Jenny Guardado FIGURE 2.Year to Year Price Differences in High versus Low Repartimiento Provinces.War periods shaded. 00 0 0 000.5 + r● 1680 1700 1720 1740 1760 Year Difference in Prices o High Reparto x Low Reparto substantial price increases are a sign of greater demand where Log(Pricei)represents the real price paid for for office. province i in bishop region j in year t;a and y cap- The empirical analysis thus follows a difference-in- tures provincial and year fixed effects,respectively.Xij differences approach by interacting the cross-sectional includes time-varying controls such as anti-governor variation in provincial profitability with plausibly ex- rebellions;Repartimientoi represents either a contin- ogenous time-variation in the likelihood of "low- uous measure of repartimiento quotas or an indicator quality"types to successfully bid for a position for provinces above or below the median of assigned during wars.Because European wars were fiscally oner- quotas (HighReparto);WarLength,is either the dura- ous to the Spanish treasury,the Crown was more tion of war in a given year,or the simple presence of in- 115.5010 likely to trade off "quality"for revenue at this time ternational conflict involving Spain (War,).Finally,Wit relative to others.More importantly,given the on- captures yearly trends for each of the seven different set and length of wars is largely driven by geopoliti- bishop regions to account for differential price trends. cal calculations in Europe they are thus less suscepti- In terms of standard errors,it is important to ac- ble to strategic timing:sales are driven by emergency count for serial correlation in office prices within a needs rather than by other considerations.In terms province.Because the number of clusters (44 or 48) of cross-sectional variation,I use the repartimiento is not large enough relative to the vector of fixed ef- quotas assigned to each province in 1754 which,for fects,time effects,and differential trends,standard er- many,was the main attraction behind serving in the rors clustered at the province level may be less cred- colonies. ible.Instead,I compute significance levels using wild By comparing provinces with high versus low repar- cluster bootstrap with the null hypothesis imposed as timiento during war and peace times,it is possible suggested by Cameron,Gehlbach,and Miller(2008). to capture shifts in the willingness to pay not driven Figure 2 provides some descriptive evidence by by province fundamentals.Furthermore,competing ex- showing the year to year difference in average office planations that either do not vary greatly within a prices for provinces with high and low repartimiento province over time (e.g.,prestige,location)or would throughout the period.As noted,price differences are not be different for positions with greater access to likely to be larger (more positive)during wartime rents precisely at times of greater scrutiny or not (e.g., periods (shaded)and smaller during peace (more altruism,career benefits),are also unlikely to be driving negative).However,given some exceptions are visi- the results.I thus estimate the following: ble,it is important to examine this relationship more systematically. Estimates from Table 1 show that provinces with Log(Priceiit)=aij+y+B(Repartimientoij x War) greater potential for profit via repartimiento fetch +Xiit +Wit+eiit, 1) higher prices relative to other provinces,particularly 976
Jenny Guardado FIGURE 2. Year to Year Price Differences in High versus Low Repartimiento Provinces. War periods shaded. substantial price increases are a sign of greater demand for office. The empirical analysis thus follows a difference-indifferences approach by interacting the cross-sectional variation in provincial profitability with plausibly exogenous time-variation in the likelihood of “lowquality” types to successfully bid for a position during wars.Because European wars were fiscally onerous to the Spanish treasury, the Crown was more likely to trade off “quality” for revenue at this time relative to others. More importantly, given the onset and length of wars is largely driven by geopolitical calculations in Europe they are thus less susceptible to strategic timing: sales are driven by emergency needs rather than by other considerations. In terms of cross-sectional variation, I use the repartimiento quotas assigned to each province in 1754 which, for many, was the main attraction behind serving in the colonies. By comparing provinces with high versus low repartimiento