Productivity Growth

Abstract

This chapter assesses long-term productivity growth and its immediate determinants in Spain. Over the last 170 years, output per hour worked dominated GDP growth, while hours worked per person shrank by one-quarter and population trebled. Half of labour productivity growth resulted from capital deepening and one-third from efficiency gains (total factor productivity). Labour productivity proceeded steadily, accelerating during the 1920s and from the mid-1950s to the late 1980s, but decelerated thereafter as capital deepening slowed down and TFP stagnated due to the fact that expanding sectors attracted less investment-specific technological progress, largely a result of institutional constraints.

Co-authored with Joan R. Rosés. An earlier version was published as L. Prados de la Escosura and J.R. Rosés (2021), “Accounting for Growth in Spain, 1850–2019”, Journal of Economic Surveys 35(3): 804–832. The estimates have been thoroughly revised and updated and, subsequently, the main text. Also, in Sect. 4.3, the sub-section on capital input has been eliminated as it overlapped with Sect. 3.3 in Chap. 3.

1 Introduction

The current productivity slowdown in advanced economies has triggered a lively debate about its causes. The long phase of robust productivity growth initiated in the aftermath of World War II, which brought about unprecedented progress in absolute and per capita GDP, has given way to a phase of deceleration in output per hour worked. Exploring the origins and drivers of such a vigorous productivity expansion may cast some light on the causes of today’s poor performance. Economic history research provides an opportunity to expand the exploration beyond the narrow time boundaries of modern national accounts.

This chapter focuses on modern economic growth in Spain, highlighting phases of fast growth and stagnation, and aims, on the one hand, to present new, consistent long-run trends in labour productivity and its drivers, including capital deepening, labour quality, and total factor productivity; and, on the other, to determine how much physical and human capital and efficiency gains have contributed to labour productivity enhancement over time and to what extent they are complementary.

The main findings are that labour productivity (measured as output per hour worked) dominated GDP long-run growth, accounting for four-fifths of the latter, while population contributed 30% and the number of hours worked per person contracted. About half of the increase in labour productivity resulted from capital deepening (that is, capital services per hour worked) and one-third from efficiency gains in the use of physical and human capital (namely, total factor productivity), while labour quality contributed the rest. The progress of labour productivity was not steady. During its phases of acceleration (the 1920s and, especially, 1954–1975), total factor productivity was its driving force, complemented by capital deepening. Since Spain’s accession to the European Union, labour productivity has sharply decelerated as capital deepening slowed down and TFP stagnated. Sustained GDP growth up to the Global Financial Crisis (1986–2007) largely resulted from an increase in hours worked per person (one-half) and to a less extent from labour productivity (less than one-third), the sluggish growth of which stemmed mostly from weak capital deepening. Institutional constraints help to explain the labour productivity slowdown.

The chapter opens by examining GDP growth and considering its proximate determinants: population, hours of work per person, and output per hour worked (Sect. 4.2). This is followed by a breakdown of the hours worked per person. Next, Sect. 4.3 investigates output per hour worked and its proximate sources, namely, intensity in the use of production factors and efficiency gains. To this end, long series of capital, land, and labour inputs are constructed, as well as factor shares in GDP to proxy their output elasticities. Section 4.4 includes a discussion of the main drivers of labour productivity.

2 GDP Growth and Its Determinants

Between 1850 and 2020, Gross Domestic Product (GDP) rose nearly 50-fold. A breakdown of GDP can be carried out using an identity,

$$ Y={LP}^{\ast } LQ/{N}^{\ast }N $$

(4.1)

Y being GDP; N, population; LQ, the number of hours worked; and LP (= Y/LQ), GDP per hour worked. Note that GDP per head, Y/N, equals LP *LQ/N.

During the last 170 years, population multiplied over three times, hours worked per person shrank by one-third, and output per hour worked rose 24-fold. GDP per head gain was lower (16-fold) though, as we have to detract the decline in hours worked person from the gains in output per hour worked.

Logarithmic rates of variation allow us to compare the pace of growth of GDP and its components over periods of different length. Thus, ln being the natural logarithm,

$$ \mathit{\ln}\left({Y}^{\mathrm{t}}/{Y}^{\mathrm{t}-1}\right)=\mathit{\ln}\left({LP}^{\mathrm{t}}/{LP}^{\mathrm{t}-1}\right)+\mathit{\ln}\left({\left( LQ/N\right)}^{\mathrm{t}}/{\left( LQ/N\right)}^{\mathrm{t}-1}\right)+\mathit{\ln}\left({N}^{\mathrm{t}}/{N}^{\mathrm{t}-1}\right) $$

(4.2)

Long-term growth in GDP (2.3% per year) appears to be largely attributable to labour productivity gains, which grew at 1.9%, compared to population, at 0.7%, and hours worked per person, which shrank at −0.3% (Table 4.1).

Table 4.1 GDP growth and its composition, 1850–2020 (annual average logarithmic rates %)

Different long phases can be distinguished, in which growth deviates from its long-run trend as a result of technological change, economic policies, and access to international markets (Fig. 4.1).

