OECD Compendium of Productivity Indicators 2025

8. Cyclical adjustment of multifactor productivity

Context and key findings

A broad-based revival in productivity growth is crucial to addressing long-term socio-economic challenges across OECD countries. Productivity growth – producing more with the same or fewer inputs – is the primary engine of sustained economic growth and rising living standards (André and Gal, 2024[1]). In today’s context, where labour force expansion is slowing and capital accumulation is constrained by high debt and uncertainty, input-driven growth is limited. That puts a spotlight on multifactor productivity (MFP) as the key engine of sustainable growth. MFP isolates the portion of productivity growth not explained by increases in labour or capital inputs – it reflects how efficiently these inputs are used (Chapter 2). This includes gains from technological progress, innovation, improved skills deployment, better resource allocation, and organisational or managerial efficiency.

Historically, MFP has demonstrated a strong correlation with overall economic growth, primarily because measured productivity often rises during economic expansions and declines during recessions. However, this procyclicality largely arises from fluctuations in factor utilisation, such as labour hoarding during downturns or intensified capital usage during expansions. Recognising this is crucial to distinguish underlying productivity trends from transient cyclical movements. Failing to account for these cyclical variations risks misinterpreting temporary shocks as structural changes, potentially leading to inaccurate assessments of longer-term productivity developments. Using a range of techniques, this chapter provides insights into the evolution of cyclically-adjusted MFP growth in the manufacturing sector for 20 OECD countries (listed in Annex Table 8.A.1) since the end of the 1990s.

Key findings

• Experimental estimates suggest that cyclically-adjusted multifactor productivity growth (MFP) averaged around 2.2% across a sample of OECD countries between 1995 and 2022, close to the average unadjusted MFP over the same period. The findings appear robust to the methodological choices and consistently highlight a trend of weak, stagnating, or continuously declining MFP growth in recent decades.

• By filtering out short-term fluctuations, cyclical adjustment offers a distinct view of productivity trends that complements the picture from unadjusted data, especially during downturns. In the 2008 crisis, for example, adjusted MFP estimates (0.1% to 2.6%) across selected OECD countries were notably higher than the unadjusted rate (-2.5%), illustrating how this alternative lens enhances analytical depth.

• Adjusted MFP trends over the last 20 years show a broad slowdown across the European Union, the United States and Japan. Growth fell from 2.4-3.7% in the European Union to 0.5-2.2% (with major divergences among countries) compared to 3-5.8% to 0.7-1.3% in the United States and from 2% to -0.8% in Japan.

Cyclically-adjusted MFP growth in the manufacturing industry

The manufacturing industry plays a crucial role in analysing Multifactor Productivity (MFP) due to its capital intensity and its central role in shaping economic progress. Yet, manufacturing productivity exhibits significant procyclicality – rising sharply during expansions and declining in recessions – due to high fixed costs, substantial economies of scale, and labour hoarding practices (Basu, Fernarld and Kimball, 2006[2]). The high degree of tradability and extensive input-output relationships in manufacturing render this part of the economy especially exposed to the spread of cyclical shocks across industries and countries (Engel and Wang, 2011[3]; Ariu, 2016[4]; Miyamoto and Nguyen, 2024[5]). This underscores the importance of distinguishing between manufacturing productivity movements driven by short-term volatility from those reflecting long-term structural change.

Between 1995 and 2022, average unadjusted MFP growth across selected OECD countries was 2.2%. To better capture the underlying evolution of productivity, two adjustment methodologies, instrumental variable (IV) regression and state-space modelling, were tested, each relying on two distinct proxies for input utilisation: capacity utilisation rates from business tendency surveys and average hours worked per worker. IV regressions show little sensitivity to the choice of proxy, with both yielding an average of adjusted MFP growth of 2.2% across selected OECD countries. State-space models produced marginally higher estimates (2.4%-2.7%) for the same countries and period. Given the traditionally large uncertainties around MFP estimates, the close alignment across methods lends confidence to the robustness of the adjusted MFP results.

Overall, these findings indicate that cyclical adjustment techniques consistently provide reliable estimates of underlying structural MFP trends, regardless of the specific modelling approach or input utilization proxy employed. The benefits of cyclically adjusting MFP are particularly evident when analysing periods of economic recession and expansion. For instance, at the peak of the 2008 financial crisis, average unadjusted MFP growth across selected OECD countries declined sharply to -2.5%. In contrast, cyclically-adjusted estimates remained notably higher: IV regression adjustments yielded estimates of 0.1% (capacity utilisation) and 1.2% (average hours worked), while state space models ranged between 1.3% and 2.6%. Conversely, during the economic expansion preceding the crisis (2002–2007), unadjusted MFP growth for selected OECD countries averaged as high as 4.1%, but cyclically adjusted estimates were notably lower: IV regression adjustments provided estimates of 2.7% (capacity utilisation) and 2.8% (average hours worked), while state-space models indicated slightly higher estimates, ranging between 3.6% and 3.7%.

