Finland: Selected Issues
Author: International Monetary Fund. European Dept.
Publication Date:
28 May 2014
eISBN:
9781498331333
Language:
English
Keywords:
ISCR; CR; firm; fund staff calculation; autoregressive output gap term; rate of growth; multivariate Augmented Phillips Curve; APC approach; long- term capacity; Output gap; Capacity utilization; Total factor productivity; Production growth
Download PDF (1.1 MB)

In recent years, the IMF has released a growing number of reports and other documents covering economic and financial developments and trends in member countries. Each report, prepared by a staff team after discussions with government officials, is published at the option of the member country.

Abstract

In recent years, the IMF has released a growing number of reports and other documents covering economic and financial developments and trends in member countries. Each report, prepared by a staff team after discussions with government officials, is published at the option of the member country.

Estimating Finland’s Potential GDP1

Finland, an economy known for its dynamic performance after the crisis of the early 1990s, is struggling to recover from the Great Recession, indicating that deeper, structural issues may be holding back growth. Has Finland’s growth potential changed? Based on a selection of approaches to estimate potential output, this note argues that (1) potential output growth has slowed or is contracting, indicating the need for structural reforms to boost TFP since the economy has failed to return to trend growth as it had in the 1990s recovery, (2) the role of Nokia in the Finnish economy has implications for steady-state growth and the need for an examination of R&D policies, and (3) that estimation of potential for Finland is particularly difficult with a high degree of variability around the estimates suggesting that policies relying on these estimates should proceed with appropriate discretion.

A. Has the Longer-Term Growth Trend Changed?

1. Finland has recovered from large negative shocks in the past. GDP levels dropped markedly after the crisis in the early 1990s. However, on the back of strong structural reforms and an ICT sector dynamically expanding around Nokia, the Finnish economy quickly resumed its pre-crisis pace. In contrast, the drop in GDP after the financial crisis and recession of 2007–2009, has turned out to be more persistent.

A01ufig01

Real GDP Growth Trends

(Bil. EUR)

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: ETLA, Haver Analytics, IMF World Economic Outlook, Statistics Finland, and Fund staff calculations.

2. This raises the question whether growth will resume its pre-crisis trend. One way to shed light on the issue is to look at the underlying or potential growth rate of the Finnish economy. In what follows, we approach this task in different ways, including (i) by assuming that potential growth can be extracted directly from GDP (univariate filtering), (ii) using a production function approach linking potential GDP to capital and labor input, among other things, (iii) and applying so-called multivariate filtering approaches that use information from variables correlated with the non-transitional component of GDP to identify its permanent or “potential” part. All of these approaches come with advantages and disadvantages, and there is considerable uncertainty surrounding their results.

B. Methodology

Hodrick-Prescott Filter (Univariate)

3. Principle. The Hodrick-Prescott (HP) filter remains one of the most widely used methods for decomposing time-series data into a trend component and a cycle component. Despite well-known weaknesses (e.g., endpoint sensitivity), the HP filter remains popular due to its simplicity as well as ease of use. Economically, it can be interpreted as a the attempt to capture the underlying factors driving potential output—such as changes in factor inputs and their utilization, including hysteresis effects on the capital stock and structural unemployment—by looking simply at the GDP outcome itself. Following Hodrick and Prescott (1997):

yt=gt+ct,fort=1…T.(1)
min{gt}t=-1T{∑t=1Tct2+λ∑t=1T[(gt−gt−1)−(gt−1−gt−2)]2}(2)

where yt is log growth decomposed into a trend component gt and a cyclical component ct in equation (1). Equation (2) describes how to obtain gt where the first term is the sum of squared deviations from trend growth and the second term penalizes variability in the trend growth. λ refers to the degree of penalty incurred by the variability term and is restricted to be greater than zero. The choice of λ is critical and remains under some debate.

