought into relief as a key dimension to consider in the functioning of organizations and markets. This paper is one of the first to study the broad impact of historical institutions on firm growth. Influential research has only recently shown that old institutions can severely constrain the human development capabilities of local populations in the very long run. The results of this paper offer an alternative view of firm stunting. The findings advance a policy channel for the long-term impact of historical institutions working steadily and deeply through the local roots and

5

attitudes towards business capital accumulation and taxes.

2

The Mita Regime

The Spanish colonization of Peru (1532–1821) required a quantity of mining workers on levels far beyond the limited local labor supply at mining sites. Mines were mostly located in relatively isolated areas of today’s Southern Peru and Bolivia. Thus the Spanish colonizers considered a broad geographic footprint for the forced labor system called the colonial mining mita (from mit’a, or “turn”, in Quechua), and implemented this remunerated, coercive, rotating system starting in 1573 under the government of Viceroy Toledo (Noejovich 2009). The mita was abolished in 1812. Under the mita regime, males of various regions across the vast extension of Peru were sent to mines in Potosi and Huancavelica, taking yearly turns over seven-year cycles. Importantly, only mita subjects from those communities demarcated by the mita geographic boundaries were impacted by the imposed regime. Communities just outside the mita boundaries were unaffected. As first proposed by Dell (2010), these institutional features lend themselves naturally to a regression discontinuity analysis employing the mita boundary as an exogenous geographic threshold dividing treatment and control groups for the study of long-term impacts. The open question this paper addresses, introducing a variety of new data sources, is whether the colonial mita institution has any bearing on the size of firms’ capital base and if so, through which channels of persistence. While historical research has mostly advanced a labor-related interpretation for the mechanisms set in motion by the mita, it is also natural that capital accumulation is relevant to assess its economic effects. For example, wage distortions and inefficient labor supply decisions were by design a direct result of the mita regime, as would be expected of any forced labor system. From a production function standpoint, such distortions likely impacted the capital accumulation decisions of individuals and organizations. For another example, consider the buyout option available to mita subjects. Instead of serving their mita time, individuals had the option to redeem themselves by paying a monetary tribute, and a common mechanism for this payment was the liquidation of their

6

capital goods. Proceeds of this liquidation were directly sent to the authorities of the mine or used for the hiring of a substitute worker. In sum, the mita system affected individuals at the treated communities, either by depleting their physical capital or by forcing them to move out to supply labor to the mines. This paper asks whether long-term causal impacts of the mita institution may be observable in the decisions of firms in treated mita regions hundreds of years later, and if so, what mechanisms may explain such influence.

3

Data

I built a data repository for this study drawing from various sources. The first and main data source is the 2008 Business Census of Peru conducted by the National Institute of Statistics and provided by the Ministry of Production. This database, to date largely unexplored for academic research, covers firms of all sizes and all industries except agriculture and financial services regardless of their tax-filing status, focusing exclusively on their operations in 2007. Firms’ locations are identified at the granular level of the census block (i.e., approximately 100 meters by 100 meters), and the block’s centroid is used for the firm-level analysis. Panel A of Table 1 reports summary statistics on these firms. Second, geographic boundary information on the colonial mita regions was obtained from Melissa Dell’s website.2 Third, geospatial characteristics at the census-block level such as elevation and slope were obtained from NASA’s Shuttle Radar Topography Mission (SRTM (2000)), matching these variables with the Peruvian national census-block-level map of 2008 developed by the National Institute of Statistics and facilitated by the Ministry of Production. Although the geographic coverage of the mita was quite extensive in colonial Peru, the areas selected for this study correspond to a region of southern Peru mostly at 3,000 meters above sea level characterized by comparable topographic conditions in the zones adjacent to the mita boundary. In terms of political 2

http://scholar.harvard.edu/dell/publications/persistent-effects-perus-mining-mita, last accessed 22 March 2017.

7

demarcations of 2008, this region corresponds to 299 districts. The regression models exclude the area of metropolitan Cusco, an important economic center that was already prosperous before the mita started (though for robustness I also show the results do not hinge on this exclusion). Figure 1 displays maps of the region studied, with the mita borders demarcated in bold lines. A fourth data source is the Census of 1572 provided by Cook, M´alaga, and Bouysse (1975). The fifth source is the 2007 Census of the Population of Peru, also conducted by the National Institute of Statistics, which includes demographic information on individuals as well as the geocoded location of each census block. Information for this population census was recorded in 2007 months before the census of businesses. The sixth data source is the 2007 Housing Census of Peru, conducted concurrently with the census of the population. The seventh data series is publicly available from Superintendencia de Banca, Seguros y AFP, consisting in the geographic location of regulated financial institutions as of 2007. The eighth source is the DMSP-OLS Nighttime Lights Time Series, containing nighttime light intensity data on fine-grained areas for year 2007 and before. The ninth source of data is the registry of transactions of the royal treasuries of the Spanish crown in Peru compiled by TePaske and Klein (1982), available on Richard Garner’s website.3 The diversity of these sources allows for a holistic view of firm growth. It also helps causal inference for a couple of reasons. First, the data sets mainly cover populations of firms and individuals, thus offering a more complete assessment of the broad heterogeneity of firms and the conditions of their local environment. Second, the main data sets are geographically referenced in a granular fashion, enabling a detailed examination of how different geographic institution regimes influence firm outcomes.

4

Specification

The long-term impacts of the mita institution can be assessed using its geographic boundary in a multidimensional discontinuity setup based on longitude and latitude (Dell 2010). 3

http://www.insidemydesk.com/hdd.html, last accessed 22 March 2017.

8

The

methodological innovation here, building on the advantage of geocoding firms at the census-block level (e.g. Black 1999), is the inclusion of numerous narrow segment fixed effects as well as a tighter window of observation on both sides of the mita boundary in a semiparametric fashion. Because firms may be highly heterogeneous, moreover, the specification seeks to control for many sources of variation. Specifically, the regression

f irm outcomeibs = α + βmitab + g(locationb ) + ηa + γXb + λj + φs + µibs

(1)

analyzes outcomes in 2007 for each firm i in census block b, whose nearest boundary segment is s, using flexible polynomials following function g of the geographic location of the firm. In total, thirty boundary segment fixed effects are employed.4 Specification (1) also includes ηa firm age fixed effects, thus accurately controlling for the life-cycle of each firm. The block-level control variables Xb included are block elevation and block slope. Importantly, the regression introduces 75 distinct industry j fixed effects to keep firms comparable across the boundary. As observations within a district may not be assumed to be independent, the standard errors of coefficients are calculated using robust clusters at the district level. To offer very close counterfactuals to the mita treatment, the semiparametric analysis can narrow the window of observation to those firms within 10 kilometers of the mita border, as will be shown in the main regression results. In equation (1), mita is a dummy variable equal to one when the firm is located on a census block inside the boundaries of the colonial mita. The regression coefficient β thus captures the discontinuous effect of the mita treatment on firm outcomes, an impact that can be interpreted to be causal because the usual identification assumptions hold given the randomness of the mita assignment (further discussed below) and the analysis of outcomes in close proximities. The flexible function g(·) takes three alternative forms: (a) cubic polynomials in the latitude and longitude of the firm; (b) cubic polynomials in the distance between the firm and Potosi, the main mining 4

Each block is assigned to the mita boundary segment that is closest to the block, and all boundary segments are of the same length. Although more than thirty segments could be introduced given the long extension of the mita boundary, the sparsity of firms in various areas near the mita boundary would leave some segments empty of nearby firms on one of the sides of the boundary. As detailed in subsection 5.5, the results do not depend on the number of segments chosen.