Fig. 4.1

Real GDP, absolute and per hour worked (2010=100) (logs)

Moderate growth took place between mid-nineteenth century and the Golden Age (1850–1953), with GDP growing at a yearly average rate of 1.5%, to which output per hour worked was the largest contributor (0.9%), followed by population (0.6%), while hours worked per person contracted mildly. Then, Spain’s Golden Age (1954–1975), witnessed a fourfold GDP growth acceleration, almost exclusively attributable to labour productivity (5.9% of 6.2% GDP growth), as population expansion was largely offset by the reduction in hours worked per person (1% against −0.7%).

The 1970s oil crises took place at the time of the transition from General Franco’s dictatorship (1939–1975) to democracy that culminated with Spain’s accession to the European Union (1985). Output per hour worked continued to thrive from 1976 to 1985, as the economic crisis and stabilisation and liberalisation reforms led to the closure of inefficient industries sheltered from competition. Labour productivity growth (5.6%) more than offset the sharp decline in hours worked per person (−3.8%), allowing mild growth in absolute and per capita GDP (2.5% and 1.8%, respectively).

Fast GDP growth (3.5% yearly) prevailed from Spain’s EU accession (1985) to the eve of the Great Recession (2007). Nearly half of this resulted from an increase in hours worked per head, since unemployment fell and new jobs were created, while labour productivity contributed only one-third.

During the Global Financial Crisis (2008–2013), GDP shrank with similar intensity to that experienced in the Great Depression (1929–1933) (−1.34% vs −1.50% per annum), second only to the sharp contraction (−6.6%) during the Civil War (1936–1939). The pace of employment destruction from 2008 to 2013 was similar to that of the ‘transition to democracy’ decade (1976–1985), with hours worked falling at −3% yearly, but labour productivity lacked the strong response of the ‘transition’ years (1.7% vs. 5.6% growth rate) and was unable to prevent a contraction in absolute and per capita GDP (−1.3% and −1.8%, respectively). In the post-Great Recession recovery (2014–2019), halted by the impact of the COVID pandemics, GDP and GDP per head grew similarly (2.6% and 2.4%), as the inflow of immigrants, the driver of population growth, was cut short, and per capita GDP growth mainly resulted from the increase in hours worked per person (about three-fourths).

A pattern can be observed since 1975: output per hour worked and hours worked per person exhibit opposite tendencies. Phases of (absolute and per capita) GDP growth acceleration and recovery (1986–2007 and 2014–2019) went hand-in-hand with rising hours worked per person through employment creation, while labour productivity growth slowed down. Conversely, phases of sluggish or negative (absolute and per capita) GDP growth, and employment destruction (1976–1985 and 2008–2013), coincided with those of labour productivity acceleration. Thus, it can be concluded that since the mid-1970s the Spanish economy has been unable to combine employment creation and labour productivity growth. This is consistent with the fact that expanding sectors that created more jobs (construction and services) had lower labour productivity relative to industry and experienced slower output per hour growth (Prados de la Escosura, 2017), which implies that they were less successful in attracting investment and technological innovation.

This paradox leads us to explore what underlies the behaviour of hours worked per person and output per hour worked.

We can break down the evolution of the number of hours worked per person (LQ/N) as follows,

$$ \left( LQ/N\right)=\left( LQ/ LF\right)\ast \left( LF/ WN\right)\ast \left( WN/N\right) $$

(4.3)

(LQ/LF) being the hours per full-time equivalent worker; (LF/WN), the ratio of full-time equivalent workers to the working age population (those aged 15–64), that is, the participation rate; and (WN/N), the share of the working age population in total population.

Thus, in rates of variation,

$$ \ln \left({\left( LQ/N\right)}^{\mathrm{t}}/{\left( LQ/N\right)}^{\mathrm{t}-1}\right)=\ln \left({\left( LQ/ LF\right)}^{\mathrm{t}}\right./\left({\left( LQ/ LF\right)}^{\mathrm{t}-1}\right)+\ln\ \left({\left( LF/ WN\right)}^{\mathrm{t}}/{\left( LF/ WN\right)}^{\mathrm{t}-1}\right)+\ln\ \left({\left( WN/N\right)}^{\mathrm{t}}/{\left( WN/N\right)}^{\mathrm{t}-1}\right) $$

(4.4)

The change in hours per full-time equivalent worker (LQ/LF), which fell from 2800 h by mid-nineteenth century to less than 1800 h in 2020, represents the main driver of hours worked per person in the long run (Table 4.2). Its contribution is especially noticeable during phases of industrialization and urbanization in the 1920s—in which the 8 h/day standard was gradually adopted—and 1959–1975. It also contributed to a lesser extent during phases of labour market adjustment and union activism such as the II Republic (1931–1936) and the ‘transition to democracy’ decade (1976–1985).