4. Application. In what follows, the HP is estimated using annual data, using two levels for λ that yield estimates of potential output of different degrees of flexibility: λ=100, as suggested by Hodrick and Prescott (1997) and λ=6.25 as in Ravn and Uhlig (2002). The single input, GDP in constant prices, is projected to 2030 using a constant rate of growth after 2019 to alleviate the endpoint bias problem.

Production Function Approach (PFA)

5. Principle. The production function approach derives potential output from a simple Cobb- Douglass production function with exogenously determined trend components. The model assumes constant labor and capital shares. Specifically,

Y∗=θK∗αL∗1−α(3)

where Y* is potential output determined by θ (total factor productivity), the smoothed real capital stock (K*), smoothed volume of labor (L*), and factor intensity α.

6. Application. The PFA is estimated on annual data. The inputs are smoothed time series of employment, computed as the share of the labor force that is employed assuming that the rate of unemployment is at a level that will keep wage inflation constant (estimated separately), net capital stock, and total factor productivity (TFP), with factor intensity calibrated to the Finnish economy. As for the HP filter, factor inputs are projected to 2030 to alleviate the endpoint bias. The PFA, although stepping beyond the simple univariate approach, is not without problems. Importantly, the rate of growth of total factor productivity needs to be estimated. Following the standard approach in this case, we estimate the underlying rate of productivity growth by applying a HP filter to the Solow- residual using (unfiltered) factor inputs and production function (3).2 In line with the discussion above, we use two sets of calibrations for λ, 100 and 6.25, resulting in measures of productivity growth of different persistence.

Multivariate Filter (MV) (Kalman State Space)

7. Principle. The state space model, an augmented version of Borio and others (2012), expands the HP filter by adding additional covariates that help identify the transitory part of GDP, albeit without the theoretical constraints featured in the Benes and others (2010) approach.3 The system of equations, used in estimating potential GDP, is as follows.

yt−yt∗=ρ(yt−1−yt−1∗)+xtβ+εt0,εt0ub˜white noiseσ0(7)
Δ2yt∗=εt∗,εt∗ud˜white nioseσ∗(8)
withλα=σ02σ∗2(9)

Building on the standard HP filter in a state-space framework, the first equation includes an autoregressive output gap term and a vector of observables xt which contains information on transitory variables. To match the frequency cutoff of the filter in (7) with that of the HP filter, the variance ratios for the state-space must be equal with those of the HP filter. Equalizing the variances is achieved by adjusting parameter λa in equation (9).

8. Application. Rather than the Bayseian approach employed by Borio and others (2013), we use maximum likelihood estimation (MLE) to estimate the model. ρ and β are estimated in a two- step procedure. First, the autoregressive parameter ρ is estimated by running an AR(1) regression on the output gap obtained from the simple HP filter. Then ρ is substituted into (7) and estimated using MLE. Following Borio and others (2013), the choice of variables in xt roughly describes the asset market, the credit cycle, and in our case, capacity utilization, as in Benes and others (2010). In particular, this model uses as inputs GDP in constant prices, the real Nokia stock price, to reflect the potential—and potentially transitory—role of Nokia for the real economy and the Finnish asset portfolio, the real short-term interest rate, real bank external assets, and capacity utilization. All time series are demeaned to reduce procyclicality and differenced to account for unit roots. Specifically, the measurement equation becomes:

yt−yt∗=β(yt−1−yt−1∗)+γ1Δrt+γ2Δbat+γ3nst+γ4Δcaput+ε4,t(10)

where y – y* refers to the output gap, r the real interest rate, ba banks’ external assets, ns Nokia stock price, capu capacity utilization, and ε is a disturbance term.

Smoothing Parameter Choice and Implication

The choice of smoothing parameters in filtering models is a widely debated topic (Maravall and del Río, 2001). Hodrick and Prescott (1997) suggest a lambda of 100 for annual data and 1600 for quarterly. Ravn and Uhlig (2002) present two approaches: the first, a time domain approach that determines lambda using the ratio of the variance of the cyclical components to the variance of the second difference of the trend component, thereby accounting for idiosyncrasies in the data; the second, a frequency domain approach building on King and Rebelo (1993) that yields the widely used lambda of 6.25 for annual data.

