Growth Recurring in a Preindustrial Economy

Abstract

On the basis of new yearly estimates of output and population, Spain’s economic performance from the late thirteenth century to mid-nineteenth century can be shown to be a succession of growing and shrinking phases without long-term net gains in average income. The simultaneous behaviour of per capita income and population is consistent with the existence of a frontier economy in which natural resources are abundant and population scarce, and precludes a Malthusian interpretation. A long phase of sustained growth and lower inequality ended in the 1570s and gave way to another period of sluggish growth and higher inequality. Growth and decline and long-term stagnation are explained by individual and collective economic decisions under institutional constraints.

Co-authored with Carlos Álvarez-Nogal and Carlos Santiago-Caballero. An earlier version appeared as L. Prados de la Escosura, C. Álvarez-Nogal and C. Santiago-Caballero (2022), “Growth Recurring in Preindustrial Spain?”, Cliometrica 16(2): 215–241. This chapter includes a revision of the estimates for population, GDP and its components, and per capita GDP.

1 Introduction

‘Prior to 1800, living standards in world economies were roughly constant over the very long run: per capita wage income, output, and consumption did not grow’ asserted Gary Hansen and Edward Prescott two decades ago.¹ This stylised fact has spread among economists in more simplified terms: income per person remained stagnant in human societies until the Industrial Revolution heralded the beginning of modern economic growth. The Unified Growth Theory’s depiction of preindustrial societies as Malthusian has reinforced this perception (Galor and Weil, 2000).²

Although the Malthusian depiction of preindustrial economies enjoys the support of distinguished scholars (cf. Clark, 2007, 2008; Madsen et al., 2019), it has recently been challenged by research in economic history. Historians are now more prone to accept a transcending of the Malthusian constraint in preindustrial Western Europe, as capital accumulation and productivity gains permitted, simultaneously, higher population and income levels, but with the caveat that such achievements were limited in scope and time (i.e. after the Black Death), and only had long term effects in the North Sea Area (Pamuk, 2007). Broadberry et al.’s (2015) ground-breaking research, for example, rejects the use of the term Malthusian to portray the early modern British economy. However, Voigtländer and Voth (2013) claim that, in north-western Europe, the Black Death brought with it an increase in the endowment of land and capital per survivor, which resulted in higher output per head within a Malthusian framework.

In an attempt to break the growth-stagnation dichotomy in preindustrial societies, historians have highlighted ‘efflorescences’ (Goldstone, 2002: 333) and ‘growth recurring’ episodes (Jones, 1988; Jerven, 2011) that feature a succession of phases of growing and shrinking output per head and only give way to modern economic growth when shrinking phases become less intense and frequent (Broadberry and Wallis, 2017). Growth driven by gains from specialisation resulting from the expansion of international and domestic markets (the so-called Smithian growth) may explain these episodes of sustained but reversible per capita income gains.

Did Smithian growth episodes take place in preindustrial Europe beyond the North Sea Area? New research suggests that they did in Iberia (Palma and Reis, 2019; Álvarez-Nogal and Prados de la Escosura, 2013), although qualitative perceptions of early modern Spain as a stagnant economy are deeply rooted (Kamen, 1978: 49; Cipolla, 1980: 250).

In this chapter, new yearly estimates of Spanish output and population for more than half a millennium are provided, which revise and improve on previous estimates. The new evidence offers empirical grounds to discuss the extent to which Malthusian efflorescences, recurring growth, or Smithian growth are defining elements of preindustrial Spain.

The chapter makes some methodological contributions to the literature on historical national accounts. It includes controlled conjectures on population and sectoral and aggregate output estimates. More specifically, it provides the first agricultural output estimates from the supply side, on the basis of a religious tax, the tithe, incurred by total production, for over 400 years, which are compared to estimates derived with a demand function for the entire time span considered by Álvarez-Nogal and Prados de la Escosura (2013). Their levels and long-run trends are rather similar, even though some significant discrepancies emerge at specific junctures. This result supports the use of the indirect demand approach to deduce trends in agricultural output.

The chapter is structured as follows. In Sect. 2.2, we construct quantitative conjectures about the population. Agricultural output is estimated, and output per head compared to earlier estimates derived with a demand approach in Sect. 2.3. Urban population estimates, adjusted to exclude those living from agriculture, are used in Sect. 2.4 to proxy trends in economic activity outside agriculture. Section 2.5 constructs aggregate output (total and per capita) estimates on the basis of the results obtained in previous sections and draws their long-run trends. In Sect. 2.6, these findings are discussed in the context of the historical debate and some conclusions extracted with regard to secular stagnation, the Malthusian model, and income distribution in preindustrial societies. Section 2.7 provides a long view of Spain’s performance in European perspective. Section 2.8 concludes.

The findings can be summarised as follows: (1) The peak average income levels reached in the late 1330s and the 1560s were only surpassed in the early nineteenth century. (2) However, preindustrial Spain’s economy was far from stagnant, exhibiting long phases of output per head growth and contraction. (3) Population and output per head moved together, at odds with the Malthusian narrative and supporting the hypothesis of Spain as a frontier economy. (4) Spain’s performance suggests Smithian growth episodes during distinctive phases: the long rise up to the Black Death, the century-long expansion up to 1570, and the sustained expansion of the eighteenth century, as larger markets favoured specialization and urbanisation. (5) Income appears less unequally distributed until the early sixteenth century and increasingly more unequally thereafter, as the relative importance of crops increased.

From these results, a puzzling question emerges: why were no significant long-run gains in living standards achieved in Spain’s frontier economy? In the absence of a persuasive Malthusian interpretation, an institutional explanation merits exploration.

2 Quantitative Conjectures on Population

Aggregate population figures for late medieval and early modern Spain consist of scattered benchmark estimates from household population surveys usually collected for taxation purposes—the so-called vecindarios (literally, neighbourhoods), that present the challenge of converting households into inhabitants-, national censuses for the late eighteenth century, and sporadic assessments for the early nineteenth century.³ Available benchmark estimates allow us to derive long run population trends, and historians have relied on baptism records to represent population dynamics.⁴

Baptism data are available from 1580 to the Peninsular War, and most regions are covered from 1700 onwards. Thus, total Spanish population can be derived by weighting each regional index by the regions’ population in a benchmark year (See Appendix A.1, Population, Estimate 1, and Fig. 2.14). However, inferring population trends from baptisms implies assuming that deaths rates maintained a stable short-term relationship with birth rates⁵ and that net migration flows were negligible over time.⁶

Álvarez-Nogal et al. (2016) attempted to reconcile population benchmarks with decadal estimates of baptisms, available since the 1520s, so the resulting estimates capture migration (forced or voluntary) and over time variations in the proportion between birth and death rates (and between births and baptised children) (Appendix, A.1 Population, Estimate 1). Unfortunately, projecting a population benchmark with baptism indices is misleading, since population is a stock variable while baptism series, as a proxy for births, represent a flow. In fact, using baptisms as measure of population amounts to proxy capital stock by investment.

Ideally, to reconstruct annual population figures we require a reliable population figure at the beginning of a benchmark year (N_t) annually adding the natural increase in population, that is, births (b_t) less deaths (d_t), less net emigration (m_t). Thus,

$$ {N}_{t+1}={N}_t+{b}_t-{d}_t-{m}_t $$

(2.1)

As there are population estimates available at various benchmarks (see Appendix, A.1 Population), all we need, then, is data on the natural increase in population (births less deaths) and net migration.

On migration, no yearly data are available and only guesstimates can be proposed. As regards emigration to the Americas, we have relied on Morner (1975: 64) who provides aggregate estimates for five periods over 1506–1670 (1506–1540, 1541–1560, 1561–1600, 1601–1625, 1626–1650) and has distributed them annually within each period.⁷ We also allowed for the outflow of Moorish population after their expulsion, which Pérez Moreda (1988: 380), estimates to be, at least, 0.3 million. Thus, we have added a figure of 60,000 emigrants for each year between 1609 and 1613 inclusively. Estimates from 1670 onwards come from Martínez Shaw (1994: 151, 167, 249) for the periods 1670–1700, 1700–1800, 1800–1830, and 1830–1850, and have been distributed annually. As regards immigration, a figure around 0.2 million has been estimated for the sixteenth century, mostly French moving to Catalonia (Pérez Moreda, 1988: 374), which we have distributed, assuming a steady inflow of 2000 people per year.

