Abstract
In light of potential changes in fiscal policy in Finland, the impact of changes in the fiscal stance on short-term economic growth is an issue of interest. This paper estimates the short-term impact of fiscal policy on growth in Finland. The estimates are based on a structural vector autoregression (SVAR) model. However, identifying the long-term impact of changes in labor taxes on employment and growth is not an easy task. Although fiscal policy influences GDP in the short-term, the impact remains small compared with results.
I. The Impact of Fiscal Policy in Finland1
A. Introduction
1. In light of potential changes in fiscal policy in Finland, the impact of changes in the fiscal stance on short-term economic growth is an issue of interest. The new government’s program includes significant expenditure increases for 2003 and 2004, at the same time that tax cuts are being considered. These actions would lower the general government surplus and possibly result in considerable fiscal stimulus. Drawing on analytical work conducted at the Ministry of Finance and the Bank of Finland, IMF staff, in the context of the 2003 Article IV consultation, consider a general government surplus of 4 percent of GDP as a medium-term norm which could be achieved by targeting an increase in the structural primary balance of ½ percent of GDP per year (while allowing the automatic stabilizers to play around the consolidation path). All this underscores why discussions of fiscal policy and its effect on the economy have come to the fore.
2. This paper estimates the short-term impact of fiscal policy on growth in Finland. The estimates are based on a structural vector auto regression (SVAR) model, following the methodology introduced by Blanchard and Perotti (2002) (hereafter called BP) in a study on the United States, and also applied by Cespedes and Hoffmaister (2003) (hereafter called CH) to Spain. Against the background of Finland’s participation in the euro area and thus the absence of an independent monetary policy, and because Finland’s economy is among the least synchronized with the core euro-area countries (Figure 1), this question takes on an added dimension.
Finland and Selected Countries: Size and Correlation of GDP Growth with the Rest of Euro Area, 1992-2003 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: WEO; IFS; and Fund staff calculations.1/ Based on real quarterly GDP data, seasonally adjusted. Ireland: 1995-2003.2/ Correlation with euro area excluding the country in question.Finland and Selected Countries: Size and Correlation of GDP Growth with the Rest of Euro Area, 1992-2003 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: WEO; IFS; and Fund staff calculations.1/ Based on real quarterly GDP data, seasonally adjusted. Ireland: 1995-2003.2/ Correlation with euro area excluding the country in question.Finland and Selected Countries: Size and Correlation of GDP Growth with the Rest of Euro Area, 1992-2003 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: WEO; IFS; and Fund staff calculations.1/ Based on real quarterly GDP data, seasonally adjusted. Ireland: 1995-2003.2/ Correlation with euro area excluding the country in question.3. While the application of the structural VAR approach to Finland is somewhat hampered by data limitations, the results suggest that fiscal policy has only a modest impact on activity. Thus, although, for example, expansionary fiscal surprises do indeed increase real GDP in the short run, the impact remains small compared to the results reported in BP and CH for Spain.
B. Empirical Evidence on the Impact of Fiscal Policy on Growth in OECD Countries
Macroeconomic models
4. Traditionally, estimates of the short-run impact of fiscal policy on growth predominantly came from simulations using structural macro models. Such studies, in which Keynesian effects play a prominent role, suggest fairly large short-term expenditure multipliers for European countries, ranging from 0.6 to 1.5 in “dollar-for-dollar” terms, and somewhat smaller revenue multipliers (see Brunila, Buti, and in‘t Veld (2002), Hunt and Laxton (2003) and, for a survey, Hemming, Rell, and Mahfouz (hereafter called HKM) (2002).2 The strength of studies based on structural macroeconomic models is also one of their major weaknesses: while structural simulations shed light on the channels via which effects may take place (or be offset), they often model these relationships based on economic theory that may not be firmly rooted in empirical estimates.
5. The study by Brunila, Buti, and in‘t Veld (2002) is of interest to the Finnish case. Using the European Commission’s Quest model, these authors compare the impact of changes in different kind of expenditure and revenue items. They measure short-term expenditure multipliers from a temporary shock in which government expenditures are increased by 1 percent of GDP (government purchases of goods and services, government investment, transfers to households, and government employment). Short-term revenue multipliers are produced by reducing labor taxes, corporate profit taxes, and value-added taxes by one percent of GDP.
6. The impact of higher government expenditures on GDP is modest because of the crowding out of private spending via higher real interest rates and leakage via imports. The extent of crowding out depends on the response of monetary policy, and leakage via imports depends, inter alia, on the openness of the economy. In the Quest model, the majority of households is assumed to be permanent income consumers, whose consumption responds only to a small extent to temporary changes in transfers or taxes. This is why changes in transfers to households and taxes have smaller effects than changes in government consumption and investment. In the case of changes in government purchases of goods and services or investment, the short-term multiplier in Finland is estimated at around 0.65. The multiplier for changes in the government wage bill is close to 1, but the multiplier for changes in transfer payments is only 0.2.
7. As was the case for temporary changes in transfers, the simulations suggest that the impact of temporary changes in labor and corporate income taxes on output is small—with multipliers of around 0.2—because the intertemporal optimizing behavior of economic agents smoothes away most of the impact.3 In the case of a permanent tax cut, the impact of a tax cut would strengthen over the medium term as the distortionary effects of taxation are reduced. For instance, a reduction in labor taxes has a direct demand effect through its impact on disposable income and a positive supply effect. The latter also allows for lower wage costs and improved competitiveness, further boosting (foreign) demand. The impact of temporary changes in indirect taxes is significantly higher than in the case of direct taxes—estimated at one-half for Finland—as private agents are assumed to bring expenditures forward in anticipation of a return to higher tax rates in subsequent years.
8. In all, in the case of temporary changes to fiscal policy, the impact of expenditure changes is found to be larger than that of revenue changes. In contrast, in the case of more permanent fiscal policy changes, the impact of expenditure changes would fade out, while the supply side effects of tax changes become more important.