Table 4.2 Growth of hours worked per head and its composition, 1850–2020 (annual average logarithmic rates %)

The participation rate (LF/WN) also made a substantial contribution to hours worked per person. During the Civil War (1936–1939), it accounted for the latter’s entire decline, while in the 1950s it mitigated its fall. From 1975 onwards, the participation rate became its main driver. Thus, LF/WN accounts for over two-thirds of the contraction in hours worked per head during the ‘transition’ decade (1976–1985) and for practically all its reduction during the Great Recession (2008–2013). In both cases, the decline was due to a dramatic surge in unemployment. In the ‘transition’ decade, the fall in hours worked per head largely resulted from the impact of the oil shocks and the exposure to international competition in industrial sectors traditionally sheltered from competition, plus the return of migrants from Western Europe. Conversely, from Spain’s EU accession (1985) up to the Global Financial Crisis (2008), the increase in the participation rate (LF/WN) was the main contributor to the increase in the number of hours worked per person, helped by rising female participation and, especially, the inflow of immigrants, which represented about 5 million people between 1996 and 2008 (Izquierdo et al., 2015: 25). Again, the rise in the participation rate, as unemployment gradually declined and immigration resumed, has been a main actor in the aftermath of the Great Recession.

Lastly, the population share of those of working age (WN/N) increased during the 1930s and 1940s and, again, between 1976 and 2007, as the dependency rate (the population of children and elderly over working age) fell, representing a demographic bonus, which prevented further decline in the number of hours worked per person during the 1930s and 1976–1985, and became its main driver in the 1940s.

What explains the evolution of output per hour worked? A growth accounting framework allows us to break down labour productivity between the contribution of factor (physical and human capital and land per hour worked) and multifactor intensity, total factor productivity that includes “changes in efficiency in the use of those inputs and changes in technology” (Bosworth and Collins, 2003: 114).

Labour productivity (LP) can be decomposed as,

$$ {LP}^{\mathrm{t}}=A{\left({KS}^{\mathrm{t}}/{LQ}^{\mathrm{t}}\right)}^{\upalpha}{\left({X}^{\mathrm{t}}/{LQ}^{\mathrm{t}}\right)}^{\upbeta}{\left({LI}^{\mathrm{t}}/{LQ}^{\mathrm{t}}\right)}^{\upgamma} $$

(4.5)

LP being labour productivity; KS, a volume index of capital services; X^t, land input; LI, labour input; and LQ, the quantity of labour (hours worked); A, total factor productivity; and α, β, and γ output elasticities to each factor of production.

Thus, to disentangle the proximate determinants of labour productivity we require volume series of capital, land, and labour inputs.

3 Factors of Production

3.1 Labour Input

The labour input is the flow of services the labour force provides for production. To compute it we begin with an estimate of the labour quantity expressed as hours worked.¹ The data for the main sectors (agriculture, forestry, and fishing, industry construction, and services) come from Prados de la Escosura (2017, updated). For the period 1850–1994, the number of hours worked is derived by allocating workers and days worked per occupied in each of the main four sectors to their subsectors and, then, multiplying the number of days worked by the average hours worked per day in each subsector on the basis of Prados de la Escosura and Rosés (2010) estimates. From 1995 onwards, the national accounts (CNE10 and CNE15) supply the hours worked by subsector.

Next, we need to allow for quality of the labour force, and here we face a choice between an income-based approached, pioneered by Jorgenson (1990), and an education-based approach inspired by Mincer (1958) (See the discussion in Oxley et al., 2008).

In the income-based approach, a labour input index results from weighting the hours worked by each category of workers within each branch of economic activity according to their share in total nominal labour earnings. The rationale is that relative wages reflect the relative productivity of workers with different attributes and, thus, any returns per worker above those received by the unskilled worker represent returns to workers’ skills (human capital). However, this approach assumes a fully competitive economy, and not complying with this assumption may result in upwards biased estimates.²

Returns to each type of worker have been taken from Prados de la Escosura and Rosés (2010) up to 1984.³ From then onwards, national accounts provide average returns per employee at a disaggregated sector level although, unfortunately, no detailed information is provided according to age, sex, and qualification within each industry.⁴ This lack of differentiation within the labour force may bias the labour input index.⁵

Returns per occupied worker have been used to weight total labour (employees and self-employed) by branch. No distinction is made between employees and self-employed in the labour force estimates for the pre-national accounts period, 1850–1953. However, national accounts distinguish between compensation of employees and gross operating surplus and mixed incomes.⁶ Part of the mixed incomes correspond to self-employed compensation. Thus, for the post-1954 years, we have estimated self-employed labour returns following the principle of opportunity cost and assuming that the self-employed labour cost equals that of the average employee in their specific industry.⁷

Thus, total labour compensation is obtained as

$$ {w}^t{L}^t=\left({w}^t{E}^t/{E}^t\right){L}^t $$

(4.6)

𝑤^𝑡𝐿^𝑡 being the total labour compensation in period 𝑡; 𝑤^𝑡𝐸^𝑡, the compensation of employees; 𝐸^𝑡, the number of employees; and 𝐿^𝑡, total employment (employees plus self-employed) in period 𝑡.