during war and peace times, it is possible to capture shifts in the willingness to pay not driven by province fundamentals. Furthermore, competing explanations that either do not vary greatly within a province over time (e.g., prestige, location) or would not be different for positions with greater access to rents precisely at times of greater scrutiny or not (e.g., altruism, career benefits), are also unlikely to be driving the results. I thus estimate the following: Log(Pricei jt) = αi j + γt + β(Repartimientoi j × Wart) + Xi jt + Wjt + i jt, (1) where Log(Priceijt) represents the real price paid for province i in bishop region j in year t; αij and γ t captures provincial and year fixed effects, respectively. Xi jt includes time-varying controls such as anti-governor rebellions; Repartimientoij represents either a continuous measure of repartimiento quotas or an indicator for provinces above or below the median of assigned quotas (HighReparto); WarLengtht is either the duration of war in a given year, or the simple presence of international conflict involving Spain (Wart). Finally, Wjt captures yearly trends for each of the seven different bishop regions to account for differential price trends. In terms of standard errors, it is important to account for serial correlation in office prices within a province. Because the number of clusters (44 or 48) is not large enough relative to the vector of fixed effects, time effects, and differential trends, standard errors clustered at the province level may be less credible. Instead, I compute significance levels using wild cluster bootstrap with the null hypothesis imposed as suggested by Cameron, Gehlbach, and Miller (2008). Figure 2 provides some descriptive evidence by showing the year to year difference in average office prices for provinces with high and low repartimiento throughout the period. As noted, price differences are likely to be larger (more positive) during wartime periods (shaded) and smaller during peace (more negative). However, given some exceptions are visible, it is important to examine this relationship more systematically. Estimates from Table 1 show that provinces with greater potential for profit via repartimiento fetch higher prices relative to other provinces, particularly 976 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800045X
Office-Selling,Corruption,and Long-Term Development in Peru TABLE 1.Office Prices and Repartimiento (1) (2) (3) DV:Log Prices(pesos) Panel A:Baseline WarLength x HighRepartimiento 0.019** 0.019** 0.016* (0.005) (0.005) (0.013) 0.0261 [0.0261 [0.0361 R-squared 0.833 0.833 0.828 Panel B:War Indicator War x HighRepartimiento 0.163* 0.164* 0.140* (0.036) (0.036) (0.066) [0.12] [0.11 [0.11] R-squared 0.831 0.831 0.827 Panel C:Continuous Repartimiento Measure WarLength x Repartimiento 0.014* 0.014* 0.011* (0.029) (0.029) (0.076) i0.0o1 i0.0o] i0.0o] R-squared 0.831 0.831 0.827 Mean DV 8.216 8.216 8.238 Observations 463 463 502 Provinces 44 44 48 Rebellion indicator No Yes Yes Provinces Bolivia No No Yes p-values in parentheses.Cluster-robust wild-bootstrap p-values in brackets All specifications include province FE,year FE,and time-trends for individual bishop regions.Bolivia provinces are four provinces ruled by the Audiencia of Charcas at the time (not Lima)but currently part of Peru.***p<0.01,**p< 0.05,p<0.1 when the Crown is less selective (during war).In terms the continuous measure of repartimiento with similar of magnitude,the coefficient of 0.019 in column 1.Panel results.In general,estimates do not change much when A,suggests that for provinces with a high repartimiento including an indicator for whether the province had a quota (above the median),an additional year at war rebellion in the year of sale(column 2),suggesting that leads to an increase in the prices of these provinces of their impact might have already been priced-in.6 Re- around 2%.Considering the average war lasts about sults also remain similar but less precise when including five years and the mean price of a Peruvian province in the sample four provinces that today are part of Peru is 5,300 pesos,this represents approximately 500 pesos but used to be part of Bolivia. or two times the yearly wage of a military captain in Parallel Trends.An important identifying assump- the Spanish army.This estimate is consistent with find- tion is that in the absence of war.price trends are par- ings from Panel B column 2,showing how more prof- allel among provinces with high and low repartimiento itable provinces exhibit 16%higher prices during war quotas.To validate this assumption,I employ a strategy times than peace.14 However,because not every year similar to the event study methodology by estimating saw the sale of provinces both above and below the me- the following: dian of the repartimiento quota,s Panel C instead uses Log(Priceiit)=aij+y 14 An alternative interpretation is that estimates are driven by low repartimiento provinces dropping prices during war.Yet,regressing B*I*HighRepartoij +xii+wit+eijt the difference in prices between war and peace on both groups of te{-4+,-3.