Baseline Model—Investing in Capacity

Now suppose the government has initial levels of fiscal and legal capacity, ${\tau _0} \lt \bar \tau $ and π ₀ < 1, and that M < απY. It can invest in higher levels of capacity, ${\pi _1} \le \bar p = 1$ and ${\tau _1} \le \bar t$ , at cost L(π ₁ – π ₀) and F(τ ₁ – τ ₀), where L′, F′ > 0 and L″, F″ > 0. The optimization function is then

(8) $$\mathop {{\rm{max}}}\limits_{G,c,{\pi _1},{\tau _1}} \alpha ln\left( G \right)\; + \;ln\left( c \right)$$

subject to

(9) $$c = \left( {1-{\tau _1}} \right){\pi _1}Y$$

(10) $$G + L\left( {{\pi _1} + {\pi _0}} \right) + F\left( {{\tau _1} + {\tau _0}} \right) = {\tau _1}{\pi _1}Y + M.$$

Using Equations (9) and (10), we can simplify Equation (8) to

(11) $$\mathop {\max }\limits_{{\pi _1},{\tau _1}} \alpha \ln \left( {{\tau _1}{\pi _1}Y + M - L\left( {{\pi _1} - {\pi _0}} \right) - F\left( {{\tau _1} - {\tau _0}} \right)} \right) + \ln \left( {\left( {1 - {\tau _1}} \right){\pi _1}Y} \right)$$ .

The first-order conditions for π ₁ and τ ₁ are

(12) $$\frac{{\alpha \left( {\tau _1^*Y - L'\left( {\pi _1^* - {\pi _0}} \right)} \right)}}{{\tau _1^*\pi _1^*Y + M - L\left( {\pi _1^* - {\pi _0}} \right) - F\left( {\tau _1^* - {\tau _0}} \right)}} + \frac{1}{{\pi _1^*}} = 0$$

(13) $$\frac{{\alpha \left( {\pi _1^*Y - F'\left( {\tau _1^* - {\tau _0}} \right)} \right)}}{{\tau _1^*\pi _1^*Y + M - L\left( {\pi _1^* - {\pi _0}} \right) - F\left( {\tau _1^* - {\tau _0}} \right)}} - \frac{1}{{1 - \tau _1^*}} = 0.$$

The first-order condition for π ₁, Equation (12), shows the government always wants to invest in legal capacity. Suppose there was a condition where the government did not want to invest in legal capacity, such that π ₁ = π ₀ < 1. Then, from Equation (12), the following identity must hold.

(14) $$\frac{{\alpha \tau _1^*Y}}{{\tau _1^*{\pi _0}Y + M - F\left( {\tau _1^* - {\tau _0}} \right)}} + \frac{1}{{{\pi _0}}} = 0$$

But since both terms are non-negative (the denominator in the first term is positive since this is just G*), then this is a contradiction. Thus, the government always wants to invest in legal capacity. Intuitively, this is because increasing legal capacity carries two benefits: increasing both the public good and private consumption in the second period.

In Online Appendix A.3, I show that these first-order conditions yield

(15) $$\frac{{\partial \tau _1^*}}{{\partial M}} \lt 0$$ ,

(16) $$\frac{{\partial \pi _1^*}}{{\partial M}} \gt 0.$$

The first statement says that when M < απY, an increase in M will decrease investments in fiscal capacity. Fiscal capacity brings the benefit of more public goods, but at the cost of less private consumption. But M provides a source of revenue for investments that do not decrease private consumption. So the government does not need more fiscal capacity. But the second statement says that more M causes the government to increase legal capacity because it is a way to transfer higher customs revenue to higher private consumption. Note that this prediction still holds under the political control case because the government always wants to increase legal capacity.

Thus, the model gives two testable predictions from a change in customs revenue. First, a significant drop in customs revenue should cause the government to invest in fiscal capacity. Second, the drop should cause the government to decrease investments in legal capacity. Yet, a casual assessment suggests that the prediction for legal capacity might not hold. While countries that depend on external revenues have low fiscal capacity, they often also have low legal capacity. If external revenues do not increase legal capacity, the model assumptions need to be reassessed.

Administrative Complementarities

The model noted previously follows the assumption in Besley and Persson (Reference Besley and Persson2009) that the cost of investing in fiscal and legal capacity depends only on how much capacity changes. But there is an argument that the administration of fiscal and legal capacity might create complementarities, such that investing in one might decrease the cost of investing in the other. For example, registering property titles (legal capacity) makes it easier to collect property taxes (fiscal capacity). In France (Johnson and Koyama Reference Johnson and Koyama2014), consolidating the tax regime (fiscal capacity) led to an increase in judges upholding the law (legal capacity). The discussion about Haiti demonstrates this principle too. The government could not improve tax collection (fiscal capacity) without better law enforcement (legal capacity), and it could not improve property registration (legal capacity) because it did not have the revenue to pay for the project (fiscal capacity). This complementarity might require new assumptions that could alter the model’s predictions.