Case studies of “expansionary” fiscal contractions
9. In circumstances of high government debt, the credibility effects of a fiscal contraction can offset (in part or fully) the traditionally assumed Keynesian effects. Indeed, the experience of certain European countries that undertook fiscal consolidation in the 1980s and 1990s in circumstances of high government debts and deficits generated a literature on “expansionary fiscal contractions.” Discussing some of the available evidence, HKM conclude that there indeed appear to have been episodes of expansionary fiscal contraction, and that some episodes share certain characteristics.
10. HKM stress, however, that caution is needed in drawing general conclusions from these experiences and point to methodological flaws such as selection bias problems (country experiences are “handpicked”), simultaneity bias (strong growth led to lower fiscal deficits), and omitted variables (not taking into account sharp devaluations that accompanied fiscal contractions and influenced growth). Indeed, although credibility effects seem plausible, it is unlikely that they have been large enough to offset the Keynesian effects of fiscal consolidation. Credibility effects can largely be captured by the reduction in long-term interest rates, the effect of which on activity can be measured separately and which is typically found to be relatively modest, compared to the Keynesian effects of fiscal policy.
Studies based on structural VAR models
11. Empirical estimates of fiscal multipliers using structural VAR models—which have a strong empirical element—typically find significant, positive multipliers. Data constraints have tended to limit their application to large industrial countries. While studies focusing on the U.S. typically find significant multipliers reinforcing Keynesian priors, recent papers on large European countries have mixed conclusions. Aarle, Garretsen, and Gobbin (2001) found considerable variation in the size and signs of multipliers for EU countries. In a study that includes Germany (as well as four Anglo-Saxon countries), Perotti (2002) finds that the effects of fiscal policy on GDP have become weaker over time; they are substantially smaller in the post-1980 sample than the pre-1980 sample. He finds that in the post-1980 sample only in Germany is the effect of government spending on GDP significantly positive on impact. But even there, the effect turns negative by the fourth quarter. His findings on the impact of taxation on GDP are mixed, but Keynesian effects seem to be present for Germany, with multipliers in the range of 0.2 to I in the first three years. In the above-mentioned study for Spain, CH find significant multipliers for both revenues and expenditures. Notwithstanding the diversity of results, in most cases the short-term fiscal multipliers estimated by structural VAR models tend to be lower than those found by simulations of macro models, but significantly higher than suggested by the literature on “expansionary fiscal contractions.”
Other methods
12. In a preliminary empirical study of the impact of fiscal policy and other factors on growth in euro area countries, IMF staff (forthcoming), using a panel of annual time series data (1980–2001), find that GDP growth has been determined mainly by changes in partner countries' import growth and the fiscal stance. The fiscal multipliers, estimated at around 0.4 to 0.5, are not as large as typically assumed in structural macro models, but larger than found in some other recent studies, including those on “expansionary fiscal contractions.” Moreover, the study suggests that fiscal multipliers have not become smaller in the 1990s. However, tentative estimations on a country-by-country basis suggest that, overall, fiscal multipliers are weaker in smaller and more open countries, as would be expected on theoretical grounds. While for Finland no statistically significant results were found in the individual country estimation, the cross-country results suggested a multiplier of around 0.4 to 0.5 for an economy with a share of total trade to GDP equal to Finland's.
C. Evidence from a Simple VAR Exercise for Finland
13. The empirical evidence on the impact of fiscal policy on output in Finland, using the structural VAR model proposed by BP and applied by CH, is based on quarterly data from 1991 onwards. The data limitations are across two dimensions. First, quarterly information on general government expenditures and revenue excludes several items of the overall fiscal flows.4 Thus, the available data were combined with quarterly national accounts data on government consumption and investment, and mapped into total revenues and expenditures data using the composition of annual total data. Second, the resulting time series are short, covering only the period 1991 to 2001. Standard unit root tests showed that transforming the fiscal data and GDP (all in real terms) into logs renders the data trend-stationary, with no clear indication of a cointegration relationship.
The model
14. Consider a trivariate system in which rt is real government revenues, gt is real government expenditures, and y, is real GDP at time t (all in logs). With the vector Xt defined as (rt, gt, yt)' the VAR can be written as
where A(L) is a lag polynominal (3 × 3) matrix containing the reduced-form dynamic effects of the system and et is a vector of serially independent reduced-form shocks to revenues, expenditures, and GDP, with E[et] = 0 and E[et et] = Ω, where Ω is the variance/covariance matrix of the reduced-form VAR surprises.5 In a structural VAR, et is interpreted as a linear combination of independently distributed structural shocks to revenues, expenditures, and GDP, that is
where B is a (3 × 3) matrix and ut is the vector of structural shocks in revenues, expenditures, and GDP.
Blanchard-Perotti identification steps
15. The identification process consists of obtaining the matrix B (see BP and CH for details). Once B is obtained, the information contained in the estimated reduced-form unexpected movements et can be used to uncover the structural shocks ut.
Imposing the assumptions of the BP approach, equation (2) can be reduced to
where ur, ug, and uy are the mutually uncorrelated structural shocks to revenue, expenditure, and GDP that are to be recovered.
16. The first line of the equations in (3) states that unexpected movements in taxes can be due to the response to unexpected movements in GDP, captured by ey, and the response to structural (or, discretionary) shocks to expenditure and revenues (respectively, ug and u1). A similar interpretation applies to unexpected movements in spending in the second line. The third line states that unexpected movements in output can be due to unexpected movements in taxes, unexpected movements in spending, or to other, structural, shocks in GDP (uy). Thus, (3) leaves us with six unknown coefficients and three equations. To be able to estimate the impact of structural, discretionary fiscal shocks on GDP, at least three of the unknown coefficients have to be pre-identified.
