Wealth and Property Taxation in the United States

It was a tax that, unusually for its time, applied to a broad set of assets rather than only land — and its administration required local assessors to place a value on nearly all property, year after year. Since colonial times U.S. local governments relied on forms of property taxation, but after the 1840s the basic principles became more uniform across states and the taxes more comprehensive, coming to be known as the General Property Tax. From the 1840s to 1930 it was a core element of the U.S. fiscal system, providing a large share of state and local revenues, with multiple layers of government — districts, municipalities, counties and states — levying their own ad valorem taxes, typically at the same rate on most property. After the 1930s its role declined as new forms of taxation replaced it, and it evolved into the modern U.S. property tax, which applies only to certain types of real estate. The paper is careful about coverage: it notes that local and state taxes encompassed by the term had varying levels of coverage of different property types, which are difficult to formally document especially at the local level, and argues rather than assumes that coverage was high enough to capture most taxable wealth.

Private, marketable, taxable assets — a definition with two explicit exclusions and one uncomfortable inclusion. It excludes public assets such as federal land, and assets of tax-exempt entities such as religious institutions and colleges. “Marketable” refers to assets that can be bought and sold, excluding human capital except under slavery: in slave states enslaved people were counted as the property of enslavers, and the returns to forced labour were recorded as returns to capital. To ensure consistency over time, the authors also construct wealth series that exclude enslaved property, and much of the paper’s regional analysis turns on the comparison between the two.

Assessors often did not value property at market price, and practices varied across time and place, so the assessed values must be scaled by an assessment ratio to recover market values. The assessment ratio is defined as the ratio of assessed wealth to market wealth. At the state level, the paper builds assessed-wealth series from a large set of detailed state reports, usually annual, that vary in format and naming across time and states. Assessment ratios come first from U.S. Census wealth data — decadal and in some cases more frequent — in which the Census carried out detailed valuation work by consulting professionals and experts and sending agents into the field; these are supplemented with information from the state reports documenting changes in assessment practices for additional years, allowing higher-frequency ratio series. Applying the ratios to assessed wealth yields annual market-wealth series, and national series are obtained by aggregating the state series. At the county level the main source is Census wealth data, which give decadal assessed values for most years and market values for some; the benchmark county series corrects assessed values using state-level ratios, with alternative series constructed for years when county-level ratios are available.

Its main advantage is that property is directly assessed rather than inferred, so it needs fewer assumptions than capitalisation or estate-multiplier methods; its main limit is that the pre-1850 data are scarce and less reliable. The paper notes that most modern U.S. wealth estimates are based on capitalised income flows using assumed rates of return, whereas direct property assessments obviate the need to infer wealth indirectly, were intended to be comprehensive with only a few documented exemptions, and are available at high frequency. It also argues the approach does not require assessments to have been made perfectly — even with imperfect valuation at assessment, the extensive information from the Census and state reports allows market values to be recovered. Against this, the authors state plainly that data before 1850 are scarce and less reliable, and although they provide the full dataset for all available years they restrict the core analysis to the post-1850 period. They also flag that estimates of real product growth before 1840 remain somewhat “conjectural” and rely on assumptions about relative labour productivity, which is why the wealth-to-GDP ratio is also plotted using two additional GDP sources.

The U.S. started wealth-poor by European standards and accumulated rapidly after the Civil War, with the ratio to GDP driven by the numerator. The wealth-to-GDP ratio was around 300% early in the nineteenth century, rose to 400% between 1850 and 1860, plummeted to 200% during the Civil War, recovered in a growth spurt to almost 500%, fell steeply to 300% with World War I, then rose to almost 600% on the eve of the Great Depression before crashing back to around 400%. Decomposing numerator from denominator, wealth per capita drives the ratio: it started low, grew slowly until the Civil War, then took off drastically from 1870 and grew much more rapidly than income per capita until the World War I crash. In international comparison the paper reports that the U.S. had lower wealth-to-GDP ratios than France and the U.K. over the nineteenth century and until the end of World War I, while cautioning that cross-country comparisons are difficult given uncertainty in historical GDP, deflators and exchange rates.

Real property was the largest category throughout, but the composition shifted away from land, and enslaved people were 15% of total U.S. wealth in 1860. Decomposing into real property, enslaved property and other personal property, land, buildings and improvements were the largest category throughout the period. For states where taxable land can be separated, the importance of land within real property declines over time: early in the century the primary source of wealth was land, abundant and low-priced compared with Europe, and policies were put in place to keep prices low and allow settlement. The paper offers a striking comparison — all real property in the U.S. represented less than 200% of GDP before the Civil War, while land alone represented 300% of national income in the U.K. Immigrants and settlers arriving in the U.S. were typically not bringing large amounts of physical property or capital, and throughout 1840 to 1940 the U.S. accumulated wealth rapidly in the form of non-land capital. The paper adds a caveat that improved and unimproved land should be distinguished: while the latter was abundant, the former was not.