We lack yearly crude birth (cbr) and death (cdr) rates for Spain prior to the 1850s, and although baptisms would roughly amount to b in expression (2.1), that is, cbr times population at the beginning of the year, assuming a fixed cdr, or a fixed cbr/cdr ratio, is unacceptable, as crude birth and death rates fluctuate widely in the short run, and even more so at times of pandemics. Fortunately, David Reher (1991) computed yearly crude birth and death rates for New Castile since 1565 (Appendix, Fig. 2.15). Hence, a possible provision of plausible conjectures on annual population levels consists of constructing alternative population estimates in which each population benchmark (N_bk) is projected forwards by adding the annual natural increase in population derived from yearly crude birth and death rates for New Castile (cbr_nct and cdr_nct), less net emigration (m_t) guesstimates. This is the procedure to adopt when we move forward (that is, when starting from, say, 1787, we want to estimate population in 1788), while we need to subtract the natural increase in population and to add net emigration in the previous year when we project population backwards (namely, when starting from 1787 we want to compute population in 1786).⁸ That is,

$$ {N}_{t+1}={N}_{bk}+{\left( cb{r}_{nct}- cd{r}_{nct}\right)}_{\ast }{N}_{bk}-{m}_{t\kern0.75em }\kern1em \mathrm{for}\;t> bk $$

(2.2)

$$ {N}_{t-1}={N}_{bk-}{\left( cb{r}_{nct-1-} cd{r}_{nct-1}\right)}_{\ast }{N}_{bk}+{m}_{t-1}\kern2.25em \mathrm{for}\;t< bk $$

(2.3)

Accepting crude birth and death rates from New Castile implicitly assumes that they are representative for the whole of Spain. Nonetheless, the crude death rate for New Castile matches the main famine mortality episodes not only for inland Spain, but for Spain as a whole.⁹ However, such an arbitrary and unrealistic assumption is largely relaxed by the procedure we propose to reconcile the resulting series. In fact, the exercise suggested by expressions (2.2) and (2.3) provides a set of population series, one for each benchmark, that do not match each other for the years in which they overlap (Appendix, Fig. 2.16). Therefore, we need to carry out a reconciliation between these alternative estimates.

A solution is to interpolate the series, accepting the levels for each benchmark-year as the best possible estimates and distributing the gap or difference between adjacent benchmark series (say, series obtained by projecting the 1752 benchmark forward, N_1752t, and the 1787 benchmark backwards, N_1787t) in the overlapping year T at a constant rate over the time span in between the two benchmark years.

$$ {N^I}_{t=}{N}_{1752t\ast }{\left[{\left({N}_{1787T}/{N}_{1752T}\right)}^{1/n}\right]}^t\kern3.5em \mathrm{for}\ 0\le t\le T. $$

(2.4)

N^I being the linearly interpolated new series, N_1787t and N_1752t the series pertaining to population obtained by projecting two adjacent population benchmarks (i.e. 1752 and 1787) with expressions (2.2) and (2.3), respectively; t, the year considered; T, the overlapping year between the two benchmarks series (say, 1787); and n, the number of years in between the two benchmark dates (that is, 35 years, 1787 less 1752, in our example).¹⁰

Figure 2.1 presents the compromise estimate along the decadal-adjusted series and the benchmarks interpolation. The comparison reveals that the main discrepancies correspond to the pre-1700 period, and while the decadal-adjusted series peaks in the 1580, the compromise series continues expanding during the first quarter of the seventeenth century, and declines thereafter, especially, in the second half of the seventeenth century, with deep contractions in the late 1640s-early 1650s and in the mid-1680s. Furthermore, the compromise series departs from the other two in the early nineteenth century as it captures the impact of the demographic crisis in the early 1800s and during the Peninsular War.

Fig. 2.1

Population: Benchmarks interpolation, decadal adjusted baptisms-based, and compromise estimates, 1400–1850 (million)

In Fig. 2.2, we present our conjectures with regard to the evolution of Spanish population that combines the compromise series since 1565 with the annual population figures obtained through the decadal adjustment (with baptism data) of the benchmarks interpolated series for the period 1520–1565, and the benchmarks interpolated series for the pre-1520 period.

Fig. 2.2

Population conjectures, 1277–1850 (million) (natural logs)

3 Agricultural Output

In pre-industrial Europe, lack of data has led to indirect estimation of agricultural output (Wrigley, 1985; Malanima, 2011; van Zanden and van Leeuwen, 2012). Using a demand function approach, Álvarez-Nogal and Prados de la Escosura (2013) computed agricultural consumption per head over 1277–1850, and assuming the net imports of foodstuffs were negligible, they used it to proxy output per head.¹¹ As this approach relies on proxies for per capita income and assumptions about income and price elasticities, it is worth exploring alternatives.

Early modern economic historians have used indirect information on a religious tax, the tithe, to draw trends in agricultural output and Álvarez-Nogal et al. (2016) we adopted this approach to infer the evolution of agricultural output in Spain between 1500 and 1800. In this section we start from this work but extend the coverage of produce and regions as well as the time span back to 1400 and forward to 1835 (See Appendix, A.2 Computing Agricultural Output Indices from Tithes).

Figure 2.3 presents output for the main crops on the basis of tithes. Cereals show a long-run expansion up to the 1570s. Wine and livestock produce, especially, shadow cereal tendencies. Wine and olive production expanded remarkably during the central decades of the sixteenth century, remaining at high output levels until 1590. Most crops fell during the early seventeenth century, recovering at a different pace between the mid-seventeenth and the mid-eighteenth centuries. In the late eighteenth century, opposite trends are found: fruits and legumes and olive oil production declined, while cereals, must, and livestock produce expanded. A fall is observed across the board in the early nineteenth century.

Fig. 2.3

Output by main produce, 1407–1814 (1790/1799=100). 11-year centred moving average (logs)

The share of each major crop in agriculture output at current prices is presented in Fig. 2.4. Cereal and animal produce are seen to be the main contributors to agricultural output, and show opposite trends, with the share of animal produce increasing and that of cereals declining up to the 1570s and in the late seventeenth and early eighteenth century, and cereals’ share expanding at the expense of animal produce in the early seventeenth and late eighteenth century.

Fig. 2.4

Output composition, 1500–1820 (%) (current prices)

We have constructed a Törnqvist index of agricultural output by weighting yearly variations in each crop’s output by the average shares in adjacent years of each crop in agriculture output, at current prices, and, then, obtaining its exponential. That is,

$$ {\mathrm{lnQ}}_{\mathrm{at}}\hbox{--} {\mathrm{lnQ}}_{\mathrm{at}-1}={\Sigma}_{\mathrm{i}}\left[{\uptheta}_{\mathrm{Qit}}\left({\mathrm{lnQ}}_{\mathrm{i}\mathrm{t}}-{\mathrm{lnQ}}_{\mathrm{i}\mathrm{t}-1}\right)\right] $$

(2.5)

with share values computed:

$$ {\uptheta}_{\mathrm{Qit}}=\frac{1}{2}\left[\left.{\uptheta}_{\mathrm{it}}+{\uptheta}_{\mathrm{it}-1}\right)\right] $$

(2.6)

Previously, current values, V, for each crop i at year t can be derived by projecting the value of each crop in 1799, V_i1799, backwards with the quantity index built on the basis of tithes, Q, and a price index, P (expressed as 1790/1799 = 1) and then, added up in order to obtain the value of total agricultural output, Va_j.

$$ V{\mathrm{a}}_{\mathrm{t}}=\Sigma {V}_{\mathrm{i}\mathrm{t}}=\Sigma {V_{\mathrm{i}1799}}^{\ast }{{\mathrm{Q}}_{\mathrm{i}\mathrm{t}}}^{\ast }{\mathrm{P}}_{\mathrm{i}\mathrm{jt}} $$

(2.7)

Later, the share of each crop, V_it/Va_t, needs to be obtained.¹²

In the evolution of agricultural output, distinctive phases can be found (Fig. 2.5). The first one was of sustained expansion that peaked in the early 1560s. A contraction between the mid-1570s and the early 1610s was followed by stagnation until 1650. A long-run expansion from the mid-seventeenth to the mid-eighteenth century, punctuated by the War of Spanish Succession (1701–1714), peaked in the 1750s, when the highest output level in four centuries was reached. Output stabilised, then, until 1790, when a decline initiated that reached a trough during the Peninsular War.