17. The response of taxes and expenditures to changes in GDP stem from the so-called automatic stabilizers and a possible discretionary adjustment of fiscal policy to cyclical conditions as measured by the reduced-form unexpected movements in GDP. Coefficients ai and b] capture both effects for revenues and expenditures, respectively. As BP argue convincingly, the use of quarterly data “virtually eliminates the second channel.” Thus, they suggest setting these coefficients equal to estimates of the revenue and expenditure elasticities. Based on recent OECD (2003) estimates for Finland, elasticities of 1 for revenues and around -0.4 for expenditure would appear to be reasonable starting points.6
18. The coefficients a2 and b2 in (3) describe possible contemporaneous impacts of structural shocks in expenditure and revenues on unexpected movements (i.e., the VAR residuals) in revenues and expenditure. How large will these effects be? The answer depends on the institutions of policy making. If tax decisions are made before expenditure decisions, a2 is zero. As suggested by BP, b2 could then be estimated freely. Alternatively, b2 could be assumed to be zero, and a2 estimated freely. As a rule, with the budget process based on the entire fiscal year (rather than quarters), both effects will be rather small, however. Thus, to economize on degrees of freedom, a2 and b2 are both set at 0.7
19. Under these assumptions, estimating the impact of discretionary changes in (or structural shocks to) fiscal policy on GDP becomes a straightforward exercise. With ai and bi known, the” cyclically adjusted” unexpected movements in revenues and expenditures can be computed as:
and
which are then used to obtain estimates of c1 and c2 in (3):
Given our assumptions, c1 and c2 capture the impact of structural or discretionary fiscal policy changes on GDP.
Results
20. To empirically identify the contemporaneous and dynamic implications of discretionary fiscal policy changes on GDP, a standard VAR is estimated with revenues, expenditure, and GDP as endogenous variables (all in real terms and logs) using four lags. In addition, the model includes seasonal dummies and a linear and non-linear trend as exogenous variables.8 The upper two panels in Figure 2 summarize the results of that exercise.
Finland: VAR Model, 1991-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.Finland: VAR Model, 1991-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.Finland: VAR Model, 1991-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.21. Using the residuals from the VAR and the OECD's (2003) revenue and expenditure elasticities (i.e., a1 = 1 and b1 = -0.4), c1 and C2 can be estimated, and the structural shocks uncovered (see lower panel in Figure 2). Table 1 summarizes the parameter values. On a dollar-for-dollar basis, the point estimate of the contemporaneous impact of a discretionary change in revenues on GDP (cj) is about -0.1; the impact of a discretionary change in government spending (c2) is about O.2.9 The direction of these effects is in line with our priors: an increase in taxes lowers GDP while higher expenditure has a positive impact on real activity, with the latter dominating the former in quantitative terms. However, the quantitative effect is notably smaller than what CH report for Spanish and BP for the U.S. data. CH report a contemporaneous effect of revenues of about -2.8 and of expenditures of about 1.4, and BP find effects of about -0.9 and 1.0, respectively.10 Moreover, in the Finnish case neither coefficient is statistically significant at conventional levels.
Finland: Summary Coefficients
The elasticity is around two times as high.
The elasticity is around two times as low.
22. The results for c1 and C2 are fairly robust with regard to different assumptions for the (exogenous) elasticities of revenues and expenditures to unexpected movements in GDP.
Assuming higher elasticities of 1.1 for revenues and -0.5 for expenditures leaves the contemporaneous impact of revenues on GDP unchanged, while the impact of expenditures on GDP increases somewhat to 0.3 on a dollar-for-dollar basis. However, neither estimate is significant even at the 10 percent level. Choosing lower elasticities, 0.9 for revenues and -0.3 for expenditures, yields an expenditure impact on GDP of just 0.1 and unchanged estimates for the impact of revenues—both insignificant.
23. Having identified the contemporaneous effects, the VAR model can be employed to take a closer look at the dynamic effects of discretionary fiscal policy on the economy. Figure 3 shows the intertemporal responses of GDP to structural shocks to revenues and expenditures. The impulse responses are expressed in dollar-for-dollar terms. The results incorporate our baseline assumptions on revenue and expenditure elasticities, but comparable figures are obtained using the alternative assumptions on elasticities discussed above.
Finland: Impulse Responses of GDP to Discretionary Fiscal Policy Shocks 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.1/ Figures show the reaction of quarterly real GDP with a 2-standard-error band. All effects are expressed as dollar-for-dollar.Finland: Impulse Responses of GDP to Discretionary Fiscal Policy Shocks 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.1/ Figures show the reaction of quarterly real GDP with a 2-standard-error band. All effects are expressed as dollar-for-dollar.Finland: Impulse Responses of GDP to Discretionary Fiscal Policy Shocks 1/
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Source: Fund staff calculations.1/ Figures show the reaction of quarterly real GDP with a 2-standard-error band. All effects are expressed as dollar-for-dollar.24. The impact of a shock to revenues on output is very modest. It peaks at about -0.25, after five quarters. This is notably lower than the results found by BP for the United States (with the strongest effect at about -0.75 after around six quarters) and it also appears to be lower than the results found by CH for Spain.11 Moreover, the impulse response function is not statistically different from zero even at its maximum.
25. The dynamic impact of an expenditure shock on GDP, while statistically significant only in the second period, is larger than that of a discretionary change in revenues. It peaks at about 0.5 in the second quarter and reaches a level close to this in the fourth. With a significant share of government expenditure directly impacting GDP through public consumption and investment, this is perhaps not surprising. But the effect remains small compared to the results reported by BP for the United States, where the impact measures close to 1 in the first quarter and rises to even higher levels around quarter fifteen. A comparison with Spain based on CH would seem to support a similar conclusion.
26. While these VAR estimates for Finland are in some ways different from VAR estimates for the United States and Spain, they are roughly consistent with simulations with macro models for Finland and with estimates from a panel of EU countries conducted at the IMF (see paragraph 12).
D. Some Considerations on the Longer-run Impact of Tax Changes
27. The limited effectiveness of revenue-centered fiscal policy is a particularly striking feature of the findings described in Section C. But due to the short-term nature of the VAR approach, VAR models are ill-equipped to capture the longer-term effects of tax changes on employment and growth, especially in less-than-fully flexible labor market environments such as in Finland, where labor demand and supply reactions are likely to take time.
28. Identifying the longer-term impact of changes in labor taxes on employment and growth is not an easy task, however. A first complication arises because tax rates are endogenous to macroeconomic and fiscal developments (Koskela and Uusitalo 2003). While there is indeed a striking correlation between the tax wedge on labor and unemployment in Finland (Figure 4), the increase in labor taxes during the early 1990s was required to balance the fiscal accounts after the (preceding) drastic increase in unemployment in the wake of the breakdown of Finland's trade with Russia and the consequences of the banking crisis. Moreover, a variety of taxes is likely to influence employment, including payroll, income, and consumption taxes. Their respective impact on the labor market will depend, inter alia, on the elasticities of labor demand and supply and the degree to which economic agents perceive the underlying fiscal decisions as permanent. This further complicates identifying the relation between taxes and employment.