9

destination at the time of colonial assignment, and (c) cubic polynomials in the shortest distance between the firm and the mita boundary. These flexible functions ensure that the effect of the mita dummy is distinguished from the potentially confounding effects of different locations in the Andean territories analyzed. A regression discontinuity design to study firms inside and outside the mita region requires that the original assignment to the mita treatment is random. Dell (2010) convincingly shows that several observables vary smoothly at the geographic threshold, thus confirming this key identification assumption in the mita context. To avoid duplication of effort, in this section I extend the existing evidence using more granular measurement to analyze whether relevant observable characteristics vary smoothly at the boundary. I calculate simple summary statistics on the main geospatial variables in narrowing windows around the mita boundary, measuring them at the level of each census block. Panel B of Table 1 reports mean values for terrain elevation and slope, calculated within 100 kilometers, 50 kilometers, and 10 kilometers of the mita boundary. Elevation is a natural property associated with climate and crop choice in the region studied. The top rows of Table 1 show that elevation is significantly different at terrains on opposite sides of the mita boundary in the case of wide windows. By contrast, when narrowing the distance to the boundary to consider only blocks within 10 kilometers and using Conley’s (1999) spatially correlated standard errors, terrain elevation is indistinguishable for mita and non-mita territories, as displayed in the last column of Panel B of Table 1 with the standard error value in brackets. These findings suggest that in the narrow window of 10 kilometers around the boundary, the mita assignment was not linked to meaningful geographic differences. Similarly, summary statistics on the slope of the terrain in blocks around the mita boundaries are displayed at the bottom of Table 1.

The slope variable captures terrain ruggedness, a

determinant of economic attractiveness. In the wider windows shown in the first columns of the table, the differences in slope across the border are statistically significant using robust standard errors, but they are insignificant using Conley spatially correlated errors in the narrow window of 10 kilometers in the last column of Table 1, and overall very small in magnitude. To make sure that

10

elevation and slope do not independently influence outcomes, hereafter all regressions will include these variables as geographic controls, and the main analyses will focus on narrow windows.

5 5.1

Results Fixed Assets

Does the colonial mita have any impact on firm investments? I first analyze whether firms operating in local geographies with a history of colonial mita treatment invest differently in productive assets than otherwise similar firms located just across the mita boundary. The business census records the amount of fixed assets across all categories of long-term productive assets of each firm. The logarithm of one plus fixed assets net of depreciation is the first dependent variable for the estimation of equation (1), and the results are displayed in Table 2. Each entry in the table represents a different regression, and only the coefficient of interest, β, the t-statistic of its standard error, and the Rsquared of the regression are shown for brevity. The results of three alternative specifications for function g(·) are displayed separately in Panels A, B, and C of Table 2. The length of the windows for the analysis of firms’ fixed assets in regard to their distance to the mita boundary are 100 kilometers, 50 kilometers, or 10 kilometers, and they are represented in the first, second, and third column of Table 2, respectively. All regressions include the geographic control variables, as well as industry, firm age, and boundary segment fixed effects detailed in Section 4 and employ clustering of standard errors by district. The first column of Table 2 reports the results of the regression discontinuity models under the three alternative cubic polynomial specifications for all firms located within 100 kilometers of the mita boundary. In all three specifications, the results indicate a strongly negative impact of the mita on the log of fixed assets, with t-statistics of -3.97, -3.22 and -3.75. When narrowing the distance to the mita boundary to consider only the 50 kilometers nearest to the boundary, the second column of Table 2 continues to show a statistically significant negative impact of the mita treatment across specifications, with t-statistics of -2.81, -1.92 and -1.80.

11

The preferred sample window, exploiting the advantage of granular geocoding of firms detailed in Section 4, sharpens the analysis to focus only on impacts within 10 kilometers on both sides of the boundary. The results using only observations in this window are in the third column of Table 2. Under all three specifications in this sample window, there is a large and statistically strong negative jump in fixed assets for firms just inside the mita region, with t-statistics equal to -2.17, -3.26, and -3.70. The economic magnitude for the mita estimate revealed by the cubic polynomial in latitude and longitude specification amounts to a size of fixed assets 87% smaller than seemingly indistinguishable firms. Taken together, these findings suggest that the equilibrium magnitude of the productive capital base at firms located in formerly mita regions is substantially smaller than that of counterfactual firms. Because the regressions include various dimensions of fixed effects as well as block-level controls, all these other dimensions of the comparison are held constant in order to focus only on the causal impact of the mita. In the long run, the colonial mita institution causes the stunting of firms in the key dimension of their productive assets. Panel A of Figure 1 shows graphical evidence on the multidimensional discontinuity of fixed assets in the microgeographies around the mita boundaries. In this chart, longitude is on the x axis and latitude on the y axis. A scattered-dot plot for the average logarithm of fixed assets for all firms in each block of the region is displayed. The size of each dot represents the number of observations in each block using a 4-point scale, which is nonlinear in order to keep dot sizes legible. The background of the plot consists of a fine grid of longitude-latitude values. For each of these granular cells on the grid, predicted values of the log of fixed assets using the results of a firm-level regression of this variable on a cubic polynomial in longitude-latitude, the mita dummy, the topographic controls, and boundary segment fixed effects are shown in six different shades of the same color, thus representing the third dimension of outcomes in this regression discontinuity setup. The graphical evidence on Panel A of Figure 1 confirms a strong difference in the logarithm of fixed assets observable precisely across the boundaries of the treated region. The lighter color for the mita areas indicates the significantly lower level of fixed assets caused by the mita treatment.

12

5.2

Inventories

I turn next to the analysis of short-term productive investments made by firms in their regular business operations: their inventory policies. Do firms with a mita heritage invest differently in this important asset class? The business census records the total commercial inventories held by firms at the end of 2007. The logarithm of one plus total commercial inventories is the dependent variable for the estimation of equation (1), and the results are displayed in Table 3. Each entry in the table is from a different regression. Panels A, B, and C report alternative cubic polynomial specifications. Different columns in Table 3 report subsamples of varying windows around the mita boundary. All the results in Table 3 indicate a statistically significant and economically meaningful negative discontinuous impact of the mita dummy on the log of inventories. The first column reports models using the 100-kilometer window around the mita boundary; under all three different polynomial specifications, the impact of the mita is negative, with t-statistics of -4.63, -4.46, and -5.65. Similarly, the second column of Table 3 displays the results of regression discontinuity models employing only firms within 50 kilometers of the mita boundary; in all models, the coefficient of interest is negative and significant, with t-statistics of -3.76, -2.24, and -3.45 and large economic magnitudes for the impact. Finally, the third column in Table 3 reports results on the sharpest discontinuity comparison between mita and non-mita region firms by narrowing the distance to 10 kilometers from the boundary. The estimates are all negative and statistically significant, with t-statistics of -2.42, -2.53, and -2.86. When compared with the summary statistics displayed in Table 1, the economic magnitude of the point estimates appears to be substantial. These regression results suggest that the mita causes drastic reductions in the equilibrium level of firms’ short-term investments. While different industries may require different optimal inventory policies, the specification accounts for this variation by including two-digit industry dummies, thus focusing on close counterfactuals.