-2,0.1.2.3,4+} provinces(excluding the constant)shows that the difference in prices is greater for high repartimiento provinces(B1=2.370 pesos)than (2) for low repartimiento ones (B2 =1,002 pesos)with the F-test of equality in the coefficients significant at the 10%level (p-value 0.08).I thank an anonymous referee for raising this possibility. 1577%of years with sales had a province both below and above the 16 Given buyers had an incentive to"know"well the province they repartimiento median sold. were purchasing,this is the most likely explanation. 977
Office-Selling, Corruption, and Long-Term Development in Peru TABLE 1. Office Prices and Repartimiento (1) (2) (3) DV: Log Prices (pesos) Panel A: Baseline WarLength × HighRepartimiento 0.019∗∗∗ 0.019∗∗∗ 0.016∗∗ (0.005) (0.005) (0.013) [0.026] [0.026] [0.036] R-squared 0.833 0.833 0.828 Panel B: War Indicator War × HighRepartimiento 0.163∗∗ 0.164∗∗ 0.140∗ (0.036) (0.036) (0.066) [0.12] [0.11] [0.11] R-squared 0.831 0.831 0.827 Panel C: Continuous Repartimiento Measure WarLength × Repartimiento 0.014∗∗ 0.014∗∗ 0.011∗ (0.029) (0.029) (0.076) [0.00] [0.00] [0.00] R-squared 0.831 0.831 0.827 Mean DV 8.216 8.216 8.238 Observations 463 463 502 Provinces 44 44 48 Rebellion indicator No Yes Yes Provinces Bolivia No No Yes p-values in parentheses. Cluster-robust wild-bootstrap p-values in brackets. All specifications include province FE, year FE, and time-trends for individual bishop regions. Bolivia provinces are four provinces ruled by the Audiencia of Charcas at the time (not Lima) but currently part of Peru. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 when the Crown is less selective (during war). In terms of magnitude, the coefficient of 0.019 in column 1,Panel A, suggests that for provinces with a high repartimiento quota (above the median), an additional year at war leads to an increase in the prices of these provinces of around 2%. Considering the average war lasts about five years and the mean price of a Peruvian province is 5,300 pesos, this represents approximately 500 pesos or two times the yearly wage of a military captain in the Spanish army. This estimate is consistent with findings from Panel B column 2, showing how more profitable provinces exhibit 16% higher prices during war times than peace.14 However, because not every year saw the sale of provinces both above and below the median of the repartimiento quota,15 Panel C instead uses 14 An alternative interpretation is that estimates are driven by low repartimiento provinces dropping prices during war. Yet, regressing the difference in prices between war and peace on both groups of provinces (excluding the constant) shows that the difference in prices is greater for high repartimiento provinces (B1 = 2,370 pesos) than for low repartimiento ones (B2 = 1,002 pesos) with the F-test of equality in the coefficients significant at the 10% level (p-value = 0.08). I thank an anonymous referee for raising this possibility. 15 77% of years with sales had a province both below and above the repartimiento median sold. the continuous measure of repartimiento with similar results. In general, estimates do not change much when including an indicator for whether the province had a rebellion in the year of sale (column 2), suggesting that their impact might have already been priced-in.16 Results also remain similar but less precise when including in the sample four provinces that today are part of Peru but used to be part of Bolivia. Parallel Trends. An important identifying assumption is that in the absence of war, price trends are parallel among provinces with high and low repartimiento quotas. To validate this assumption, I employ a strategy similar to the event study methodology by estimating the following: Log(Pricei jt) = αi j + γt + τ∈{−4+,−3,−2,0,1,2,3,4+} βτ ∗ Iτ ∗ HighRepartoi j + xi jt + wjt + i jt (2) 16 Given buyers had an incentive to “know” well the province they were purchasing, this is the most likely explanation. 977 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800045X