Fig. 2.5

Agricultural Output Törnqvist Index, 1402–1835: Level and Hodrick-Prescott Trend. (1790/1799=100) (natural logs). Sources: See the text

If we now focus on agricultural output per person (Fig. 2.6, continuous line), two main phases can be identified: a high plateau covering the fifteenth century and up to early 1570s, and a low plateau spanning between the early seventeenth century and the 1750s, with a transitional phase of decline, between the late 1570s and the 1620s, in between, in which output per person shrank by one-third. A new phase of severe contraction is apparent from the 1750s to the Peninsular War, representing one-fourth of the initial level.

Fig. 2.6

Agricultural output and consumption per head Törnqvist Indices, 1277–1850: Levels and Hodrick-Prescott Trend (1790/1799=100) (natural logs). Sources: Text and Álvarez-Nogal and Prados de la Escosura (2013)

How does the new tithes-based agricultural output per head compare to the consumption per head estimates derived with the demand approach? Both series present roughly the same trends since the early sixteenth century (Fig. 2.6). However, some differences emerge. While the demand approach series were already on high plateau since 1400, the tithes-based series show lower levels and higher volatility up to the 1500s. The shift from a high to a low path of output per head is also common to both estimates, reaching a trough in the early seventeenth century, but the tithes-based series present a sharper and neater decline, starting in the mid-late 1570s. Lastly, although the lower plateau covers roughly the same period in the two set of estimates, the post-1650 recovery is stronger and exhibits less volatility in the tithes-based ones.

It is worth noting that the parallel behaviour of the demand-approach and tithes-based series supports the view that crop and livestock destruction appears as the main factor behind the sharp decline in tithes collection during the Peninsular War, rather than peasants’ lack of compliance with the religious tax. However, Fig. 2.6 also shows that the tithes-based output departs sharply from the demand approach estimates from 1819 onwards, and the fact that the years between 1820 and 1833 correspond to a period of peace, suggests that it is non-compliance with the religious tax that explains the widening gap between the two indices. The so-called Trienio Liberal (1820–1823), a phase of liberalisation, weakened Ancien Régime institutions and discouraged tithe compliance (Anes and García Sanz, 1982; Canales, 1982; Torras, 1976). The bottom line is, therefore, that the parallel trends of the tithe-based and the demand approach estimates endorse the use of tithes as a reliable indicator of agricultural output tendencies until 1818. Moreover, our findings challenge the dismissal of the demand approach as simple controlled conjectures. Lacking direct sources of agricultural production, as it is often the case in preindustrial societies, the demand approach appears to provide a reasonable procedure to infer agricultural output trends.

Since our goal here is to provide the best possible estimate for long-run agricultural output, we propose a new index that accepts the demand approach estimates for 1818–1850 and the tithe-based ones for 1402–1818, and projects its level for 1402 back to 1277 with the demand approach index (dotted and dashed lines in Fig. 2.7).¹³

Fig. 2.7

Agricultural Output Törnqvist Index (spliced), 1277–1850: Level and Hodrick-Prescott Trend (1850/1859=100) (natural logs). Sources: Text

4 Output in Non-agricultural Activities: Urbanization as a Proxy

A reconstruction of trends in industrial and services output is beyond the scope of this chapter. It would require a thorough investigation of industrial output, sector by sector, most probably on the basis of a variety of indirect indicators among which taxes merit analysis. In the case of services, the prospects of obtaining a proper assessment of output are even bleaker. A crude short cut to proxy trends in economic activity outside agriculture is urbanization, more specifically, the use of changes in the urbanization rate (ratio between urban and total population) to infer trends in non-agricultural output per head.¹⁴ In this section, we follow Álvarez-Nogal and Prados de la Escosura (2013) and improve on their estimates by including additional urbanization benchmarks and better population data.

We have adopted the definition of ‘urban’ population as dwellers in towns of 5000 inhabitants or more.¹⁵ However, a caveat is necessary. Urban population has been accepted here as a proxy for output in non-agricultural activities after excluding those living on agriculture. The reason is that the existence of ‘agro-towns’ (namely, towns in which a sizable share of the population was dependent on agriculture for living) appears to be a feature of pre-industrial Spain. ‘Agro-towns’ have their roots in the Reconquest. In a frontier economy, towns provided security and lower transactions costs during the re-population following the southwards advance (Ladero Quesada, 1981; Rodríguez Molina, 1978). In the thirteenth century, Christian settlers from Aragon, Catalonia, and Southern France acquired farms but preferred to live in towns (MacKay, 1977: 69). It has been claimed that, in southern Spain, “agro-towns” were the legacy of highly concentrated landownership after the acceleration in the pace of the Reconquest and the Black Death, which increased the proportion of landless agricultural workers (Vaca Lorenzo, 1983; Valdeón Baruque, 1966), although Cabrera (1989) attributes the rise of latifundia to the generalization of the seigniorial regime during the fourteenth and fifteenth centuries. In our estimates, ‘agro-towns’ appear as mainly located in Andalusia, and since the late eighteenth century, also in Murcia and Valencia. Thus, we have computed trends in the rate of adjusted urbanization—that is, the share of non-agricultural urban population in total population—in an attempt to capture those in industry and services output per head (See Appendix, A.4 Adjusted Urban Population).¹⁶

Notwithstanding the existence of ‘agro-towns’, urban economic activity was closely associated to industry and services. In sixteenth-century Old Castile, Yun-Casalilla (2004) calculates, only 1 in 12 in the urban labour force worked in agriculture. Pérez Moreda and Reher (2003: 129) suggest, for 1787, a similar proportion of farmers in Spain’s urban population.¹⁷ Moreover, the rural population carried out non-agricultural activities (storage, transportation, domestic service, construction, light manufacturing) especially during the slack season in agriculture (Herr, 1989, López-Salazar, 1986).¹⁸

Spain’s urban population, adjusted to exclude population living on agriculture, has been computed at benchmark years for the period 1530–1857 (Correas, 1988; Fortea, 1995). Total and adjusted urban population levels for 1530 were projected backwards with Bairoch et al. (1988: 15–21) estimates.¹⁹ The urban population for Spain in 1530, 1561, and 1646 has been inferred from data for the Kingdom of Castile (Fortea, 1995). Adjusted urbanization rates, that is, urban population not living on agriculture expressed as a share of total population, are presented at benchmark years in Table 2.1. Annual figures of ‘adjusted’ urbanization rates have been derived via linear interpolation of the benchmark estimates.

Table 2.1 Adjusted urbanization rates, 1277–1857: Benchmark estimates (%)

The accelerated expansion of the early 1500s slowed down in its second half of the century and was reversed during the first half of the seventeenth century. Then, urbanization recovered slowly, accelerating after the War of Succession to surpass the late sixteenth-century peak by the second half of the eighteenth century. Interestingly, these figures are at odds with the rather stable rate of urbanization (around 20%) widely used in estimates by Bairoch et al. (1988).