Southern wealth looked comparable to other regions before the war only because enslaved people were counted as property; removing them reveals the South as wealth-poor beforehand and stagnating afterwards. Enslaved people accounted for over 40% of total property in Southern states, and in Georgia, Alabama and Florida they represented more than 50% of total property in 1860. Excluding enslaved property, the South appears poorer than other regions and not accumulating wealth at their rate even before the Civil War; while other regions’ wealth-to-income ratios grew after the war, the South’s stagnated, and it remained lower in wealth until 1940. The paper quantifies the war’s destruction state by state: in Texas, where enslaved people represented 35% of total property, property excluding enslaved property declined by 51% between 1860 and 1870; in Mississippi, where the share was 44%, property excluding enslaved property was 53% lower in 1870 than in 1860. The relation between the enslaved share and the decline is described as noisy but increasing. For comparison, property per capita in Philadelphia more than doubled over the same decade.

Enough to reverse the appearance of persistence: the state rank-rank correlation between 1850 and 1870 is 0.57 excluding enslaved property, and 0.04 including it. Excluding enslaved people from personal property, there was strong persistence in state ranks even after the Civil War, with a rank-rank correlation of 0.73 between 1850 and 1860 and 0.57 between 1850 and 1870. Including enslaved people reduces the 1850–1870 correlation to 0.04. The paper makes a parallel point for white residents: including enslaved property, white residents in Southern states appeared more than twice as wealthy as those in non-Southern states and saw their property per capita fall by 75% during the Civil War; excluding it, white residents had similar levels of property per capita in Southern and non-Southern states before the war, with a clear divergence and much slower growth afterwards. Black residents had significantly higher property per capita in non-Southern than Southern states, but even in non-Southern states their property was drastically lower than that of white residents.

Effective property tax rates in the South almost tripled in about five years, peaking at 1.2% in 1870, then reverted to around 0.6% as Reconstruction ended. Before the Civil War, effective tax rates in Northern states were twice as high as in Southern states, which the paper reads as reflecting lower investments in public goods and infrastructure in the South. Confronted with a decline in the property tax base and needs to invest in public goods like public schools, newly elected Republican legislators in the South pushed for higher property tax rates during Reconstruction. The paper describes the sudden increase as met by backlash that triggered political violence, especially against black politicians, and reports that as Democrats regained control of the South — ending Reconstruction and enabling the institution of the Jim Crow regime — tax rates quickly reverted to a much lower level than in Northern states.

By four separate measures it did not decline: dispersion, concentration, rank persistence, and the speed of convergence. First, on σ-convergence, the yearly standard deviation of log property per capita across states remains roughly constant, with a similar pattern across counties. Second, the share of total national wealth held by the top 10% of richest counties increased from 1860 to 1930, reaching about 70% of total U.S. property by the end of the period. Third, county-level rank-rank correlations of property per capita starting from 1870 are 0.79 over ten years and remain high at 0.67 over the entire 60-year period, with high persistence at the state level too — places that started poorer remained poorer. The paper also compares the county distribution of property per capita in the 1920s to household income today from the Opportunity Atlas and reports a rank-rank correlation of 0.6 between the two.

Slower in wealth than in income, and slower still because of the South. Regressing 60-year growth in property per capita on initial 1870 property, the speed of convergence without controls is β = 0.011; excluding Southern counties it is 0.028; with the full set of geographic, demographic and occupational controls it rises to β = 0.024 with an R² of 0.60. At the state level the same exercise yields an even smaller β = 0.007 over 1870–1930. Comparing directly against income data over 1880–1920, the paper reports that β estimates are 2–2.5 times higher using income data without controls, and 1.5 times higher with controls; the estimates from Barro et al. (1991) are somewhat lower than those from IPUMS income data but still show faster convergence unless controls are included. The paper’s summary is that the U.S. experienced limited spatial convergence from 1870 to 1930, largely driven by Southern states.

Geography predicts levels much better than growth; demography — especially literacy — predicts both. Geographic characteristics explain 21% of initial property per capita in 1870 but only 9% of subsequent conditional growth. Climate is an important predictor of initial wealth: a one-standard-deviation-higher July temperature, characteristic of Southern counties, is associated with 25% lower initial wealth, and more abundant winter precipitation with significantly lower initial wealth and slightly lower growth. Topography, captured by elevation and ruggedness, is negatively related to 1870 wealth but not significantly correlated with wealth growth; better soil productivity and lower distance to the coast are significantly positively correlated with long-run growth. Demographic variables explain 20% of the variance in 1870 property and 4% of conditional growth, with the literacy rate — a proxy for local human capital — exhibiting the highest correlation and alone explaining 10% of the variance in initial property. Counties with higher population in 1870 were wealthier and grew faster, which the paper links to scale effects in innovation and growth. Occupational shares explain 12% of the variance in initial property and 3% of long-run growth: counties specialised in public administration, mining and commerce were significantly richer in 1870, while more agricultural counties were significantly poorer and accumulated property significantly more slowly.