Moreover, while inventories may reflect

momentary decisions in the life cycle of a firm, when I alternatively use as the dependent variable the beginning-of-the-year-2007 inventories, or the simple average between inventories at the beginning

13

of the year and end of the year, the results of these untabulated regressions show a similar causal impact of the mita dummy. Thus, the findings that the institutional imprint reduces inventories paints a consistent picture with the findings on fixed assets and reveals another dimension of firm stunting attributable to the colonial institution. A graphical examination of firm inventories across the geographic boundaries of the mita confirms the regression results.

Panel B of Figure 1 shows graphical evidence on the

multidimensional discontinuity of commercial inventories in the regions of interest, following the same specification as that of the chart in Panel A described above. The chart shows a strong difference in the logarithm of inventories observable across the mita borders.

5.3

Commercial name

Firms may also invest in intangible assets to obtain returns on them over different horizons (Tadelis 1999). Specifically, a local firm may seek differentiation through establishing a commercial name for the business or its products. The choice of a commercial name, going beyond a generic denomination, is recorded in the business census. I thus analyze the dummy equal to one when the business has a commercial name as the dependent variable of interest. The decision to register a commercial name appears to be nontrivial in this context, as less than one third of the firms in the region studied show a value of one for this dummy. Table 4 reports the estimation results of equation (1), displaying as in the case of the previous tables various specifications and sample windows. The coefficient of interest captures the causal impact of the mita on the decision to invest in the intangible form of a commercial name. Table 4 shows that across all specifications and window lengths, the discontinuous impact of the mita on having a commercial name is strongly negative, with large economic magnitudes compared to the unconditional mean displayed in Table 1. In the first column, which keeps only observations within a 100-kilometer distance of the mita boundary, the coefficient of interest is negative and statistically significant at the 1% level, with t-statistics of -2.86, -2.79 and -3.22. The impact of the mita on the commercial name dummy persists at similar levels of statistical and

14

economic significance, with t-statistics of -4.17, -2.54, and -2.72, when narrowing the sample to focus only on firms within 50 kilometers of the boundary. In the preferred tests using the narrow distance of 10 kilometers from the boundary, the impact of the mita on the intangible investment in a commercial name is again seen as strongly negative and statistically sharp across specifications. The t-statistics take the values of -4.36, -3.19, and -4.15. The magnitudes of more than a 20% reduction in the probability of having a commercial name across the specifications are substantive, especially in a context with an unconditionally low propensity to employ a commercial name. These findings suggest that the negative impact of the mita is noticeable not only on the level of physical investments but also on the dimension of intangible, softer investments made by the firm. While having a commercial name is probably not necessary to operate — as reflected by the large proportion of firms without one — it seems to be directly associated with the stunting of business development caused by some historical connection with the mita regime. Operating a firm generically or anonymously appears to be a long-run consequence of the mita regime. Panel C of Figure 1 displays evidence on the discontinuous impact of the mita on having a commercial name amongst firms close to the boundary. The three-dimensional chart indicates a sharp pattern of lower propensity to invest in a commercial name within mita regions, consistent with the regression results.

5.4

Business tax registration

An important feature of the 2008 census of businesses is its broad coverage of firms, regardless of their status of formal registration with the tax authority. Specifically, all Peruvian firms and individuals that are registered for business income tax purposes hold a business tax identification number, and this decision is recorded in the census. Are firms within mita boundaries equally likely to hold a business tax ID number as those just outside mita boundaries? Table 5 reports the results of models following equation (1). The coefficient of interest shows the causal impact of the mita on

15

the decision to formally register a firm with the tax authority and thus pay business income tax. The results of the 100-kilometer window show evidence of a lower likelihood of holding a business tax ID in Panels A and C of Table 5, contrasting with no significance in Panel C. The t-statistics take values of -2.37, 0.36, and -2.37. When narrowing the sample distance to within 50 kilometers of the mita boundary, the estimates take negative signs in all specifications, while resulting only statistically significant in Panel A. The t-statistics on the mita variable are -1.94, -0.55, and -1.29. In the preferred tests using only observations within 10 kilometers of the mita boundary, all specifications show a statistically significant negative impact of the mita on the likelihood of having a business tax ID number. The t-statistics are -2.10, -1.73 and -2.16. The economic significance of these estimates is large, implying a probability of tax registration at least 11% lower than the counterfactual firms across the mita boundaries. These results on tax registration propensity are indicative of the habits and incentives of business owners in mita regions, to be further analyzed in Section 6.

5.5

Robustness

The generally robust results shown for the impact of the mita on firm outcomes are further probed in a variety of ways now summarized, with detailed tables omitted for brevity. First, the cubic polynomial specifications for function g(·) in equation (1) for the models of interest are changed to quartic or quadratic polynomials, leaving the results largely unchanged. Second, the distance for the narrow windows (and firms) close to the mita boundary is varied to 15 kilometers (4,749 firms) or 5 kilometers (630 firms), as compared to the current focus on 10 kilometers (4,026 firms). Unsurprisingly, given the very similar sample size, the widening of the window to 15 kilometers has no noticeable impact on the results, while the excessive narrowing to a few hundred firms reduces the power of the log of fixed assets regressions without changing the direction of the impact and leaves the results on inventories and commercial name largely similar. The stability in the number of firms around 10 kilometers is nonetheless reassuring that this distance is appropriate for

16

measurement, as done in the main results displayed. A third robustness check changes the number of segments s of the mita boundary that leads to geographic fixed effects in specification (1). Increasing the number to 60 segments or reducing it to twenty or just even four segments leaves the results unaffected. A fourth test includes metropolitan Cusco in the estimation sample, or alternatively changes the definition of Cusco to exclude the Lucre and Oropesa districts from the definition of the metropolitan area that is excluded from the main tests, leaving the results unaltered. A fifth set of checks excludes from the estimation sample those areas that are close to a mita boundary segment that includes a river, without changing the results.

5.6

Does human stunting account for business stunting?

Dell (2010) showed two human development consequences of the mita: (a) reduced levels of household consumption in 2001, as measured by the 2001 ENAHO survey of households, and (b) children stunting in 2005, as measured by a school census. Those findings suggest the tantalizing possibility of linking the human development impacts of the mita with the stunting of business firms as of 2007. Put differently, if human development outcomes are so negative in the long run, then the business stunting consequences might be construed as a mere side effect of the mita. Importantly, my design seizes the granular nature of business outcomes within 10 kilometers of the boundary for inference, which makes a match with the less granular development outcomes studied by Dell difficult. Specifically, for the narrow bands in my design, only 12 districts of the census zones analyzed have household consumption data in the less inclusive 2001 ENAHO survey, thus precluding an effective match. A feasible inquiry can be implemented with the 2005 census of school children’s height, available for all districts, with the caveat of not having geocoded locations for the domicile of those children. I thus employ the children stunting ratio at the district level for 2005 as a control variable in equation (1) and redo all the 2007 investment regressions, that is, the impact of the mita on fixed assets, inventories, holding a commercial name, and having a business tax ID. Notice that

17

this alternative specification introduces an average human development outcome (as of 2005) in a regression explaining business outcomes (as of 2007) of the colonial mita. If the explanatory power of the mita on business impacts runs through the human development proxy of stunting, then the independent content of the mita dummy should be largely diminished or even disappear. Yet in untabulated regressions, I essentially find the same firm-level investment impacts of the mita as those reported in the preceding subsections. The results of this alternative design indicate a robust negative long-run impact of the colonial mita institution on physical and intangible investments of firms beyond its effect on individuals’ health. When conceptualizing this broad development indicator as a proxy for the level of demand in the local population, the distinct impacts of the mita on firms suggest a supply-side mechanism to the long-term impacts on firm stunting. What explains such persistent effects? I explore this question next.