To look at administrative complementarities, I modify the cost functions. The cost of increasing one aspect of capacity depends not just on the size of the change but also on the level of the other aspect of state capacity. For example, if the government wants to increase revenues from property taxes, it has to register and protect properties. Or there could be more explicit complementarities if the government uses the same office to both register properties and collect revenues. So the cost of investing in fiscal capacity is F(τ ₁ – τ _{0 ,} π ₁) with F _π < 0, F _ππ > 0, F _πτ < 0. Similarly, the cost of investing in legal capacity is L(π ₁ – π _0, τ ₁) with L _τ < 0, L _ττ > 0, L _πτ < 0. Adding these cost functions to the government’s optimization problem means the government seeks to maximize

(17) $$\mathop {\max }\limits_{{\pi _1}{\tau _1}} \left( {\alpha \ln \left( {{\tau _1}{\pi _1}Y + M - L({\pi _1} - {\pi _0},{\tau _1}} \right) - F\left( {{\tau _1} - {\tau _0},{\pi _1}} \right)) + \ln \left( {\left( {1 - {\tau _1}} \right){\pi _1}Y} \right)} \right).$$

In Online Appendix A.4, I show that the complementarities imply that we do not know the signs of $\frac{{\partial \pi _1^*}}{{\partial M}}$ or $\frac{{\partial \tau _1^*}}{{\partial M}}$ without further knowledge about the cost functions. Thus, complementarities could reverse the relationship between legal capacity and external revenues, such that more external revenues could lead to lower legal capacity.

The intuition is simple to grasp if we walk through a change in external revenues. When external revenues increase, the government wants to decrease fiscal capacity. But a decrease in fiscal capacity raises the cost of investing in legal capacity, which pushes the government to decrease investments in legal capacity. If the government cannot collect sufficient tax revenue, it cannot pay surveyors or judges who settle property disputes. This introduces tension. The government wants to decrease fiscal capacity to increase private consumption. But lower fiscal capacity discourages investments in legal capacity, and lower legal capacity results in lower private consumption. Thus, the government must balance these trade-offs between the savings from decreasing capacity investments and the loss to private consumption. But we can only understand these trade-offs with knowledge of the cost functions.

The two models produce testable predictions. If investments in fiscal and legal capacity are independent of each other, then a drop in external revenues should cause the state to increase investments in fiscal capacity and decrease investments in legal capacity. But if the drop in external revenues causes the government to increase both fiscal and legal capacity, then there is evidence that the investments in capacity are complementary.

Fiscal Capacity before and after Mobilization

While the Haitian government focused on improving internal revenues under the U.S. occupation, its focus became sharper when there were external revenue shocks. When coffee revenues dropped in 1929, the government increased tax collections (Schmidt Reference Schmidt1971, p. 196). Then, in response to the Great Depression, it again increased internal tax collections (Schmidt Reference Schmidt1971, p. 221) and modified several internal taxes, including a gasoline tax, a stamp tax, and income taxes.Footnote ⁴ The government did not respond to subsequent fluctuations in the 1930s with large reforms, though that might be because the earlier reforms anticipated a period of greater variation. But at the end of 1941, when the United States mobilized in response to Pearl Harbor, the Haitian government faced a budget crisis.

As America’s entrance into the war restructured Haiti’s trade, customs revenues fell precipitously. Customs receipts in 1941–42 were lower than at any point in the previous 20 years, including every year of the Depression (Banque Nationale de la Republique d’Haiti 1942, p. 62). The fall came because the war diverted high-tariff imports like cars and cement. Unlike previous revenue shortfalls in the decade, customs remained persistently low for years (see Table 1). Anticipating a sustained shock, President Élie Lescot’s administration rushed to find new sources of revenue. “Before many weeks of war had passed, it became evident that new methods would have to be devised and special arrangements made in order to enable the country to ride out the storm” (Banque Nationale de la Republique d’Haiti 1942, p. 2). The administration appealed to the United States for help, received a line of credit from the Export-Import Bank, and passed a special tax on the country’s largest export, coffee (Banque Nationale de la Republique d’Haiti 1942, pp. 2–3, 7). But its efforts were insufficient.

Table 1 EXPORT VALUE AND TAXES COLLECTED BY HAITIAN GOVERNMENT, 1932–1948

Notes: Figures are in thousands of gourdes and are adjusted for inflation (constant 1930 Gourdes). Export value indicates the value of all exports from Haiti; external revenues are tax revenues collected on exports and imports at customhouses, and internal revenues are taxes collected on economic activity within the country. The income tax is part of internal revenues.

Sources: Annual Reports of Haiti Bureau du Representant Fiscal, 1932–1948.

Their next step was unsurprising given the baseline model: the drop in customs stimulated an expansion of fiscal capacity. Despite the drop in trade taxes, the economy as a whole was performing well. GDP estimates are unavailable for this period, but Table 1 shows the value of exports was over 40 percent higher in 1945 than it was in 1941. This strong economic performance meant that increasing internal revenues was a shrewd policy for a desperate government. In 1942, the government, in partnership with the U.S. fiscal representative, resorted to reforming the income tax.Footnote ⁵ It increased rates for almost all brackets (the median rate increase was 70 percent, see Table 2) and changed the top tax rate from 6 percent to 15 percent. Under the old law, the government treated business income (from sociétés anonymes) differently than personal income, but with the reform, it treated the two sources the same. If the government could enforce the law, the reform would significantly increase internal revenues.

