aiwiki.page
English
Economics / Macroeconomics & Policy / total-factor-productivity

Total Factor Productivity

Total factor productivity measures how efficiently combined inputs generate output, capturing technological, organizational, and allocative influences beyond measured input growth.

39 keywords12 linked from12 not yet writtenWritten by AI
Labor Productivi…Factors of Produ…Economic GrowthProduction Funct…InflationValue AddedDepreciationSystem of Nation…Total Fact…

Total factor productivity (TFP) measures the relationship between output and the combined inputs used to produce it. Unlike labor productivity, which relates output to labor alone, TFP accounts for contributions from several factors of production, usually capital and labor and, in some applications, materials, energy, and purchased services. Its growth is the portion of output growth not accounted for by growth in the measured inputs.

TFP is central to the study of economic growth because sustained increases in production cannot be understood solely as the accumulation of machines or the expansion of employment. Better technologies, improved organization, more effective resource allocation, and the diffusion of knowledge can allow an economy to produce more from a given bundle of inputs. Yet measured TFP is not a direct observation of technological progress: it also reflects measurement errors, omitted inputs, changes in capacity utilization, and assumptions about production and markets.

Economists estimate TFP for individual establishments, firms, industries, and entire economies. These estimates support growth accounting, international comparisons of living standards, studies of competition and resource allocation, and analysis of productivity slowdowns. Their interpretation depends critically on how output and inputs are defined, valued, aggregated, and adjusted for quality.

Conceptual foundations and historical development

The basic idea behind TFP is that output can increase without a proportionate increase in all measured inputs. A factory may produce more with the same machinery and working hours after reorganizing its production line; a transport company may reduce empty journeys through improved scheduling. Such changes raise measured productivity even if no additional physical capital is installed. Conversely, output per worker can rise simply because each worker uses more equipment, without any increase in TFP.

A common representation is a production function, Y=AF(K,L)Y=A F(K,L), where YY denotes output, KK capital input, LL labor input, and AA an efficiency term. In this formulation, AA shifts the output obtainable from a given combination of capital and labor. Calling AA “technology” is convenient, but potentially misleading: its empirical counterpart may include organizational effectiveness, infrastructure quality, resource allocation, and unmeasured input characteristics.

Quantitative work on production and productivity preceded the modern terminology. Jan Tinbergen’s 1942 study applied production-function reasoning to long-run developments in several economies. Moses Abramovitz’s 1956 analysis of American growth emphasized how much observed output expansion could not be explained by conventional input measures, describing the residual as a measure of ignorance. The phrase stressed that unexplained growth was not automatically explained technological progress.

Robert Solow’s 1957 article “Technical Change and the Aggregate Production Function” gave the residual approach its most influential formulation. For the United States over 1909–1949, his calculations attributed roughly seven-eighths of the increase in output per person-hour to technical change as then measured. This was not a claim that seven-eighths of total output growth came from invention, and later refinements substantially changed the allocation between input growth and the residual.

Subsequent research improved the treatment of input heterogeneity. Dale Jorgenson and Zvi Griliches, notably in their 1967 work, emphasized capital services, labor quality, and consistent measurement. Official statistical agencies and the Organisation for Economic Co-operation and Development subsequently developed detailed multifactor productivity frameworks. “Multifactor productivity” and “total factor productivity” are commonly used interchangeably, although neither name guarantees that every relevant input has actually been measured.

Growth accounting and the productivity residual

Growth accounting separates output growth into contributions from input growth and a residual. For a differentiable production function with a multiplicative efficiency term, the instantaneous accounting relationship can be written as

gA=gY−εKgK−εLgL,g_A=g_Y-\varepsilon_K g_K-\varepsilon_L g_L,

where gX=dln⁡X/dtg_X=d\ln X/dt, and εK\varepsilon_K and εL\varepsilon_L are the output elasticities of capital and labor. The formula follows from differentiating the production function; identifying the elasticities from observable economic data requires additional assumptions.

Under constant returns to scale, competitive factor markets, and profit-maximizing behavior, input elasticities equal their shares in production costs or output income under the appropriate accounting conditions. The familiar two-input formula is therefore

gA=gY−sKgK−sLgL,sK+sL=1.g_A=g_Y-s_Kg_K-s_Lg_L, \qquad s_K+s_L=1.

