Kingdom of the Netherlands
Netherlands: Selected Issues
Author: International Monetary Fund
Publication Date:
03 Jun 2008
eISBN:
9781451829594
Language:
English
Keywords:
ISCR; CR; export; Netherlands; SITC; Germany; CIT rate; business cycle; export behavior; TOT shock; cyclical export; growth advantage over Germany; Export performance; Labor costs; Terms of trade; export growth advantage; product SITC
Download PDF (1.9 MB)

Overall competitiveness of the Dutch economy seems adequate, but domestically produced exports have lost market share recently. Over the past three decades, globalization has greatly influenced economies as countries have become more integrated. Empirical studies on business cycles synchronization and transmission of shocks among countries have provided conflicting results. In its descriptive part, this study concludes that Dutch export competitiveness is not a problem so far. This also finds that the Netherlands is relatively more exposed to supply-driven shocks while Germany is more exposed to demand-driven shocks.

Abstract

Overall competitiveness of the Dutch economy seems adequate, but domestically produced exports have lost market share recently. Over the past three decades, globalization has greatly influenced economies as countries have become more integrated. Empirical studies on business cycles synchronization and transmission of shocks among countries have provided conflicting results. In its descriptive part, this study concludes that Dutch export competitiveness is not a problem so far. This also finds that the Netherlands is relatively more exposed to supply-driven shocks while Germany is more exposed to demand-driven shocks.

I. Maintaining Competitiveness in the Global Economy: Dutch Export Performance1

A. Introduction

1. Overall competitiveness of the Dutch economy seems adequate, but domestically produced exports have lost market share recently. With a large current account surplus, robust exports, and strong growth, external Dutch competitiveness would appear to be satisfactory. In addition, the external sector is contributing positively to economic growth. After the sizable real appreciation in 2001-03, REER measures based on different price indices have been relatively stable, although ULC-based measures suggest a growing gap with some competitors.2 Manufacturing unit labor costs have fallen for the last four years, but at a lower pace than for some trading partners. Yet, multilaterally-consistent measures of equilibrium exchange rates suggest that the real exchange rate for the Netherlands is broadly in equilibrium. Domestically produced exports’ growth was in line with adjusted market growth until 2000.3 Afterward, the Dutch market share declined. With foreign trade representing about 80 percent of GDP, the loss of market share has raised concerns about Dutch competitiveness.

uA01fig01

Real Effective Exchange Rates

(ULC based, 2002=100)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

2. Past staff analysis indicated that, while aggregate measures of competitiveness showed no sign of worsening, disaggregated trade data suggested some deterioration.4 More recently, some other observers have also suggested that the country has not benefited fully from the opportunities offered by the rapid economic growth of emerging Asia and the enlargement of the EU. A flexible economy should be able to reorient the destination of its exports and product mix toward fast-growing economies and sectors. In addition, staff also found that TFP growth in the Netherlands was associated with certain structural features of the labor market. In particular, Dutch TFP growth decelerated as the secular fall in the ratio of the minimum wage to the median wage and in union density tapered off after 1998.5 Similarly, staff analysis found a strong negative effect of changes in the ratio of the minimum wage to the median wage and in union density on TFP growth, both key features of the Dutch labor market.

3. Building on past staff research, this paper performs a descriptive analysis of Dutch export data and also analyzes quantities and price developments following shocks to the economy. The analysis of export behavior is done using data by destination and by SITC, Revision 3, product classification. Notably, this analysis distinguishes between the cyclical and the trend components of the series. Next, this chapter analyzes the behavior of prices and quantities following a domestic and a foreign shock to the Dutch economy. In particular, the paper compares and contrasts the reaction of Dutch and German macro and trade variables to shocks to unit labor costs in manufacturing and to terms of trade.

4. Over the past three decades, globalization has greatly influenced economies as countries have become more integrated. Integration has occurred through intensive trade of goods and services (Imbs, 2004), and financial services (Brook et al, 2003). Economies have benefited from trade and foreign direct investment (FDI). However, globalization can make countries more vulnerable to external shocks as well as crises can be severe and contagion can spread rapidly. The high degree of economic and financial integration, stresses the importance of product- and factor-market flexibility. Economies’ ability to absorb domestic- and foreign-origin shocks takes paramount importance, even more so when countries’ policy menu is restricted, for example by participation in a currency area. Not surprisingly, competitiveness issues have been taken to the front line of the economic and political debate in Europe.