Table 2 HAITIAN INCOME TAX SCHEDULE, 1932 AND 1942

Notes: Figures are in Haitian gourdes, unadjusted for inflation.

Sources: Rates from 1932 come from Le Moniteur, 29 September 1932. Rates from 1942 come from Le Moniteur, 25 May 1942.

To collect more revenue, the Haitian Internal Revenue Service (HIRS) had to overcome two barriers. First, it needed more personnel. From 1926 to 1931, HIRS employed 80–90 rural agents responsible for collecting communal taxes. By 1933, the number of employees had jumped to 151, but the HIRS administrators had little confidence in the new hires. “Due to the fact that there are many districts in which receipts are so sparse that it is not possible to pay local agents adequate salaries for their collection, the Internal Revenue Service is still considerably handicapped in getting honest and efficient local officers. The turnover is consequently very large among these local agents,” (Haiti Bureau du Representant Fiscal 1933, pp. 128–29). Moreover, only a fraction of the agents worked in the regions outside of the western population center: the three regions bordering the Dominican Republic in the east employed 36 agents total in 1933, and only 17 in 1931. After 1933, HIRS did not report personnel data, but its nominal expenditures on wages and salaries soared after the reform (see Figure 8). Specifically, in the first year of the reform, “the staff of traveling inspectors was considerably increased” (Banque Nationale de la Republique d’Haiti 1943, p. 12). In a rural country with few major population centers, traveling inspectors were crucial to collecting taxes.

Second, HIRS needed more physical capital. To help the newly hired traveling inspectors, HIRS purchased new automobiles (Banque Nationale de la Republique d’Haiti 1943, p. 12). Prior to the reform, HIRS did not have much physical infrastructure. “The Internal Revenue Service maintains agencies throughout the Republic, and many of these agencies in the past have been poorly housed or completely lacking in Government owned quarters” (Banque Nationale de la Republique d’Haiti 1944, p. 11). This changed with the reform. “A program of construction of internal revenue offices was undertaken in the course of the year” (idem.). Furthermore, HIRS had historically struggled to track who owed taxes, but with the new budget “the Internal Revenue Service also undertook the construction of an archives building” (Banque Nationale de la Republique d’Haiti 1943, p. 12), which was completed the next year (Banque Nationale de la Republique d’Haiti 1944, p. 11).

These investments contributed to increasing tax revenues not just in the year the reform passed but in subsequent years as well. Before the reform, real income tax revenues averaged 551,000 gourdes per year (see Table 1).Footnote ⁶ But in 1943, the first year the reform was fully implemented, these revenues jumped to 1,868,000 gourdes, more than three times the pre-reform average. By 1948, revenues had increased sevenfold to 4,220,000. The reform had transformed the role of the income tax. Before the reform, income tax revenues were about 10 percent of all internal taxes. After the reform, the income tax became the most important source of internal revenue, comprising 50 percent of total internal taxes by 1948. The substantial growth in income taxes suggests the reform did more than just increase tax rates; it collected taxes more efficiently. But since tax revenues can grow for multiple reasons, I decompose the growth to find how much fiscal capacity has contributed.

Legal Capacity in Haiti before U.S. Mobilization

Like with all countries, measuring legal capacity in Haiti is difficult, but I approach this challenge by looking at the government’s land rental program. This government program leased public land to farmers and residents. It let tenants choose the property, cultivate their own crops, keep all income generated from the land, and have exclusive access to it. Thus, the program reflected the government’s legal capacity because the leases granted the tenant property rights. Protecting property rights on the rented properties was not a public good; only one person could hold the title, and that person could prevent others from using the property. Since the program provided exclusive rights, the tenants paid rent on the lease. It did not, however, give tenants alienation rights, so the tenant could not sell the property or use it as collateral. Even though the property rights were incomplete, the rental program gave the tenant a set of property rights that required legal capacity to protect.

Enforcing property rights sometimes stretched the state’s legal capacity. The demands on capacity can be seen in a 1928 report that listed some of the cases that came before the state (Haiti Bureau du Representant Fiscal 1928, pp. 73–75):

The example exhibits the legal capacity needed to establish a claim: adjudicating competing claims, physically searching the property for the original boundary, convincing the claimants to accept the decision, and investigating fraudulent titles.

One sign of insufficient legal capacity is the recovery rate for rent. From 1930 to 1932, the average recovery rate for all land rental revenue was 51 percent (see Table 3). But rates varied significantly across the 10 administrative regions. Average recovery rates in three regions were at or above 70 percent, while they were less than 40 percent in three other regions. These rates could be low for a few reasons. First, the rental program might not have had the capacity to collect the revenues. A second reason could be that the program’s administrators did not know who rented which plots or how much they owed. A program with insufficient capacity to track rents does not have the capacity to enforce claims. Finally, it might not have had sufficient capacity to evict tenants who do not pay rent. But if the program did not have the capacity to remove tenants, it could not enforce property rights.

Table 3 FRACTION OF LAND RENTS RECOVERED BY DISTRICT, 1930–1932

Notes: The recovery rate in a district is defined as the fraction of rents due that were collected on time that year. The total recovery rate accounts for all rents due across the country.