Finland: Labor Taxes and Unemployment, 1979-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: OECD; and Fund staff calculations.1/ Income tax plus employee and employer contributions in percent of labor costs for single persons without children.Finland: Labor Taxes and Unemployment, 1979-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: OECD; and Fund staff calculations.1/ Income tax plus employee and employer contributions in percent of labor costs for single persons without children.Finland: Labor Taxes and Unemployment, 1979-2002
Citation: IMF Staff Country Reports 2003, 326; 10.5089/9781451813210.002.A001
Sources: OECD; and Fund staff calculations.1/ Income tax plus employee and employer contributions in percent of labor costs for single persons without children.29. But even in the face of these difficulties, there seems to be a significant and positive longer-term correlation between labor taxes and unemployment (and a negative correlation with employment) in Finland. Roughly half of a change in income tax rates is translated to a change in labor costs, which, in turn, influence employment. By comparison, the labor-cost and employment effect of changes in payroll taxes (e.g., social security contributions for employers) seems to be somewhat higher, as more than half of these taxes are shouldered by employers (Koskela and Uusitalo 2003). Honkapohja and others (1999), using industrial-level time series data, find that the long-run elasticity of employment to changes in labor cost is about -0.7 in Finland. This implies that a 1 percentage point decrease in labor taxes would increase employment by about 0.4 percent. While there are some indications that the longer-term impact of tax changes in Finland (as well as in other Nordic economies) could be somewhat weaker than in other countries (Daveri and Tabellini, 2000), the order of magnitude of these effects is clearly nonnegligible.12
30. The discussion holds a crucial message for policy makers, putting the VAR-based findings into perspective: even though tax cuts might have only limited impact as a short-term macroeconomic policy tool, cutting taxes on labor can very well have a significant positive impact on employment and output in the longer run—but this requires keeping the public finances healthy so that tax cuts are not seen as being only of a temporary nature.
E. Concluding Remarks
31. The main message of this study is that, at least over the short run, discretionary fiscal measures have only a modest impact on the Finnish economy. While fiscal policy influences GDP in the short-run in line with our priors, the impact remains small compared to results reported from related studies on the U.S. and Spain. This difference in policy impact, often explained by the relative smallness and openness of the Finnish economy, is in line with findings in other recent studies conducted by the European Commission and the IMF.
32. A caveat to these findings is that the applicability of the structural VAR approach to the Finnish case is somewhat limited by the lack of data. Comprehensive high frequency fiscal data are not readily available and, to the extent it can be constructed, covers only a relatively short period marked by notable real shocks that are hard to model endogenously. This calls for a degree of caution in interpreting the results.
33. A second qualification to the VAR-based approach—whose importance to policy makers is hard to overplay—concerns its short-term perspective. Labor market decisions can play an important part in the transmission of fiscal policy decisions to the real economy, but they also might require time, especially in less flexible institutional environments. And indeed the discussion of existing Finnish evidence reveals that the impact of tax changes, which the VAR-model characterizes as rather small, is more prominent in the longer run. Thus, while the short-term impact of tax cuts seems small, they would seem to be a more effective tool to fostering growth over longer time periods, especially when undertaken in the context of expenditure consolidation to ensure that the tax cuts are not seen as reversible and that a sustainable fiscal position is obtained.
References
Aarle, B., H. Garretsen, and N. Gobbin, 2001, “Monetary and Fiscal Policy Transmission in the Euro area: Evidence from a Structural VAR Analysis,” paper presented at a Vienna Institute for International Economic Studies seminar, Vienna, Austria, January.
Blanchard, O., and R. Perotti, 2002, “An Empirical Characterization of the Dynamic Effects of Changes in Government Spending and Taxes on Output,” The Quarterly Journal of Economics (November).
Brunila, A., M. Buti, and J. in‘t Veld, 2002, “Fiscal Policy in Europe: How Effective are Automatic Stabilizers?,” European Economy (September),(Brussels: European Commission).
Céspedes, L., and A. Hoffmeister, 2003, “Fiscal Policy and Macroeconomic Volatility in Spain: An Empirical Assessment”, Spain: Selected Issues, IMF Country Report No. 03/41 (Washington: International Monetary Fund).
Daveri F., and G. Tabellini, 2000, “Unemployment, Growth and Taxation in Industrial Countries,” Economic Policy, (April), Vol. 15, No. 30, pp. 47–104, (Oxford, U.K. and Boston, USA: Blackwell Publishers Ltd.)
Hemming, K., and Mahfouz, 2002, “The Effectiveness of Fiscal Policy in Stimulating Economic Activity—A Review of the Literature,” IMF Working Paper 02/208 (Washington: International Monetary Fund).
Honkapohja, S., E. Koskela, and R. Uusitalo, 1999, “Employment, Labor Taxation, and the Balance of the Public Sector,” Finnish Economic Journal, Vol. 95, pp. 74–95.
Hunt, B., and D. Laxton, 2003, “Some Simulation Properties of the Major Euro Area Economies in MULTIMOD,” IMF Working Paper 03/31 (Washington: International Monetary Fund).
International Monetary Fund, “Adopting the Euro in Central Europe: an Examination of Policy Issues,” (forthcoming: Washington).
Koskela, E. and R. Uusitalo, 2003, “The Un-intended Convergence: How the Finnish Unemployment Reached the European Level,” Mimeo, January (Helsinki: University of Helsinki).
Perotti, R., 2002, “Estimating the Effects of Fiscal Policy in OECD Countries”, paper presented at an ISOM conference, Frankfurt, Germany, June.
Organization for Economic Cooperation and Development, 2003, Economic Outlook, Sources and Methods, No. 73, June (Paris).
Prepared by Helge Berger and Louis Kuijs.
A multiplier in “dollar-for dollar terms” measures the ratio of the (absolute) amount of additional activity in terms of currency units generated by one unit of additional expenditure.