6

Mechanisms

I now analyze what mechanisms may explain the long-term effects of the mita on businesses. To do this I employ three novel data sets with different levels of granularity and time coverage. Taking advantage of the richness of these sources, I test for three broad classes of mechanisms. First, I inquire whether the local individual and household conditions in areas close to the mita border as of 2007 indicate a discontinuous difference consistent with either constraints or choices. Second, I analyze the financial and economic environment surrounding business firms to assess whether stark differences in development across mita boundaries are an external constraint for growth. Third, I trace three centuries of local colonial tax collections in districts across mita boundaries to see whether an imposed tax policy is consistent with long-term attitudes behind firm stunting.

6.1

Local individual and household conditions

The first mechanism I explore in order to understand why contemporary firms may differ across the colonial mita boundary is to inquire whether the local populations as of 2007 — encompassing

18

both business owners and customers5 — show discontinuously different personal features across the boundary. To conduct this analysis, regression models following equation (1) using individuals instead of firms, and controlling for individuals’ age, are implemented. No firm-level controls are employed. The results are displayed in the left-hand side of Table 6. Because the models of interest focus on observations within 10 kilometers of the mita boundaries, only results for this narrow sample window are discussed here for brevity. One key characteristic of the local populations analyzed is their tendency to invest in human capital, as proxied by school attendance. School attendance (a dummy) pertains to children under 15 years of age, as measured by the census. The first column of Table 6 shows a significantly positive impact of the mita on the level of school attendance of the local populations, with small yet meaningful point estimates against the unconditional mean of 0.82 for the dependent variable in the sample employed. The t-statistics on the mita variable are 2.98, 2.22, and 2.73. Mita-region children are thus more likely to attend school than those of non-mita regions, suggesting that the human capital prospects in mita regions are not at a disadvantage. Panel A of Figure 2 shows a graphical representation of the discontinuity results. The regression results suggest that the lower business investment of firms in mita regions is not associated with a lower level of investment in human capital and may instead be consistent with a more education-oriented attitude of the surrounding communities. The population census also records whether individuals choose to be covered by any kind of insurance (e.g., medical, employment-based), and the negative version of this decision is the dependent variable I analyze next.

Conceptually, individuals choosing not to be covered by

insurance can be viewed as having a different risk attitude from those insured; in the case of Peru, the poorest individuals in 2007 were broadly covered by government-paid insurance, so the uninsured may not be confounded with the poorest in my analysis. The results across all specifications in the second column of Table 6 indicate that individuals close to the mita border are significantly more likely to have no insurance of any kind. The economic magnitude of this impact 5

For confidentiality, the population census never identifies individuals; hence, no individual-level match is attainable with the business census.

19

is substantial when compared with the unconditional mean of 0.40 for the dependent variable in the sample analyzed. The t-statistics on the mita coefficient are 3.33, 4.23, and 4.73. Panel B of Figure 2 displays a graphical version of the discontinuity. These strong no-insurance findings suggest that local populations in mita regions appear to be more willing to individually face their own personal risks, forfeiting the financial and psychological benefits of an insurance instrument. One last individual dimension I analyze involves the choice of marital status, which could be argued as a potentially important dimension conditioning attitudes toward investment. Generally, with a lower propensity to personal commitments in the local population, investment incentives could be seen as less attractive. The findings shown in the third column of Table 6 lend some support to this possibility. Specifically, the causal impact of the mita on whether the individual is single appears to be positive. The coefficient across all specifications has a positive sign, and is statistically significant in two specifications (t-statistics of 1.96 and 1.71). The economic significance is moderate when compared with the unconditional mean of 0.35 for the single marital status dummy in the sample analyzed. Panels C of Figure 2 displays the graphical representations of the discontinuity. Taken together, the results on how the personal conditions of the local populations hosting business firms vary across the mita boundary provide a nuanced characterization of a broad array of choices and conditions possibly behind business investment outcomes. The housing census of 2007, conducted at the same time as the census of individuals, offers additional information to inquire whether the business impacts of the mita might be related to wealth or, equivalently, to the lack of house collateral for financial resources (e.g. de Soto 2000). The analysis is conducted at the household level, thus implying a smaller number of observations than that of the analysis of individuals. The fourth column of Table 6 reports the results of the test measuring the impact of the mita on whether the household owns the house where it resides. Across all specifications, there is no evidence of a negative impact of the mita on house ownership, with even some marginally significant evidence that the impact might be positive (t-statistics of 1.34, 1.25 and 1.76). It is thus clear that home ownership levels are not lower in mita regions, suggesting that the relevant mechanism is not one of constraints.

20

To further explore whether the wealth of the household is significantly different in mita regions, I employ a principal components analysis based on 22 home-specific variables available in the housing census (e.g. Maccini and Yang 2009). The first (or most important) component of the total variance of the housing variables explains a relatively high proportion of that variance. This first component, labeled house index and proxying for household wealth, is used as the dependent variable for the test reported in the fifth column of Table 6. No significant impact of the mita variable on the house index is found, with t-statistics of -1.17, -0.40, and -1.58. Viewed as a whole, the results based on household conditions point in the same direction as the results at the individual level, suggesting a choice rather than a constraint behind the striking patterns of business stunting in mita regions. Qualitative evidence underscores the cultural roots behind individual attitudes in these local populations, which may act as a motivation consistent with the individual and business choices documented so far. The business owners I spoke with while visiting five mita and non-mita areas within 10 kilometers of the mita boundary revealed varying levels of trust in the way they conducted business transactions. When discussing their willingness to trust customers and business partners6 in their regular operations, a concern about expropriation and loss of assets somewhat more pronounced in mita areas suggested the existence of deeply rooted attitudes among these business owners.7

6.2

The financial and economic environment

The main results of the study, based on the census of Peruvian firms of 2007, indicate that mitaimpacted firms are smaller, less prone to registering with the tax authority, and less likely to employ a commercial name than seemingly indistinguishable counterfactual firms. The methodological focus on a granular analysis of firms, while clearly advantageous, begs the question of whether the 6

Throughout my interviews with business owners, I stayed away from asking them about their trust in government and the tax institution, in order not to be perceived as a tax supervisor. See subsection 6.3 for an analysis of colonial taxation in these regions. 7 See Guiso, Sapienza, and Zingales (2009), Algan and Cahuc (2010), Dohmen et al. (2011), and Pierce and Snyder (2017) for some prior studies investigating the history background of these attitudes across different settings.