5 Aggregate Output

The next stage is to construct an index of aggregate output (Q). Rather than estimating long-run output with fixed weights, which introduces an index number problem, as it implicitly assumes that relative prices do not change over time, we have computed a Törnqvist index in which real GDP is obtained by weighting yearly output variations in agriculture, Q_at, and industry and services, proxied by ‘adjusted’ urban population, N´_{urb-nonagr t}, with the average, in adjacent years, of the shares of agriculture, θ_Qat, and non-agricultural activities, θ_Qi+st, in GDP at current prices.²⁰ That is,

$$ \ln {Q}_{\mathrm{t}}\hbox{--} \ln {Q}_{\mathrm{t}-1}={\uptheta}_{\mathrm{Qat}}\left(\ln {Q}_{\mathrm{at}}\hbox{--} \ln {Q}_{\mathrm{at}-1}\right)+{\uptheta}_{\mathrm{Qi}+\mathrm{st}}\left(\ln {N}_{urb-{nonagr}_{\mathrm{t}}}^{\prime }-\ln {N}_{urb-{nonagr}_{\mathrm{t}-1}}^{\prime}\right) $$

(2.8)

where agricultural, θ_Qat, and non-agricultural, θ_Qi+st, share values are computed as:

$$ {\uptheta}_{\mathrm{Qat}}=\frac{1}{2}\left[\left.{\uptheta}_{\mathrm{at}}+{\uptheta}_{\mathrm{at}-1}\right)\right]\mathrm{and}\;{\uptheta}_{\mathrm{Qi}+\mathrm{st}}=\frac{1}{2}\left[\left.{\uptheta}_{\mathrm{i}+\mathrm{st}}+{\uptheta}_{\mathrm{i}+\mathrm{st}-1}\right)\right] $$

(2.9)

and, then, Q_t is obtained as its exponential.

In order to get sector shares in current GDP, θ_it, current values, V, for each sector i at year t are derived by projecting each sector’s value added average in 1850/1859, V_i1850/9, backwards with the quantity, Q, and price P_, indices previously built for each sector, Q_at and P_at for agriculture, and N´_{urb-nonagr t} (‘adjusted’ urban population) and P_i+st, for industry and services, respectively, (expressed as 1850/1859 = 1) and, then, added up to attain the value of total output, V._t

$$ {\mathrm{V}}_{\mathrm{a}\mathrm{t}}={\mathrm{V}}_{\mathrm{a}1850/9}{\mathrm{Q}}_{\mathrm{a}\mathrm{t}}{\mathrm{P}}_{\mathrm{a}\mathrm{t}} $$

(2.10)

$$ {\mathrm{V}}_{\mathrm{i}+\mathrm{s}\mathrm{t}}={\mathrm{V}}_{\mathrm{i}+\mathrm{s}1850/9}{N}_{urb-{nonagr}_{\mathrm{t}}}^{\prime }{\mathrm{P}}_{\mathrm{i}+\mathrm{s}\mathrm{t}} $$

(2.11)

$$ \mathrm{V}{.}_{\mathrm{t}}={\mathrm{V}}_{\mathrm{at}}+{\mathrm{V}}_{\mathrm{i}+\mathrm{st}} $$

(2.12)

Later, the shares of agricultural and non-agricultural activities were obtained, respectively, as θ_Qat = V_at/V_t. and θ_Qi + st = V_i + st/V_t.

As regards price indices, the price index already built in the section on agriculture has been accepted. For non-agricultural activities, an unweighted Törnqvist index was computed with industrial goods and consumer price indices and nominal wages.²¹ This amounts to allocating one-third of the weight to industry (the industrial price index) and two-thirds to services (nominal wage and consumer price indices), which represents a good approximation to these sector shares in non-agricultural output in the 1850s (Prados de la Escosura, 2017). (For the source of prices see Appendix, A.3 Commodity and Factor Price Indices.)

What does the long run evolution of total output show? Distinctive phases can be observed (Fig. 2.8). Three phases of expansion: (1) between 1277 (the earliest date for which we have estimates) and the early 1340s, whose origins possibly go as far back as to the late eleventh century; (2) from the 1470s to 1570, disrupted in the early decades of the sixteenth century; and (3) from the mid-seventeenth to mid-nineteenth century, interrupted during the Spanish Succession (1701–1714) and Napoleonic (1793–1815) Wars. Two phases of sustained decline complete the picture: the first one, triggered by the Black Death (1348), very intense until the 1370s, followed by stagnation until the first quarter of the fifteenth century; and a second one, from the late sixteenth to the mid-seventeenth century.

Fig. 2.8

Real GDP Törnqvist Index, 1277–1850: Level and Hodrick-Prescott Trend (1850/1859=100) (natural logs). Sources: See the text

If we now turn to output per head, its evolution follows a wide W shape, with phases of growth which peak in 1341, 1566, and 1850, separated by deep contractions in the late fourteenth and early seventeenth century (Fig. 2.9). Each phase of expansion up to the Napoleonic Wars (1277–1341, 1472–1566, and 1643–1850) shows similar trend growth but, as output per head declined sharply during shrinking episodes, each subsequent phase of growth started from a lower level and, hence, evolved along a lower path, with the result that, in the very long run, the trend growth rate is practically nil and per capita income levels hardly change at all (Table 2.2, Panel A).

Fig. 2.9

Real GDP per head Törnqvist Index, 1277–1850: Level and Hodrick-Prescott Trend (1850/1859=100) (natural logs). Sources: See the text

Table 2.2 Output and population trend growth, 1277–1850 (%)^a (annual average logarithmic rates)

Trend growth rates²² for the new estimates (Table 2.2) show that in phases of economic expansion and contraction, total output responded more than proportionally to population and confirm the view that output per head and population trends were directly associated.

When we compare the new index of output per head to earlier estimates by Álvarez-Nogal and Prados de la Escosura (2013), it is noticeable that in the new series, the economic collapse in the late sixteenth century began earlier, in the 1570s, not in the late 1580s, and was deeper. Nonetheless, the use of supply and demand methods to assess trends in agricultural production provides similar long-term results in both levels and trends over 1402–1818 (Fig. 2.10).²³ This key methodological finding supports the use of an indirect approach such as a demand function when no sources for a direct estimation are available.²⁴

Fig. 2.10

Real GDP per head, 1277–1850: New and Álvarez-Nogal and Prados de la Escosura (2013) Törnqvist Indices: Level and Hodrick-Prescott Trend (1850/1859=100) (logs). Sources: See the text and Álvarez-Nogal and Prados de la Escosura (2013)

6 Interpreting the Results: Evidence and Hypotheses

Are there any lessons to be drawn from the new quantitative evidence on preindustrial Spain’s performance? Some stylised facts about preindustrial societies can perhaps be put to the test. An initial example is that of stagnant average incomes. Although living standards did not experience a noticeable improvement over the very long run, the expansive and contracting phases in the W-shaped evolution of Spain’s real output per head contradict this view (Fig. 2.9). Instead, our results lend support to the idea of growth recurring over six centuries. Moreover, Broadberry and Wallis (2017) claim that, as shrinking phases become shorter and less frequent after growing phases, modern economic growth emerges, appears to be confirmed by Spain’s early nineteenth century experience.

A second stylised fact is the Malthusian nature of preindustrial economies. Trends in Spanish population and per capita income, expressed in logs, are offered in Fig. 2.11.²⁵ Population and real output per head expanded simultaneously up to the Black Death, during the late fifteenth and the sixteenth century, and from the early eighteenth to the mid-nineteenth century; conversely, population and income per person shrank in the late fourteenth and in the early seventeenth centuries. How can we explain these results, at odds with the Malthusian view? A plausible explanatory hypothesis is the existence of a frontier economy, resource abundant in preindustrial Spain, but how long did Spain remain a frontier economy? Labour productivity moved together with the labour force in agriculture, so when population and labour declined or grew, labour productivity did so too, and this pattern, which applied not only to Habsburg Spain but also to Bourbon Spain, may have lasted until the mid-nineteenth century. Furthermore, land rent and labour productivity in agriculture also moved together (Álvarez-Nogal et al., 2016: 466–467). Moreover, the fact that in Spain the Black Death was not the watershed that it constituted in central and Western Continental Europe and the British Isles may be explained by its specific traits. In Western Europe, by wiping out between one-half and one-third of the population, the Black Death reduced demographic pressure on resources, raised land- and capital-labour ratios, and led to higher returns to labour vis-à-vis land or capital and higher relative prices for non-agricultural goods. Cheaper capital and labour scarcity led to lower interest rates and higher wages that incentivised physical and human capital accumulation and stimulated labour saving technical innovation and female participation (Pamuk, 2007). The fact that factor proportions in post-Plague Western Europe were apparently similar to pre-Plague Spain’s helps to explain why the negative economic consequences of the Black Death, despite its comparatively milder demographic impact, prevailed in Spain during the late fourteenth and early fifteenth century. In Spain, population density before the Plague (8.9 inhabitants per square kilometre in 1300) was much lower than in most Western European countries after the Plague in 1400 (Álvarez-Nogal et al., 2020) and the Plague destroyed a pre-existing fragile equilibrium between population and resources (Álvarez-Nogal and Prados de la Escosura, 2013).²⁶ Furthermore, the collapse in the late sixteenth century and its lasting effects do not adjust to the Malthusian narrative.²⁷ The fall in real output per head that, in its early stage (−0.65% over 1567–1610), was as sharp as the one associated with the Black Death (−0.67% from 1342 to 1377), seems crucial to Spain’s falling behind. From 1570 to 1650, while population stagnated and per capita income shrank, the economy shifted from commercial and trade-oriented to inward looking and rural.