5. Empirical studies on business cycles synchronization and transmission of shocks among countries have provided conflicting results. Most findings show increasing synchronization of economic variables across countries (Nadal De Simone, 2002, Bordo and Helbling, 2003, Kose et al, 2005). According to an alternative view, however, despite large increases in trade and financial openness, G-7 business cycles may have become less synchronized, for instance, because trade flows lead to increased specialization of production. (Stock and Watson, 2003, Kose and Yi, 2006). Other studies have emphasized the sources of shocks, their spillovers, and channels of their transmission from one country or region to another. Recent examples include the study of the monetary transmission mechanism in the Euro area by Ciccarelli and Rebucci (2006), and Canova, Ciccarelli, and Ortega (2007). Similarly, Canova and Ciccarelli (2006) find a positive and significant effect of U.S. GDP growth shocks on France and Italy, but a negligible effect on German GDP growth. Given that the VAR methodology used in those studies has some limitations—the most conspicuous being that it cannot accommodate a large panel of series without the risk of running short of degrees of freedom—Stock and Watson (2002) use the approximate structural dynamic factor model on a large panel of developed countries’ variables and, like Kabundi and Nadal De Simone (2007) and Eickmeier (2007), find that U.S. demand shocks and EU supply shocks have a positive and significant effect on French and German output (Appendix II).

6. In its descriptive part, this study concludes that Dutch export competitiveness is not a problem so far. However, (1) Dutch exports trend has been somewhat below Germany’s in the 2000s. (2) Dutch traditional lead over Germany in exports of manufactured goods, machinery and transport equipment was lost in the 2000s.

7. In its analytical part, this study finds that the Netherlands is relatively more exposed to supply-driven shocks while Germany is more exposed to demand-driven shocks. Following an increase in unit labor costs in manufacturing (ULCM), the adjustment in the Netherlands is comparatively less flexible than in Germany. The Netherlands adjusts relatively less via price and wage changes, and more via employment changes. (4) The same features are also evident when the two countries are faced with an upward, supply-driven, terms of trade (TOT) shock. (5) The Netherlands profits relatively less from a demand-driven increase in the TOT, but accordingly, suffers a relatively lower output fall following a demand-driven ULCM increase.

8. The remainder of the paper is as follows. Next section discusses the data and the procedure adopted to make them stationary. Section III describes the cyclical and trend behavior of trade variables; section IV discusses the behavior of major macro and trade variables following a shock to ULCM and to the TOT. Section V summarizes the paper and draws policy implications.

B. Data and Treatment of Non Stationarity

9. This study uses two large data panels. The first one comprises 396 quarterly macroeconomic series and 106 series of trade by country (for a total of series N = 502). Trade series include imports and exports to the euro area, the EU, accession countries, Canada, the United States, the United Kingdom, Japan, China, Asia, Latin America, and the rest of the world. The second data panel contains 396 quarterly macroeconomic series and 110 series of trade by SITC, Revision 3, category of products (for a total of series N = 506). The sample period is 1981:Q1–2006:Q4 (i.e., T = 104). The countries are France, Germany, Japan, the Netherlands, the United Kingdom, and the United States. In addition, a set of global variables is included, containing such items as crude oil prices, a commodity industrial inputs price index, world demand, and world reserves. Variables have been seasonally adjusted.

10. For estimation purposes, series are treated as covariance-stationary. Instead of applying unit roots tests to determine the degree of integration of the series and then difference or detrend them depending respectively on whether they are I(1) or I(0) with a deterministic trend, the Corbae-Ouliaris Ideal Band-Pass Filter was used to make the series stationary (Corbae and Ouliaris, 2006).6 There are several reasons for this approach. First, available unit root tests have low power and often the decision on the degree of integration of the series has to be based on subjective judgment. Second, first differencing removes a significant part of the variance of economic time series. Third, the Corbae and Ouliaris filter is consistent, is not subject to end-point problems, and has no finite sampling error.

C. Stylized Facts

11. Dutch and German exports display a similar cyclical behavior by destination and by product, although Germany’s is more volatile. The business cycle in main trading partners correlates well with exports. The U.S. driven early-1980s recession, the European 1993 recession, and the bursting of the stock market “bubble” at the end of the 1990s are clearly correlated to exports behavior (IMF Country Report No 05/401). In general, the Dutch export cyclical component is less volatile than Germany’s, which may be associated with the product composition of both countries exports; in other words, German exports have a higher short-term elasticity with respect to output. Divergences in recent trade performance between both countries seem unrelated to the cyclical part of trade flows. Instead, albeit recent, export trend growth developments, in both export destination and product composition, may suggest that the Dutch economy would need to become more flexible to preserve its relative competitive position.