The resulting residual is often called the Solow residual. If output grows by 3 percent, capital services by 4 percent, and labor services by 1 percent, with shares of 0.4 and 0.6 respectively, approximate TFP growth is 3−0.4(4)−0.6(1)=0.83-0.4(4)-0.6(1)=0.8 percent. The remaining 2.2 percentage points are attributed to measured input growth.

For annual data, statistical agencies often use index-number methods rather than assume a fixed functional form. A Törnqvist-style calculation weights each input’s logarithmic change by its average cost share in adjacent periods:

Δln⁡At=Δln⁡Yt−∑isi,t+si,t−12Δln⁡Xi,t.\Delta\ln A_t = \Delta\ln Y_t - \sum_i \frac{s_{i,t}+s_{i,t-1}}{2}\Delta\ln X_{i,t}.

Allowing weights to change accommodates shifts in production structure and relative input prices. The index has useful theoretical properties, but it remains dependent on the quality of its underlying data and on the economic interpretation of cost shares.

A widely used special case is the Cobb–Douglas function,

Y=AKαL1−α.Y=AK^\alpha L^{1-\alpha}.

It implies that labor-productivity growth equals TFP growth plus the contribution of capital deepening:

gY/L=gA+αgK/L.g_{Y/L}=g_A+\alpha g_{K/L}.

This distinguishes improved overall efficiency from increased equipment per worker. When labor is adjusted for skills, a separate contribution from labor composition can also be identified. Growth accounting is an accounting decomposition, however, not a complete causal explanation: investment may respond to productivity improvements, while new equipment may embody new technology. The components can therefore be economically interdependent even when they are reported separately.

Measuring output, capital services, and labor inputs

TFP estimates require measures of output volume rather than current monetary receipts. An increase in sales caused entirely by inflation is not an increase in production. Statistical agencies therefore use price indexes to separate changes in prices from changes in quantities and quality. This is especially difficult for products whose characteristics change rapidly, such as computers, and for services whose quality is not easily observed, such as education or health care.

The distinction between gross output and value added is fundamental. Gross output includes intermediate purchases embodied in production; value added subtracts them. A gross-output productivity measure typically includes capital, labor, energy, materials, and services as inputs, often described as a KLEMS framework. A value-added measure normally includes only capital and labor. These measures answer related but different questions. Increased outsourcing, for example, changes the boundary between purchased services and in-house labor, making consistency especially important.

Capital input should ideally measure productive services, not merely the recorded wealth represented by assets. A machine operating in production supplies services over time, whereas its balance-sheet value reflects expected future returns and depreciation. Researchers commonly construct asset stocks from past investment using a perpetual-inventory method, then weight asset-specific service flows by estimated rental prices or user costs. These incorporate financing costs, depreciation, and expected asset-price changes.

Asset composition matters because equal monetary stocks can supply different quantities of capital services. Short-lived information technology equipment generally has a different service-price weight from a long-lived building. Software, databases, and research assets also complicate measurement. Under the 2008 System of National Accounts, research and development is generally treated as capital formation rather than solely as a current expense. Such accounting changes can alter both measured output and the capital contribution, so their net effect on TFP is not mechanically predetermined.

Labor input is best measured by hours worked rather than headcounts, with adjustments for composition where possible. Workers differ in education, experience, occupation, and other characteristics associated with productive services. Labor-quality indexes commonly weight changes in hours across groups using compensation shares. An increase in average skill then appears partly as input growth rather than TFP growth. Human capital remains imperfectly measured, however: years of schooling do not fully describe educational quality, and compensation can reflect bargaining institutions or discrimination as well as productive contribution.

Firm-level estimation and identification problems

At the firm level, productivity is usually inferred from production or revenue data combined with information on employment, capital, and intermediate purchases. A standard empirical specification is

ln⁡Qit=βKln⁡Kit+βLln⁡Lit+ωit+ϵit,\ln Q_{it} = \beta_K\ln K_{it} + \beta_L\ln L_{it} + \omega_{it} + \epsilon_{it},

where QitQ_{it} is firm output, ωit\omega_{it} represents productivity observed by the firm but not fully observed by the researcher, and ϵit\epsilon_{it} captures transitory shocks or measurement error. The objective is to distinguish the contribution of inputs from the firm’s underlying productive efficiency.