21

broader economic environment where these firms operate may constitute a first-order constraint hindering their growth. In this analysis, one has to be aware of a circular logic: firms constitute an essential pillar of the economic environment. Yet the analysis of the geographic environment surrounding firms (e.g. Felkner and Townsend 2011, Brown, Cookson, and Heimer 2016) rests on the merits of alternative factors (e.g., banks) that, given their historical imprint, may condition firm growth. To address this set of issues with a methodology analogous to the one so far employed, I construct a grid of one-square-kilometer areas in all the regions studied, and for each grid I record key variables related to the economic environment faced by firms.8 The results of multidimensional regression discontinuity models at the level of each one-square-kilometer area of this grid are presented in Table 7. First, using publicly available information on the stock of all regulated financial institutions (labeled as banks for simplicity here) at any point during 2007, I employ the logged number of bank branches as the dependent variable for the analysis. The first column of Table 7 reports the results, indicating no statistical impact of the mita on the presence of banks. The t-statistics are -1.34, -0.81, and -0.99. In general, the narrow region studied (both inside and outside mita boundaries) has very few bank branches present. The implication of the test is that mita-impacted firms are not at a disadvantage with respect to the counterfactual case. A graphical version of the discontinuity is displayed in Panel A of Figure 3. Second, employing nighttime light intensity data in year 2007, I construct the logarithm of one plus the light intensity continuous variable for each of the one-square-kilometer areas within 10 kilometers of the mita boundary. The first column of Table 7 shows that this proxy for local economic activity is not significantly different across the mita boundary. The t-statistics on the mita coefficient are -0.22, 0.47, and 0.19. Finally, to proxy for the economic dynamism and growth of the local environment, I take the difference between the log of light intensity in 2007 and the log of light intensity in 1992, and use it as the dependent variable for the model displayed in the 8

Those one-square-kilometer areas that occupy both mita and non-mita sectors (i.e., areas intersected by mita borders) are excluded from the analysis for clarity.

22

third column of Table 7. No distinguishable pattern across mita boundaries is found for the local 1992-2007 growth as proxied by changes in light intensity. The t-statistics for the mita variable are 0.03, 0.84, and 0.62. The graphical versions of the discontinuity are displayed in Panels B and C of Figure 3. These findings based on a well-known proxy for economic activity suggest that the local environment is not a salient constraint for the investment of mita-region firms.

6.3

Colonial taxation from the 16th to the 18th century

I next explore a mechanism tracing further back in time to analyze the economic persistence of the mita. A key archival data source employed by colonial history researchers to describe patterns of fiscal operations in America is provided by TePaske and Klein (1982). Acclaimed for its breadth of coverage, this systematic registry of the decentralized network of royal treasuries (cajas reales) in Peru has had relatively little influence on the debate about the impacts of the mita (see Iyer (2016) for an exception). The tax records in this source correspond to the fiscal revenue of corregimientos, administrative jurisdictions of the Spanish Crown in its colonies governed by a state authority. A clear advantage of analyzing the transactions recorded by the royal treasuries is that they span three centuries (i.e., from 1571 to 1757) and deal with economic matters potentially impacted by the mita,9 thus offering a long-term perspective for why the mita influenced capital accumulation outcomes. A disadvantage is that the royal treasuries were geographically more aggregated (only four of them existed in the area of study) than the hundreds of granular locations I have analyzed so far. To gain insight consistent with a regression discontinuity framework, my examination requires focusing on the geography-specific transactions of TePaske and Klein’s (1982) registry. Specifically, the registry records the amount of inflows and outflows of each royal treasury, the period covered by each transaction (often spanning 12 months), and in some transactions also a string with the name of the community in which the transaction originated. Many transactions do not have microgeographic detail and are simply recorded in a centralized fashion at the level of 9

The mita started in 1573; for clarity, I thus exclude a handful of observations from this source available from 1571– 1573. Using a 1572 census, Dell (2010) shows that taxation levels were indistinguishable across the mita boundary before the mita started, an important precedent for my findings.

23

each of the four royal treasuries of the area of interest. After discarding those transactions that had no microgeographic reference and using Cook, M´alaga, and Bouysse (1975) to match transaction strings with real location names in the colonial period, all matched locations are further classified into one of the current districts of Peru. Moreover, the amount of each transaction is converted into a silver currency denomination using the rates detailed by TePaske and Klein (1982) and Cook, M´alaga, and Bouysse (1975). For the analysis, I build a database with 489 district-year observations on the yearly individual taxes levied at local geographies within 50 kilometers of the mita border covering three centuries. Observations at this level of analysis are geographically more aggregated than those in all previous tests of the paper, so I follow closely Dell’s (2010) district-level design to estimate the causal impact of the mita treatment on tax collection outcomes as a channel of persistence over three centuries. Analogously to the more granular specification employed before, the district-level equation

taxesdst = α + βmitad + g(locationd ) + γXdt + ηs + λt + µdst

(2)

details the discontinuous impact of the mita on the tax ratio of district d in year t, controlling for four boundary segment s dummies, a flexible polynomial function g(·) taking various forms with distance from the district capital as the input, year t fixed effects, and the geographic district-level controls of mean slope and elevation. Standard errors are clustered at the level of each district. Table 8 reports the results of the estimation. Each entry in the table represents a different regression, and only the coefficient of interest, β, the t-statistic of its standard error, and the R-squared of the regression are shown for brevity. Panels A, B, and C of Table 8 report different specifications of the polynomial function g(·). I first focus on the impact of the mita on the logarithm of local taxes levied. The first column of Table 8 shows a positive and statistically significant influence of the mita on the logarithm of taxes collected across all periods studied and across all specifications. The t-statistics are 2.39, 2.06, and 1.69, and the magnitudes of the coefficient of interest are large. Moreover, the fact that all

24

regressions control for year fixed effects isolates the mita impact from other secular movements in economic conditions. In essence, the models estimate a positive causal impact of the mita treatment on the taxes paid in proximate communities on opposing sides of the mita boundary within the same year, after controlling for a number of competing factors. In the second column of Table 8, the same regression model is extended to introduce interactions of the mita dummy with dummies for the 17th and 18th centuries in order to see whether the impact of the mita on taxation changed over time. Yet as displayed in the second column of Table 8, no significant evidence is found of a change in the higher taxation level for mita regions as the time progressed; in two of the three specifications, the overall impact of the mita on taxes remains positive and statistically significant, and the point estimates are broadly similar to those displayed in the first column. Taxes in colonial times were collected on an individual basis using the number of indigenous males between the ages of 17 and 50 as the tax base. It is thus important to check whether the impact of the mita on taxes is simply an artifact of differing taxable population sizes across mita boundaries. The third and fourth columns of Table 8 uses the same tax variable as in the prior models, this time scaled by the number of indigenous males as of 1572 in the taxable age range. The results in the third column shown a positive impact of the mita on populationscaled taxes, with t-statistics of 1.85, 2.07, and 1.60, thus showing statistical significance in two of three specifications. Moreover, when introducing the interactions with the 17th and 18th century dummies, the conclusions remain qualitatively unchanged.10 Panels A and B of Figure 4 display evidence on the multidimensional discontinuity of taxes in all districts with transactions matched in the royal treasuries registry. Different shades of background color of the landscape indicate higher or lower values for the tax variables averaged over each district over the whole period analyzed. The black dots represent the capital of each district contributing observations to the analysis. The three-dimensional chart gives evidence on the discontinuously higher tax pressure in mita regions. 10

The scaling factor in the denominator is the male population before the mita. If arguments about the diminishing population of mita communities after the mita was enacted are invoked (Cook 2004), the results would represent a conservative estimate for the positive impact of the mita on tax pressure.