12. Dutch competitiveness in the 2000s has had divergent trends in terms of both geographic distribution and product composition. Starting in 2003-04, the Netherlands lost its trade export growth advantage over Germany vis-à-vis the euro area, the EU, Japan, and the UK. In contrast, and against some observers’ views, it seems that the Netherlands has taken advantage of the eastward expansion of the EU starting in 2000 and of Chinese rapid growth after 2003 (Table I-1). Finally, the Netherlands has lost its traditional growth advantage over Germany in manufactured goods starting in 2000, and machinery and transport equipment, miscellaneous manufactured goods, and crude materials except fuels, starting in 2003 (Table I-2). With the exception of animal and vegetable oils, fats and waxes, a similar pattern is present in all other SITC, Revision 3, categories.

Table I-1.

Trend Exports per Region 1/

(Average annual percent change)

article image

Numbers in bold indicate a higher growth rate of Dutch trend exports.

Table I-2.

Trend Exports per Product SITC 1/

(Average annual percent change)

article image

Numbers in bold indicate a higher growth rate of Dutch trend exports.

13. In sum, the descriptive analysis suggests that there has been since 2000-03 an underperformance of Dutch exports relative to its own past and relative to Germany. The change in export performance is recent, but raises some questions about the future competitiveness of the Dutch economy. Importantly, these developments seem unrelated to the exchange rate given the growth deceleration pattern observed in terms of export destination.

uA01fig02

Cyclical Exports from the Netherlands and Germany to the US and the UK

(million euros)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

uA01fig03

Cyclical Exports from the Netherlands and Germany: Manufactured Goods, and Machinery and Transport Equipment

(million euros)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

uA01fig04

Trend Exports from Netherlands and Germany to Euro Area and Accesion Countries

(average annual percent change)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

uA01fig05

Trend Exports from Netherlands and Germany to Japan and China

(average annual percent change)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

uA01fig06

Trend Exports from Netherlands and Germany to the EU and Asia

(average annual percent change)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

uA01fig07

Trend Exports from Netherlands and Germany to the US and the UK

(average annual percent change)

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

D. Dynamic Behavior

The model and shocks identification procedure

14. To gain further insight into the dynamic behavior of the Dutch economy, this study uses a large dimensional approximate dynamic factor model in the tradition of Stock and Watson (2002). See Appendix II for description of the model.7 The intuition behind the approximate dynamic factor model analysis is simple and can be summarized as follows. A vector of time series can be represented as the sum of two latent components, a common component and an idiosyncratic component. The common component, which is a linear combination of common factors, is driven by few common shocks, which are the same for all variables. Nevertheless, the effects of common shocks differ from one variable to another due to different factor loadings. In contrast to standard common component analysis, the idiosyncratic component is driven by idiosyncratic shocks, which are specific to each variable. The static factor model used here differs from the dynamic factor model in that it treats lagged or dynamic factors as additional static factors. Thus, common factors include both lagged and contemporaneous factors.

15. The estimation process comprises estimating the common components and identifying a reduced number of structural shocks that explain the common components of the variables of interest. The identification of structural shocks is done by focusing on the reduced form VAR residuals. Following Eickmeier (2007), the identification scheme has three features: (1) it maximizes the explained variance of the forecast error of the chosen variable and calculates impulse-response functions; (2) it assumes that identified shocks are linearly correlated to a vector of fundamentals; and (3) it identifies orthogonal shocks by rotation using a sign-identification strategy that imposes inequality restrictions on the impulse-response functions of variables based on a typical aggregate demand and aggregate supply framework.8 Only those shocks that have a structural meaning are chosen.