Jenny Guardado FIGURE 3. Coefficients from Table 2 ears before ears before befor into nto ears ear 95%Cl ●Coefficient where I is a set of seven dummy variables equal to 1 if sion of time-varying measures of revenue collection at t years had passed since the start of war(-4+<t <4 the caja level17 change little the baseline estimates.The +)where-4+refers to more than 4 years before a war latter reduces concerns that higher prices are mostly and 4+refers to more than 4 years into the war.The driven by reported fiscal revenue. year immediately before a war is left as the comparison Robustness.Other robustness checks allow me to as- group and assigned zero in Figure 3 for visualization sess the sensitivity of the baseline results.First,in Table purposes.If the coefficient of the periods prior to a war A.3 of the Appendix,I provide a nonparametric way (B-4+,B-3 and B-2)are not significantly different from of evaluating the relationship of interest by splitting 21105.501090 zero,the parallel trends assumption is likely to hold. the continuous measure of repartimiento quotas into Results presented in Figure 3 and Table 2 show that quartiles and interacting them with the length of Eu- prices in provinces with high repartimiento versus oth- ropean wars.As noted,coefficients become larger and ers did not significantly differ in the years immediately more precise the higher the value of the conditioning before a war,but that the divergence actually took variable,consistent with the findings from the linear place during war times as can be seen in the increase models. in the size of the coefficient in the first year of war and Second,in Table A.4 I limit the sample to only those remaining positive in subsequent years.It is only after conflicts in which Spain was theoretically less suscepti- more than four years at war that the difference exhibits ble to manipulate the onset and length of wars.Specif- greater statistical precision. ically,"succession"wars,where the timing is driven by Time-Varying Confounders.In addition to checking the sudden death of a monarch.18 Results show that the for pretreatment trends,it is important to rule out other findings are not driven by potentially endogenous entry factors unrelated to extraction that would be driving into wars. the increase in office prices during wars in Europe. Given the main effect is an interaction term,it is To do so,I interact traits such as the distance of the susceptible to potential outliers.Therefore,I estimate province to the capital (Lima),the elevation of the the preferred specification(column 2 of Table 1)while province,and an indicator for the presence of a bishop leaving out two provinces at the time (out of 44)to seat with the length of war to examine the stability of assess the sensitivity of the results to changes in the the baseline coefficients.Table 3 shows that the com- bined effect of repartimiento quotas and wartime in Europe is still the main determinant of the increase in 17 Peru was fiscally organized in 13 cajas or revenue-collection office prices rather than reflecting"better"provinces hese are the War of the Spanish Succession (1704-1714)and in terms of location,elevation,distance to capital,or a War of the Polish(1733-1738)and Austrian Succession (1740-1748), bishop seat.Furthermore,Panel B shows that the inclu- which overlaps with Jenkins'Ear War(1738-1739). 978
Jenny Guardado FIGURE 3. Coefficients from Table 2 -.5 0 .5 Change in Log(Office Prices) 4+ years before 3 years before 2 years before 1 year before 1 year into 2 years into 3 years into 4+ years into 95 % CI Coefficient where Iτ is a set of seven dummy variables equal to 1 if τ years had passed since the start of war (−4 + ≤τ ≤ 4 +), where −4+ refers to more than 4 years before a war and 4+ refers to more than 4 years into the war. The year immediately before a war is left as the comparison group and assigned zero in Figure 3 for visualization purposes. If the coefficient of the periods prior to a war (β−4 +,β−3 and β−2 ) are not significantly different from zero, the parallel trends assumption is likely to hold. Results presented in Figure 3 and Table 2 show that prices in provinces with high repartimiento versus others did not significantly differ in the years immediately before a war, but that the divergence actually took place during war times as can be seen in the increase in the size of the coefficient in the first