Fig. 2.11

GDP per head and population Hodrick-Prescott Trends, 1277–1850: (1850/1859=100) (logs). Sources: See the text

Long-run performance has been discussed, so far, in average terms, but how were the gains and losses over successive growing and shrinking phases of per capita income distributed among social groups? The Williamson Index, defined here as the nominal (that is, current price) ratio between output per head and unskilled wage rates and expressed with 1790/1799=100, makes it possible to draw trends in inequality. The rationale underlying the Williamson Index is that GDP captures the returns to all factors of production, while the unskilled wage only captures the returns accruing to one factor, raw labour.²⁸ This way, average returns are compared with returns to unskilled labourers, that is, those at the middle of distribution are compared with those at the bottom. We cannot establish precisely, however, how close to the absolute poverty line unskilled wages were, although attempts to compute welfare ratios (namely, the ratio between a male labourer’s yearly returns and the cost of maintaining his family) suggest that unskilled workers were living close to subsistence in early modern Spain (Allen, 2001; but see López Losa and Piquero Zarauz, 2021). The new Williamson Index improves on the one used in Álvarez-Nogal and Prados de la Escosura (2013) by employing current prices and, hence, avoiding the distortions introduced by the use of different deflators for GDP and wages (see Appendix, A.3 Commodity and Factor Price Indices, for the sources of wages), and more reliable GDP estimates.

Inequality trends followed those of GDP per head, expanding and contracting accordingly. Two phases in the evolution of income distribution can be distinguished, however. One of lower inequality, from the late thirteenth century (and probably earlier) up to the early sixteenth century, and another, of higher inequality, from the mid-sixteenth century onwards (Fig. 2.12), which presents an upward trend and matches the experience of early modern Europe (Hoffman et al., 2002; Alfani, 2021).

Fig. 2.12

Nominal Williamson Index and real GDP per head Hodrick-Prescott Trends, 1277–1850 (1790/1799=100) (natural logs). Sources: See the text

7 Spain in an International Perspective

How did Spain perform internationally? Angus Maddison (1995, 2006) compared average incomes across countries and over time in a common monetary unit and at constant prices. Maddison’s set of international estimates of real income per head in 1990 Geary-Khamis dollars international prices resulted from projecting per capita GDP levels in 1990 dollars, expressed in purchasing power parity (PPP) terms—that is, adjusted for differences in price levels across countries-, back and forth with volume indices taken from historical national accounts. Although Maddison’s approach has been widely used, it can certainly be challenged. Its main shortcoming derives from the severe index number problem it introduces in the comparisons, since the basket of goods and services produced and consumed in 1990, and their prices, become less and less representative as one moves back and forth in time.²⁹

If, with all the caveats about the reliability of income levels derived with a remote benchmark, we follow Maddison’s approach and express product per head in 1990 Geary-Khamis (G-K) dollars, Spain’s average income ranged between G-K 1990 $600–1100 over half a millennium.³⁰ As the absolute poverty line was set by the World Bank at 1985 $1 a day per person, that is, 1990 $426, preindustrial Spain’s average income always remained above the absolute poverty line, more than doubling it in the early fourteenth century, in the late fifteenth and the sixteenth century and, again, since the late eighteenth century (See Appendix, Table 2.3).³¹

How does Spain compare to major economies in preindustrial Western Europe? At the time of the Black Death, average income levels in Spain were above those of the North Sea Area (Netherlands and the United Kingdom) and France (Fig. 2.13). Then, in 1560s, at the peak of its expansion, Spain’s per capita GDP still remained ahead the U.K and France’s, but way below that of the Netherlands. The collapse from the 1570s represented a watershed and Spain fell behind during the seventeenth century. In the early eighteenth century and the post-Napoleonic Wars economic recovery, Spain partially caught up with France but not with the U.K., and its growth was not strong enough to prevent another episode of falling behind during the early nineteenth century.

Fig. 2.13

Real GDP per head Hodrick-Prescott Trends 1270–1850: European Perspective ($1990) (logs). Sources: Spain, see the text; France, Ridolfi and Nuvolari (2020); Netherlands, van Zanden and van Leeuwen (2012); United Kingdom, Broadberry et al. (2015)

8 Concluding Remarks

In this chapter, we have attempted to make the most of scattered data. The results, conjectural as they may be, offer some preliminary conclusions and hypotheses for further research.

• 1. Our aggregate output estimates revise and improve on previous work by (Álvarez-Nogal and Prados de la Escosura, 2013; Álvarez-Nogal et al. 2016). In particular, our agricultural output estimates based on tithes largely confirm those previously obtained with a demand approach. This represents a relevant methodological finding for the reconstruction of historical national accounts: the use of indirect methods such as a demand function to assess trends in agricultural output is warranted in the absence of direct sources.

• 2. Although no significant long-term change in per capita output emerges over more than half a millennium, Spain’s preindustrial economy was far from stagnant and long phases of absolute and per capita growth and decline alternated. ‘Smithian’ and ‘growth recurring’ episodes seem to be present in Spain’s performance.

• 3. Population and output per head moved together, at odds with the conventional depiction of preindustrial societies as Malthusian. This finding is consistent with the high land-labour ratios found in a frontier economy.

• 4. In a frontier economy, living standards are usually relatively high and incomes not very unequally distributed. These features seem to reflect Spain’s experience until the early sixteenth century.

• 5. If we project Spain’s per capita income trend growth during 1470–1570 until the onset of the Napoleonic Wars, we obtain similar levels to the U.K.’s. Why was Spain’s performance up to the 1570s cut short, giving way to a sustained falling behind? Why did Spain never return to the virtuous path initiated in the late fifteenth and consolidated during the sixteenth century? Conventional Malthusian narratives do not appear persuasive in a context of simultaneous growth or decline of population and per capita income. The answer seems to be in policymakers’ economic decisions and new incentives. The long-run unintended consequences of Spain’s attempt to preserve its European Empire provides an explanatory hypothesis that needs to be explored. Sustained increases in fiscal pressure on dynamic urban activities to finance imperial wars in Europe triggered de-urbanisation and led to a collapse in average real incomes, from which early modern Spain never fully recovered. Furthermore, post-1570s Spain appears to present a mirror image of the North Sea Area’s experience where the pull of urban demand triggered an agricultural revolution, as peasants had an incentive to raise their purchasing power to access the new urban goods and services.

Appendix

See Table 2.3.

Table 2.3 GDP, population, and inequality, 1277–1850

1.1 A.1 Population

Benchmarks

The benchmark levels used have been 1100, 1300, 1347, 1435, 1492, 1506, 1508, 1530, 1591, 1646, 1712–1717, 1752, 1787, 1797, 1821, 1833, and 1850. The main source are Pérez Moreda (1988, 2002). Benchmark estimates have been derived as follows.

• 1000. We assumed that Portugal’s population represented the same proportion of total Iberia’s population as in 1300, and the resulting figure was subtracted from Iberia’s to obtained that of Spain, 3.75 million.

• 1300. Population figures for Aragon and Castile kingdoms, 4 million (Pérez Moreda, 2002) have been increased with Pérez Moreda (1988) conjectures on Nazri Granada and Navarre’s population in 1300 (0.4 and 0.1 million, respectively) reaching a total of 4.5 million for present-day Spain.