Figure 8.1. Comparison of cyclically-adjusted MFP growth

Average across selected OECD countries, Percentage points

Note: This chart compares average Multifactor Productivity (MFP) growth estimates for 18 OECD countries (Australia, Austria, Germany, Estonia, Finland, France, Spain, the United Kingdom, Italy, Japan, Lithuania, Israel, Luxembourg, the Netherlands, Norway, Slovenia, the United States, and Slovak Republic) using different cyclical adjustment methods. “GFC” refers to the 2008 Global Financial Crisis, while “COVID-19” indicates the pandemic-related economic downturn in 2020. Adjustments are made using two modelling approaches: Instrumental Variable (IV) regression and state-space modelling. Both methods utilise proxies for input utilisation—either capacity utilisation (“Cap. Util.”) based on business tendency surveys, or average hours worked per worker (“Avg. Hours”).

Source: (Arend, Botev and Jaax, forthcoming[6]).

Statlink

During the more recent period of volatility associated with the COVID-19 pandemic, the two adjustment techniques diverged significantly. While average unadjusted MFP growth rates across selected OECD countries dropped to -1%, IV regression adjustments intensified the procyclical effects, producing significantly lower estimates ranging from -3.3% to -6.7%. Conversely, the state space model continued to filter out cyclical volatility effectively, yielding positive estimates between 1.3% and 1.9% (Figure 8.1). Visual analysis of country-specific adjusted series further highlights inherent methodological trade-offs (Figure 8.2). Although IV regression adjustment aligns closely with actual MFP growth series and thus offers more conservative estimates, it struggles during exceptional periods, such as the COVID-19 pandemic. This limitation arises from instruments selected primarily based on long-term cyclical correlations, which might fail to adapt adequately to extraordinary circumstances. Conversely, state space model adjustments, assuming a more pronounced cyclical component, provide greater adjustments during volatile periods, effectively isolating only very long-term productivity trends.

Figure 8.2. Comparison of cyclically adjusted MFP growth for France

MFP growth, Percentage points

Note: The blue lines represent the cyclically adjusted MFP series based on average hours worked per worker, while the dark blue lines represent the cyclically adjusted MFP series based on capacity utilisation rates. The dashed lines indicate the cyclical adjustment using instrumental variables, whereas the dot-dashed lines show the adjustment using a state-space model. The red line represents the unadjusted MFP series.

Source: (Arend, Botev and Jaax, forthcoming[6]).

Statlink

What does cyclical adjustment reveal about long-term trends in MFP growth?

Adjusted MFP data reveal diverse long-term productivity trends across major economies (Figure 8.3). In Europe, adjusted MFP growth slowed from 2.4-3.7% in 2005 to between 0.5% and 2.2% in 2019. This masks significant internal divergence: Germany’s adjusted MFP growth fell sharply from 1.7% in 2005 to -1.6% in 2019, whereas Italy maintained a consistently low level, barely shifting from 0.1% to 0% over the same period. Lithuania, by contrast, experienced sustained strong growth, only moderating slightly from 4.5-4.9% in 2005 to 3-4% in 2019.

The United States showed a clear downward trend, though adjusted MFP growth remained positive and above most Western European countries, declining from 3-5.8% in 2005 to 0.7-1.3% in 2019. Similarly, Japan experienced a decline in adjusted MFP growth, falling from around 2% in 2005 to -0.8% in 2019.

Figure 8.3. Adjusted MFP time series for selected OECD countries, 1995-2022

Note: The dark blue lines represent cyclically adjusted MFP growth based on capacity utilisation (CU) rates, while the lighter blue lines represent adjustments based on average hours worked per worker. The dashed lines indicate cyclical adjustments using instrumental variable (IV) regression, whereas the dot-dashed lines represent adjustments derived from a state-space model. The red line shows the unadjusted MFP series. The shorter IV regression series for Norway ends in 2020 due to data limitations in the instrumental variable for monetary policy shocks.

Source: (Arend, Botev and Jaax, forthcoming[6]).

Statlink

References

[1] André, C. and P. Gal (2024), “Reviving productivity growth: A review of policies”, OECD Economics Department Working Papers, No. 1822, OECD Publishing, Paris, https://doi.org/10.1787/61244acd-en.

[6] Arend, T., J. Botev and A. Jaax (forthcoming), “Cyclical adjustments of multifactor productivity”, OECD Statistics Working Papers.

[4] Ariu, A. (2016), “Crisis-proof services: Why trade in services did not suffer during the 2008–2009 collapse”, Journal of International Economics, Vol. 98, pp. 138-149, https://doi.org/10.1016/j.jinteco.2015.09.002.

[2] Basu, S., J. Fernarld and M. Kimball (2006), “Are Technology Improvements Contractionary?”, American Economic Review, Vol. 96/5, pp. 1418-1448, https://doi.org/10.1257/aer.96.5.1418.

