Methodology

The multivariate filtering approach (denoted throughout as MVF) estimates the potential output based on a model that captures relationships between actual and potential GDP, unemployment, and inflation (Blagrave, et al., 2015). In this approach, the potential output is thought of as the level of output that can be achieved if nominal prices and wages were fully flexible (Okun, 1962). It uses actual inflation, inflation expectations, and the deviation of the unemployment rate from its estimated equilibrium level (unemployment gap) to distinguish between demand and supply shocks. Bayesian techniques are used to estimate a simultaneous system of equations that describe the evolution of three key variables: output, inflation, unemployment, and their trends. Two economic relations are imposed:

• The Phillips curve, linking the output gap (an unobservable variable y_t) to the observable inflation π_t:

  $\begin{array}{cc}{\pi }_t=\lambda {\pi }_{t+1}+(1-\lambda ){\pi }_{t-1}+\beta y_t+{\varepsilon }_t^{\pi }& \left(1\right)\end{array}$
• The Okun’s law, governing the relationship between cyclical unemployment and output gap:

  $\begin{array}{cc}{u}_t={\tau }_2{u}_{t-1}+{\tau }_1{y}_t+{\varepsilon }_t^u& \left(2\right)\end{array}$

where y_t is the output gap, π^t+1 inflation expectation, and u_t is the unemployment gap (the deviation of unemployment rate from its equilibrium level). Additional seven equations describe the evolution of output and unemployment rates as the sum of their respective stochastic trends and shocks (see Blagrave et al., 2015, for a full set of equations).

The method has the advantage of relatively limited data requirements and robustness to end-of-sample problems. In its basic form, the technique is relatively straightforward to implement requiring only a few variables (real GDP, unemployment, and inflation). In addition, the filter could use medium-term growth and inflation forecasts to improve the accuracy of estimates in the outer years and address the common end-of-sample problem prevalent in filtering techniques.

Application to CESEE

Applying the MVF approach to CESEE countries using standard measures of price inflation may be misleading. At the core of the method is the New Keynesian Phillips curve postulating that if the output gap is positive over time, prices will begin to rise in response to demand pressures and vice versa. Imported inflation has played a significant role in explaining the variation of headline inflation in the region in recent years (Iossifov and Podpiera, 2014). As such, headline inflation may not provide a complete picture for judging the cyclical position of CESEE economies. Even core inflation is still contaminated by external shocks given the high pass-through of global commodity prices in CESEE and the high share of imported inflation in core inflation (Iossifov and Podpiera, 2014). Therefore, in addition to headline and core inflation, we also use wage inflation to better gauge domestic demand pressures. Indeed, using wage inflation in the MVF suggests less slack in several CESEE countries in 2014–15 (Figure 1; Appendix A, Figure A.1).⁶

CESEE EU: Comovement of Cyclical Variables

Citation: IMF Working Papers 2017, 037; 10.5089/9781475578980.001.A001

Sources: Haver Analytics and IMF staff calculations.Note: CE-5 = Czech Republic, Poland, Hungary, Slovakia and Slovenia. SEE EU = EU countries in South-Eastern Europe.

• Download Figure
• Download figure as PowerPoint slide

View Full Size

CESEE EU: Comovement of Cyclical Variables

Citation: IMF Working Papers 2017, 037; 10.5089/9781475578980.001.A001

Sources: Haver Analytics and IMF staff calculations.Note: CE-5 = Czech Republic, Poland, Hungary, Slovakia and Slovenia. SEE EU = EU countries in South-Eastern Europe.

• Download Figure
• Download figure as PowerPoint slide

CESEE EU: Comovement of Cyclical Variables

Citation: IMF Working Papers 2017, 037; 10.5089/9781475578980.001.A001

Sources: Haver Analytics and IMF staff calculations.Note: CE-5 = Czech Republic, Poland, Hungary, Slovakia and Slovenia. SEE EU = EU countries in South-Eastern Europe.