16. Given recent Dutch ULCM developments and that the Netherlands is the quintessential case of a small open economy, the dynamics of shocks to ULCM and to TOT is analyzed. The choice of shocks seems also relevant given the discussion in the previous section. As in standard macroeconomic models, an increase in ULCM can be interpreted as the result of a fall in labor productivity or an increase in labor compensation. The former is going to be interpreted as a supply shock and the latter as a demand shock. This is consistent with the empirical observation that real wages are procyclical. Similarly, a rise in the TOT can result from a deterioration of the country’s competitiveness related to structural factors, or alternatively, from strong world demand for the country’s products. If the shock is persistent, it will result in an increase in consumption (and investment) and the current account will move into deficit. Instead, if the TOT increase is due to strong world demand for the small country’s products, given the transient nature of the shock, consumers will largely save the windfall and the current account will move into surplus. The table above displays the sign restrictions for shock identification, imposed contemporaneously and during the first year after the shock.

article image

Estimation

17. The estimation procedure comprises several steps. The first step of the estimation is the determination of the number of factors. As mentioned above, the series are assumed to follow an approximate dynamic factor model. Using Bai’s and Ng’s (2002) selection criteria, four factors were chosen. The identification of the structural shocks followed the approach of the structural VAR literature. No identification technology is completely foolproof. While the identification technology is flexible enough not to require special restrictions to disentangle common shocks from the contemporaneous transmission of regional or country-specific shocks, it does require additional work to confirm the nature and source of shocks. The study does not restrict the impact effect of the shock. In addition, after identifying two shocks and giving them an economic interpretation, the same analysis was done on a data set containing only Dutch variables. It showed that the resulting impulse-responses are similar to those of the broader data set, supporting the identification of the shocks.

18. Only two structural shocks could be identified for each variable of interest (ULCM and TOT). The identification procedure proposed by Uhlig (2003) was applied to the common components of the Netherlands’s and Germany’s ULCM and TOT. The objective was to find a reduced number of structural shocks that maximize the explanation of its forecast error variance over 20 periods. As noted above, sign restrictions on impulse response functions were used to provide economic meaning to the structural shocks. Following Peersman (2005), sign restrictions were applied to the first two principal component shocks taking pairs of shocks: a supply shock and a demand shock. The bootstrap was done with the objective of removing the possible bias in the VAR coefficients which can arise from the small sample size. The impulse-response functions are calculated for the first five years to display the cyclical pattern associated with the structural shocks.9 Both the median response and a 90 percent bootstrapped confidence band are estimated.10

Results

19. Econometric results are presented in the form of variance decomposition.11 Tables I-3 to I-6 show the variance decomposition and the forecast error variance of the common components (henceforth, error variance) of Dutch and German variables explained by the two identified shocks to ULCM and TOT, respectively. These two shocks suffice to explain 99 percent of the error variance of the common components of Dutch and German ULCM and TOT over 20 quarters. The variance shares of ULCM common components are high, as they reach about 68 percent and 73 percent for the Netherlands and for Germany, respectively. In contrast, the variance shares of TOT common components are smaller, especially for the Netherlands: up to 20 percent and 43 percent for the Netherlands and Germany, respectively. This indicates that Dutch TOT are more heavily influenced by idiosyncratic factors than Germany’s. As in Kabundi and Nadal De Simone (2008), the TOT are relatively less significant channels of shock transmission.

20. The main result of the econometric exercise is that the Netherlands is relatively more exposed to supply-driven shocks, while Germany is more exposed to demand-driven shocks. In addition, supply shocks to ULCM are relatively more important than demand shocks; the opposite is true for TOT shocks. That feature is consistent with the real business cycles literature that stresses the importance of productivity-driven shocks as the most significant source of business cycle fluctuations. Thus, the effect of supply shocks is more persistent.12

21. Following a supply-driven (productivity) increase in ULCM, Dutch output falls driven by the decline in consumption, investment and exports, a fall that is larger than in Germany. This is the result of a larger appreciation of the real exchange rate and a milder fall in the CPI in the Netherlands than in Germany. Real compensation of employees is stable in the Netherlands, while it falls in Germany. Dutch total employment falls for a longer period of time than German employment. The dollar value of exports to all destinations increases more in Germany than in the Netherlands; the value of Dutch exports to China actually falls. The same results apply in terms of the euro value of exports per product, especially for mineral fuels and lubricants, and chemicals and related products (Figure I-1, for the Netherlands, and Figure I-2 for Germany).

Figure I-1:
Figure I-1:
Figure I-1:

ULCM, the Netherlands

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

Figure I-2:
Figure I-2:
Figure I-2:

ULCM, Germany

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

22. Therefore, following an increase in unit labor costs in manufacturing (ULCM), the Netherlands displays less flexibility to adjust than Germany. This is especially the case if the increase in ULCM is due to a fall in productivity (supply-driven). The Netherlands adjusts relatively less via price and wage changes, and more via employment and output changes. There seems to be a relatively larger downward rigidity of wages in the Netherlands. The same features are also evident when the two countries are faced with an upward, supply-driven, TOT shock—not shown here to conserve space.