year of war and remaining positive in subsequent years. It is only after more than four years at war that the difference exhibits greater statistical precision. Time-Varying Confounders. In addition to checking for pretreatment trends,it is important to rule out other factors unrelated to extraction that would be driving the increase in office prices during wars in Europe. To do so, I interact traits such as the distance of the province to the capital (Lima), the elevation of the province, and an indicator for the presence of a bishop seat with the length of war to examine the stability of the baseline coefficients. Table 3 shows that the combined effect of repartimiento quotas and wartime in Europe is still the main determinant of the increase in office prices rather than reflecting “better” provinces in terms of location, elevation, distance to capital, or a bishop seat. Furthermore, Panel B shows that the inclusion of time-varying measures of revenue collection at the caja level17 change little the baseline estimates. The latter reduces concerns that higher prices are mostly driven by reported fiscal revenue. Robustness.Other robustness checks allow me to assess the sensitivity of the baseline results. First, in Table A.3 of the Appendix, I provide a nonparametric way of evaluating the relationship of interest by splitting the continuous measure of repartimiento quotas into quartiles and interacting them with the length of European wars. As noted, coefficients become larger and more precise the higher the value of the conditioning variable, consistent with the findings from the linear models. Second, in Table A.4 I limit the sample to only those conflicts in which Spain was theoretically less susceptible to manipulate the onset and length of wars. Specifically, “succession” wars, where the timing is driven by the sudden death of a monarch.18 Results show that the findings are not driven by potentially endogenous entry into wars. Given the main effect is an interaction term, it is susceptible to potential outliers. Therefore, I estimate the preferred specification (column 2 of Table 1) while leaving out two provinces at the time (out of 44) to assess the sensitivity of the results to changes in the 17 Peru was fiscally organized in 13 cajas or revenue-collection regions. 18 These are the War of the Spanish Succession (1704–1714) and the War of the Polish (1733–1738) and Austrian Succession (1740–1748), which overlaps with Jenkins’ Ear War (1738–1739). 978 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800045X
Office-Selling,Corruption,and Long-Term Development in Peru TABLE 2.War and Office-Prices (1) (2) (3) DV:Log Prices(Pesos) Panel A 4+years before war x HighReparto 0.145 0.150 0.102 (0.393) (0.379) (0.545) 「0.4821 0.4481 0.5941 3 years before war x HighReparto 0.003 0.003 0.048 (0.987 (0.988) (0.785) f0.948 0.951 「0.831 2 years before war x HighReparto -0.002 -0.019 0.135 (0.992) (0.936) (0.522) f1.001 0.9641 0.5161 1 year after war x HighReparto 0.132 0.132 0.130 (0.415) (0.414) (0.426) f0.5721 [0.5681 「0.4921 2 years after war x HighReparto 0.162 0.163 0.151 (0.281) (0.280) (0.319) [0.4181 [0.4181 「0.4241 3 years after war x HighReparto 0.071 0.071 0.062 (0.722) (0.723) (0.759) [0.79 [0.79] [0.8181 4+years after war x HighReparto 0.228* 0.228** 0.220* (0.048) (0.049) (0.057) [0.26] [0.262] [0.234 Mean DV 8.216 8.216 8.238 Observations 463 463 502 R-squared 0.832 0.832 0.828 Provinces 44 44 48 Rebellion indicator No Yes Yes Provinces Bolivia No No Yes p-values in parentheses.Cluster-robust wild-bootstrap p-values in brackets. All specifications include province FE,year FE,and time-trends for individual e2sanepoetarcm88ogeamop0baHpe60ereg 0.05,*p<0.1 sample.As shown in Figure A.5 of the Appendix,the bility (Duke,Marquis,Count,etc.),to a nobility order t-statistic is always well-above conventional levels of (Knight of the Santiago or Calatrava order),or have statistical significance and coefficients remain similar pursued a career in the army(captain,sergeant,among to the baseline.An additional exercise in Table A.5 others). shows that the result is robust to including flexible As discussed more extensively above,in eighteenth province-specific trends(as opposed to regional ones) century Spain social status was a sign (if imperfect)of to account for potential omitted variables. economic status,social capital and connections to the In sum,results show that buyers paid more for more royal court in Madrid,as well as of lineage and a fam- profitable positions