In the absence of liquidity constraints, temporary tax cuts that are later reversed would not have any effect on spending.
The available quarterly fiscal data accounts for 34 percent of expenditures, and data on revenues that accounts for 84 percent of the total.
Sometimes et is referred to as the vector of unexpected movements or surprises based on the reduced-form VAR model.
See CH for comparable results for Finland. The empirical section below comments on the robustness of these assumptions.
The assumption is in line with preliminary estimates of (3) which found no robust evidence that either a2 or b2 is statistically and economically significant.
To help the model capture some of the exogenous structural changes influencing GDP growth during the 1990s—that is, the breakdown of Finland's trade relations with the former Soviet Union, the banking crisis in the early 1990s, and the rise of the export-driving ICT sector—both a linear time and a non-linear time trend are included. The latter is a Hodrick-Prescott approximation of actual GDP growth with the smoothing factor set to 1,000.
The (log-) estimates of the coefficients are transformed into dollar-for-dollar based on sample means of the revenue-to-GDP and expenditure-to-GDP ratios, respectively. The dynamic results discussed below are treated equivalently.
Comparisons with BP are based on their model assuming a deterministic trend in the VAR (as in our model). The data definition used by BP differs from ours, but comparable results following their specifications can be obtained. All CH results have been transformed into dollar-for-dollar terms assuming a revenue-to-GDP and expenditure-to-GDP ratio of 0.45.
The comparison for Spain is less straightforward, since CH present nonstandardized impulse responses. However, based on the time profile of the GDP reaction to the tax shock and the high contemporaneous impact described above, the dynamic effect would clearly appear to exceed the one reported in Figure 3.
There are some indications that the longer-term impact of tax changes in Finland (as well as in other Nordic economies) could be somewhat weaker than in other countries (Daveri and Tabellini, 2000). Koskela and Uusitalo (2003) argue this could be due to the more centralized bargaining systems in the Nordic region which might make wage formation less sensitive to changes in taxation compared to less centralized systems. The explanation is not fully compelling, however. In the Finnish case, for example, the government has used the centralized wage negotiations of recent years to condition tax policy on wage behavior, fostering wage moderation through promises of tax cuts on labor.
APPENDIX I: The Baseline NAWRU Model
The filtering process yielding the NAWRU is based on the unobserved components approach, see Kuttner (1994), Denis, Mc Morrow, and Roeger (2002), and Planas and Rossi (2003). The starting point for the first component of the bivariate model is the definition:
which decomposes the observable unemployment rate Ut in a cyclical component, Ct, and a non-cyclical component, . Additional exogenous regressors—M ≤ 3—are assigned to the latter component, such that
where Tt represents the underlying long-term trend, or NAWRU. Without additional exogenous regressors, i.e., M = 0, the non-cyclical component and the NAWRU-trend coincide.
The trend component is modeled according to its statistical properties, i.e. no economic information (e.g., on structural breaks) is included. The most general specification (see further below) is given by a random walk with drift, where the drift term μt is itself a random walk (and the trend Tt, hence, a second-order random walk):
Both errors, zt and at, are n.i.i.d; if Var(a,) = 0, the model collapses to a 1st order random walk with drift. On the other hand, the cyclical component in (1) is specified as an AR(N) process:
where N ≤ 2. To guarantee stationarity of the cyclical component, it must hold that Σ φn < 1.
The second component of the generic model is given by
and L is the lag operator.
This Phillips curve relationship links the change in wage inflation to (i) exogenous determinants of wage inflation, Xu such as (changes in) labor productivity or (changes in) the terms of trade, with 0 ≤ L ≤ 10 in the empirical application; (ii) autoregressi ve terms of the wage inflation (with 0 ≤ S ≤ 2); (iii) the rf-th difference of the lagged observed unemployment rate Ut; (iv) the cyclical unemployment component Ct, (with 0 ≤ R ≤ 4). and (v) an error term, which can have a MA(I) structure, I ≤ 3.
In the estimation, the most generic model is used: with respect to (A1), the specification chosen is a bivariate autoregressive model, with the trend expressed as a second order random walk, hence, Var(a,) ≠ 0 in (A2). Furthermore, an AR(2) specification is selected for the cyclical component in (A3), as indicated by preliminary tests (not reported here). Experiments with additional exogenous regressors in (A1), that is, M > 0, have resulted in a deterioration of the statistical fit.
Regarding (A4), the second difference of the lagged first series (Δ2Ut-1), as well as the contemporary cyclical component of unemployment were included, that is, d = 2 and R = 0. The choice of a 2nd order RW specification in (A2) implies for (A4) that d = 2, that is, the lagged unemployment series regressor enters in second difference in order to obtain a stationary regressor. In (A4), no exogenous regressors were employed in the baseline case.35
Table A1 illustrates the impact of different assumptions on the ARMA structure of the Phillips curve equation (A4) on key results of the model, including the estimated coefficient fio, the implied 2002 NAWRU, as well as selected test statistics. The significance of the estimator βo—which multiplies the contemporaneous cyclical unemployment component—indicates whether changes in wage inflation respond to the general economic environment as represented by the cyclical unemployment component. It is expected to be negative and significant, reflecting the dampening effect of an adverse economic environment on the size of wage increases. The estimate of the NAWRU in 2002 (as opposed to the official unemployment rate of 9.1 percent) is reported to allow a plausibility check of the results. Whether the model gives a statistically acceptable description of the endogenous series’ first two moments is checked by means of a Ljung-Box residual test statistic, with the null hypothesis being that the residuals are white noise.
Descriptive Results for the NAWRU Model (Simulated Annealing Algorithm)
beta0 represents the coefficient of the cyclical component of unemployment in the Phillips curve.