A Törnqvist index of labour input (LI) is then computed,

$$ \ln \left({LI}^{\mathrm{t}}/{LI}^{\mathrm{t}-1}\right)=\Sigma {\overline{v}}^{\mathrm{i},\mathrm{t}}\ln \left({LQ}^{\mathrm{i},\mathrm{t}}/{LQ}^{\mathrm{i},\mathrm{t}-1}\right) $$

(4.7)

where LQ^l,t is the quantity of labour (hours worked) in branch i and v^{i, t} = ½(v^{i, t − 1} + v^{i, t}) the 2 year average share of each branch in total labour compensation (𝑤^𝑡𝐿^𝑡), being 𝑣^i,t ₌ 𝑤^i𝑡𝐿^i𝑡 / 𝑤^𝑡𝐿^𝑡. Then, the labour input index is obtained as the exponential.

An index of labour quality (H) that measures the labour input’s composition effect can be derived as the ratio between the labour input and labour quantity indices.⁸

$$ {H}^{\mathrm{t}}={LI}^{\mathrm{t}}/{LQ}^{\mathrm{t}.} $$

(4.8)

Our alternative education-based labour input combines the quantity of labour (hours worked) with an estimate of the quality of labour on the basis of school attainment. Up to 2000, data on average years of schooling for working age population (15–64 years) derive from Prados de la Escosura and Rosés (2010), who draw on Núñez’s (2005) education attainment estimates, completed for 2000–2010 with Barro and Lee’s (2013, updated) 5-year benchmark estimates, linearly interpolated, and UNESCO data, from 2010 onwards.

Following Bosworth and Collins (2003) and Lee and Lee (2016), labour quality is derived by combining years of schooling with the rate of return of education.⁹ Rates of return tend to be higher in early phases of development, but decline as economies develop. However, since private rates of return overestimate social rates of returns, it seems reasonable to adopt low values for the rate of return over time, and 7% per year of education has been chosen.¹⁰

$$ \mathrm{Thus}, EDU={\left(1+r\right)}^s $$

(4.9)

r being the rate of return and s the average years of schooling.

Then, the education-based labour input index is derived as the product of the labour quantity and labour quality indices.

An important caveat is that the education approach only considers levels of quantitative achievement (number of years of schooling), without any adjustment for the quality of education received. It ignores experience, on-the-job training, and informal education, as well as differences in the rate of return between different types of education. It also neglects the fact that education can be pursued as consumption, not as investment for production. Furthermore, in early stages of economic development, labour skills are largely dependent on experience and on-the job training, while formal education contributes more to labour quality in later phases.¹¹

A comparison of the alternative labour input indices derived with income- and education-based labour quality shows a similar evolution although the education-based series exhibit faster growth over time (Fig. 4.2). However, if we focus on labour quality, substantial differences emerge between the income- and education-based estimates (Fig. 4.3). Education-based labour quality accelerated in the late nineteenth century before flattening until the mid-1920s, when another spurt took place. Following the fall in the aftermath of the Civil War (1936–1939), there was steady growth that only slowed down during the Great Recession. Conversely, income-based labour quality improved moderately until 1920, when it accelerated until the eve of the Civil War. The post-1950 recovery, which only matched the pre-war level in 1960, gave way to an improvement until 1990, although it decelerated in the 1980s, and has flattened during the last three decades. In a nutshell, the main difference between the two outcomes of the two approaches is that, in the education-based labour input, labour quality has made a substantial contribution since the mid-twentieth century while, according to the income-based labour input, the contribution of labour quality was significant only during the 1920s and early 1930s and between 1950 and the mid-1980s (Table 4.3).

Fig. 4.2

Labour input: income- and education-based estimates (2010=100) (logs)

Fig. 4.3

Labour quality: income- and education-based estimates (2010=100) (logs)

Table 4.3 Labour input growth, 1850–2020 (annual average logarithmic rates %)

A challenge is posed by these opposite trends between the income- and education-based labour quality estimates. Which one better reflects the evolution of human capital? Both the income- and the education-based approaches have serious shortcomings. The fully competitive economy assumption in the income-based approach, if relaxed, would imply that labour quality is upwards biased in the resulting estimates, as part of it would simply represent the market power effect of higher income members in the labour force. In turn, ignoring experience, informal education and on-the-job training would bias upwards the growth of education-based estimates of labour quality, as compulsory and universal formal education (not just primary and secondary) has increased the number of years of schooling since the mid-twentieth century. Moreover, it could be argued that education is a high-income elastic good whose consumption demand must have increased substantially over the last 30 years as per capita income has doubled since Spain’s accession to the EU (1985), without necessarily having a significant impact on the quality of labour.¹² Therefore, although the actual evolution of labour quality might lie somewhere between the two alternative estimates, the income-based approach, though possibly downward biased, seems to provide a less distorted picture.¹³