• Download Figure
• Download figure as PowerPoint slide

Methodology

The multivariate filter with financial frictions developed by Berger et al. (2015) takes into account credit growth and house prices, in addition to inflation and capacity utilization, to obtain sustainable output estimates (denoted throughout as MVF-FIN).⁷ The main idea behind this approach is to use the link between financial and output cycles as an additional information to identify demand and supply shocks. When inflation and inflation expectations are aligned but financial variables depart from their sustainable level, the filter assigns a larger portion of the actual output to the cyclical component. In this way, it offers a reduced-form approach to capturing the link between financial cycles and output gaps. The model is estimated using maximum likelihood in the state-space context.

Specifically, the sustainable output is estimated by decomposing observed GDP series into two unobservable components, the cycle (in the broadest sense capturing all transitory movements of GDP) and the trend (or sustainable) output:

where y_t and $y_t^{*}$ represent observed and sustainable output, respectively, and c_t is the business cycle. Mirroring the HP filter, the model is estimated with a constraint on the variance ratio λ.

In addition, the model takes into account information from a set of observable variables, xt, which could be correlated with the output gap. To match the dynamic properties of the latter, the approach allows for the current output gap to depend on its lagged values:

where the variance ratio $\frac{\text{Var}(y-y*)}{\text{Var}({\mathrm{\Delta }}^{2}y*)}$ is constrained to match the one implied by equations (3)-(5). In the empirical application, x_t includes credit growth, consumer price inflation, and, where available, house price inflation, and a survey-based measure of capacity utilization.

Application to CESEE

Estimating potential output using MVF-FIN is relevant for CESEE countries because most of them have experienced strong credit expansion before the financial crisis and rapid credit slowdown/contraction afterwards (Figure 1). However, a special challenge in the case of CESEE is to disentangle financial cycles from financial deepening.

To address this challenge, we use credit gap estimates from the IMF (2015) to ensure that the structural increase in credit stocks in CESEE countries is not considered part of credit cycle.⁸ The differences between output gap estimates using original credit data and adjusted credit data are substantial. The results suggest that ignoring the structural increase in credit due to financial deepening would have led to an underestimation of potential output during boom years. Similarly, after the crisis, it would have resulted in overestimation of potential output. This is particularly prominent for the Baltic states and SEE-EU countries that have experienced strong credit booms and busts (Appendix A, Figure A.2).

Methodology

The production function approach (denoted throughout as PF) focuses on the supply side—production factors—to decompose output into structural and cyclical components. In its full specification, it takes into account the cycles in labor input (including average hours worked), capital (including capacity utilization), and the total factor productivity (TFP), which is obtained as a residual.

We use a standard Cobb-Douglas production function approach (as in e.g. Christiansen, 2013; Johnson, 2013; Epstein and Macchiarelli, 2010; Konuki, 2008; Moore and Vamvakidis, 2007; Denis et al., 2006, and Faal, 2005). What remains unexplained after accounting for capital and labor inputs and their intensities is attributed to the TFP, denoted by A_t (the Solow residual). The specification for the quarterly real output Y_t is the following:

where:

• K denotes the capital stock, which is derived using the usual perpetual inventory model (Epstein and Macchiarelli, 2010, and Teixeira de Silva, 2001) as K_t = (1-ρ)K_t-1+I_t, where ρ is the depreciation rate calibrated using the historical average rate and I stands for investment. The capital stock of the initial year is taken from the Penn World Tables.

• CU is a survey based measure of capacity utilization.

• L is the number of employed persons, using the national accounts concept.

• AHW is the average hours worked from the national accounts.

• α stands for the labor share in the production function. Labor share is calculated as the ratio of compensation of employees to gross value added.⁹

The TFP, which is computed as a residual, includes human capital, as its contribution to the TFP growth before and after the crisis is small.¹⁰ Due to data constraints, the TFP in some countries also includes capacity utilization and average hours worked.

The potential output is derived using calibrated labor share, smoothed trends in unemployment and TFP, and the actual capital stock. Unemployment rates, average hours worked, and TFP are smoothed using HP filter (smoothing factor 1600). Capacity utilization series are demeaned. Employment consistent with NAIRU is derived using actual labor force and the smoothed unemployment rate.