Descriptive Results for the NAWRU Model (Simulated Annealing Algorithm)
| Model Specification: ARMA(s,i) | ||||||
|---|---|---|---|---|---|---|
| 1,1 | 1,2 | 1,3 | 2,1 | 2,2 | 2,3 | |
| beta0 1/ | 0.0215 | 0.0196 | 0.0194 | 0.0130 | -0.0093 | -0.0101 |
| t-value | 2.32 | 1.97 | 1.93 | 1.64 | -2.13 | -2.33 |
| NAWRU 2002 | 9.07 | 9.00 | 9.00 | 8.94 | 8.31 | 8.33 |
| Ljung-Box p-value residuals from (A1) | 0.03 | 0.03 | 0.03 | 0.02 | 0.15 | 0.15 |
| Ljung-Box p-value residuals from (A4) | 0.19 | 0.80 | 0.92 | 0.91 | 0.21 | 0.27 |
| -2*loglikeiihood | 111.78 | 107.25 | 107.20 | 106.62 | 101.78 | 101.33 |
| R2 (uncentered) | 0.21 | 0.31 | 0.31 | 0.35 | 0.43 | 0.44 |
beta0 represents the coefficient of the cyclical component of unemployment in the Phillips curve.
Descriptive Results for the NAWRU Model (Simulated Annealing Algorithm)
| Model Specification: ARMA(s,i) | ||||||
|---|---|---|---|---|---|---|
| 1,1 | 1,2 | 1,3 | 2,1 | 2,2 | 2,3 | |
| beta0 1/ | 0.0215 | 0.0196 | 0.0194 | 0.0130 | -0.0093 | -0.0101 |
| t-value | 2.32 | 1.97 | 1.93 | 1.64 | -2.13 | -2.33 |
| NAWRU 2002 | 9.07 | 9.00 | 9.00 | 8.94 | 8.31 | 8.33 |
| Ljung-Box p-value residuals from (A1) | 0.03 | 0.03 | 0.03 | 0.02 | 0.15 | 0.15 |
| Ljung-Box p-value residuals from (A4) | 0.19 | 0.80 | 0.92 | 0.91 | 0.21 | 0.27 |
| -2*loglikeiihood | 111.78 | 107.25 | 107.20 | 106.62 | 101.78 | 101.33 |
| R2 (uncentered) | 0.21 | 0.31 | 0.31 | 0.35 | 0.43 | 0.44 |
beta0 represents the coefficient of the cyclical component of unemployment in the Phillips curve.
The table points to an unsatisfactory description of the data in the first four ARMA specifications (in particular with regard to the unemployment rate, see the first Ljung-Box test statistic). In addition, the positive value for β0 is clearly counterintuitive. The ARMA (2,2) and (2,3) specifications instead yield a reasonable statistical description: normality assumptions on both equations are not significant at a 10 percent level. The estimated β0 is significant for both models. Increasing the number of moving average terms in (A4) raises the t-value (in absolute terms). At 8.3 percent, the NAWRU derived for 2002 (in both specifications) is about 0.8 percentage points lower than observed unemployment. Using the ARMA(2,3) model, the final specification of (A4) is hence:
Figure 6 in the main text illustrates the NAWRU implied by (A4), and (A1), (A2), and (A3), specified as:
APPENDIX II: Additional Technical Restrictions of the Baseline Model
Additional restrictions imposed during the estimation process relate to parameters of the maximization technique (simulated annealing), and other parameters. In addition, if exogenous regressors were to be employed, boundaries on these variables in both equations (Al’) and (A4) could be imposed as well.
Additional Technical Assumptions and Restrictions
Restrictions not applicable due to the choice of trend specification (second order random walk).
Additional Technical Assumptions and Restrictions
| A. Assumptions on the “Simulated Annealing” Parameters | |||
|---|---|---|---|
| Initial temperature | 1 | ||
| Temperature redution factor | 0.8 | ||
| Tolerance | 0.005 | ||
| Number of cycles | 15 | ||
| Number of iterations | 50 | ||
| Maximum number of evaluations | 1000000 | ||
| B. Restrictions/Boundaries on “Other” Parameters | |||
| lower bound | upper bound | estimated model | |
| Equation (A1) | |||
| Drift | -0.5 | 0.5 | n.a. 1/ |
| AR1 | -1.96 | 1.96 | 1.32 |
| AR2 | -0.97 | 0.97 | -0.82 |
| Cycle MAI | -1.96 | 1.96 | n.a. 1/ |
| Cycle MA2 | -0.97 | 0.97 | n.a. 1/ |
| Trend innovation variance | 0.01 | 0.3 | 0.01001 |
| Trend slope variance | 0.75 | 0.95 | 0.75 |
| Cycle innovation variance | 0 | 0.406 | 0.31 |
| Equation (A2) | |||
| Intercept | -1 | 1 | -0.004 |
| Gamma - lag 1 | -2 | 2 | -0.003 |
| Beta - lag 0 | -2 | 2 | -0.01 |
| AR1 | -1.96 | 1.96 | -0.14 |
| AR2 | -0.97 | 0.97 | -0.85 |
| MAI | -0.97 | 0.97 | 0.38 |
| MA2 | -0.97 | 0.97 | 0.9 |
| MA3 | -0.97 | 0.97 | -0.18 |
| Innovation variance | 0 | 0.000474 | 0.0003 |
Restrictions not applicable due to the choice of trend specification (second order random walk).
Additional Technical Assumptions and Restrictions
| A. Assumptions on the “Simulated Annealing” Parameters | |||
|---|---|---|---|
| Initial temperature | 1 | ||
| Temperature redution factor | 0.8 | ||
| Tolerance | 0.005 | ||
| Number of cycles | 15 | ||
| Number of iterations | 50 | ||
| Maximum number of evaluations | 1000000 | ||
| B. Restrictions/Boundaries on “Other” Parameters | |||
| lower bound | upper bound | estimated model | |
| Equation (A1) | |||
| Drift | -0.5 | 0.5 | n.a. 1/ |
| AR1 | -1.96 | 1.96 | 1.32 |
| AR2 | -0.97 | 0.97 | -0.82 |
| Cycle MAI | -1.96 | 1.96 | n.a. 1/ |
| Cycle MA2 | -0.97 | 0.97 | n.a. 1/ |
| Trend innovation variance | 0.01 | 0.3 | 0.01001 |
| Trend slope variance | 0.75 | 0.95 | 0.75 |
| Cycle innovation variance | 0 | 0.406 | 0.31 |
| Equation (A2) | |||
| Intercept | -1 | 1 | -0.004 |
| Gamma - lag 1 | -2 | 2 | -0.003 |
| Beta - lag 0 | -2 | 2 | -0.01 |
| AR1 | -1.96 | 1.96 | -0.14 |
| AR2 | -0.97 | 0.97 | -0.85 |
| MAI | -0.97 | 0.97 | 0.38 |
| MA2 | -0.97 | 0.97 | 0.9 |
| MA3 | -0.97 | 0.97 | -0.18 |
| Innovation variance | 0 | 0.000474 | 0.0003 |
Restrictions not applicable due to the choice of trend specification (second order random walk).