At the heart of the debate is the length of the cycle. Ravn and Uhlig’s (2002) lambda of 6.25 for annual and 1600 for quarterly data (consistent with Hodrick and Prescott, 1997) implies a cycle length of around 9.8 years and yields estimates that reflect the lower level of smoothing. The standard Hodrick and Prescott (1997) lambda of 100 for annual data implies a much longer cycle of approximately 19.8 years whose estimates are reflective of the much higher level of smoothing. The corresponding lambda for quarterly data is 25,199.

The estimates produced by the three models are remarkably sensitive to the choice of the smoothing parameter. In this paper, we have used both the lambda values of 6.25 as well as 100 for annual data and correspondingly 1600 and 25,199 for quarterly data. The table below shows the extent to which the choice of smoothing parameter affects the level of potential output and the output gap and by implication, the estimates of structural balance and fiscal impulse.

Effect of Smoothing on Potential Estimates

article image
Sources: Fund staff calculations.Note: The difference is calculated as the absolute value of the high smoothing estimate less the low smoothing estimate.

2013 data is 2012 for the multivariate filter.

C. Results

9. All models estimate the current growth rate of Finnish potential output as low. In particular, the HP filter and production function approach estimates potential growth in 2013 at around 0.4 and 0.5 percent, respectively, while the multivariate approach finds potential output growth on a declining path at –0.4 percent (see text figure and Figures 1–3). However, all approaches see average potential output growth at around 0.3-0.4 percent over the 2009–13 period. The Augmented Phillips curve approach by Benes and others (2010) provides similar point estimates (Box 2).

Figure 1.
Figure 1.

Hodrik-Prescott (HP) Filter Approach

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: IMF World Economic Outlook and Fund staff calculations.
Figure 2.
Figure 2.

Production Function Approach (PFA)

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: Eurostat, Haver Analytics, IMF World Economic Outlook, OECD, Statistics Finland, and Fund staff calculations.
Figure 3.
Figure 3.

Multivariate Filter Approach

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: Eurostat, Haver Analytics, and Fund staff calculations.
A01ufig02

Growth of Potential Using Various Methods

(Percentage points)

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: Haver Analytics, IMF World Economic Outlook and Fund staff calculations.Note: The chart shows annual potential growth based on different models and for a range of assumptions about the smoothness of potential output.

10. The assumed smoothness of potential output is a key determinant of these results. Depending on the assumed smoothness, the estimated growth rate of potential will vary widely—especially when looking at longer periods of time. In particular:

  • The Hodrick-Prescott filter (Figure 1) smoothes the path of GDP growth but fails to account for the structural breaks. The output gap is currently negative and is expected to close under both scenarios in the medium term (2017-18), however, potential growth under the low smoothing assumption (6.25) changes from 0.2 to 1.9 whereas under the high smoothing assumption (100) it increases from 0.6 to just 0.9. This suggests that much weaker GDP growth is required under the high smoothing assumption.

  • The production function approach shows very similar characteristics (Figure 2). Under the low smoothing parameter (6.25), potential growth falls into negative territory from 2009–2013 while under the high smoothing parameter (100) it remains solidly positive despite slowing over the same period. Likewise, the output gap under the low smoothing parameter closes in 2014, while the high smoothing parameter series does not close until 2016.

  • The Multivariate approach suggests that Finland’s economy was growing at above potential for much of the period following the 1990s recession and before the Great Recession (Figure 3). One interpretation is that the model identifies much of the growth generated by the booming ICT sector as transitory, resulting in a lower estimate of potential growth. As a consequence, the model estimates a very low growth rate also for 2013. However, the model is quite sensitive to the level of smoothing and in the period 2007–2009 varies by almost 2.5 percentage points.