Application to CESEE

Not accounting for intensities of production factors may potentially understate booms and busts. Besides the cycle in employment, the measures of intensity of the input utilization— namely capacity utilization and average hours worked—have cyclical components as well, contributing to the overall business cycle. Figure 1 shows that cycles in average hours worked and capacity utilization are synchronized, exacerbating the potential output identification problem if not accounted for explicitly:

• Capacity utilization increases during booms, when growing output is hitting production limits and falls during busts when capacity becomes excessive for a new lower level of output. Figure 1 shows the fluctuations in capacity utilization throughout the business cycle. This suggests that not accounting for such cyclical components in capacity utilization may overstate potential output growth during booms and understate during busts—making output gap less volatile.

• The average hours worked measure exhibits cyclical fluctuations as well. During booms, incumbent workers are working more hours while during busts their hours are reduced. Figure 1 shows that this is most pronounced in the Baltic states and SEE-EU countries.

Figure A.3, in Appendix A, shows a comparison between output gaps derived using production function approach with and without adjusting capacity utilization and average hours worked for the CESEE EU countries. The output gap amplitude appears to be smaller by 2–8 percentage points in the case of production function without adjustments for capacity utilization and average hours worked.

Factors of production:

Before the crisis, most countries in our sample appear to have enjoyed strong TFP growth and some countries also dynamic capital accumulation, while labor played a smaller role (Figure C.1, Appendix C). On average, capital accumulation contributed more than 2 percentage points to the region’s potential growth during 2005–09. In Bulgaria, Estonia, and Latvia, capital contribution was even higher—about 4 percentage points. Average estimated TFP contribution to the potential growth was about 2 percentage points during 2005–09. In Latvia, Lithuania, Serbia, and Slovakia, estimated TFP contribution was above average close to 3 percentage points. In general, labor played a smaller role, although in some countries it was still a considerable source of growth. In addition, some countries such as Romania and Moldova experienced outward migration, which gave rise to a negative contribution of labor to potential growth.

After the crisis, most countries experienced a significant TFP growth slowdown, although factors contributing to the postcrisis potential growth slowdown varied across countries of the region (Figure 3). Contribution of capital accumulation was mixed. In 7 out of 18 countries in our sample (including Bulgaria, the Baltics, and Slovenia), the contribution of capital accumulation to growth halved after the crisis. These are the countries that experienced rapid credit growth before the crisis fueled by external borrowing. These countries underwent a significant adjustment in their current account balances after the onset of the crisis, mainly on account of reduction in investments. In contrast, countries like the Czech Republic or Poland, which had not experienced excessive borrowing during precrisis years, continued to record strong contributions to the potential growth from capital accumulation.¹⁵ Romania also experienced excessive credit boom financed by external borrowing, but a big part of credit was used to finance consumption. Therefore, an increase in the saving rate after the crisis played a big part in the current account adjustment in Romania.

External versus country-specific factors:

Based on our analysis, external/common factors appear to have been closely correlated with potential growth in CESEE countries excluding the CIS and Turkey, while in the CIS and Turkey idiosyncratic/country-specific factors have been more important, particularly after the crisis (Figure 4).¹⁶

• Euro area growth: The above average potential growth in the region before the crisis appears to have been associated with a strong precrisis potential growth in the euro area. This strong correlation persisted during the crisis as well, when potential growth in the region fell short of its long-run average over 2000–15. Among other factors, this may reflect the expansion of the technological frontier, which stalled after the crisis (see IMF, 2016a).

• Other common factors: For CESEE excluding the CIS and Turkey, “other common factors” exhibited strong co-movement with the region’s potential growth before the crisis. This “other common factors” component grew much faster before the crisis, but slowed notably after the crisis, before accelerating again more recently. The region’s potential growth mimicked the observed pattern in “other common factors”. The positive co-movement before the crisis may reflect the expansion of global supply chains, including rapid integration of several CEE countries into the German supply chain, and growing trade links with the rest of Europe, on the back of the EU accession (8 CESEE countries joined the EU in 2004 and another two countries joined the EU in 2007). After the crisis, the significant negative contribution of “other common factors” could be due to a slowdown in global trade, structural bottlenecks, and the slow resolution of crisis legacies (NPLs, debt overhang) that held back investment in many CESEE countries as discussed in the IMF (2015). In more recent years, increased absorption of EU structural funds could be behind positive contribution of “other common factors”.