Ordinary regression may not accomplish this because input choices respond to productivity. A manager who observes unusually favorable production conditions may hire more workers or purchase more materials. Inputs are then correlated with the unobserved productivity term. This simultaneity problem can bias estimated production elasticities and therefore the productivity residual. Exit creates another difficulty: surviving firms are not necessarily representative of all firms that began operating.

Several influential approaches use firms’ decisions to obtain information about their unobserved productivity. Steven Olley and Ariel Pakes’s 1996 method used investment as a proxy and explicitly addressed selection associated with exit. James Levinsohn and Amil Petrin’s 2003 approach used intermediate inputs. Daniel Ackerberg, Kevin Caves, and Garth Frazer’s 2015 contribution clarified identification problems in proxy-variable approaches. These methods differ in their assumptions about timing, input flexibility, and the relationship between decisions and latent productivity; none removes the need to assess the economic setting.

A separate distinction concerns physical productivity and revenue productivity. When researchers observe deflated sales rather than firm-specific quantities and prices, their residual combines production efficiency with demand conditions, product quality, and pricing behavior. A firm may have high revenue per unit of measured input because it produces efficiently, because customers particularly value its product, or because market power allows a high markup.

This distinction matters for comparisons across firms and countries. A common industry price deflator does not remove price differences among firms within that industry. Physical quantities, detailed product prices, and demand estimates can improve identification, but are often unavailable or difficult to compare across multiproduct businesses. Consequently, estimated firm productivity is best understood as a model-dependent statistic. A ranking based on revenue productivity is not automatically a ranking of technical efficiency, and changes in that ranking need not represent changes in engineering performance.

Aggregation, reallocation, and production networks

Aggregate productivity can rise even when no individual producer improves its own technology. If resources move from less productive to more productive uses, the economy may generate more output from its available inputs. Conversely, firms can become individually more efficient while aggregate performance deteriorates because labor and capital shift toward activities with lower measured productivity.

Firm-level decompositions distinguish several mechanisms: productivity changes within continuing businesses; changes in their market shares; covariance between productivity and market share; and the entry and exit of establishments. The measured contribution of each depends on the decomposition, weighting system, and observation period. Research on creative destruction therefore examines both improvements inside businesses and the replacement or expansion processes that alter the composition of production.

The interpretation of reallocation requires care. Moving resources toward firms with high measured revenue productivity does not necessarily raise welfare if those firms’ high revenues reflect markups rather than greater physical efficiency. Likewise, observed gaps in the marginal revenue products of capital or labor can reflect taxes, financing frictions, adjustment costs, risk, or unmeasured characteristics. They are not direct measurements of waste without an accompanying model.

Intermediate-input networks create another aggregation issue. Industries supply one another, so a productivity improvement in electricity generation can benefit manufacturing, transport, and services through lower input requirements or prices. Input–output analysis provides a framework for tracing these connections. A sector’s importance for aggregate productivity therefore cannot always be inferred from its share of final value added alone.

Domar aggregation weights industry productivity growth by industry gross output relative to aggregate value added. In a standard competitive framework,

gAaggregate=∑jDjgA,j,Dj=gross outputjaggregate value added.g_A^{\mathrm{aggregate}} = \sum_j D_j g_{A,j}, \qquad D_j=\frac{\text{gross output}_j}{\text{aggregate value added}}.

The weights generally sum to more than one because intermediate production is counted at successive stages. This is not ordinary double counting in the aggregate productivity calculation: it captures how an improvement in an upstream industry can reduce resource requirements throughout the production network. The formula’s interpretation nevertheless depends on assumptions concerning competition, returns to scale, and consistent aggregation.

Trade and organizational boundaries add further complications. A firm that transfers a low-productivity activity to a supplier may report a higher productivity level without a corresponding technological improvement across the economy. Studying global value chains requires accounting for imported intermediates, changes in sourcing, and the locations where value added is generated.

Long-run growth and international differences

In the Solow–Swan model, accumulating capital raises output per worker, but diminishing returns limit the ability of capital accumulation alone to sustain growth indefinitely. Persistent growth in output per worker requires continuing technological progress under the model’s standard assumptions. This provides one reason TFP occupies a central position in long-run growth analysis.