during war times relative to peace, ily reputation.19 In fact,the laws governing the Indies the key question is then:Why? established that individuals of high-social status were the best suited to serve as corregidores in the colonies (Lohmann Villena 1957,100).In line with this,Panel A Mechanism:Negative Selection of Colonial of Table A.6 shows that the Crown preferred to appoint Officials? those with military careers or nobility titles,whereas those lacking social status were more likely to access Does the increase in office prices reflect a change in office via purchase.Such a preference is optimal in a the type of officials in provinces with greater oppor- context of poor information and monitoring since it tunities for extraction versus others?Although mea- suring "quality"in this context is no easy task,I pro- vide evidence based on one dimension:the social status 19It was quite common for individuals to be rewarded by the Crown for services or military accomplishments done by relatives(e.g.,par- of purchasers.That is,whether they belong to the no- ents,uncles,grandparents). 979
Office-Selling, Corruption, and Long-Term Development in Peru TABLE 2. War and Office-Prices (1) (2) (3) DV: Log Prices (Pesos) Panel A 4+ years before war × HighReparto 0.145 0.150 0.102 (0.393) (0.379) (0.545) [0.482] [0.448] [0.594] 3 years before war × HighReparto 0.003 0.003 0.048 (0.987) (0.988) (0.785) [0.948] [0.95] [0.83] 2 years before war × HighReparto –0.002 –0.019 0.135 (0.992) (0.936) (0.522) [1.00] [0.964] [0.516] 1 year after war × HighReparto 0.132 0.132 0.130 (0.415) (0.414) (0.426) [0.572] [0.568] [0.492] 2 years after war × HighReparto 0.162 0.163 0.151 (0.281) (0.280) (0.319) [0.418] [0.418] [0.424] 3 years after war × HighReparto 0.071 0.071 0.062 (0.722) (0.723) (0.759) [0.79] [0.79] [0.818] 4+ years after war × HighReparto 0.228∗∗ 0.228∗∗ 0.220∗ (0.048) (0.049) (0.057) [0.26] [0.262] [0.234] Mean DV 8.216 8.216 8.238 Observations 463 463 502 R-squared 0.832 0.832 0.828 Provinces 44 44 48 Rebellion indicator No Yes Yes Provinces Bolivia No No Yes p-values in parentheses. Cluster-robust wild-bootstrap p-values in brackets. All specifications include province FE, year FE, and time-trends for individual bishop regions. Bolivia provinces are four provinces ruled by the Audiencia of Charcas at the time (not Lima) but currently part of Peru. ∗∗∗ p < 0.01, ∗∗ p < 0.05, ∗ p < 0.1 sample. As shown in Figure A.5 of the Appendix, the t-statistic is always well-above conventional levels of statistical significance and coefficients remain similar to the baseline. An additional exercise in Table A.5 shows that the result is robust to including flexible province-specific trends (as opposed to regional ones) to account for potential omitted variables. In sum, results show that buyers paid more for more profitable positions during war times relative to peace, the key question is then: Why? Mechanism: Negative Selection of Colonial Officials? Does the increase in office prices reflect a change in the type of officials in provinces with greater opportunities for extraction versus others? Although measuring “quality” in this context is no easy task, I provide evidence based on one dimension: the social status of purchasers. That is, whether they belong to the nobility (Duke, Marquis, Count, etc.), to a nobility order (Knight of the Santiago or Calatrava order), or have pursued a career in the army (captain, sergeant, among others). As discussed more extensively above, in eighteenth century Spain social status was a sign (if imperfect) of economic status, social capital and connections to the royal court in Madrid, as well as of lineage and a family reputation.19 In fact, the laws governing the Indies established that individuals of high-social status were the best suited to serve as corregidores in the colonies (Lohmann Villena 1957, 100). In line with this, Panel A of Table A.6 shows that the Crown preferred to appoint those with military careers or nobility titles, whereas those lacking social status were more likely to access office via purchase. Such a preference is optimal in a context of poor information and monitoring since it 19 It was quite common for individuals to be rewarded by the Crown for services or military accomplishments done by relatives (e.g., parents, uncles, grandparents). 979 Downloaded from https://www.cambridge.org/core. Shanghai JiaoTong University, on 26 Oct 2018 at 03:53:05, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1017/S000305541800045X