25

In sum, the regression analysis of rich tax levying data over three centuries shows a suggestive pattern for the economic environment in mita regions triggered by the exogenous imposition of the regime, an environment that remained fairly stable over time due to the central government’s fiscal pressure. The mita appears to have triggered a geographically uneven tax extraction system that differentially affected some communities for an extended period. Regions that were fortunate enough not to be chosen for the mita enjoyed a lighter tax burden in the ensuing centuries. For regions treated with the mita and with the subsequent heavier taxation enacted there, to the extent that the rents of investments did not fully flow to their original owners but were diverted in part to government coffers, attitudes towards capital accumulation likely developed differently.

7

Conclusion

In this paper, I use novel data from contemporary Peru to study how the colonial mita labor system (1573–1812) caused long-run impacts on firm investments. Regression discontinuity models exploiting the granular geographic information of the business census show severely lower levels of fixed assets and inventories, a lower likelihood of having a commercial name, and lower level of registration with the tax authority for firms within mita boundaries compared with nearly identical firms just outside. I also employ novel data with different levels of granularity and time coverage to test for three broad classes of mechanisms. First, I find that key conditions of the local populations within mita boundaries — a higher likelihood of school attendance, a lower likelihood of insurance coverage, and statistically indistinguishable house wealth — compared with those outside boundaries are consistent with a mechanism in which the culture and attitudes of individuals lead to lower business capital accumulation and more tax avoidance. Second, I find no statistical difference in the presence of banks and no difference in the nighttime light intensity when comparing mita and non-mita areas, thus confirming that the external environment does not appear to be differentially hindering firm growth. Third, when analyzing a centuries-long registry of treasury transactions, I find a heavier tax burden for mita regions, thus offering the past experiences of disadvantageous taxation as a likely channel of persistence for the long-term capital accumulation impacts of the

26

mita. My results, if correct, point to a nuanced role of historical institutions in explaining economic growth. The undeniable importance of firms for economic development has led scholars to examine many drivers of business investments. The range of explanations offered so far in the literature has broadened the palette of prescriptions available to policymakers, yet the role of historical institutions has been largely underexplored. Influential research has recently shown that longdated institutions can severely constrain the human development capabilities of local populations today. This paper is one of the first to study the broader impact of historical institutions on firm growth working through a policy channel (taxes) affecting individual choices and attitudes towards investment and taxation in the very long term. Firm stunting is thus found to be quite different from human stunting. A deeper understanding of how and why firms grow in their institutional context is likely to have substantive welfare implications.

References Acemoglu, Daron, Simon Johnson, and James Robinson, 2002, Reversal of fortune: Geography and institutions in the making of the modern world income distribution, Quarterly Journal of Economics 117, 1231–1294. Alesina, Alberto, and Dani Rodrik, 1994, Distributive politics and economic growth, Quarterly Journal of Economics 109, 465–490. Algan, Yann, and Pierre Cahuc, 2010, Inherited trust and growth, American Economic Review 100, 2060–2092. Banerjee, Abhijit, and Lakshmi Iyer, 2005, History, institutions, and economic performance: The legacy of colonial land tenure systems in India, American Economic Review 95, 1190–1213. Besley, Timothy, and Torsten Persson, 2013, Taxation and development, in Martin Feldstein Alan Auerbach, Raj Chetty, and Emmanuel Saez, ed.: Handbook of Public Economics, vol. 5 of Handbook of Public Economics . chap. 2, pp. 51 – 110 (Elsevier). 27

Black, Sandra E., 1999, Do better schools matter? Parental valuation of elementary education, Quarterly Journal of Economics 114, 577–599. Bloom, Nick, and John Van Reenen, 2010, Why do management practices differ across firms and countries?, Journal of Economic Perspectives 24, 203–224. Brown, James R., J. Anthony Cookson, and Rawley Z. Heimer, 2016, Law and finance matter: Lessons from externally imposed courts, Review of Financial Studies 30, 1019–1051. Buera, Francisco, and Yongseok Shin, 2013, Financial frictions and the persistence of history: A quantitative exploration, Journal of Political Economy 121, 221–272. Cabral, Lu´ıs M.B., and Jos´e Mata, 2003, On the evolution of the firm size distribution: Facts and theory, American Economic Review 93, 1075–1090. Chirinko, Robert, and Daniel Wilson, 2008, State investment tax incentives: A zero-sum game?, Journal of Public Economics 92, 2362–2384. Conley, Timothy, 1999, GMM estimation with cross sectional dependence, Journal of Econometrics 92, 1–45. Cook, Noble D., 2004, Demographic Collapse: Indian Peru, 1520-1620 (Cambridge University Press). , Alejandro M´ alaga, and Th´erese Bouysse, 1975, Tasa de la Visita General de Francisco de Toledo (Universidad Nacional de San Marcos). de Soto, Hernando, 1989, The Other Path (Harper & Row: New York). , 2000, The Mystery of Capital: Why Capitalism Triumphs in the West and Fails Everywhere Else (Basic Books). Dell, Melissa, 2010, The persistent effects of Peru’s mining Mita, Econometrica 78, 1863–1903. Djankov, Simeon, Caralee McLiesh, and Andrei Shleifer, 2007, Private credit in 129 countries, Journal of Financial Economics 84, 299–329. 28

Dohmen, Thomas, Armin Falk, David Huffman, and Uwe Sunde, 2011, The intergenerational transmission of risk and trust attitudes, Review of Economic Studies 79, 645–677. Engen, Eric, and Jonathan Skinner, 1996, Taxation and economic growth, National Tax Journal pp. 617–642. Fafchamps, Marcel, and Bart Minten, 2002, Returns to social network capital among traders, Oxford Economic Papers 54, 173–206. Felkner, John S., and Robert M. Townsend, 2011, The geographic concentration of enterprise in developing countries, Quarterly Journal of Economics 126, 2005–2061. Garicano, Luis, Claire Lelarge, and John Van Reenen, 2016, Firm size distortions and the productivity distribution: Evidence from France, American Economic Review 106, 3439–3479. Gourio, Fran¸cois, and Nicolas Roys, 2014, Size-dependent regulations, firm size distribution, and reallocation, Quantitative Economics 5, 377–416. Guiso, Luigi, Paola Sapienza, and Luigi Zingales, 2009, Cultural biases in economic exchange?, Quarterly Journal of Economics 124, 1095–1131. Hopenhayn, Hugo, 2014, Firms, misallocation, and aggregate productivity: A review, Annual Review of Economics 6, 735–770. Hsieh, Chang-Tai, and Peter Klenow, 2009, Misallocation and manufacturing TFP in China and India, Quarterly Journal of Economics 124, 1403–1448. Iyer, Apsara, 2016, Agricultural and mining labor interactions in Peru: A long-run perspective (1571-1812), manuscript. Kirchler, Erich, 2007, The Economic Psychology of Tax Behaviour (Cambridge University Press). Kleven, Henrikand, and Mazhar Waseem, 2013, Using notches to uncover optimization frictions and structural elasticities: Theory and evidence from Pakistan, Quarterly Journal of Economics 128, 669–723. 29