References
Apel, M, and P. Jansson, 1997, “System Estimates of Potential Output and the NAIRU,” Sveriges Riksbank Working Paper No. 41 (Stockholm: Sveriges Riksbank).
Artus, J. R., 1977, “Measures of Potential Output in Manufacturing for Eight Industrial Countries, 1955–78,” Staff Papers, Vol. 24, pp. 1–35. (Washington: International Monetary Fund).
Baxter, M, and R. King, 1999, “Measuring Business Cycles: Approximate Band-Pass Filters for Economic Time Series,” Review of Economics and Statistics, Vol. 81, pp. 575–93.
Beveridge, S., and C.R. Nelson, 1981, “A New Approach to Decomposition of Economic Time Series into Permanent and Transitory Components with Particular Attention to Measurement of the ‘Business Cycle’,” Journal of Monetary Economics, Vol. 7, pp. 151–74.
Beyer, A., and R. E. A. Farmer, 2002, “Natural Rate Doubts,” ECB Working Paper 121 (Frankfurt: European Central Bank).
Blanchard, O., and D. Quah, 1989, “The Dynamic Effects of Aggregate Demand and Supply Disturbances,” American Economic Review, Vol. 79, pp. 655–73.
Box, G. E. P., and G. M. Jenkins, 1976, Time Series Analysis, Forecasting and Control (San Francisco: Holden Day).
Brunila, A., J. Hukkinen, and M. Tujula, 1999, “Indicators of the Cyclically Adjusted Balance: The Bank of Finland’s Experience,” Bank of Finland Discussion Paper 1/99 (Helsinki: Bank of Finland).
Bums, A. F., and W. C. Mitchell, 1946, Measuring Business Cycles (New York: National Bureau of Economic Research).
Canova, F., 1999, “Does Detrending Matter for the Determination of the Reference Cycles and Selection of Turning Points,” Economic Journal, Vol. 109, pp. 126–50.
Cheung, Y. W., and K. S. Lai, 1993, “Finite Sample Sizes of Johansen’s Likelihood Ratio Tests for Cointegration,” Oxford Bulletin of Economics and Statistics, Vol. 55, pp. 313–28.
Corbae, D., and S. Ouliaris, 2002, “Extracting Cycles from Non-stationary Data,” (unpublished: Austin, Texas and Singapore: University of Texas and National University of Singapore).
Corbae, D., S. Ouliaris, and P.C.B. Phillips, 2002, “Band Spectral Regression with Trending Data,” Econometrica, Vol. 70, pp. 1067–109.
De Masi, P. R., 1997, “IMF Estimates of Potential Output: Theory and Practice,” IMF Working Paper No. 97/177 (Washington: International Monetary Fund).
Denis, C., K. Mc Morrow, and W. Roeger, 2002, “Production Function Approach to Calculating Potential Growth and Output Gaps—Estimates for the EU Member States and the U.S.,” European Commission Economic Papers, No. 176 (Brussels: European Commission).
Everaert, L., and F. Nadal De Simone, 2003, “Capital Operating Time and Total Factor Productivity,” IMF Working Paper No. 03/128 (Washington: International Monetary Fund).
Gylfason, T., 1998, “Unemployment, Efficiency, and Economic Growth: The Case of Finland,” Ekonomiska Samfundets Tidskrift, Vol. 3, pp. 193–203.
Harvey, A.C., and A. Jaeger, 1993, “Detrending, Stylized Facts and the Business Cycle,” Journal of Applied Econometrics, Vol. 8, pp. 231–47.
Hodrick, R.J., and E.C. Prescott, 1997, “Post-war U.S. Business Cycles: An Empirical Investigation,” Journal of Money, Credit, and Banking Vol. 29, pp. 1–16.
Jalava, J., and M. Pohjola, 2001, “Economic Growth in the New Economy,” World Institute for Development Economics Research Discussion Paper No. 2001/5 (Helsinki: United Nations University).
Johansen, S., 1995, Likelihood based Inference on Cointegration in the Vector Autoregressive Model (Oxford: Oxford University Press).
Kuttner, K. N., 1994, “Estimating Potential Output as a Latent Variable,” Journal of Business and Economic Statistics, Vol. 12, pp. 361–8.
Planas, C., and A. Rossi, 2003, “Program GAP - Version 2.2, Technical Appendix” (unpublished: Ispra, Italy: European Commission, Joint Research Centre).
Rasi, C.M., and J. M. Viikari, 1998, “The Time-Varying NAIRU and Potential Output in Finland,” Bank of Finland Discussion Paper No. 6/98 (Helsinki: Bank of Finland).
Ross, K., and A. Ubide, 2001, “Mind the Gap: What is the Best Measure of Slack in the Euro Area?” IMF Working Paper No. 01/203 (Washington: International Monetary Fund).
Scacciavillani, F., and P. Swagel, 1999, “Measures of Potential Output: An Application to Israel,” IMF Working Paper No. 99/96 (Washington: International Monetary Fund).
Schreiber, S., and J. Wolters, 2003, “What’s Wrong with the (German) NAIRU?” (unpublished: Berlin and Frankfurt: Free University Berlin and University of Frankfurt).
Wagner, N., 2001, “A Note on Finland’s New Economy” in Finland: Selected Issues, IMF Country Report No. 01/215 (Washington: International Monetary Fund).
Prepared by Helge Berger and Andreas Billmeier. The authors would like to thank the Finnish authorities for their interest in the subject—which served as a catalyst for this paper.
The full-capacity stock of capital is usually approximated by the actual stock of capital. For a more elaborate approach, using French data on capital operating time, see Everaert and Nadal De Simone (2003).