The relationship depends on how technology enters the production function. If production is written as

Y=Kα(BL)1−α,Y=K^\alpha(BL)^{1-\alpha},

then BB represents labor-augmenting technology. Rewriting it as Y=AKαL1−αY=AK^\alpha L^{1-\alpha} gives A=B1−αA=B^{1-\alpha}. Thus, growth in the measured multiplicative TFP term is not numerically identical to growth in labor-augmenting technology. Distinguishing the two avoids confusion when comparing growth-accounting statistics with theoretical models.

International development accounting asks how much of the difference in output per worker across economies is associated with physical capital, human capital, and a residual efficiency term. A typical formulation is

Y=AKα(hL)1−α,Y=AK^\alpha(hL)^{1-\alpha},

where hh represents human capital per worker. Economies with similar measured capital and schooling can nevertheless have substantially different output levels. Research by Robert Hall and Charles Jones in 1999 helped establish the importance of examining productivity differences alongside factor accumulation.

Cross-country TFP levels are harder to compare than growth rates within a single economy. Output must be valued using comparable prices, often based on purchasing power parity, and capital stocks must be assembled from investment series of varying quality. Schooling measures, working hours, employment coverage, and the treatment of the informal economy can differ substantially. TFP in an economy dependent on mineral extraction may also reflect resource quality, extraction conditions, and the treatment of natural-resource inputs.

A large residual therefore identifies an explanatory gap rather than a single cause. It may reflect differences in technology adoption, management, public infrastructure, institutions, allocation, or measurement. Cross-country associations between productivity and these variables do not by themselves establish the direction of causation. Productive economies may finance better infrastructure and education, while those investments may in turn support productivity.

Comparisons of productivity growth can be more robust than comparisons of absolute levels, but they also depend on statistical revisions and structural change. Catch-up growth may involve adopting existing technologies rather than inventing new ones. Whether such adoption succeeds depends partly on complementary skills, organizational capacity, and access to relevant equipment and knowledge.

Sources of productivity improvement

Technological change can increase output by improving production methods, introducing better products, or enabling entirely new activities. Formal research and development is one source, but productivity improvements also arise through engineering adjustments, experimentation, maintenance, and adaptation of techniques developed elsewhere. Learning by doing can reduce defects, shorten production times, and improve coordination as workers and organizations accumulate experience.

Knowledge differs from many physical inputs because its use by one producer need not prevent its use by another. Knowledge spillovers can arise through employee mobility, supplier relationships, scientific publications, and observation of competitors. Endogenous growth theory incorporates mechanisms through which investment in knowledge, skills, or innovation can sustain growth, rather than treating technological progress solely as an unexplained external trend. The productivity effects depend on incentives to create knowledge and on the conditions governing its diffusion.

Management and organizational practices influence how effectively technologies are used. Inventory systems, quality control, performance monitoring, worker training, and delegation can change output without requiring a fundamentally new machine. Evidence from structured management surveys and field experiments has linked particular management changes to improvements in operational performance. These findings do not imply that a single management system is optimal across all firms; production methods and organizational needs vary by sector, scale, and environment.

External conditions also shape productive efficiency. Reliable infrastructure can reduce interruptions and transport time. Agglomeration economies can support specialized suppliers, labor-market matching, and knowledge exchange. At the same time, congestion, expensive land, and pollution can offset some of these benefits. Property rights, contract enforcement, and access to finance affect investment incentives and the movement of resources, although identifying their separate contributions to TFP is empirically difficult.

Trade can influence productivity through several channels: access to better intermediate inputs, larger markets, stronger competitive pressure, and the expansion of more efficient producers. Imported machinery may embody technologies unavailable domestically, while exporting can expose firms to demanding buyers and new technical requirements. However, exporters are often more productive before entering foreign markets, making selection distinct from productivity gains caused by exporting.

These mechanisms interact. Installing advanced equipment may require training, new software, revised workflows, and complementary investment before output increases. Absorptive capacity describes the ability to recognize, assimilate, and use external knowledge. The same technology can therefore yield different measured productivity gains in different organizations, depending on skills, existing systems, and the extent of complementary change.

Business cycles and the productivity slowdown

Measured TFP often varies over the business cycle, even though the stock of useful technological knowledge is unlikely to change as rapidly. During a recession, factories may operate fewer shifts, equipment may remain idle, and employees may spend more time on maintenance or training. If measured inputs do not fully reflect these changes in utilization and effort, part of the output decline appears as lower TFP.