[9] Comin, D. et al. (2025), “Revisiting Productivity Dynamics in Europe: A New Measure of Utilization-Adjusted TFP Growth”, Journal of the European Economic Association, https://doi.org/10.1093/jeea/jvaf003.

[3] Engel, C. and J. Wang (2011), “International trade in durable goods: Understanding volatility, cyclicality, and elasticities”, Journal of International Economics, Vol. 83/1, pp. 37-52, https://doi.org/10.1016/j.jinteco.2010.08.007.

[7] Inklaar, R. (2006), “Cyclical Productivity in Europe and the United States: Evaluating the Evidence on Returns to Scale and Input Utilization”, Economica, Vol. 74/296, pp. 822-841, https://doi.org/10.1111/j.1468-0335.2006.00554.x.

[8] Martin, J. and K. Jones (2022), “An occupation and asset driven approach to capital utilisation”, Paper presented at the 2022 CRIW Conference on Technology, Productivity and Economic Growth, https://conference.nber.org/conf_papers/f158768.pdf.

[5] Miyamoto, W. and T. Nguyen (2024), “Changing Global Input-Output Linkages and Demand Spillover”, IMF Economic Review, Vol. 72/3, pp. 1125-1151, https://doi.org/10.1057/s41308-024-00259-6.

[10] Planas, C., W. Roeger and A. Rossi (2013), “The information content of capacity utilization for detrending total factor productivity”, Journal of Economic Dynamics and Control, Vol. 37/3, pp. 577-590, https://doi.org/10.1016/j.jedc.2012.09.005.

Annex 8.A. Methodology

How to read the indicators

Cyclically-adjusted estimates of MFP growth require two design choices. First a proxy for the intensity of how inputs are used needs to be chosen. The literature has identified two main options: average hours worked per worker and self-reported capacity utilisation rates from business surveys. Second, the non-cyclical part of MFP growth needs to be extracted. This can be achieved either through instrumental variable regressions or through the use of state-space models and Kalman filters (Annex Figure 8.A.1). This chapter applied both approaches to gain insights into the evolution of cyclically-adjusted MFP growth in the manufacturing industry for 20 OECD countries for the time period 1995-2022.

Annex Figure 8.A.1. Main steps for the cyclical adjustment technique

Source: (Arend, Botev and Jaax, forthcoming[6]).

For the first step, while average hours worked per worker are frequently employed to adjust MFP, capacity utilisation rates from business surveys are seen to be more robust (Inklaar, 2006[7]; Martin and Jones, 2022[8]). Surveys provide a more direct measure of capacity utilisation than hours per worker as they are unaffected by changes in the composition of the workforce and changes in the relative cost of labour and capital.

For the second step, the IV regression approach is the most common in the literature (Comin et al., 2025[9]). The instruments chosen for the adjustment are country-specific oil price indices, economic policy uncertainty, monetary policy shocks and financial stress indicators. Overall, this approach performs well across the sample, with 15 out of 20 country-manufacturing panel IV regressions producing statistically significant results at the 90% confidence level.

In addition to the IV approach, a state-space framework to model the link between capacity utilisation surveys or average hours worked and the cyclical part of MFP growth was also employed. It allows estimating shifts and discrete breaks in MFP growth over time, which seems especially relevant in a period where MFP growth is slowing down and possibly affected by major economic shocks (Planas, Roeger and Rossi, 2013[10]). The model performed well for some countries for which the IV approach failed to provide statistically significant estimates (Annex Table 8.A.1).

Annex Table 8.A.1. Best cyclical-adjustment approach by country

Proxy / Modelling technique	Instrumental variable regression	State-space model
Capacity utilisation rates	AUT, CAN, DEU, EST, FIN, FRA, HRV, ITA, JPN, LTU, NLD, SVK, SVN,	AUS,
Average hours worked per worker	ESP, NOR	USA, GBR, LUX, ISR,

Note: The best approach here is determined by a combination of factors including statistical significance, strength of the instruments, residual correlation with GDP growth and visual analysis.

Source: (Arend, Botev and Jaax, forthcoming[6]).

Performance across models is robust

Both the instrumental variables approach and the state-space models each perform relatively well, in that they eliminate most of the correlation of MFP with GDP growth. However, they are highly dependent on specific assumptions and modelling choices.

Instrumental variable approaches primarily suffer from the challenge of identifying valid instruments. A good instrument must be strongly correlated with the cyclical component but uncorrelated with the underlying structural shocks or productivity innovations. In practice, finding such instruments can be difficult to find, and weak instruments can result in biased and inconsistent estimates as the results for the recent COVID-19 period demonstrate.

State-space models, while more flexible, rely on strong parametric assumptions, which may not align with the complex realities of productivity cycles. Misspecification of these assumptions or of the underlying state and measurement equations can seriously bias the estimated trend and cycle components.