Source: Recovery rates come from the Annual Reports of Haiti Bureau du Representant Fiscal, 1930–1932.

The Trujillo Massacre and Legal Capacity

Although there is evidence that Haiti’s legal capacity was low, that does not mean it was not optimal. In the model, the optimal level of legal capacity is π = 1. But it is unclear what π = 1 means in the real world. For example, while it seems that optimal protection would include always respecting property rights, there are many examples where economic development came from violating rights (Lamoreaux Reference Lamoreaux2011). To test the model, we need to know if legal capacity is below the optimal level. While we might not know what is optimal, we can observe whether the legal capacity constraint is binding.

To show legal capacity was below Haiti’s optimum before U.S. mobilization, I use a demand shock to the land rental program caused by the Trujillo Massacre. In October 1937, President Rafael Trujillo of the Dominican Republic ordered the Dominican army to massacre ethnic Haitians living in his country. The massacre was focused along the border in the north-western region of the Dominican Republic, but the entire country felt its effects. From the 1936 to 1950 census, the Dominican Republic lost at least 30,000 Haitians (Palsson Reference Palsson2023b), of whom about 12,000 died in the massacre (Vega Reference Vega1995). The survivors flooded into Haiti. Many of the refugees were repatriated Haitians, but a significant portion were Dominicans of Haitian descent who had never been to Haiti. While the magnitude of the refugee shock is unknown, Palsson (Reference Palsson2023b) suggests that it increased the population of districts near the refugee camps by 8 percent.

Despite the large refugee population movements, Haiti’s government did little to support them or confront the Dominican Republic. Led by President Lescot’s predecessor, President Stenio Vincent, the government started refugee camps near the border in the North and South (see Figure 2). The government intended to use the camps to coordinate aid, but it failed to provide the promised services (Pierre-Charles Reference Pierre-Charles1965, pp. 111–12). Not only did Vincent’s administration provide inadequate support to the refugees, but it also avoided confronting Trujillo. Vincent did not strengthen border security or threaten the Dominican Republic (Smith Reference Smith2009, pp. 31–32). Even when Vincent got Trujillo to agree to pay a meager $750,000 indemnity (about $14 million in 2020), he later settled for $525,000 ($10 million in 2020), of which little went to the refugees (Heinl, Heinl, and Heinl 2005, p. 482).Footnote ⁷ In fact, President Vincent went out of his way to avoid conflict with Trujillo by appealing to the United States for mediation (Roorda Reference Roorda1996). There is a compelling case that his administration did not respond because he was protecting rents for himself and the elite, but this view is incomplete without also understanding state capacity.Footnote ⁸

Figure 2 TAX PRECINCT BOUNDARIES AND REFUGEE CAMP LOCATIONS

Notes: The map represents the 10 tax precincts during this period. The two colored precincts contain the refugee camps and are the treated precincts in the synthetic control analysis in the section “Evidence for Administrative Complementarities.”

Sources: Camp locations from Derby and Turits (Reference Derby and Turits1993).

Although the government failed to adequately provide aid through its refugee camps, the camps were not the only form of aid the refugees relied on. The refugees also applied for properties under the rental program (Palsson Reference Palsson2021). This demand for property will help test the state’s legal capacity.

Property Requests

The data come from public land rental notifications published in the government’s gazette, Le Moniteur. By law, the program had to publish a notification in Le Moniteur once it approved a lease in case there were competing claims. From 1930 to 1949, it published 8,554 notifications. Each notification contains key descriptive information about the requested land, such as the district (commune) where it was located, when it was requested, and when it was approved.

The details in the notifications allow me to calculate a proxy for legal capacity. Since the notification gives the date the property was requested and the date it was approved, I calculate the processing delay for each property, which is a common measure of effective property rights systems (De Soto Reference Soto2000). Looking at delays over time shows that prospective tenants had to wait significantly longer once the refugees arrived (see Figure 3). Before the massacre, the average delay was below 10 months. But after the refugees came in 1937, there was an unambiguous increase. Delays peaked at 40 months for requests in 1939. Then they decreased steadily until they returned to pre-massacre levels in 1945.

Figure 3 AVERAGE DELAY BETWEEN REQUEST FOR A RENTAL PROPERTY AND APPROVAL BY YEAR OF REQUEST, 1930–1949

Notes: Notifications in Le Moniteur reported when a property was requested and when it was approved. The delay is calculated as the difference in months between these two dates. The dashed lines are a 95 percent confidence interval from a pooled regression with delays as the dependent variable and year-dummies as the only explanatory variable.

Sources: Data from issues of Le Moniteur, 1930–1949.

The trends in Figure 3 suggest that capacity was strained and relieved, but it is hard to tell the timing of the improvements in capacity. The peak in 1939 seems to indicate that the investments in capacity happened quickly. But a 40-month delay from 1939 means the properties were approved in the middle of 1942, which means the improvement in capacity coincided with U.S. mobilization.

Fiscal Revenues

I also collect data on fiscal outcomes from the Fiscal Department’s Annual Reports. From 1932 to 1949, the reports contained consistent and detailed information on tax collections across the country’s 10 precincts (shown in Figure 2). They also include data on HIRS budgets and personnel expenditures.