23. The Netherlands profits relatively less from a demand-driven increase in the TOT (accordingly, it enjoys a relatively lower fall in output following a demand-driven increase in ULCM). This is the result of a larger appreciation of the real exchange rate and a smaller fall in the CPI in the Netherlands than in Germany. In addition, while real compensation to employees falls in Germany, it actually increases in the Netherlands and this despite the fall in labor productivity. Accordingly, total employment increases relatively less in the Netherlands. The dollar value of Dutch exports by destination increases, except the value of exports to China. The increase in exports values is, however, larger for Germany. The euro value of Dutch exports increases less than the value of German exports, especially in animal and vegetable oils and fats, chemicals and related products, manufactured goods, and machinery and transport equipment. The value of Dutch exports of fuels and lubricants actually falls, while it is flat for Germany. See Figure I-3 for the Netherlands and Figure I-4 for Germany.

Figure I-3:
Figure I-3:
Figure I-3:

TOT, the Netherlands

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

Figure I-4:
Figure I-4:
Figure I-4:

TOT, Germany

Citation: IMF Staff Country Reports 2008, 172; 10.5089/9781451829594.002.A001

24. Summarizing, overall, the Netherlands adapts less quickly to upward output pressures due to strong world demand. Following an increase in demand-driven TOT, the Netherlands is relatively less flexible to adjust than Germany; it has to undergo a larger change in quantities as prices and wages seem more rigid downward.

E. Conclusion and Policy Implications

25. The Dutch economy is significantly affected by economic activity in the rest of the world. In recent years, its export performance relative to a major trading partner, Germany, has deteriorated somewhat. The question posed in this paper is whether the Netherlands may be suffering from a competitiveness problem. The short answer is no. However, the analysis suggests that while the loss of market share in the 2000s has been contained, competitiveness problems may appear in the near future to the extent that labor, product and service markets require more flexibility to cope with foreign and domestic shocks in an increasingly globalized economy.

26. Export competitiveness is not a problem so far. However, the recent underperformance of the Dutch economy in certain products, and regarding some export market destinations, does not seem to be related to the economy’s relative cyclical position, but to the trend growth of its exports. While the Dutch trend overall export growth is somewhat below Germany’s in the 2000s, the Netherlands has preserved its relative advantage in terms of trend export growth vis-à-vis Germany in animal and vegetable oils, fats and waxes. It lost its traditional lead over Germany in exports of manufactured goods, machinery and transport equipment. So policies to prevent erosion of the competitive position of the Dutch economy may be required.

27. The analysis indicates that the Netherlands is relatively more exposed to supply-driven shocks; Germany is instead more exposed to demand-driven shocks. Following an increase in ULCM, the Netherlands is less flexible in adjusting than Germany, especially if the increase in ULCM is due to a fall in productivity (namely, supply-driven shock). The Netherlands adjusts relatively less via price and wage changes, and more via employment and output changes. The same features are also evident when the two countries are faced with an upward, supply-driven, TOT shock. Finally, the Netherlands profits relatively less from a demand-driven increase in the TOT; accordingly, it suffers a lower fall in output following a demand-driven increase in ULCM.

28. It is difficult to overestimate the relevance of product and factor markets flexibility for the open Dutch economy. The importance of trade flows and relative price changes in the international transmission of disturbances, as well as the policy constraints imposed by the euro area, highlight the relevance of product and factor markets flexibility. The Netherlands will benefit from further structural reforms that increase good, service, and labor markets flexibility. In particular, from policies that boost productivity via research and development, that reduce its relatively high EPL, and raise labor market participation.

Table I-3.

Forecast Error Variance of the Common Components of Netherlands Variables Explained by the Supply and Demand Shock to Unit Labor Costs in Manufacturing, 1981-2006 1/

article image

Forecast horizon is 20 quarters and refers to the levels of the series. Confidence intervals are constructed using bootstrapping methods. The bootstrap was made up of 500 draws. The bootstrap is done with the objective of removing the possible bias in the VAR coefficients which can arise from the small sample size.

Table I-4.