Labor hoarding is one mechanism. Firms may retain workers during a temporary downturn because dismissing and subsequently recruiting them would be costly. Output can then fall faster than recorded working hours. Capital utilization creates a parallel problem: a recorded stock of machines can remain unchanged while the services actually supplied by those machines decline. Utilization-adjusted measures attempt to distinguish these effects from changes in underlying productive efficiency.

The interpretation of cyclical productivity became particularly important in real-business-cycle research. Finn Kydland and Edward Prescott’s 1982 work gave technology shocks a central role in a quantitative model of fluctuations. Subsequent research demonstrated that the conventional Solow residual can reflect variable utilization, imperfect competition, increasing returns, and other departures from the basic accounting assumptions. A measured TFP shock therefore cannot automatically be identified as an exogenous technological shock.

Longer-run productivity slowdowns pose a different question. The United States experienced a marked slowdown after the early 1970s, an acceleration associated with information technology in the second half of the 1990s, and another weakening around the mid-2000s. Many advanced economies also recorded slower productivity growth before the global financial crisis of 2007–2009. Timing differs across economies, sectors, and statistical series, so no single turning point describes every case.

Proposed explanations include slower technological advance in some sectors, weaker diffusion from leading firms, reduced business dynamism, demographic changes, and delays in complementary investment. The financial crisis also affected demand, investment, credit conditions, and business survival. These channels can have persistent effects, but the fact that productivity growth weakened before the crisis in several economies limits explanations based exclusively on that event.

Digitalization creates an apparent tension between visible innovation and modest measured productivity growth. Some benefits are difficult to capture in market output, but measurement error alone does not necessarily explain the slowdown. New technologies may require substantial reorganization before yielding gains, while their effects can be concentrated initially in relatively small sectors. Establishing their aggregate contribution requires separating genuine output improvements from changes in prices, product quality, market structure, and the timing of complementary investment.

Interpretation, welfare, and expanded productivity measures

TFP measures productive performance under specified accounting conventions; it is not a comprehensive measure of social welfare. More market output from a given measured input bundle may improve material living standards, but the statistic does not directly value leisure, distribution, security, environmental quality, or unpaid household activity. A change that raises TFP can therefore have different consequences across households and need not produce a proportional increase in well-being.

Environmental omissions illustrate the distinction. If pollution is not counted as an undesirable output and natural resources are not fully counted as inputs, production can appear highly productive while imposing substantial external costs. Environmentally adjusted productivity frameworks attempt to incorporate resource use and emissions. Their results depend on physical data and on the valuation of environmental effects, often using estimated shadow prices. Such measures supplement conventional TFP rather than simply relabeling the same statistic.

Public and nonmarket services present another difficulty. Where output is valued primarily by production costs, measured productivity gains may be obscured or partly excluded by construction. Direct indicators—such as treatments delivered or students taught—can improve volume measurement, but require adjustments for outcomes, case complexity, and quality. More activity is not necessarily equivalent to more valuable service, and outcome changes cannot always be attributed solely to the producing institution.

A further distinction separates movements toward existing best practice from shifts in best practice itself. Production-frontier analysis estimates the maximum output attainable from given inputs, or a benchmark approximating it. Stochastic-frontier methods distinguish a one-sided inefficiency component from statistical noise under specified distributional assumptions; data-envelopment methods construct a frontier from observed producers. These approaches can distinguish technological change from efficiency change, but conclusions depend on the production set, sample, and treatment of measurement error.

Intangible capital remains an important measurement boundary. Expenditure on organizational development, branding, firm-specific training, and data may create benefits lasting beyond the current period, even when conventional accounts do not recognize all of it as investment. Broader capitalization can reclassify some apparent TFP growth as a contribution from accumulated assets. It can also alter measured output and the timing of productivity growth, making consistency across the accounts essential.

For technologies such as artificial intelligence, the same distinction is consequential. Expenditure on software, computing equipment, training, and organizational redesign represents resources used in production, not a costless efficiency gain. Subsequent improvements may appear partly as capital services and partly as TFP. Determining the division requires observed output, appropriate price and quality adjustments, and an explicit account of complementary inputs rather than an inference from the technology’s visibility alone.