The Annual Reports report data on property-related tax revenues. Although Haiti did not have a land tax, the government received revenue from public land rentals and fees for registering mortgages and property transfers. The revenue trends are displayed in Figure 4. Nominal receipts increased gradually through the 1930s, but there was a clear break in the trend after U.S. mobilization. When the revenues are adjusted for inflation, they are much lower in the post-mobilization period. But this was by design: by law, the government fixed the rent on its leases for 10 years. Nominal rents could only increase if fees increased or if the government improved rental recovery rates, reevaluated rental properties at their 10-year mark, or leased more land. Since these last three are functions of legal capacity, this is additional evidence that the government’s capacity increased after mobilization. But I can explore the hypothesis further using the disaggregated revenues across the 10 precincts.

Figure 4 GOVERNMENT REVENUES FROM ALL PROPERTY RENTALS AND FEES, 1930–1948

Notes: The revenues are the total receipts collected from property transfer fees and public land rentals. Real revenues have been adjusted for inflation using the Bulmer-Thomas (Reference Bulmer-Thomas2012) price index.

Sources: Annual Reports of Haiti Bureau du Representant Fiscal, 1930–1948.

External Revenue Shock Increased Fiscal Capacity

The data in Table 1 show that income taxes increased after the tax reform. This suggests the tax reform worked and may be evidence it increased fiscal capacity too. But since it was inspired by concerns about WWII mobilization, there could be other coincident factors from the war that increased income tax revenues. For example, President Lescot also responded to the war by developing rubber exports to meet global shortages (Smith Reference Smith2009, pp. 44–47). While his plan ultimately failed, it reflects how the war introduced possible confounding factors.

To uncover the link between the external revenue shock and fiscal capacity, I decompose tax revenue growth. Note that tax revenue in year s is defined as R _s = t _s p _s Y _s , where Y _s is income, p _s is property right protection, and t _s is the realized tax rate. Since revenues depend on the government’s ability to collect taxes, the realized tax rate is the official tax rate ( $t_s^o$ ) times the share of income that the government can observe (x _s ). For example, if the official tax rate is 10 percent but 50 percent of income is hidden from the government, then the realized tax rate is 5 percent. Thus, revenues can be expressed as ${R_s} = t_s^o{x_s}{p_s}{Y_s}$ , and changes in R _s can be decomposed as

(18) $${\rm{\Delta }}\ln {R_{s0}} = {\rm{\Delta }}\ln t_{s0}^o + {\rm{\Delta }}\ln {Y_{s0}} + {\rm{\Delta }}\ln \left( {{x_{s0}}{p_{s0}}} \right),$$

where the s0 subscript indicates the difference between year s and the base year. Thus, changes in tax revenue are the sum of changes in the official tax rate, changes in income, and changes in how much income the government observes and protects. This last term, Δln(x _s0 p _s0), reflects changes in state capacity that are not captured by the change in the official tax rate.

Although the capacity term is unobservable, I can estimate ΔlnR _s0, ${\rm{\Delta }}\ln t_{s0}^o$ , and ΔlnY _s0, and then infer Δln(x _s0 p _s0) as the residual. I can directly calculate ΔlnR _s0 because I observe how much income tax is collected every year. While I know how tax rates changed, I cannot directly calculate ${\rm{\Delta }}\ln t_{s0}^o$ because the tax change differed across brackets (see Table 2) and I do not know the distribution of income. But I can calculate an upper bound by assuming the entire distribution experienced the largest change, which was the highest tax bracket going from 6 to 15 percent. Since this change is larger than if economic growth shifted the entire income distribution up to the next bracket, I am confident it is an upper bound estimate. Estimating ΔlnY _s0 is harder because we do not have GDP estimates for Haiti over this period. But calculating ΔlnY _s0 is the same as estimating how tax revenues would have grown without the tax reform, and this counterfactual is easy to estimate using synthetic control methods (Abadie, Diamond, and Hainmueller Reference Abadie, Diamond and Hainmueller2010). I explain the synthetic control process in Online Appendix Figure A1. In calculating all of these terms, I use inflation-adjusted figures and use 1941 as the base year.

Figure 5 shows the decomposition of tax revenue growth into the changes from economic growth, tax reform, and capacity expansion. In 1943, the first full year of the tax reform, the reform ( ${\rm{\Delta }}\ln t_{s0}^o$ ) accounted for at most 52 percent of the change in revenues, with economic growth (ΔlnY _s0) accounting for another 31 percent. But that leaves at least 17 percent of the 1943 growth unaccounted for. In fact, in all years, the tax reform and economic growth explained about 80 percent of the growth. This unexplained 20 percent is evidence that the fiscal response to WWII detailed previously—hiring more personnel, purchasing cars, and constructing tax offices—increased state capacity.

Figure 5 DECOMPOSITION OF HAITI’S INCOME TAX REVENUE GROWTH, 1941–1948

Notes: The solid black line shows the change in income tax revenue (ΔlnR _s0) relative to 1941. The light gray area shows the portion of the change in revenue that is attributed to economic growth (ΔlnY _s0), calculated by synthetic control as shown in Appendix Figure A1. The dark gray area shows how much is attributed to the 1942 tax reform ( ${\rm{\Delta }}\ln t_{s0}^o$ ), calculated using the tax brackets from Table 2. The remaining portion, in white, is the residual and is attributed to changes in capacity.