Forecast Error Variance of the Common Components of Germany Variables Explained by the Supply and Demand Shock to Unit Labor Costs in Manufacturing, 1981-2006 1/

article image

Forecast horizon is 20 quarters and refers to the levels of the series. Confidence intervals are constructed using bootstrapping methods. The bootstrap was made up of 500 draws. The bootstrap is done with the objective of removing the possible bias in the VAR coefficients which can arise from the small sample size.

Table I-5.

Forecast Error Variance of the Common Components of Netherlands Variables Explained by the Supply and Demand Shock to Terms of Trade, 1981-2006 1/

article image

Forecast horizon is 20 quarters and refers to the levels of the series. Confidence intervals are constructed using bootstrapping methods. The bootstrap was made up of 500 draws. The bootstrap is done with the objective of removing the possible bias in the VAR coefficients which can arise from the small sample size.

Table I-6.

Forecast Error Variance of the Common Components of Germany Variables Explained by the Supply and Demand Shock to Terms of Trade, 1981-2006 1/

article image

Forecast horizon is 20 quarters and refers to the levels of the series. Confidence intervals are constructed using bootstrapping methods. The bootstrap was made up of 500 draws. The bootstrap is done with the objective of removing the possible bias in the VAR coefficients which can arise from the small sample size.

Appendix I

29. Let us assume that Xt is an I(1) process with ΔXt = νt such that νt has a Wold representation. The spectral density of νt is fνν(λ)>0, for all λ. The discrete Fourier transform of Xt for λt ≠ 0:

wx(λs)=11−eiλxwv(λs)−eiλx1−eiλx(Xn−X0)n1/2,

where λs=2πsn,s=0,1,…,n−1,

article image
are the fundamental frequencies. The second term makes it clear that the Fourier transform is not asymptotically independent across fundamental frequencies because the second term is a deterministic trend in the frequency domain with a random coefficient (Xn−X0)n1/2
article image
. Unless that term is removed, it will produce leakages into all frequencies λt ≠ 0:, even in the limit as n → ∞. Sacrificing a single observation, instead of estimating the random coefficient a-la-Hannan (1970), Corbae and Ouliaris (2006) show that by imposing that (Xn – X1) = (Xn – X0) will produce an estimate that will have no finite sampling error, has superior endpoint properties, and has much lower mean-squared error than popular time-domain filters such as HP or B-K. In addition, in contrast to B-K, it is consistent. This is the ideal band-pass filter used in the paper.

Appendix II

30. This study uses a large dimensional approximate dynamic factor model in the tradition of Stock and Watson (1998 and 2002). In contrast to the models of Sargent and Sims (1977) and Geweke (1977), it admits the possibility of serial correlation and weakly cross-sectional correlation of idiosyncratic components, as in Chamberlain (1983) and Chamberlain and Rothschild (1983). Similar models have recently been used by Giannone, Reichlin, and Sala (2002), Forni and others (2005), and Eickmeier (2007).

31. A vector of N time series Yt = (ylt, y2t,…,yNt)’ with T observations can be represented as the sum of two latent components, a common component Xt = (xlt, x2t,…,xNt)’ and an idiosyncratic component Ξt = (ε1t, ε2t,…εNt)’

Yt=Xt+ΞtYt=CFt+Ξt(1)

where Ft = (flt, f2t,…,frt)’ is a vector of r common factors, and C = (c’1,c’2,…,c’N)’ is a N × r matrix of factor loadings, with r<<N13 The common component Xt, which is a linear combination of common factors, is driven by few common shocks, which are the same for all variables. Nevertheless, the effects of common shocks differ from one variable to another due to different factor loadings. The idiosyncratic component is driven by idiosyncratic shocks, specific to each variable. The static factor model used here differs from the dynamic factor model in that it treats lagged or dynamic factors Ft as additional static factors. Thus, common factors include both lagged and contemporaneous factors.14

32. Using the law of large number (as N → ∞), the idiosyncratic component, which is weakly correlated by construction, vanishes; and therefore, the common component can be easily estimated in a consistent manner by using standard principal component analysis. The first r eigenvalues and eigenvectors are calculated from the variance-covariance matrix cov(Yt) and define the N × r matrix V; and since the factor loadings C = V, equation (1) becomes,

Ft=V′Yt,(2)

and the common component X, can be written as,

Xt=VV′Yt,(3)

From (1), the idiosyncratic component is,

Ξt=Yt−Xt,(4)

33. From all the more or less formal criteria to determine the number of static factors r, the Bai and Ng (2002) information criteria was followed. As in Forni and others (2005), Ft was estimated by an autoregressive representation of order 115:

Ft=BFt−1−ut,(5)

where B is a r xr matrix and ut a r×t vector of residuals.