Kumar, Krishna, Raghuram Rajan, and Luigi Zingales, 1999, What determines firm size?, NBER working paper. La Porta, Rafael, Florencio Lopez de Silanes, Andrei Shleifer, and Robert W. Vishny, 1997, Legal determinants of external finance, Journal of Finance 52, 1131–1150. Laeven, Luc, and Christopher Woodruff, 2007, The quality of the legal system, firm ownership, and firm size, Review of Economics and Statistics 89, 601–614. Leal-Ord´on ˜ez, Julio C., 2014, Tax collection, the informal sector, and productivity, Review of Economic Dynamics 17, 262–286. Loayza, Norman V., 1996, The economics of the informal sector: A simple model and some empirical evidence from Latin America, in Carnegie-Rochester conference series on public policy , vol. 45 pp. 129–162. Elsevier. Lucas, Robert E., Jr., 1978, On the size distribution of business firms, Bell Journal of Economics 9, 508–523. Macchiavello, Rocco, and Ameet Morjaria, 2015, The value of relationships: Evidence from a supply shock to Kenyan rose exports, American Economic Review 105, 2911–2945. Maccini, Sharon, and Dean Yang, 2009, Under the weather: Health, schooling, and economic consequences of early-life rainfall, American Economic Review 99, 1006–1026. McMillan, John, and Christopher Woodruff, 1999, Interfirm relationships and informal credit in Vietnam, Quarterly Journal of Economics 114, 1285–1320. Michalopoulos, Stelios, and Elias Papaioannou, 2014, National institutions and subnational development in Africa, Quarterly Journal of Economics 129, 151–213. Noejovich, H´ector, 2009, La transici´ on del sistema prehisp´anico al sistema econ´omico colonial, in Compendio de Historia Econ´ omica del Per´ u . , vol. 2 (Banco Central de Reserva del Per´ u).

30

Nunn, Nathan, 2008, The long-term effects of Africa’s slave trades, Quarterly Journal of Economics 123, 139–176. , 2012, Culture and the historical process, Economic History of Developing Regions 27, S108–S126. Pierce, Lamar, and Jason Snyder, 2017, The historical slave trade and firm access to finance in Africa, Review of Financial Studies forthcoming. Qian, Jun, and Philip Strahan, 2007, How laws and institutions shape financial contracts: The case of bank loans, Journal of Finance 62, 2803–2834. SRTM National Aeronautics and Space Administration and the National Geospatial Intelligence Agency, 2000, Shuttle Radar Topography Mission Global 30m Ellipsoidal, Available at OpenTopography. Tadelis, Steven, 1999, What’s in a name? Reputation as a tradeable asset, American Economic Review 89, 548–563. TePaske, John, and Herbert Klein, 1982, The Royal Treasuries of the Spanish Empire in America (Duke University Press). Wasylenko, Michael, 1999, Taxation and economic development: The state of the economic literature, Public Administration and Public Policy 72, 309–328.

31

Table 1:

Summary Statistics

Observations in Panel A are at the firm level for all firms in the 299 districts of the study zone as covered by the 2008 census of firms. Log of fixed assets is the logarithm of one plus fixed assets net of depreciation as of 31 December 2007 expressed in Peruvian soles. Log of inventories is equal to the logarithm of one plus the total amount of commercial good inventories as of 31 December 2007 expressed in Peruvian soles. The commercial name dummy equal to one when the firm responded to the census question about the commercial name with a valid answer. The business tax ID number dummy is equal to one when the firm responded in the affirmative to the census question about having a Registro Unico de Contribuyentes (RUC) number. Observations in Panel B are at the census-block level. Geospatial metrics are calculated using a resolution of 30 meters by 30 meters from SRTM (2000); the geocoded centroid of each census block is obtained by averaging the coordinates of the block’s borders. The observations employed for each test are restricted to a range of kilometers or districts around the mita borders. Robust standard errors are shown in parentheses, and Conley standard errors in brackets. ***, **,* stand for a significant difference of means at the 1%, 5% and 10% level, respectively.

Panel A: Firm variables Log of fixed assets Log of inventories Has a commercial name (1/0) Has a business tax ID number (1/0)

N.obs. 54675 54678 54678 54678

Mean 4.86 4.44 0.21 0.44

Std.dev. 3.69 3.86

Min. 0.00 0.00 0.00 0.00

Max. 21.89 19.28 1.00 1.00

Panel B: Geospatial metrics and tests < 100 km of Mita Boundary Inside Outside s.e.

Sample Falls Within... < 50 km of Mita Boundary Inside Outside s.e.

< 10 km of Mita Boundary Inside Outside s.e.

Elevation

3650

3257

(6.64)∗∗∗ [150.5]∗∗∗

3531

3240

(8.72)∗∗∗ [141.08]∗∗

3410

3002

(18.2)∗∗∗ [267.89]

Slope

7.28

7.43

(0.09)∗ [1.50]

8.54

7.67

(0.118)∗∗∗ [1.48]

10.1

10.53

(0.26)∗ [1.42]

22168

6844

10366

5020

1485

1654

Sample size

32

Table 2:

Fixed Assets

Observations are at the firm level for firms covered by the census in the study sample. The models estimate equation (1) introducing all controls and fixed effects. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Log of Fixed Assets Sample within:

< 100 km of boundary

< 50 km of boundary

< 10 km of boundary

Panel A. Cubic Polynomial in Latitude and Longitude −1.421∗∗∗ (−3.97) 0.15

Mita R2

−1.137∗∗∗ (−2.81) 0.20

−0.866∗∗ (−2.17) 0.25

Panel B. Cubic Polynomial in Distance to Potosi Mita R2

−1.116∗∗∗ (−3.22) 0.14

−0.569∗ (−1.92) 0.20

−1.370∗∗∗ (−3.26) 0.24

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

−0.897∗∗∗ (−3.75) 0.14

−0.508∗ (−1.80) 0.20

−1.572∗∗∗ (−3.70) 0.24

289 39665

185 16137

43 4026

33

Table 3:

Inventories

Observations are at the firm level for firms covered by the census in the study sample. The models estimate equation (1) introducing all controls and fixed effects. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Log of Inventories Sample within:

< 100 km of boundary

< 50 km of boundary

< 10 km of boundary

Panel A. Cubic Polynomial in Latitude and Longitude −1.428∗∗∗ (−4.63) 0.24

Mita R2

−1.318∗∗∗ (−3.76) 0.29

−1.543∗∗ (−2.42) 0.38

Panel B. Cubic Polynomial in Distance to Potosi Mita R2

−1.320∗∗∗ (−4.46) 0.23

−0.582∗∗ (−2.24) 0.28

−1.602∗∗ (−2.53) 0.36

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

−1.541∗∗∗ (−5.65) 0.24

−0.876∗∗∗ (−3.45) 0.29

−1.252∗∗∗ (−2.86) 0.37

289 39667

185 16137

43 4026

34

Table 4:

Commercial Name

Observations are at the firm level for firms covered by the census in the study sample. The models estimate equation (1) introducing all controls and fixed effects. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Has a Commercial Name Sample within:

< 100 km of boundary

< 50 km of boundary

< 10 km of boundary

Panel A. Cubic Polynomial in Latitude and Longitude −0.131∗∗∗ (−2.86) 0.11

Mita R2

−0.224∗∗∗ (−4.17) 0.13

−0.332∗∗∗ (−4.36) 0.16

Panel B. Cubic Polynomial in Distance to Potosi Mita R2

−0.119∗∗∗ (−2.79) 0.11

−0.134∗∗ (−2.54) 0.13

−0.233∗∗∗ (−3.19) 0.15

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

−0.121∗∗∗ (−3.22) 0.10

−0.130∗∗∗ (−2.72) 0.12

−0.202∗∗∗ (−4.15) 0.16

289 39667

185 16137

43 4026

35

Table 5:

Business Tax ID Number

Observations are at the firm level for firms covered by the census in the study sample. The models estimate equation (1) introducing all controls and fixed effects. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Has a Business Tax ID Number Sample within:

< 100 km of boundary

< 50 km of boundary

< 10 km of boundary

Panel A. Cubic Polynomial in Latitude and Longitude −0.118∗∗ (−2.37) 0.16

Mita R2

−0.118∗ (−1.94) 0.19

−0.234∗∗ (−2.10) 0.26

Panel B. Cubic Polynomial in Distance to Potosi Mita R2

0.018 (0.36) 0.15

−0.030 (−0.55) 0.19

−0.136∗ (−1.73) 0.26

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

−0.107∗∗ (−2.37) 0.15

−0.073 (−1.29) 0.18

−0.119∗∗ (−2.16) 0.28

289 39667

185 16137

43 4026

36

Table 6:

Local Individual and Household Conditions

Observations are either at the individual level (left-hand side panel) or at the household level (right-hand side panel). Only observations within 10 kilometers of the mita boundaries are employed in the estimation. The models estimate an equation analogous to (1). All regressions at the individual level control for the individual’s age. No firm-level controls are employed. Attending school is a dummy, measured only for individuals under 15 years of age. Has no insurance is a dummy calculated over any type of public or private insurance of any kind. Single is a marital-status dummy only recorded for individuals of age 12 or older. Own house is a dummy. House index is the first (most important) component of a principal components analysis. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Individual Is Attending School

Has No Insurance

Household Is Single

Owns House

House Index

Panel A. Cubic Polynomial in Latitude and Longitude 0.024∗∗∗ (2.98) 0.25

Mita R2

0.079∗∗∗ (3.33) 0.12

0.007 (0.79) 0.40

0.086 (1.34) 0.09

−0.166 (−1.17) 0.45

0.057 (1.25) 0.09

−0.068 (−0.40) 0.45

Panel B. Cubic Polynomial in Distance to Potosi 0.034∗∗ (2.22) 0.25

Mita R2

0.080∗∗∗ (4.23) 0.12

0.016∗ (1.96) 0.40

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

0.033∗∗∗ (2.73) 0.25

0.090∗∗∗ (4.73) 0.12

0.013∗ (1.71) 0.40

0.071∗ (1.76) 0.09

−0.287 (−1.58) 0.45

82 66423

84 213007

84 150654

84 51369

84 51369

37

Table 7:

Economic Environment in 1-square-kilometer Areas

Observations are at the 1-square-kilometer-area level for all areas within 10 kilometers of the mita border. The models estimate an equation analogous to (1). No firm-level controls are employed. Log number of bank branches and log light intensity are measured in 2007. Log growth of light intensity is the difference between the log of light intensity in 2007 and the log of light intensity in 1992 for the area. t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Log Number of Bank Branches

Log Light Intensity

Log Growth of Light Intensity

Panel A. Cubic Polynomial in Latitude and Longitude −0.000 (−1.34) 0.00

Mita R2

−0.005 (−0.22) 0.21

0.001 (0.03) 0.20

Panel B. Cubic Polynomial in Distance to Potosi −0.000 (−0.81) 0.00

Mita R2

0.011 (0.47) 0.20

0.015 (0.84) 0.18

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita R2 N. clusters (districts) Sample size

−0.000 (−0.99) 0.00

0.005 (0.19) 0.20

0.011 (0.62) 0.18

89 21297

89 21297

89 21297

38

Table 8:

Three Centuries of Colonial Taxes at the District Level

Observations are at the district-year level. Only districts with capitals within 50 kilometers of the mita border are included in the sample. The models estimate equation (2) introducing all controls and fixed effects. The first and second column display models using the yearly amount of taxes (from TePaske and Klein (1982)) as the dependent variable, in logarithms. The third and fourth column show models using taxes scaled by the adult male population of each local community from the 1572 census (from Cook, M´ alaga, and Bouysse (1975)). t-statistics based on standard errors clustered by district are in parentheses. ***, **,* stand for significance at the 1%, 5% and 10% level, respectively.

Dependent Variable: Log Taxes

Taxes scaled by pre-Mita 1572 Adult Males

Panel A. Cubic Polynomial in Latitude and Longitude 1.286∗∗ (2.39)

Mita Mita × 17th-century Mita × 18th-century R2

0.66

1.543∗∗ (2.54) −0.517 (−0.77) −0.891 (−1.06) 0.66

1.689∗ (1.85)

0.41

2.226∗∗ (2.40) −1.298 (−1.63) −1.319 (−1.41) 0.42

Panel B. Cubic Polynomial in Distance to Potosi Mita

1.392∗∗ (2.06)

Mita × 17th-century Mita × 18th-century R2

0.47

1.413∗ (1.94) 0.119 (0.13) −0.417 (−0.44) 0.47

1.661∗∗ (2.07)

0.38

1.996∗ (2.00) −0.791 (−0.95) −0.864 (−0.79) 0.39

Panel C. Cubic Polynomial in Distance to Mita Boundary Mita

1.196∗ (1.69)

R2

0.45

1.374 (1.63) −0.126 (−0.14) −0.857 (−1.06) 0.45

N. clusters (districts) Sample size

33 489

33 489

Mita × 17th-century Mita × 18th-century

39

1.239 (1.60)

0.41

1.603 (1.68) −0.687 (−0.76) −0.894 (−0.94) 0.41

33 489

33 489

Figure 1:

Three-dimensional Discontinuity Graphs of Firm Outcomes

This figure shows plots of the main variables on a Cartesian map of longitude and latitude, as described at the end of each subsection with regression results. The size of each dot represents the number of observations in each block using a 4-point scale.

40

Figure 2:

Three-dimensional Discontinuity Graphs of Local Conditions

This figure shows plots of the main variables on a Cartesian map of longitude and latitude, as described at the end of each subsection with regression results. The size of each dot represents the number of observations in each block using a 4-point scale. Panels A through D are based on individuals. Panels E and F are based on households.

41

Figure 3:

Three-dimensional Discontinuity Graphs of the Economic Environment

This figure shows plots of the main variables on a Cartesian map of longitude and latitude, as described at the end of each subsection with regression results. No dots are displayed because all areas in the geographic grid contribute only one observation to the analysis.

42

Figure 4: Three-dimensional Discontinuity Graphs of District-level Colonial Taxation

This figure shows plots of the main variables on a Cartesian map of longitude and latitude, as described at the end of each subsection with regression results. Each dot represents the location of district capitals with tax data available for the analysis.

43