Source: Income tax receipts collected from Annual Reports of the Fiscal Representative, 1941–1948.

These results confirm that $\frac{{\partial \tau _1^*}}{{\partial M}} \lt 0$ , which is consistent with the baseline model. Since the baseline model is supported by the literature, this is not a surprising result. But it is still helpful for the augmented model, where we were unsure about the sign. Next, we want to investigate the sign of $\frac{{\partial \pi _1^*}}{{\partial M}}$ .

External Revenue Shock Increased Legal Capacity

The income tax revenue results are consistent with both the baseline and augmented models’ predictions about fiscal capacity. To distinguish between the two, I need to look at how mobilization affected legal capacity. If the baseline model holds, I should see that the shock to external revenues caused the government to invest less in legal capacity $$\left( {{{\partial \pi _1^*} \over {\partial M}} \gt 0} \right)$$ . But if the augmented model is true, I could observe an increase in legal capacity $$\left( {{{\partial \pi _1^*} \over {\partial M}} \lt 0} \right)$$ if the legal capacity cost curve depends on the level of fiscal capacity. I present evidence that the fall in external revenues led to greater investments in legal capacity. This suggests that there are administrative complementarities between fiscal and legal capacities, which I explore in the next subsection.

My first evidence that legal capacity increased after U.S. mobilization is that HIRS processed requests faster. In Figure 3, I showed that delays increased after the refugees’ arrival and that the delays shrank after U.S. mobilization. Here, I want to dissect the effects of the refugee shock and U.S. mobilization using a hazard model. I estimate a Cox hazard model, where the hazard for property request i being processed is given by:

(19) $${\lambda _i}\left( {t|{X_i}\left( t \right)} \right) = {{\rm{\lambda }}_0}\left( t \right){\rm{exp}}\left( {\beta '{X_i}\left( t \right)} \right)$$ ,

where λ() is the hazard function (failure is defined as the government processing the request); λ ₀ is the base rate hazard; t is the number of months in the queue; and X are the included covariates for property request i. I include covariates on permanent features of the property request—its type (urban or rural) and the number of properties in the queue when the property was requested—as well as time-varying features—a dummy for whether t is after the massacre and another dummy for whether t is after U.S. mobilization.

The hazard model captures the effects of the external revenue shock and the influx of refugees with the time-varying indicators. The indicators can vary because the shocks occur while the property request sits in a queue. So if a request was submitted in July 1937, the massacre indicator would equal zero, but if the request was still in the queue in November 1937, the indicator would equal one. Similarly, the U.S. mobilization indicator is equal to zero for months before January 1942 and one for January and all months after. Thus, the hazard model accounts for the events happening while the request is still processing, which is important for capturing the dynamics of the investments.

The primary interest of the analysis is to see how the revenue shock and refugee influx affected the processing time for requests. From the hazard model, I can derive survival curves (where death means the property was processed) evaluated at three different periods: before the massacre, between the massacre and U.S. mobilization, and after U.S. mobilization.Footnote ⁹ I plot the three survival curves in Figure 6. Before the massacre, the graph indicates there was a 20 percent chance approval would take longer than eight months. Then it shows processing times increased substantially between the massacre and U.S. mobilization. The probability that approval would take longer than eight months increased to 85 percent, and there was a 20 percent chance it would take longer than four years. But once U.S. mobilization forced external revenues to decline, the curve returned to pre-massacre levels.

Figure 6 PROBABILITY THAT PROPERTY IS STILL PENDING GIVEN THE NUMBER OF MONTHS SINCE REQUESTED

Notes: Survival curves were derived from a Cox proportional hazard model that controlled for property type, the number of properties in the program’s queue at the time of the request, and dummy variables for when the massacre and reform occurred.

Source: Data from issues of Le Moniteur, 1930–1949.

The curve’s return to the pre-massacre survival rates, even as more requests enter, suggests there might be an equilibrium level of state capacity relative to the demands placed on it. Before the refugees arrived, the government was in a capacity-request equilibrium. But it left the equilibrium when the refugee shock flooded it with requests beyond its capacity to process. Because the government did not expand capacity, delays increased. But once WWII altered external revenues, it expanded capacity, and the government returned to the pre-refugee capacity-request equilibrium.

An additional piece of evidence that legal capacity increased after U.S. mobilization comes from the demarcation of property. For a state to protect property, it must demarcate the land so it knows what it is protecting. While some states demarcate properties through geographic coordinates, the most common system is metes and bounds, which defines properties by their local environment. HIRS used the metes and bounds system, defining properties by features such as roads and, most commonly, by who occupied adjacent properties. But states struggle to provide full property protection under the metes and bounds system because of its dependence on local knowledge and its vague definitions of property boundaries (Libecap and Lueck Reference Libecap and Lueck2011). And states with low capacity might struggle even more under a metes and bounds system (Dimitruk, Du Plessis, and Du Plessis Reference Dimitruk, Du Plessis and Du Plessis2021).