• 1347. Pérez Moreda (2002) assumes 0.5% population growth over 1300–1347, reaching 5.1 million. We find this assumption on the high side, as qualitative evidence suggests substantial population losses due to bad harvests and famines in the early fourteenth century (Valdeón, 1969; Ladero Quesada, 1981; Vaca, 1983). Instead, we have accepted Pérez Moreda’s growth assumption but excluding years of famine (1301, 1309–1311, 1331–1347) for which no population growth was assumed. The resulting figure, 5.0 million, would imply a yearly growth rate of 0.2% over 1300–1347.

• 1351. As the Black Death had a dramatic impact on the population within a short period of time, 1348–1350, we hypothesise a 25% contraction between 1347 and 1351, in line with the regional evidence available (Castán Lanaspa, 2020; Furiò, 2013; Pérez Moreda, 1988, 2002).

• 1435. Population was obtained by adding up the estimate for Christian Spain in 1435, 3.8 million (Pérez Moreda, 2002), and Granada and Navarre’s population, 0.3 million, c. 1420 (Pérez Moreda, 1988).

• 1492 onwards. Pérez Moreda (2002) estimates for 1492, 1506, and 1508 include the entire population of present day Spain. Population growth between c. 1492 and 1500 was offset by the decline resulting from the Jew population expelled after 1492, that Pérez Moreda (1988: 368) estimates in 0.15 m., and Muslim emigration to North Africa during the Granada war and after the conquest of the Nazri Kingdom by the Catholic Kings (1492), that altogether could be estimated in 0.3 million. The figures for 1530 and 1591 from Pérez Moreda (2002, 1988: 372) and the one for 1646 from Reher (personal communication).

• 1712–1717. Pérez Moreda (1988: 384), on the basis of Bustelo (1973, 1974) provides a 7.7–8.15 million range. We have been accepted the upper bound for 1717, which appears to be consistent with the available estimates for mid-eighteenth century.

• 1752–1850. The figures for 1752 come from the Ensenada population census (Pérez Moreda, 1988: 385). Figures for 1787 (10.4 million) and 1797 (10.5 million) from Floridablanca and Godoy population censuses have been raised to 11.0 and 11.5 million, respectively, following Bustelo’s (1972) proposal. Pérez Moreda assumes zero net growth between 1797 and 1815. Estimates for 1821 and 1833, from Pérez Moreda (1988: 402). The latter has been increased by 5% to offset its underestimate. The estimate for 1850 from Prados de la Escosura (2017).

Alternative Yearly Estimates

Estimate 1

Baptism indices are yearly available for practically all regions between 1700 and 1809, although its coverage declines as one moves back to 1580 and from 1809 onwards.³² An annual national index can be derived by weighting each regional baptism index, B_rt, expressed as 1790–1799=1, by the average of regional population in 1787 and 1797 censuses, N_r1787-97 .³³

$$ {B}_{.t}=\varSigma\;{N}_{r1787-97\ast }{B}_{rt}\kern2.5em \mathrm{for}\ 0\le t\le T $$

(2.13)

Figure 2.14 presents annual population estimates derived from baptism indices along those obtained through log-linear interpolation of each pair of adjacent benchmark estimates. It can be observed that, from the early seventeenth to the late eighteenth century, the baptism-based series shadows the interpolated series but at a lower level. It also reveals the high volatility of baptism series that precludes inferring yearly population levels from it.³⁴

Estimate 2

This estimate offers an alternative solution to the one used in the main text—expression (2.4)—, as a variable-weighted geometric average has been computed for each pair of estimates previously derived using adjacent benchmarks, in which the closest benchmark series gets a larger weight.

$$ {N}_d={\left({X}_d\right)}^{\left(\mathrm{n}-\mathrm{t}\right)/\mathrm{n}}\ast {\left({Y}_d\right)}^{\mathrm{t}/\mathrm{n}}\kern2em \mathrm{for}\ 0\le t\le T $$

(2.14)

Being N the population at decadal estimates d, X and Y, the values corresponding to the projection of each adjacent benchmark (initial and final) figures (i.e., 1700 and 1750) with baptism decadal indices, respectively; and n the number of years in between 0 and T.

See Figs. 2.14, 2.15 and 2.16.

Fig. 2.14

Population: Benchmarks’ interpolation and estimates derived by projecting regional 1787–1797 population average with baptism series, 1400–1850 (million)

Fig. 2.15

New Castile crude birth and death rates, 1565–1850 (0/00). Source: Reher (1991)

Fig. 2.16

Population, 1565–1850: Alternative Benchmarks Projected with Reher’s New Castile crude birth and death rates and compromise estimates (linear interpolation) (million)

1.2 A.2 Computing Agricultural Output Indices from Tithes

Tithe records go back to the Middle Ages but the dearth of written sources reduces the time span in which they are available. In Spain, tithes can be traced back to the early fifteenth century for cereals and olive oil and to the end of the century for wine, while for fruits and vegetables and livestock tithes already exist for the sixteenth century. In Roman Catholic countries tithes did not disappear until the French Revolution and the Napoleonic Wars. In the case of Spain, tithes persisted until the 1830s (Canales, 1982), but its reliability to capture output tendencies after 1808 is hampered by lack of compliance as a result of the Peninsular War and the institutional collapse of the Ancien Régime.

The translation of tithes into output trends raises some questions. Tithes were imposed on farming and livestock production and although, nominally, represented 10% of total production, in practice, its share fluctuated and was usually smaller. Collection procedures, whether direct or rented out to private agents, and the payment system (in kind or cash) changed over time and varied across regions. Also, the resistance of peasants to pay the tax varied, as did the tax exemptions of specific producers, and the opportunities for evasion resulting from the emergence of new crops. Does all this render tithes questionable as a proxy for output tendencies?

In favour of the use of tithes it can be asserted, though, that in late medieval and early modern Spain, where different fiscal systems operated, tithes provided homogeneous information across regions. Moreover, tithes were computed on total output, with the local priest acting as supervisor and making public the names and amounts paid by each producer. The latter also found in its publicity a guarantee of property rights on the harvested land (Santiago-Caballero, 2011, 2014). Lastly, the diversity of tithe beneficiaries multiplied the accounting records available allowing a direct comparison between alternative sources. All this has led historians to depict tithes as a fixed proportion of total production from which output trends can be inferred (García Sanz, 1979).

Unlike most studies we have chosen national rather than a regional or local approach. Thus, aggregates for main crops have been constructed on the basis of an extensive dataset of tithe series at regional and local levels. We have been able to gather tithe records from as early as the fourteenth century.³⁵

The choice of a procedure to aggregate multiple series into homogenous and continuous series was a key decision.³⁶ When the sources made it possible, our favoured approach has been working on the series at a local level. The first step has been establishing whether the series are complete on an annual basis. In most of the cases we found gaps in the records that ranged from just 1 year to longer periods of time. The way in which we have dealt with missing values depended on the amount of information lost and on the availability of sources. If the number of missing observations was small, we derived them by extrapolating the results from series in the same region that presented a similar behaviour due to analogous climatic and soil conditions. In order to obtain the best estimation, we used as proxy the series that were geographically close to the one to be estimated. Missing years were interpolated using the available series that showed a higher correlation in the years around the missing values.³⁷ In our opinion, when the amount of years to be estimated was manageable, this procedure offers the most reliable way to filling the gaps in the series and provides the best possible estimations.

If the number of missing values was large or the existence of alternative local series scarce, we have relied on alternative methods. In these cases, we filled the missing values using the average weight that the local series to be estimated did represent in the aggregate provincial sample.³⁸ However, we were aware of the fact that the weights of the series within the sample changed over time and, therefore, that we had to make adjustments to calculate missing years in the same location that were separated by long periods of time. For that reason we decided to re-calculate the weight of the municipality around each gap. The periods used to estimate the weights therefore varied within the same municipality depending on the years that had to be estimated, a fact that adds robustness to our estimation. Once we had estimated the missing years for all the local series, we simply aggregated them in order to generate the provincial series. When local series from different authors for the same province and period were available, we used the overlapping periods in order to splice them and derive a single series. We also followed the same process in those cases in which the series came from the same source but different local series were available for different periods of time, and we spliced them through on the basis of the overlapping years.

As a result of a long and detailed process we derived series at provincial or regional level that were, then, combined in order to obtain national aggregates for the main crops: cereals, wine, olive oil, legumes, fruit, and animal produce (including wool and silk).