3.2 Capital Input

3.2.1 Land Input

According to the OECD Manual (OECD, 2009), only land under dwellings and other construction and cultivated land should be considered as sources of capital services. Although land under structures is assumed to evolve as structures do and is, therefore, included under capital, agricultural land—a non-produced asset that suffers no depreciation—is considered to be an independent factor of production that provides a flow of services into production, an established practice in historical studies.¹⁴

Assessing the actual amount of land currently in agricultural use represents a challenge, and even more difficult is the valuation of land. Lack of annual data on land used prior to 1958, has forced us to accept the data at available scattered benchmarks and derive yearly figures through interpolation. For 1850–2000, Prados de la Escosura and Rosés (2009) estimates have been accepted, but without any adjustment for the agricultural economic cycle; from 2000 onwards these estimates are completed with data taken from official surveys on dry and irrigated land by type of use (Encuesta sobre superficies y rendimientos de cultivos en España, ESYRCE, 2023). Prices of different types of land for 1931 and 1985 are taken from Prados de la Escosura and Rosés (2009), and those for 2017 come from the Encuesta de Precios de la Tierra (2023).

A land input index has been obtained, weighting hectares of land assigned to different types of cultivation over 1850–1931, 1931–2000, and 2000–2020 by their average prices in 1931, 1985, and 2017, respectively. The resulting indices were then spliced into a single Laspeyres index.

Land input expanded in the late nineteenth and early twentieth century, and after declining during the Civil War, recovered in the 1940s. However, hardly any growth is observed thereafter and its contraction over 1986–2007 was partly reversed after the Great Recession (Table 4.4). Land input per hour worked exhibits negative growth except for 1890–1920 and during phases of employment destruction (1976–1985 and 2008–2013).

Table 4.4 Land input growth, 1850–2020 (annual average logarithmic rates %)

4 Proximate Determinants of Labour Productivity Growth

To establish the contribution of each factor of production to aggregate productivity growth, we need to weight their growth by their output elasticities. Under perfect competition and constant returns to scale, the values of these elasticities correspond to factor shares in GDP.¹⁵ Although the Spanish economy was far from fully competitive over time, we follow the usual practice (OECD, 2019) and accept this oversimplifying assumption, although it will bias our total factor productivity estimates.¹⁶

The labour share has been obtained by dividing total labour compensation (see the subsection on labour input above) by GDP at market prices.¹⁷ Then, the share of other factors, that is, 1 less the labour share, needs to be distributed between capital and land. Lack of information on land rents forces us to estimate land compensation as a residual, assuming that the difference between agricultural value added and labour outlays accrued to land property. However, this estimate provides an upper bound for the land share as it assumes no returns to capital in agriculture.¹⁸ The share of capital was, then, derived as a residual after subtracting labour and land returns from GDP.

Although, on average, factor shares conform to the stylised fact of two-thirds corresponding to labour and one-third to property owners (capital and land), factor shares are far from stable over time, contradicting Kaldor’s (1957: 592) stylised fact (Fig. 4.4). Labour and capital shares evolved as mirror images. Capital compensation increased its contribution to GDP, while labour reduced it, between 1880 and World War I and from 1960 onwards, and during a short episode in the late 1940s and early 1950s. Conversely, while the capital share declined in the interwar years (1919–1935) and, again, in the late 1950s, the labour share rose.

Fig. 4.4

Three factor shares (% GDP)

We can now compute the proximate sources of labour productivity growth using a Törnqvist index,

$$ \ln \left({LP}^{\mathrm{t}}/{LP}^{\mathrm{t}-1}\right)={\displaystyle \begin{array}{l}\Sigma {\overline{v}}^{\mathrm{k},\mathrm{t}}\left[\ln \left({KS}^{\mathrm{t}}/{KS}^{\mathrm{t}-1}\right)-\ln \left({LQ}^{\mathrm{t}}/{LQ}^{\mathrm{t}-1}\right)\right]\\ {}+\kern2px \Sigma {\overline{v}}^{\mathrm{x},\mathrm{t}}\left[\ln \left({X}^{\mathrm{t}}/{X}^{\mathrm{t}-1}\right)-\ln \left({LQ}^{\mathrm{t}}/{LQ}^{\mathrm{t}-1}\right)\right]\\ {}+\kern2px \Sigma {\overline{v}}^{\mathrm{l},\mathrm{t}}\left[\ln \left({LI}^{\mathrm{t}}/{LI}^{\mathrm{t}-1}\right)-\ln \left({LQ}^{\mathrm{t}}/{LQ}^{\mathrm{t}-1}\right)\right]+\ln \left({TFP}^{\mathrm{t}}/{TFP}^{\mathrm{t}-1}\right)\end{array}} $$

(4.16)

where ${\overline{v}}^{\mathrm{i},\mathrm{t}}=\frac{1}{2}\left({v}^{\mathrm{i},\mathrm{t}-1}+{v}^{\mathrm{i},\mathrm{t}}\right)$ the 2 year average share of each factor of production in GDP at market prices.