One of the main purposes of the rental notifications was to demarcate properties, but HIRS often provided incomplete descriptions. The most common incomplete description was to say that the adjacent property was occupied by “Qui de droit,” that is, “Whoever owns it.” This could be an admission that the HIRS agent could not find the property’s owner. But it could also indicate that the agent decided that the benefit of finding the owner was not worth the effort. This incomplete demarcation, however, creates insecurity for the tenant. Thus, reducing incomplete demarcations is a sign of greater legal capacity.

Figure 7 shows the proportion of property boundaries with incomplete demarcation by request year. I define a boundary as incompletely demarcated if HIRS reported the occupant as “Qui de droit.” For each property, HIRS reported four boundaries. Before the massacre, between 1 and 2 percent of the boundaries were incompletely demarcated. After the massacre, this figure peaks at 4 percent, at the same time that properties experienced the longest delay between request and approval (see Figure 3). This is further evidence that legal capacity was strained during this period. But after U.S. mobilization, the proportion of incomplete boundaries rapidly decreased until it approached zero. Legal capacity had expanded.

Figure 7 PROPORTION OF PROPERTY BOUNDARIES WITH INCOMPLETE DEMARCATION, 1930–1949

Notes: A neighbor is unknown if the notification in Le Moniteur names the neighbor as “Qui de droit” or “Whoever owns it.” Each property has four neighbors (one for each cardinal direction). The gray area indicates the 95 percent confidence interval for proportion of neighbors who are unknown. Confidence intervals come from a pooled regression with the proportion of unknown neighbors as the dependent variable and year-dummies as the only explanatory variable.

Sources: Data from Le Moniteur, 1930–1949.

Contrary to the baseline model, the external revenue shock led to greater legal capacity $\left( {{{\partial \pi _1^*} \over {\partial M}} \lt 0} \right)$ . The higher capacity let the state reduce processing delays after U.S. mobilization and define properties better. In the augmented model, this is possible if investments in fiscal capacity significantly reduce the cost of investing in legal capacity. I explore this possibility next.

Institutional Evidence

First, Haiti’s appropriations law treated internal revenues and customs revenues differently. HIRS’s budget came entirely from internal taxes—10 percent of all internal revenues and 15 percent of all communal taxes collected by HIRS (Banque Nationale de la Republique d’Haiti 1942, p. 15). None of it came from customs revenue. From 1934 to 1941, the nominal budget was flat, as shown in Figure 8a. In that same figure, the budget is almost indistinguishable from expenditures before 1942 because HIRS spent its entire allocation. This provides a partial answer for why the state did not invest in legal capacity when the refugees started coming at the end of 1937: HIRS was already spending all of its budget and could not make room for expanding state capacity. The budget could have expanded if the government had changed the appropriations law to divert customs revenues to address the refugee crisis, but a reform would have required a costly political process. Since HIRS controlled its share of internal revenues, the nominal budget would naturally expand if internal revenues increased. Indeed, it doubled by 1944 after the external revenue shock triggered the income tax reform. As the budget expanded, so did expenditures and state capacity.

Figure 8 BUDGET, EXPENDITURES, AND WAGES FOR THE INTERNAL REVENUE SERVICE, 1930–1948

Note: Budget indicates the budget of the Haitian internal revenue service, and expenditures indicate how much of the budget was spent. Wages are a subset of expenditures. Inflation adjustment in Panel (b) comes from the Bulmer-Thomas (Reference Bulmer-Thomas2012) price index.

Source: Data come from the Annual Reports of Haiti Bureau du Representant Fiscal, 1930–1948.

A second reason legal capacity increased after U.S. mobilization was because the same administration collected taxes and processed land rentals. Much of the increase in HIRS expenditures went to personnel. Figure 8a shows HIRS spent a significant share of its budget on wages, and those expenditures increased after the reform. When the government expanded personnel to collect income taxes, those same personnel could be used to run the land rental program.

Surprisingly, nominal expenditures increased legal capacity even though real expenditures were falling. After adjusting for inflation using Bulmer-Thomas’s (2012) price index, we see that real expenditures began falling in 1940 (see Figure 8b). This is puzzling because we have evidence showing that the government was hiring more personnel, buying vehicles, building tax offices, collecting extra tax revenues, and delivering better results for property rights all while real expenditures were lower. There are two potential solutions to this puzzle. One is that the Bulmer-Thomas index, which is derived from Dominican data, overstates inflation in Haiti. Another is that wages are sticky, and higher nominal expenditures on wages meant HIRS was hiring more workers at lower real wages. This explanation is supported by Bulmer-Thomas, “Until 1940, with the possible exception of Cuba, nominal wage rates exhibited very little variation in the Caribbean….Thus, an increase in current revenue is likely to have translated not only into an increase in real revenue but also real expenditures because it could purchase more teachers, soldiers, nurses, clerks, and so on” (pp. 570–71). Indeed, this hypothesis is supported in Figure 8b by the gap between the budget and expenditure. In 1943 and 1944, the real value of the HIRS budget was higher than it was in all previous years, which means HIRS could have paid higher real wages (which one imagines the workers would have demanded) but did not. Unfortunately, while sticky wages are an attractive hypothesis, we do not have data on HIRS employment to explore this further. Hopefully, future work can resolve this puzzle.