34. Once a decision is taken on the process followed by the common factors, structural shocks have to be identified by focusing on the reduced form VAR residuals of (5). Following Eickmeier (2007), the identification scheme has three steps. First, maximize the explained variance of the forecast error of the chosen variable and calculate impulse-response functions. Of interest here are unit labor costs in manufacturing (ULCM) and terms of trade (TOT). So, using ULCM as an example, a few major shocks driving them are identified.16 This implies maximizing the explanation of the chosen variance of the k-step ahead forecast error of ULCM with a reduced number of shocks.17 To this end, k–step ahead prediction errors ut are decomposed into k mutually orthogonal innovations using the Cholesky decomposition of the variance-covariance matrix of the ut residuals. The lower triangular Cholesky matrix A is such that ut = Avt and E(vtvt′)=I

article image
. Hence,

cov⁡(ut)=AE(vtvt)A′=AA′.(6)

35. The impulse-response function of yit for the identified shock in period k is obtained as follows:

Rik=CiBkA,(7)

with ci the ith row of factor loadings of C and with a corresponding variance-covariance k matrix Σj=0kRijRij′

article image
for yit or the k–step ahead prediction error of yit.

36. Second, the identified shocks are assumed to be linearly correlated to a vector of fundamentals. These fundamental forces ωt =(ωlt,ω2t,…, ωrt)’ behind Dutch ULCM are correlated to the identified shocks through the r × r matrix Q18. Thus,

vt=Qωt,(8)

37. The intuition behind the procedure is to select Q in such a way that the first shock explains as much as possible of the forecast error variance of the Netherlands’ ULCM common component over a certain horizon k, and the second shock explains as much as possible of the remaining forecast error variance. Focusing on the first shock, the task is to explain as much as possible of its error variance

σ2(k)=Σj=0k(Rijql)(Rijql)′,(9)

where i is, in our example, the Dutch ULCM, and ql is the first column of Q. The column ql is selected in such a way that q1′σ2q1

article image
is maximized, that is

σ2(k)=Σj=0k(Rijq1)(Rijq1)′=q′1Sikq1

where Sik=Σj=0k(k+1−j)Rij′Rij,

article image

38. The maximization problem subject to the side constraint q’1q1 = 1, can be written as the Lagrangean,

L=q1′Sikq1−λ(q1′q1−1)(10)

where λ is the Lagrange multiplier. From (10), q1 is the first eigenvector of Sik with eigenvalue λ and, therefore, the shock associated with q1 is the first principal component shock. Q is the matrix of eigenvectors of S, (q1, q2, …, qr), where ql (l=1,…,r) is the eigenvector corresponding to the lth principal component shock. Along the lines of Uhlig (2003), Eickmeier (2007), and Altig and others (2002), in this paper it is posed: k = 0 to k = 19, i.e., five years, which covers short- as well as medium-run dynamics.

39. Up to now, the principal component orthogonal shocks are identified up to a rotation using a Monte Carlo technique. If two shocks are identified, for example, following Canova and de Nicoló (2003), the orthogonal vector of fundamental forces ωt = (ωlt,ω2t)’ is multiplied by a 2 × 2 orthogonal rotation matrix P of the form:

p=[cos(θ)−sin(θ)sin(θ)cos(θ)],

where θ is the rotation angle; θ ∈ (0,π), produces all possible rotations and varies on a grid. If θ is fixed, and q=5, there are q(q– 1)/2 bivariate rotations of different elements of the VAR. Following the insights of Sims and Zha (1999), and as in Peersman (2005), Canova and de Nicoló (2003), Eickmeier (2007), Kabundi and Nadal De Simone (2007), the number of angles between 0 and x is assumed to be 12—as explained in footnote 16, this paper uses a grid of 30 degrees. This implies 6, 191, 736, 421x1010 (1210) rotations. Hence, the rotated factor wt = Pωt still explains in total all the variation measured by the first two eigenvalues. This way the two principal components ωi are associated to the two structural shocks wi through the matrix P, and the impulse-response functions of the two structural shocks on all the fundamental forces can be estimated.