Total factor productivity (TFP) growth is, then, derived as a residual,

$$ \ln \left({TFP}^{\mathrm{t}}/{TFP}^{\mathrm{t}-1}\right)={\displaystyle \begin{array}{l}\ln \left({LP}^{\mathrm{t}}/{LP}^{\mathrm{t}-1}\right)\hbox{--} \{\Sigma {\overline{v}}^{\mathrm{k},\mathrm{t}}[\ln \left({KS}^{\mathrm{t}}/{KS}^{\mathrm{t}-1}\right)\\ {}-\ln \left({LQ}^{\mathrm{t}}/{LQ}^{\mathrm{t}-1}\right)]+\Sigma {\overline{v}}^{\mathrm{x},\mathrm{t}}[\ln \left({X}^{\mathrm{t}}/{X}^{\mathrm{t}-1}\right)\\ {}-\ln \left({LQ}^{\mathrm{t}}/{LQ}^{\mathrm{t}-1}\right)]\}\end{array}} $$

(4.17)

and the TFP index is obtained as its exponential.

Table 4.5 presents the breakdown of the average logarithmic growth rate of GDP per hour worked into the contribution of factor accumulation and efficiency gains (total factor productivity) and offers two alternative estimates of TFP growth derived with income- and education-based labour quality series, respectively. Figure 4.5 provides the yearly evolution of TFP using both indices.¹⁹

Table 4.5 Labour productivity growth and its sources, 1850–2020 (annual average logarithmic rates %)

Fig. 4.5

Total factor productivity: alternatively estimated with income- and education-based labour quality (2010=100) (logs)

From 1850 to 2020, capital deepening contributed over half the growth of labour productivity and efficiency gains about one-third, with the remainder attributable to labour quality. A glance at the evolution of labour productivity makes it possible to distinguish different phases of growth, three of them with TFP significant contributions. Between the mid-nineteenth century and World War I, a phase of sustained progress from 1850 to the early 1890s gave way to another of sluggish performance until 1919. Efficiency gains account for the growth differential between the two phases. While capital contribution was steady during these 70 years, TFP only expanded from 1850 to 1892, providing half the growth of labour productivity (slightly less when education-based labour quality is used in the computation).

The 1920s witnessed a vigorous performance of labour productivity, trebling its pre-1890 growth. Capital deepening doubled its pace and contributed about one-third of labour productivity growth. However, TFP was the main driver, with its contribution ranging from over half to nearly two-thirds of labour productivity growth (depending on whether it is derived with income- or education-based labour quality). During the 1930s, TFP collapse accounted almost exclusively for the decline in labour productivity growth. TFP made also the largest contribution to its post-Civil War recovery.

Output per hour worked grew exceptionally fast from 1954 to 1985 (5.8%), a period that encompasses the Golden Age and the ‘transition to democracy’ decade. Efficiency gains contributed half of its growth and physical capital accounted for another two-fifths, with the rest attributable to labour quality. A closer look reveals that during the Golden Age (1954–1975) TFP contributed over half labour productivity growth, and over one-third in the ‘transition to democracy’ decade, while the contribution of capital deepening rose from over one-third in the Golden Age to half in the ‘transition’ years.

Then, between Spain’s accession to the EU (1985) and the eve of the Global Financial Crisis (2007), labour productivity growth shrank to less than one-fifth compared to 1954–1985, becoming largely extensive, rather than intensive. Capital deepening accounted for the sluggish output per hour growth and TFP did not contribute at all. Sluggish labour productivity growth played, thus, a secondary role in a long phase of robust (absolute and per capita) GDP growth (3.5% and 2.8%) that was driven by the increase in hours worked per person resulting from higher employment (Table 4.1).

The Great Recession (2008–2013) was another episode in which capital drove the mild acceleration in labour productivity growth, while TFP growth was negative. In the post-Global Financial Crisis years, capital deepening prevented negative labour productivity growth. When only the education-based labour quality is considered, human capital made a contribution that cancelled negative TFP growth.

As human capital is a major factor in narratives of economic growth, the role of labour quality in Spain’s long run growth merits some comments. If we follow the education-based approach, labour quality added to labour productivity growth from the mid-twentieth century onwards, and has made a significant contribution since Spain’s accession to the European Union (1985), second only to capital deepening. Such an optimistic outcome needs to be set against reservations with regard to educational attainment as a measure of human capital; in particular, the demand for said attainment as a high-income elastic consumption good. The income-based approach, although upwards biased as it assumes perfect competition, suggests, instead, that labour quality contributed to labour productivity growth during the Golden Age and the ‘transition to democracy’ decade, but not thereafter, which sounds a more persuasive narrative.

We have replicated the growth accounting exercise using only two factors of production, as is conventionally the case (assuming that the share of capital is 1 less the share of labour), in order to provide a robustness test for our results. Figure 4.6 presents the evolution of TFP that results from growth accounting exercises with two and three factors of production for both estimates with income- and education-based labour quality. Both sets of estimates follow the same pattern, but the two-factor estimates present a higher level relative to 2010, the benchmark year. This implies slightly slower TFP growth, which results from the fact that capital input, which grows much faster than land input, receives a larger weight (as it includes the land share in GDP) in the growth accounting exercise (Table 4.6). An implication of this comparison is that growth accounting exercises for developing economies that neglect the land input tend to over-exaggerate the share of capital and, hence, underestimate TFP growth.