It is for cereals for which the availability of data is wider over space and time with different series covering Andalusia (three out of four provinces, Seville—which included also Cadiz and Huelva—, Cordoba, and Granada, which included Malaga), Extremadura, Murcia, New Castile, Old Castile-Leon (including Burgos—which also included Rioja and Santander—, Leon—which included Asturias—, Palencia, Segovia, Soria, Valladolid, and Zamora), Galicia, Basque provinces, and the Canaries, within the Kingdom of Castile; plus Aragon, Balearics, Catalonia, and Valencia, in the Kingdom of Aragon; plus the Kingdom of Navarre.

As for wine, tithes information was restricted to Andalusia (Seville, Cadiz, Huelva, and Cordoba), Murcia, Old Castile (Rioja, Segovia, and Santander), Basque, Navarre, Aragon, and Catalonia). These regions represented, nonetheless, the main producing areas.

In the case of olive oil information only related to Andalusia (Seville and Cordoba), Extremadura, Balearics, Catalonia, and Navarre. Again, these were the main producers in early modern Spain.

Information about tithes on legumes and fruit is scant and we only managed to get tithes for Balearics and Catalonia, Valencia, and Navarre. These areas represent, nonetheless, above 40% of the value of production in the 1799 Census.

In the case of animal produce, tithes for livestock and wool, are available for Old Castile (Segovia and Soria), Extremadura, Murcia, Navarre, Aragon, and Valencia.

In all cases, we had to interpolate missing values with the help of the geographically closer series. We then constructed regional series by assuming that series for missing provinces evolve alongside those for which data were available. Alternatively, missing values for odd years were log-linearly interpolated.

Weighting provincial series for each crop poses a major challenge. The 1799 Census of Fruits and Manufactures provides the only available estimate of quantities and values of agricultural and industrial goods for early modern Spain. It has a poor reputation largely due to Josep Fontana’s (1967) severe critique. Nonetheless, Fontana largely exonerated cereal production from his criticism and suggested a correction for olive oil output. Unfortunately there is no alternative to the 1799 Census. A possibility would be to derive weights from the highly reputed Cadastre of Ensenada for the 1750s, but it only covers the Kingdom of Castile, leaving aside the Kingdom of Aragon (including Aragon, Balearics, Catalonia, and Valencia) and the Kingdom of Navarre. Furthermore, no distinction is made in the Cadastre’s “respuestas generales” (aggregate results) by crops, only between crops and animal produce (Matilla Tascón, 1947; Grupo ’75, 1977).

We have re-computed the value of total output for the 1799 benchmark by, firstly, correcting olive oil production, as suggested by Fontana (1967); then, valuing each crop at a single price derived as the weighted average of provincial prices. Using a single set of prices helps to correct for the risk of spurious provincial prices (as pointed out by Fontana), while provides us with consistent estimates. Furthermore, it implies a purchasing power parity adjustment across Spanish provinces. The value of agricultural output c. 1799 resulted from aggregating the value of each crop obtained by multiplying its quantity by the average national price. We used, then, provincial (regional) shares in the value of each main crop in 1799 as weights to construct national volume indices for each of them, expressed using 1790/1799 as 100.

The valuation of livestock output in the 1799 Census raises a problem as the livestock total (number of different type of cattle) is mixed with animal produce (i.e., wool). The total value of animal output should then be reduced, in principle, to offset this exaggeration. However, livestock figures are grossly underestimated in the 1799 Census. The data from the 1750s Cadastre of Ensenada for the Kingdom of Castile roughly doubles the 1799 Census figures for the Castilian provinces (García Sanz, 1994). Since there no evidence of a major decline in Castilian livestock during the late eighteenth century exists, such a discrepancy evidences under-reporting in the 1799 Census.³⁹

A detailed list of the sources used can be found in L. Prados de la Escosura, C. Álvarez-Nogal and C. Santiago-Caballero (2022), “Growth Recurring in Preindustrial Spain?”, Cliometrica 16(2): 215–241, Supplementary file1 (DOCX 402 KB).

1.3 A.3 Commodity and Factor Price Indices

Agricultural Prices

For each main crop, prices for 1276–1500 derive from Argilés (1999), for Catalonia (Lérida), Zulaica (1994) and Hamilton (1936), for Aragon, and Hamilton (1936) for Valencia and Navarre, Izquierdo Benito (1983), for Toledo, and Alonso Casado (1991, 2009), for Burgos. Prices for 1501–1800, come from Felíu (1991), for Catalonia, and from Hamilton (1934, 1947), and Hamilton’s unpublished manuscript working sheets (kindly provided by Robert Allen) for Andalusia, New and Old Castile, and Valencia. From 1800, prices comes from Felíu (1991), for Catalonia, up to 1808; Morilla (1972) and Ponsot (1986) for Andalusia; and Llopis Agelán (1980) for wool in Guadalupe. Prices for each produce have been weighted by the regional shares in each main produce’s production by 1799 in order to derive prices at national level.

Industrial Prices

An unweighted Törnqvist index of manufacturing prices (building materials—timber, plaster, lime, tiles, nails—, fuel—coal, wood—, paper, parchment, textiles—cloth, linen, silk—, wax) for 1276–1500 was constructed on the basis of those we had previously built on the basis of original data, for Aragon, 1276–1429 (Zulaica Palacios, 1994), and 1429–1500 (Hamilton, 1936); Toledo, 1401–1475 (Izquierdo Benito, 1983); and Burgos, 1390–1500 (MacKay, 1981; Casado Alonso, 1985, 1991). For the period 1501–1860, we have used an aggregate manufacturing price index kindly supplied by Joan Rosés.

Consumer Price Index

A CPI for 1276–1501 was constructed as a weighted average of agricultural (0.75) and industrial (0.25) Törnqvist price indices, except for Valencia (Allen, 2001). For 1501–1860, a Törnqvist index was derived from regional CPIs: Catalonia, 1501–1807 (Felíu, 1991), and 1830–1860 (Maluquer de Motes, 2005); Valencia, 1501–1785 (Allen, 2001); New Castile (Reher and Ballesteros, 1993), Old Castile, 1518–1650 (Llopis Agelán et al., 2000) and 1751–1860 (Moreno Lázaro, 2002).

Wage Rates

Unweighted Törnqvist indices of nominal wage rates for masons, bricklayers, tilers, and carpenters were computed from the following sources: Aragon, 1277–1423 (Zulaica Palacios, 1994) and 1423–1497 (Hamilton, 1936); Lérida, 1361–1500 (Argilés, 1999); Valencia, 1413–1500 (Allen, 2001) in the Kingdom of Aragon; Toledo, 1401–1475 (Izquierdo Benito, 1983); and Burgos, 1390–1500 (MacKay, 1981; Casado Alonso, 1985, 1991) in the Kingdom of Castile. For 1501–1860, the sources used were: Catalonia (Felíu, 1991; Maluquer de Motes, 2005), New Castile (Reher and Ballesteros, 1993), Old Castile (Moreno Lázaro, 2002), and Valencia (Allen, 2001).

1.4 A.4 Adjusted Urban Population

In order to distinguish those in the urban population who depended on industrial and service activities, an arithmetical exercise has been carried out. Wrigley (1985) assumed that, in pre-industrial Europe, all agricultural population lived in rural areas so to derive the population related to non-agricultural activities, to those living towns, the rural population not involved in agricultural activities should be added. Therefore, the crucial distinction to make was between the agricultural and non-agricultural shares of rural population. However, in preindustrial Spain, the existence of ‘agro-towns’ (namely, towns in which a sizable share of the population was dependent on agriculture) is assumed. Hence, the challenge is to establish which share of rural and urban population lived on agriculture.

In order to distribute rural and urban population into agricultural and non-agricultural we start by comparing the share of the economically active population (L) occupied in agriculture (L_ag/L), and the share of total population (N) living in rural areas (N_rur/N). If the ratio between these two shares [(L_ag/L):(N_rur/N)] is above one, this would mean that part of the population living in towns worked in agriculture. Conversely, a ratio below one suggests that part of those living in the countryside work for industry and services.