Relatively little empirical work has been done comparing different output gap estimation methods. However, Brunila, Hukkinen, and Tujula (1999) briefly describe the approaches used by the Bank of Finland when assessing cyclically-adjusted budget measures: a Hodrick-Prescott filter, and the production function approach implicit in the Bank’s econometric model BOF5 (which does not incorporate such considerations as the natural rate of unemployment). Other work on potential output and the output gap in Finland includes Gylfason (1998), who uses a broken linear trend to account for a structural shift towards slower economic growth in the early 1970s, and Rasi and Viikari (1998) who apply an unobserved components method developed by Apel and Jansson (1997) to the Finnish data (potential output and the natural rate of unemployment are the unobserved variables estimated simultaneously). De Masi (1997) reviews related research done at the IMF.
In fact, Ross and Ubide (2001) singled out the quadratic trend as the best methodology (out of many) to forecast both business cycle turning points and the inflation in the euro area.
See, for example., Harvey and Jaeger (1993) for an overview of the shortcomings. Ross and Ubide (2001) discuss alternative approaches to determine the parameter λ endogenously.
In (2), the second difference of the trend is not defined around the first and the final observation, hence the different summation bounds between the value function and punishment term.
Estimation spans the period 1960-2002.
In what follows, given the very similar results for the time series ending in 2005 and 2008, the 2005 end point results are reported when comparing the HP filter to other trend estimation techniques.
In this formulation, p refers to the number of autoregressive lags, d refers to the order of integration, and the third parameter, q, gives the number of moving average lags; the series is assumed to be integrated of order one, that is, d = 1.
This implies, in particular for higher order models, that the filtered trend can be more volatile than the original series.
Note that various information criteria (Akaike, Schwarz, Hannan-Quinn, forecast prediction error) pointed to an ARIMA(2,1,2) specification (based on Box-Jenkins estimation). This specification, however, resulted in unrealistic volatility of the trend component around the peak-trough period in the late 1980s/early 1990s.
A so-called exact band-pass filter acts in principle as a double filter: it eliminates frequencies outside a range, here the business cycle frequency. For estimation purposes, however, these filters are usually spelled out in the time domain, since integrated series—such as real GDP—could not be handled by traditional frequency domain approaches, see Baxter and King (1999). They argue that upfront detrending of the series in order to apply discrete Fourier transforms involve a discretionary choice of the detrending method, whereas the symmetric moving average approximation would successfully remove any deterministic or stochastic trends up to second order.
See Corbae, Ouliaris, and Phillips (2002) for the analysis of the asymptotic case.
Given that the data are annual, a periodicity for the business cycle between 2 and 8 years has been assumed. Experiments with somewhat longer and shorter cycles yielded broadly similar results.
Blanchard and Quah (1989) also show that small violations of the identification scheme (e.g., lasting effects on output stemming from nominal shocks through a wealth effect) are of minor consequence.
The VAR model underlying the estimation includes, in addition to a constant, four lags of the endogenous variables, as indicated by information criteria. No residual autocorrelation was present in the specification chosen. Other specifications were tested, but dismissed, mostly on statistical grounds.
Early work on the production function approach includes Artus (1977). Subsequent research has refined the approach in various directions, see, for example, De Masi (1997).
Of course, the choice of a filter to detrend the unemployment rate and TFP adds an element of discretion.
See Jalava and Pohjola (2001), and Wagner (2001).
Looking ahead, demographic developments will play a significant role in Finland, with the baby boomer generations expected to retire soon. In a longer-term analysis, this kind of information could be taken into account using the production function approach.
Important shortcomings of the approach include the dependence on a number of crucial assumptions, for example, (constant) shares of capital and labor, and the functional form of the production relationship (number of input factors, returns to scale). In addition, data requirements can pose significant problems to any production function approach: in particular, the capital stock is difficult to measure consistently.
See Denis, Mc Morrow, and Roeger (2002). This methodology substitutes for more “traditional” approaches—such as the Hodrick-Prescott filter—and, at the same time, unifies the Commission’s efforts toward a consistent representation of inflationary pressures in the member countries.
Note that these conclusions also hold for CPI inflation instead of wage inflation.
With three lags, no significant residual autocorrelation emerged, whereas more parsimonious models reveal problems of autocorrelation at the first lag (statistics not reported).
In a model with a restricted trend, all variables (including the trend) appeared to be excludable from the system.
On theoretical grounds, the unemployment rate is bounded by the interval (0;1) and hence not truly 1(1). The fact that it cannot grow out of bounds in the long run, however, does not preclude it from behaving like an integrated process in the shorter run, as evidenced by the test statistics. The stationarity tests presented above do not allow for a structural break in the series analyzed. The strong rise of the unemployment rate in the early 1990s—as described above—could be viewed as such a break. This proposition is not investigated further since a stationarity result for the unemployment rate when allowing for a break in the series even underscores the case for the NAWRU approach, see Schreiber and Wolters (2003).
A more detailed description of the model set-up can be found in Appendix I. In the terminology of the European Commission, the NAWRU model is known as the “GAP model.” As the analysis in the appendix shows, both the assumed representation of wage inflation and the inclusion of additional regressors can have substantial impact on trend unemployment.
Here, the model abstracts from arguments related to the role of trade unions, and centralized wage bargaining.
This parallels earlier results: in Jalava and Pohjola (2001), it is found that the importance of “multi-factor productivity,” a concept similar to total factor productivity in the present paper, almost doubled in the second half of the 1990s, see also Wagner (2001).
See Table 4 in Wagner (2001) for a qualitatively similar result.
Assuming that TFP is entirely associated with “technological progress” may bias the interpretation of the results to the extent, for example, that there are errors in measuring the capital stock—though the results of Jalava and Pohjola (2001) suggest good reasons to believe in the association.
See Wagner (2001).
The simple linear, quadratic, and exponential trend measures have been omitted.
Maximization of the likelihood function was carried out by two algorithms. While the simulated annealing algorithm is slower than a Newton type algorithm, it is more likely to identify a global maximum. Since our experiments showed that local maxima posed a problem using the Newton-type algorithm, the simulated annealing algorithm was applied in the estimation.