Fig. 4.6

Total factor productivity: estimated with three and two factors of production and income- and education-based labour quality (2010=100) (logs)

Table 4.6 Labour productivity growth and its sources, 1850–2020: two factors of production (annual average logarithmic rates %)

How do our results for the evolution of the TFP compare with earlier studies? Figure 4.7 compares our new estimates, derived with both the income- and education-based labour quality with those by Prados de la Escosura and Rosés (2009) for 1850–2000, derived with income-based labour quality, and Bergeaud et al. (2016), updated estimates, using 2000 as reference. These two series present a close evolution until the last quarter of the twentieth century, as they rely on the same sources.²⁰ When compared to our new estimates, a similar evolution is observed but both earlier estimates grow faster during the 1960s and early 1970s and, in the case of Prados de la Escosura and Rosés (2009) the growth differential with our new estimates continues during the ‘transition to democracy’ years. It is also worth mentioning that Bergeaud et al. series present a sharp deceleration after 1986 but still some progress, unlike the stagnation or negative TFP growth in the rest of the estimates.

Fig. 4.7

Long run trends in total factor productivity: comparative estimates (2000=100) (logs). Note: New estimates derived with income- and education-based labour quality

Another possible comparison regarding the post-1950 era is provided in Fig. 4.8, which presents the new estimates together with those provided by the Penn World Tables 10.0 (Feenstra et al., 2015, updated) for the post-1954 era, and the Conference Board (2022) from 1990 onwards, in which TFP is derived using education-based labour quality. The Conference Board’s TFP series closely match our own education-based estimates, while the Penn World Tables series adopt an intermediate position between two new set of estimates. Although there are noticeable differences in the pace of growth, these trends largely coincide, with the PWT10.01 series showing, like the new estimates with income-based labour quality, sustained TFP growth until 1989 and, then, mild but steady decline until 2013, while the Conference Board series stresses the post-1990 fall, as do the new TFP estimates derived with education-based labour quality.

Fig. 4.8

Total factor productivity since 1950: alternative estimates (2010=100) (logs). Note: New estimates derived with income- and education-based labour quality

How does Spain compare to other countries during phases of TFP acceleration such as the 1920s or the Golden Age (1950–1973)? Although methodological differences may bias the results, a face value comparison provides some informative results.²¹ In the 1920s, when contrasted with other Peripheral European countries, TFP growth appears more intense in Spain than Portugal and Turkey, but less than in Italy. Portugal’s yearly growth was below 1% and Turkey’s was negative, while in Italy and Spain growth reached 2.5% and 1.9–2.2% (depending on the use of income- or education-based labour quality), respectively.²² Moreover, TFP grew faster in Spain than in the U.K. and the U.S. However, from 1850 to 1890, the previous phase of TFP acceleration, Spain’s TFP growth was lower than in the U.K. but higher than in the U.S. and Italy.²³

In the Golden Age, the yearly rate of growth in Spain (2.9–3.2% from 1954 to 1975) was, again, above those of Portugal (1.5%) and Turkey (0.8%), but below Italy’s (4.0%), although Spain TFP’s behaved better than Italy’s in the late 1970s and 1980s.²⁴ Spain also exhibited faster TFP growth than the leading socialist countries of Central and Eastern Europe—Czechoslovakia, Hungary, and Poland—, which grew at 1.3%, 2.1%, and 1.9%, respectively, from 1950 to 1970 (Vonyó and Klein, 2019: 335). If we extend the comparison to South East Asia, where TFP acceleration started after 1960, we observe that Spain’s rate of growth (2.4–2.6% in the years 1959–1985) was higher than in Hong-Kong, South Korea, and Taiwan, 2.3%, 1.7%, and 2.1%, respectively, from 1966 to 1991 (Young, 1995: 672). Lastly, if the contrast is carried out with the advanced economies, it emerges that TFP grew faster in Spain than in the U.S. (2.1%) and the U.K. (1.9%), similarly to Germany and Japan (3.3% and 3.2%), but slower than in France (3.6%).²⁵

It can therefore be concluded that Spain compared to the best performers during phases of generalised TFP growth acceleration such the 1920s and the years 1950–1975.

If we now turn to the long phase of TFP deceleration since 1986, what explains the shift from efficiency gains to capital deepening as labour productivity’s main driver? The fact that TFP growth halted helps explain the shift, but why did this happen to TFP? A convergence hypothesis can be considered. As TFP grew sharply over three decades (Fig. 4.8), Spain moved closer to the technological frontier and achieving further efficiency gains became more difficult. Furthermore, once-and-for-all structural change associated with the shift of resources from sectors of low or slow growing productivity to those of high, or fast growing productivity (i.e. labour moving from agriculture into manufacturing) had already taken place by the time Spain joined the EU. Thus, Spain’s potential for catching up would have been exhausted, and TFP growth slowed down, adjusting to its pace in advanced economies.