40. A sign-identification strategy is followed to identify the shocks. The method was developed by Peersman (2005). This strategy imposes inequality sign restrictions on the impulse response functions of variables based on a typical aggregate demand and aggregate supply framework.19 Only those rotations among all possible q × q rotations that have a structural meaning are chosen.20 The following table displays the sign restrictions for the identification of shocks that are imposed contemporaneously and during the first year after the shock.

article image

References

  • Altig, D., L.J. Christiano, M. Eichenbaum, and J. Linde, 2002, “Technology Shocks and Aggregate Fluctuations,” manuscript.

  • Bai, J., and S. Ng, 2002, “Determining the Number of Factors in Approximate Factor Models,” Econometrica, 70(1), pp. 191–221.

  • Bordo, M.D., and T. F. Helbling, 2004, “Have National Business Cycles become more synchronized?” in Siebert, Horst (ed.) Macroeconomic Policies in the World Economy, Springer Verlag (Berlin and Heidelberg).

    • Search Google Scholar
    • Export Citation

  • Brooks R., K. Forbes, and A. Mody, 2003, “How strong are Global Linkages?” Manuscript, http://www.imf.org/external/np/res/seminars/2003/global/pdf/over.pdf.

    • Search Google Scholar
    • Export Citation

  • Canova, F. and M. Ciccarelli, 2006, “Estimating Multi-Country VAR Models,” ECB Working Paper 603.

  • Canova, F., M. Ciccarelli, and E. Ortega, 2007, “Similarities and Convergence in G-7 Countries,” Journal of Monetary Economics, 53(3), pp. 850–878.

    • Search Google Scholar
    • Export Citation

  • Canova, F., and G. de Nicoló, 2003, “On the Sources of Business Cycles in the G-7,” Journal of International Economics, 46, pp. 133–66.

    • Search Google Scholar
    • Export Citation

  • Chamberlain, G., 1983, “Funds, Factors, and Diversification in Arbitrage Pricing Models,” Econometrica 51, pp. 1281–304.

  • Chamberlain, G., and M. Rothschild, 1983, “Arbitrage, Factor Structure and Mean-Variance Analysis in Large Markets,” Econometrica 51, pp. 1305–24.

    • Search Google Scholar
    • Export Citation

  • Ciccarelli, M. and A. Rebucci, 2006, “Has the transmission mechanism of European monetary policy changed in the run-up to EMU?” European Economic Review, 50(3), pp. 737–776.

    • Search Google Scholar
    • Export Citation

  • Corbae, D., and Ouliaris, S., 2006, “Extracting Cycles from Nonstationary Data,” Econometric Theory and Practice: Frontiers of Analysis and Applied Research, Cambridge University Press, D. Corbae, S. Durlauf and B. Hansen, eds.

    • Search Google Scholar
    • Export Citation

  • Daal, Wendly, Francisco Nadal De Simone, and Sibel Yelten, 2006, Kingdom of the Netherlands—Netherlands: Selected Issues, IMF Country Report No. 06/284 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

  • Eickmeier, S., 2007, “Business Cycle Transmission from the U.S. to Germany—A Structural Factor Approach,” European Economic Review, 51, pp. 521–551.

    • Search Google Scholar
    • Export Citation

  • Eickmeier, S., and J. Breitung, 2006, “How synchronized are new EU member states with the euro area? Evidence from a structural factor model,” Journal of Comparative Economics, Vol. 34(3), pp. 538–563.

    • Search Google Scholar
    • Export Citation

  • Forni, M., D. Giannone, M. Lippi, and L. Reichlin, 2005, “Opening the Black Box: Structural Factor Models with Large Cross-Sections,” manuscript.

    • Search Google Scholar
    • Export Citation

  • Geweke, J., 1977, “The Dynamic Factor Analysis of Economic Time Series,” in Latent Variables in Socio-Economic Models; edited by D.J. Aigner and A.S. Golberger (Amsterdam: North-Holland), p. 19.

    • Search Google Scholar
    • Export Citation

  • Giannone, D., L. Reichlin, and L. Sala, 2002, “Tracking Greenspan: Systematic and Unsystematic Monetary Policy Revised,” CEPR Working Paper 3550.

    • Search Google Scholar
    • Export Citation

  • Hannan, E.J., 1970, “Multiple Time Series”, New York: Wiley.

  • Hofman, David, Francisco Nadal De Simone, and Mark Walsh, 2005, Kingdom of the Netherlands—Netherlands: Selected Issues, IMF Country Report No. 05/225 (Washington: International Monetary Fund).