Augmented Phillips Curve Approach

The Augmented Phillips Curve (APC) approach follows Benes and others (2010). It comprises the output gap, the employment gap, and the manufacturing capacity utilization gap, each one with an identifying equation: equation (1) is founded in the new classical augmented Phillips Curve model and relates inflation to the output gap; equation (2) uses a dynamic Okun’s law as a basis for establishing the relationship between output and employment; and equation (3) employs the general framework of Okun’s law to describe the relationship between output and capacity utilization.

πt=πt−1+βyt+Ω(yt−yt−1)+εtπ(1)

where π refers to current core inflation, y represents the output gap, and ε is a disturbance term;

ut=ϕ1ut−1+ϕ2yt+εtu(2)

where u is the current unemployment rate, y is the output gap, and ε, a disturbance term;

ct=γ1ct−1+γ2yt+εtc(3)

where c is the current manufacturing capacity utilization rate, y is the output gap, and ε, a disturbance term.1

Application. The APC approach is estimated using quarterly data through the third quarter of 2013. The inputs to the model are: GDP, CPI, core CPI in level and growth, unemployment rate, capacity utilization, and long term inflation expectations.2 The model is then estimated using Bayesian Regularized Maximum Likelihood with priors to ensure reasonable estimates. Forecasts are then generated to 2019Q4. The model uses the priors reported in Benes and others (2010), but Finland specific steady state priors and labor share assumption. Based on data 1990–2013, the steady state growth rate of output is estimated at about 1.7 percent.

Results. The point estimates from the APC approach are broadly comparable with the HP-filter and the Production function approaches, suggesting that potential GDP grew by about 0.4 percent over the 2009–13 period and 2013 (Figure 4).3

1 See Benes and others (2010) for the estimating equations.2 For inflation expectations we use the 10 year expectations for Germany from Consensus Forecasts.3 However, the APC approach does not lend itself easily to varying the signal-to-noise ratio as the other models discussed in this paper.
Figure 4.
Figure 4.

Multivariate Augmented Phillips Curve (APC) Approach

Citation: IMF Staff Country Reports 2014, 140; 10.5089/9781498331333.002.A001

Sources: Bank of Finland, Consensus Forecasts, Eurostat, Haver Analytics, Statistics Finland, and Fund staff calculations.

D. Conclusion

11. Estimates of potential output for Finland are an important part of the toolkit for policymakers—but they come with a degree of uncertainty. As this paper illustrates, the use of different methodologies and assumptions can lead to different results. Under the HP, PFA, and multivariate approach, the choice of smoothing can just as reasonably produce a negative growth rate as well as a positive one. Likewise, output gap estimates, critical to fiscal policy, should also be carefully considered.

12. However, there are indications that Finnish potential output growth is low at this juncture. From 1997–2007, potential growth, independent of the choice of smoothing, averages 3.2 percent per year. In 2013, that average has dropped to 0.2 with several of the models producing negative growth. This result indicates that the lack of a recovery in Finland is largely structural in nature. Therefore, any indication that the output gap is closing is due to falling potential rather than a pickup in growth.

13. This points to the advantages of structural reforms aiming to enhance Finland’s long- term capacity. In particular, TFP enhancing measures could be crucial in helping the economy recover despite the time it takes to implement them. This would require steps to adjust policies, especially policies to encourage R&D, to the post-Nokia era as well as additional effort to achieve innovation and growth in smaller firms outside the existing ICT cluster.

References

  • Ali-Yrrkö, Jyrki (ed.), 2010, “Nokia and Finland in a Sea of Change,” Taloustieto Oy, Helsinki, Finland.