Table 4.7 compares levels of output per hour worked in 1990 (expressed in 2019 EKS US dollars) in OECD countries (ranked from top to bottom) with their TFP growth rates since 1990 using the Conference Board (2022) dataset. In both periods considered, that of expansion, 1990–2007, and 1990–2019, Spain had the poorest TFP performance, and all countries with higher initial levels of output per hour worked than Spain in 1990 exhibit faster TFP growth in both periods. Such results refute, therefore, the convergence hypothesis.²⁶

Table 4.7 Labour productivity in 1990 (2019 EKS US$) and TFP growth 1990–2019 (%)

Alternative explanations have been put forward to explain why during the last three decades labour productivity growth has slowed down in Spain and become extensive rather than intensive. It has been hypothesised that, as resources were re-allocated towards sectors that attracted less innovation (from traded to non-traded sectors, i.e. low skill services and construction), aggregate efficiency declined. Specifically, Díaz and Franjo (2016) blamed investment in residential structures, stimulated by favourable relative prices and subsidies, together with low investment specific technical change (ISTC), for the TFP slowdown. Pérez and Benages (2017) stressed the low investment in intangibles and the excess capacity and limited use of their capital by predominantly small firms. The picture was completed by Cuadrado et al. (2020) who pointed to the limited exploitation of new technologies because of workers’ low skills. The recovery of the share of structures in net capital stock and its substantial contribution to total value of capital services in the early twenty-first century support these assertions (Chap. 2). Moreover, the low ISTC is consistent with the deceleration of capital ‘quality’ since 1990 (Fig. 4.5).

García-Santana et al. (2020) offered a nuanced view of the TFP slowdown in which it is allocative inefficiency across firms, rather than across sectors, that accounts for the deceleration.²⁷ Moreover, they found that government regulation (cronyism) is its ultimate determinant. Looking at the context in which this misallocation has taken place, Gopinath et al. (2015) argued that, by lowering interest rates and encouraging an inflow of capital, the adoption of the Euro may have been partly responsible for the allocation of capital to less productive firms and, hence, for the low TFP growth.

Furthermore, companies’ low expenditure on research and development and low investment in intangible capital, which hampers TFP (Corrado et al., 2013), are associated with regulatory restrictions on competition in product and factor markets (Alonso-Borrego, 2010). Specifically, retail trade regulation, the costs of company creation, lack of flexibility in the labour market, bankruptcy legislation and judicial procedures all militate against competition (Mora-Sanguinetti and Fuentes, 2012).

5 Concluding Remarks

The current productivity slowdown has stimulated research on the causes of growth. This chapter has explored long-term growth and its proximate sources in Spain. Labour productivity dominated GDP long-run growth. Half the increase in labour productivity came from capital deepening and one-third from efficiency gains. In phases of labour productivity acceleration, total factor productivity was its driving force and a complementarity existed between capital deepening and efficiency gains. Moreover, Spain was among the best performers during phases of generalised TFP acceleration such as the 1920s and the Golden Age.

Since the mid-1970s, the Spanish economy has been unable to combine employment creation with labour productivity growth and capital deepening, a finding consistent with the fact that expanding sectors that created more jobs experienced slower output per hour growth, as they were less successful in attracting investment and technological innovation. During the ‘transition to democracy’ decade (1976–1985), labour productivity continued to thrive, since deep structural change and industrial re-structuring eliminated sheltered low-productivity industries.

Labour productivity slowdown only began after Spain’s accession to the European Union, associated with deceleration in capital deepening and TFP stagnation. GDP growth became extensive, largely depending on the increase in hours worked per person as employment grew until the Global Financial Crisis. Capital misallocation, low investment in intangibles and ISTC negatively affecting capital deepening and TFP growth resulted from obstacles to competition in product and factor markets, subsidies, and cronyism.

So do restrictions to economic freedom, regulation and worsening property rights, in particular, help explain the poor labour productivity performance during the last three decades? Furthermore, does economic freedom constitute an ultimate determinant of capital deepening and TFP growth over the long run? Answering these questions require further research.

Appendix

See Tables 4.8, 4.9, 4.10, 4.11 and 4.12.

Table 4.8 Real GDP and its composition, 1850–2020 (2010=100)

Table 4.9 Hours worked per person and its composition, 1850–2020 (2010=100)

Table 4.10 Labour input and its composition, 1850–2020 (2010=100)

Table 4.11 Factor shares (% GDP), 1850–2020 (current prices)

Table 4.12 GDP per hour worked and its proximate determinants, 1850–2020 (2010=100)

See Figs. 4.9 and 4.10.

Fig. 4.9

Labour quality: new income- and education-based estimates compared to PWT10.01 and conference board education-based estimates (2010=100) (logs)

Fig. 4.10

Total factor productivity: alternatively estimated with VICS ex-ante exogenous and ex-post endogenous and income-based labour quality (2010=100) (logs)