The paper reads migration as a convergence force operating through dilution. Counties that experienced higher ten-year population growth, and those with a higher share of foreigners, had lower property per capita in 1870, and over the full period lagged higher population growth is associated with lower wealth growth in the following decade. The interpretation offered is that richer places see inflows of migrants, but on average these newcomers have lower wealth and dilute wealth per capita over the next decade. Note this sits alongside a distinct finding in the other direction: conditional on population size, a higher share of foreigners is significantly positively associated with higher long-run growth.

Yes — the occupational structure of counties evolves the way the country-level literature describes. As a county’s property per capita increases, the fraction employed in agriculture declines steadily and the fraction in services increases, while manufacturing follows a characteristic hump shape, first increasing and then decreasing as counties grow richer. The paper reads this as evidence that structural transformation away from agriculture is a relevant pattern of development even at the local labour market level, and notes that this non-monotone pattern for manufacturing is why the linear regressions do not detect a precise manufacturing effect.

Counties with a larger share of enslaved property in 1860 were poorer in 1870 and accumulated property significantly more slowly over the next 60 years, conditional on the full set of controls — an association of substantial magnitude that survives restricting to the South. The reported magnitude is that a 10 percentage point increase in the share of enslaved property in total property, conditional on the initial 1870 property level, reduces the growth rate of property in the next 60 years by 5 percent. Restricting the sample to Southern counties only — where the non-Southern zeros cannot drive the result — there is still a strong negative association, smaller in magnitude, and robust to the county-level geographic, demographic and occupational controls. This is framed as a correlation throughout; the paper describes the exercise as exploring how reliance on enslavement correlates with wealth accumulation in the decades following abolition.

Largely not: initial inequality mediates at most one-sixth of it. Sokoloff and Engerman (2000) argued that after abolition enslavement remained detrimental to long-run development because it increased initial economic inequality, delaying modern economic growth. Consistent with that argument’s first step, the fraction of enslaved property is positively correlated with higher initial wealth inequality. But a strong negative and significant correlation between enslavement and growth remains even when controlling for initial inequality, and the estimated correlation between the 1860 enslaved-wealth fraction and future growth is little affected by introducing county-level inequality controls. The paper’s stated conclusion is that the impact of enslavement on the slow convergence of the U.S. South extended beyond wealth inequality after the Civil War, with systemic policies and the Jim Crow regime playing critical roles.

A robust negative correlation, measured within the same country and state so that institutional and cultural factors are held fixed: a 10 percentage point higher top-10% wealth share in 1870 is associated with 20 percent lower property growth over the subsequent 60 years. The paper positions this against a vast literature on inequality and growth that relies mainly on cross-country correlations, and identifies its own key advantage as granularity — measuring the relationship across places within the same country and state. Illustrating the raw pattern, highly unequal counties with top-10% shares close to 100% in 1870, such as Baton Rouge, LA or Charleston, SC, had almost 70 percent lower growth of property per capita over the next 60 years, while counties such as Douglas, NE or Larimer, CO, where the initial top-10% wealth share was about 75%, experienced higher growth. The relationship remains highly significant after adding the full array of geographic, demographic, occupational and enslaved-property-share controls.

Human capital accumulation: slower growth in local literacy accounts for about 20% of the association. A mediation analysis adding changes in population composition, education and occupational structure between 1870 and 1930 identifies the change in the local literacy rate as the most important mediator — lower growth of literacy rates in areas with higher inequality alone accounts for 20% of the association between higher inequality and lower long-run growth. The paper connects this to earlier work (Ramcharan 2006; Acemoglu et al. 2007) suggesting a negative correlation between 1860 land-ownership inequality and school enrolment or education expenditures, and states its results confirm that a lower rate of human capital accumulation is a strong mediator of the inequality–growth link.

Both an extension of the data backwards and several unexplored uses of it. The property tax data are described as especially useful post-1850, when quality and availability are much better, which is why the main analysis is restricted to that period — but the dataset contains substantial information for many states before 1850, which the paper suggests could be used to construct better national and sub-national measures of economic activity before 1840, a period American historians often call a “statistical dark age”. Other directions named are comparing wealth-based and income-based determinants of economic activity, studying the effects of local wealth on outcomes such as innovation or education, and performing a finer analysis of different types of wealth using the additional detail in the tax records.