However, deriving the ratio between the agricultural, L_ag, and the rural economically active populations, L_rur (L_ag /L_rur) requires further adjustment which allows for urban-rural differences, firstly, in the proportion of total population (N) in working age, or potentially active population (PAP), and, then, in the share of the working age population (PAP), which is economically active (L).

Fortunately, we have information on the PAP/N ratio in both rural and urban areas by region for 1787 (Marcos Martín, 2005). This ratio (computed—due to the census distribution by age cohorts—as population ages 16–50 over total population) differs by region (i) between urban (PAP/N)_{urb i_1787} and rural (PAP/N)_{rur i_1787} areas, being larger in urban areas, but showing low dispersion in both cases.⁴⁰

The implication is that using rural and urban population without previously adjusting for age composition biases the results against agricultural employment, as, on average, the rural (PAP/N)_rur ratio is 87.5% of the urban one. Unfortunately, no yearly data on the PAP/N ratio are available for Spain, except for New Castile, for which Reher (1991) computed it from the late sixteenth century onwards.⁴¹ Thus, we are forced to proxy long-run changes in Spain’s PAP/N by those in New Castile’s (NC) (PAP/N)_{NC_t}.⁴²

Thus, we derived the urban and rural working age at each benchmark year t as follows,⁴³

$$ \boldsymbol{PA}{\boldsymbol{P}}_{{\boldsymbol{urb}}_{\mathbf{i}\mathbf{t}}}^{\prime }={{\boldsymbol{N}}_{{\boldsymbol{urb}}_{\mathbf{i}\mathbf{t}}}}^{\ast}{{\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{urb}}_{\mathbf{i}\_\mathbf{1787}}}}^{\ast}\left({\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{N}\boldsymbol{C}}_{\mathbf{t}}}/{\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{N}\boldsymbol{C}}_{\_\mathbf{1787}}}\right) $$

(2.15)

$$ \boldsymbol{PA}{\boldsymbol{P}}_{{\boldsymbol{rur}}_{\mathbf{i}\mathbf{t}}}^{\prime }={{\boldsymbol{N}}_{{\boldsymbol{rur}}_{\mathbf{i}\mathbf{t}}}}^{\ast}{{\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{rur}}_{\mathbf{i}\_\mathbf{1787}}}}^{\ast}\left({\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{N}\boldsymbol{C}}_{\mathbf{t}}}/{\left(\boldsymbol{PAP}/\boldsymbol{N}\right)}_{{\boldsymbol{N}\boldsymbol{C}}_{\_\mathbf{1787}}}\right) $$

(2.16)

Then, in order to arrive to figures for economically active urban (L_{urb it}) and rural (L_{rur it}) populations at each benchmark we needed to derive the relevant L/PAP ratios. Alas, we were only able to compute the L/PAP ratio for 1787 without being able to distinguish between urban and rural ratios. Hence, we estimated figures of urban and rural EAP for every benchmark year as

$$ {\boldsymbol{L}}_{{\boldsymbol{urb}}_{\mathbf{i}\mathbf{t}}}^{\prime }=\boldsymbol{PA}{{\boldsymbol{P}}_{{\boldsymbol{urb}}_{\mathbf{i}\mathbf{t}}}^{\prime}}^{\ast}{\left(\boldsymbol{L}/\boldsymbol{PAP}\right)}_{\mathbf{i}\_\mathbf{1787}} $$

(2.17)

$$ {\boldsymbol{L}}_{{\boldsymbol{rur}}_{\mathbf{i}\mathbf{t}}}^{\prime }=\boldsymbol{PA}{{\boldsymbol{P}}_{{\boldsymbol{rur}}_{\mathbf{i}\mathbf{t}}}^{\prime}}^{\ast}{\left(\boldsymbol{L}/\boldsymbol{PAP}\right)}_{\mathbf{i}\_\mathbf{1787}} $$

(2.18)

Next, we compared the economically active population occupied in agriculture (L_ag), with that living in rural areas (L´_rur). If L_ag > L´_rur it can be presumed that part of the population living in towns worked in agriculture. Conversely, if L_ag < L´_rur the implication is that those living in the countryside allocated part of their working time to industry and services. This way, we distributed the rural (L´_rur) and urban (L´_urb) economically active populations into agricultural (_ag) and non-agricultural (_nonag) occupations and reached a figure for urban non-agricultural labour (L´_{urb-nonag it}).

$$ {L}_{ru r- nona{g}_{it}}^{\prime }={L}_{ru{r}_{it}}^{\prime}\hbox{--} {L}_{a{g}_{it}}\kern1em \mathrm{if}\kern0.37em {L}_{ru{r}_{it}}^{\prime }>{L}_{a{g}_{it}},0 otherwise $$

(2.19)

$$ {\boldsymbol{L}}_{\boldsymbol{rur}\hbox{-} {\boldsymbol{ag}}_{\mathbf{it}}}^{\prime }={\boldsymbol{L}}_{{\boldsymbol{rur}}_{\mathbf{it}}}^{\prime }-{\boldsymbol{L}}_{\boldsymbol{rur}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime } $$

(2.20)

$$ {L}_{urb-a{g}_{it}}^{\prime }={L}_{a{g}_{it}}\hbox{--} {L}_{ru{r}_{it}}^{\prime}\kern1.75em \mathrm{if}\;{L}_{a{g}_{it}}>{L}_{ru{r}_{it}}^{\prime },0 otherwise $$

(2.21)

$$ {\boldsymbol{L}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime }={\boldsymbol{L}}_{{\boldsymbol{urb}}_{\mathbf{it}}}^{\prime }-{\boldsymbol{L}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{ag}}_{\mathbf{it}}}^{\prime } $$

(2.22)

Thus, economically active population outside agriculture is obtained as

$$ {\boldsymbol{L}}_{{\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime }={\boldsymbol{L}}_{\boldsymbol{rur}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime }+{\boldsymbol{L}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime } $$

(2.23)

Moreover, we can estimate the adjusted urban population in towns of 5000 or more inhabitants (excluding those living on agriculture), by re-scaling the resulting figures for urban economically active population outside agriculture with the activity rate (L/N),

$$ {\boldsymbol{N}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime }={\boldsymbol{L}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}^{\prime }/\left({\boldsymbol{L}}_{{\boldsymbol{urb}}_{\mathbf{it}}}^{\prime }/{\boldsymbol{N}}_{{\boldsymbol{urb}}_{\mathbf{it}}}\right), $$

(2.24)

Thus, we can obtain an adjusted rate of urbanization (Ua_it) that partly offsets at least the upward biased effect of the agro-towns:

$$ {\boldsymbol{Ua}}_{\mathbf{it}}={\mathbf{100}}^{\ast }{\boldsymbol{N}}_{\boldsymbol{urb}\hbox{-} {\boldsymbol{nonag}}_{\mathbf{it}}}/{\boldsymbol{N}}_{\mathbf{it}} $$

(2.25)

Regrettably, though, we lack data to compute the share of labour in agriculture (L_ag /L) at each benchmark year. For L_ag evidence can only be obtained for 1857 and 1787, from population census and for 1752, restricted to the Kingdom of Castile, from the Cadastre of Ensenada (Grupo ’75, 1977).⁴⁴ Wrigley (1985) and Allen (2000) also faced this shortcoming, and Wrigley assumed that, in early sixteenth century England and France, up to 80% of the rural labour force was in agriculture and reduced arbitrarily this figure over the three following centuries. Allen (2000) accepted the same percentage for most European countries circa 1500 and interpolated the years up to the first one (1800) for which he had estimates. In the case of Spain, we assumed a fixed 80% share of EAP in agriculture and interpolated log-linearly the shares between 1530 and 1787 and 1787 and 1857.⁴⁵

However, efficiency changes resulting from variations in the composition of labour by economic sectors and in the dependency rate could affect our proposed measure. Thus, we have carried out a sensitivity test by estimating the intersectoral shift effect that results from changes in the shares of industry and services in non-agricultural employment and in the productivity gap between industry and services. Furthermore, we have allowed for changes in the potentially active to total population ratio (PAP/N) that could also affect our index. Fortunately trends in the proposed index of output outside agriculture do not appear to be significantly altered by either demographic or output composition changes during the early modern era.⁴⁶