Despite the puzzle of real expenditures, Haiti’s institutional arrangements support the case for administrative complementarities between fiscal and legal capacity.

Empirical Evidence

While the institutional evidence for complementarities is strong, we can get further evidence by looking at rental collections after U.S. mobilization. The rental collections show how complementarities go both ways.

While rental revenues should increase after a wave of new property approvals, the program could have not collected more rent for two reasons. First, as seen in Table 3, the program historically had low collection rates. With how easily it was overwhelmed by the requests, the program might have also struggled to collect the new rents. Second, it could not increase rental revenues if tenants left their leases. If tenants felt that the rent did not justify the protections they received, they could exercise their outside option and squat on unoccupied land. This is especially true for the ones who waited years to get their requests approved and might have found attractive outside options while waiting. If the program tried to improve its efficiency by collecting rents without protecting claims, it would fail. Thus, higher revenues are evidence that it was providing legitimate property protections.

To see how U.S. mobilization affected rental revenues, I can use the fact that the rented properties were disproportionately in areas with refugees. If legal capacity improves, then I expect rent collection to increase in these areas. To test this hypothesis, I compare rent collections in tax precincts with and without refugees using a synthetic control analysis (Abadie, Diamond, and Hainmueller Reference Abadie, Diamond and Hainmueller2010). I use synthetic control because the data on rent collections are available only at the tax precinct level, of which there are ten in the country during this period, of which only two hosted refugee camps (see Figure 2). In cases like this with small sample sizes, synthetic control improves on difference-in-differences by weighting control observations to best match the treatment group. Although I focus on the results from the synthetic control analysis, Online Appendix A1 compares the synthetic control results to difference-in-differences, revealing similar results.

Because I have limited data, constructing the synthetic control is straightforward. I compare rent collections in precincts with refugees (treated precincts) to precincts without refugees (control precincts). Because the precincts are treated by the refugees before U.S. mobilization, I use October 1937 as the treatment date. This lets me look at how the precincts behaved when the refugees arrived and if there was a change in 1942. I estimate the synthetic control weights from taxes collected from October 1933 to September 1937. Tax receipts are adjusted for inflation, transformed with a logarithmic transformation, and normalized to zero in 1937 (the financial year before the refugees arrived).

The synthetic control analysis shows that the refugees had a large and sustained impact on public land rental receipts, and the effect was magnified by the U.S. mobilization’s effect on capacity. Tax receipts for the refugee precincts and the synthetic control are plotted in Figure 9a. From 1933 to 1937, the two groups followed similar patterns. But after 1937, the precincts diverged. Between the refugees’ arrival in 1938 and mobilization at the end of 1941, rental revenues in refugee precincts were about 20 percent higher than non-refugee precincts (p-values are significant and reported in Online Appendix Table A1). After U.S. mobilization, refugee precincts increased yet again by about 20 percent, widening the gap between the two types.

Figure 9 SYNTHETIC CONTROL ANALYSIS OF THE EFFECT OF REFUGEES AND U.S. MOBILIZATION ON GOVERNMENT REVENUES FROM LAND RENTALS AND PROPERTY TRANSFER FEES, 1930–1948

Notes: Figures display a synthetic control analysis of the effect of the refugee shock and U.S. mobilization on receipts from property rentals, the variable of interest, and property transfer fees, the placebo, the treatment, and synthetic control units. The treated units are the two tax precincts that hosted refugee camps (see Figure 2). The dark vertical line indicates when treatment was assigned in the synthetic control analysis (when the refugees arrived). The lighter vertical line indicates the 1942 U.S. mobilization, though the analysis did nothing to account for it.

Sources: Data come from the Annual Reports of Haiti Bureau du Representant Fiscal, 1930–1948.

The evidence in Figure 9a is consistent with legal capacity increasing after U.S. mobilization. But the analysis might be confounded by a separate economic shock that also increased land values and was coincident with both the timing of the reform and the location of the refugees. For instance, U.S. mobilization may have increased demand for goods produced in the refugee precincts, which subsequently increased land values. This concern is valid but fails to explain why there are unmistakable shifts not just when the United States mobilized but also when the refugees first came. Regardless, to address the concern, I do a placebo synthetic control analysis, replacing the dependent variable with documentary recording fees, that is, fees collected from recording mortgages and property transfers. This analysis is a great placebo test because, like the rental receipts, the fees are related to the value of land, but the refugees did not have the assets to get mortgages or buy property. Thus, it should not show any differences between precincts.

Figure 9b displays the treatment and synthetic control units using property transfer receipts as the dependent variable. The patterns in property transfer receipts are distinct from the patterns found in land rental receipts. They show no evidence of a shock to recording fees that was unique to refugee precincts, neither following the refugees’ entrance nor after U.S. mobilization. The post-1942 patterns are similar across all precincts.

This empirical exercise demonstrates how fiscal and legal capacity reinforce each other. The tax reform increased HIRS’s budget, which led to more properties being processed. Because more properties were occupied, HIRS collected more rent. By expanding one, it improved the other. While these complementarities should encourage governments to invest in capacity, external revenues may be high enough to impede investing in either.