  • Benes, Jaromir, Kevin Clinton, Robert Garcia-Saltos, Marianne Johnson, Douglas Laxton, Peta Manchev, and Troy Matheson, 2010, “Estimating Potential Output with a Multivariate Filter,” IMF Working Paper No. 10/285 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Billmeier, Andreas, 2004, “Measuring a Roller Coaster: Evidence on the Finnish Output Gap,” IMF Working Paper No. 04/57 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Borio, Claudio, Piti Disyatat, and Mikael Juselius, 2013, “Rethinking Potential Output: Embedding Information about the Financial Cycle,” Bank of International Settlements Working Papers No. 404.

    • Search Google Scholar
    • Export Citation

  • Christiansen, Lone E., 2013, “Potential Output and Unemployment,” Sweden, 2013 Article IV Consultation, Selected Issues, IMF Country Report No. 13/277, Chapter I.

    • Search Google Scholar
    • Export Citation

  • Cotis J.-P., J. Elmeskov and A. Mourgaine, 2005, “Estimates of potential output: benefits and pitfalls from a policy perspective,” in L. Reichlin (ed.), Euro area business cycle: stylised facts and measurement issues, CEPR.

    • Search Google Scholar
    • Export Citation

  • Harvey, A. C., and A. Jaeger, 1993, “Detrending, Stylized Facts and the Business Cycle,” Journal of Applied Econometrics, Vol. 8(3), pp. 231–247.

    • Search Google Scholar
    • Export Citation

  • Hodrick, Robert and Edward Prescott, 1997, “Post-war US business cycles: an empirical investigation,” Journal of Money, Credit, and Banking, Vol. 29, pp. 1–16.

    • Search Google Scholar
    • Export Citation

  • International Monetary Fund, 2012, “Finland. Staff Report for the 2012 Article IV Consultation,” IMF Country Report No. 12/253, August.

    • Search Google Scholar
    • Export Citation

  • King, Robert G., and Sergio T. Rebelo, “Low Frequency Filtering and Real Business Cycles,” Journal of Economic Dynamics and Control 17:1/2 (1993), 207–231.

    • Search Google Scholar
    • Export Citation

  • Kydland F. and C. Prescott, “Business cycles: real facts and a monetary myth”, Federal Reserve Bank of Minneapolis, Quarterly Review (Spring), pp 3–18, 1990.

    • Search Google Scholar
    • Export Citation

  • Maravall, A., and A. del Río, 2001, “Time aggregation and the Hodrick-Prescott filter,” Bank of Spain Working Paper 01/08.

  • Mrkaic, M. (2014). Note on Potential Output Estimation. Unpublished.

  • OECD, 2012, “OECD Economic Surveys: Finland 2012,” OECD Publishing (Paris: Organization for Economic Cooperation and Development) http://dx.doi.org/10.1787/eco_surveys-fin-2012-en

    • Search Google Scholar
    • Export Citation

  • Ravn, Morten O., and Harald Uhlig, 2002, “On Adjusting the Hodrick-Prescott Filter for the Frequency of Observations,” The Review of Economics and Statistics, Vol. 84(2), pp. 371–376

    • Search Google Scholar
    • Export Citation
1

Prepared by Thomas Dowling.

2

NAIRU is obtained from the APC approach described below. Employment at NAIRU is calculated, then the entire series is smoothed using the HP filter. This approach is taken due to the high level of volatility in the Finnish employment data.

3

See Mrkaic (2014). See Box 2 for an application of the model to Finland.

Same Series

Other IMF Content

Other Publishers

Cover IMF Staff Country Reports
Finland: Selected Issues
Author: International Monetary Fund. European Dept.
Volume/Issue:
Volume 2014: Issue 140
Publisher:
International Monetary Fund
ISBN:
9781498331333
ISSN:
1934-7685
Pages:
60
DOI:
https://doi.org/10.5089/9781498331333.002

Real GDP Growth Trends

(Bil. EUR)

Hodrik-Prescott (HP) Filter Approach

Production Function Approach (PFA)

Multivariate Filter Approach

Growth of Potential Using Various Methods

(Percentage points)

Multivariate Augmented Phillips Curve (APC) Approach