European Economic Review 49 (2005) 1543 – 1569 www.elsevier.com/locate/econbase

The monetary transmission mechanism: Evidence from the industries of &ve OECD countries Luca Dedolaa , Francesco Lippia; b;∗ a Bank

of Italy, Rome, Italy London, UK

b CEPR,

Received May 2002; accepted October 2003

Abstract This paper studies the monetary transmission mechanism using disaggregated industry data from &ve industrialized countries. Our goal is to document the cross-industry heterogeneity of monetary policy e2ects and relate it to industry characteristics suggested by monetary transmission theories. Sizable and signi&cant cross-industry di2erences in the e2ects of monetary policy are found. Such di2erences swamp the hardly detectable cross-country variability. Sectoral output responses to monetary policy shocks are systematically related to the industry output durability, &nancing requirements, borrowing capacity and &rm size. These &ndings are consistent with a quantitatively non-negligible role of &nancial frictions in the monetary transmission. c 2003 Elsevier B.V. All rights reserved.  JEL classication: E52; E32; G32 Keywords: Monetary policy transmission; Balance sheet data

1. Introduction The study of the macroeconomic e2ects of monetary policy has long been the subject of a vast and growing literature, stemming from the enduring contribution of Friedman and Schwartz (1963). In line with that tradition, most empirical studies rely on aggregate data. This paper shows that integrating the aggregate evidence with information based on disaggregated data provides useful information. We report new evidence on the monetary transmission mechanism based on the e2ects of monetary policy shocks on the industrial activity of 21 manufacturing sectors in &ve OECD countries (France, Germany, Italy, the UK and the US). The goal is twofold. First, we document the cross-industry heterogeneity of the output response to unanticipated monetary policy. ∗

Corresponding author. Research Department, Banca d’Italia, via Nazionale 91, 00184 Roma, Italy. Tel.: +39-0647922580; fax: +39-0647923723. E-mail address: [email protected] (F. Lippi). c 2003 Elsevier B.V. All rights reserved. 0014-2921/$ - see front matter
doi:10.1016/j.euroecorev.2003.11.006

1544

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

Second, we provide evidence that links these responses in a systematic fashion to industry characteristics suggested by monetary transmission theories. The stylized facts that emerge from this analysis are useful for economists seeking to build better models of the monetary transmission mechanism. Disaggregated data (at the industry-level) can be more useful in understanding the monetary transmission mechanism than aggregate data because of two reasons. First, several determinants of monetary policy e2ectiveness suggested by economic theory (e.g. interest-rate demand sensitivity, &nancial requirements of the production process, &rm access to external funds) vary more across sectors within a country than across the aggregate data of developed countries. This suggests that monetary policy should have important distributional e2ects, which can most easily be detected by exploiting the wide cross-sectional variation in disaggregated data. For instance, the di2erent impacts of monetary policy on the spending components of output (e.g. investment and durable versus non-durable consumption) have been documented by Bernanke and Gertler (1995). The information provided by this heterogeneity, which may be useful in understanding the monetary transmission mechanism, is lost with aggregation. Second, the cross-industry dimension allows us to make progress on some identi&cation problems beleaguering the study of the monetary transmission, because it naturally o2ers a richer set of controls. For instance, a common strategy to test for the e2ects of credit market frictions consists in contrasting small and large &rms’ responses to a monetary shock based on the presumption that the former are more likely to be constrained by such frictions, e.g. Kashyap and Stein (1994) and Gertler and Gilchrist (1994). However, the di2erential reactions of small and large &rms might reHect that small &rms are concentrated in cyclically sensitive industries (Eichenbaum, 1994). By exploiting cross-industry observations, drawn from a number of countries, we are able to relate the intensity of the output response of a given industry to sectoral features, e.g. &rm-size, while controlling for other characteristics (e.g. output durability, &nancing requirements and industry &xed e2ects) that might explain the size distribution of &rms across sectors. We begin by measuring the e2ects of unanticipated monetary policy on industrial output by means of a structural VAR that is applied to 21 manufacturing industries in each of the &ve countries considered using monthly data. By focusing on the effects of policy shocks, the VAR approach is well suited to analyzing questions on the monetary transmission. Moreover, its widespread use makes our results comparable to previous studies. The results of this measurement exercise highlight signi&cant cross-industry di2erences of policy e2ects. A clear &nding emerges: the cross-industry di2erences are much larger than the cross-country di2erences. A decomposition of the industry e2ects into industry-speci&c and country-speci&c components reveals that the former component explains almost all of the variability, leaving basically no role to “national” factors. This con&rms the importance of disaggregated data for the study of the transmission mechanism. 1 1

A recent body of empirical work, produced by economists of the Eurosystem has, in the same spirit of this study, investigated micro data on banks and &rms in order to gain insights on the transmission mechanism (see Angeloni et al., 2002).

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1545

After documenting the industry e2ects of monetary policy we use an original &rmlevel database to measure some industry features which, according to monetary transmission theories, are likely to a2ect the size of the policy e2ect. We then analyze the relationship between the industry e2ects of monetary policy and these industry features. 2 The analysis reveals that the e2ects are systematically related to factors that a2ect both industry demand and supply sensitivity to interest rate changes, namely the industry output durability, &nancing requirements (i.e. short-term debt and working capital) and borrowing capacity (&rm size, leverage and interest rate burden). 3 The latter two classes of features are suggested by theories that emphasize the importance of the “user cost of capital”—a cost-channel e2ect of monetary policy—and of &nancial frictions, highlighting a possible ampli&cation of the traditional e2ects. The quantitatively signi&cant role of the latter features is of interest because, as e.g. Bernanke (1993) has argued, it is not the existence of a credit channel e2ect that is in doubt but rather its quantitative importance in the overall context of policy transmission. 4 Our analysis is closely connected to Barth and Ramey (2001) and Peersman and Smets (2002). The &rst paper &nds signi&cant di2erential e2ects of monetary policy shocks across the US manufacturing industries and shows that these di2erences can, in turn, be traced back to di2erential “cost” and “demand” e2ects of monetary policy. The authors explore the link between a measure of industry interest expense and the response of industry prices in two cross-sectional snapshots of the data, the 1974 and 1990 recessions. In this paper, we o2er more systematic evidence linking the di2erences recorded across industries to a wider set of industry characteristics measured over a longer time period. Peersman and Smets (2002) estimate the e2ects of monetary policy on 11 industries of seven euro area countries, also &nding considerable cross-industry heterogeneity in policy sensitivity that is statistically related to di2erences in output durability, &nancial structure and &rm size. Di2erently from our approach, they focus on the e2ects of monetary policy shocks common across countries and the degree of asymmetry of industry policy e2ects across recessions and booms. Our paper is also related to Carlino and DeFina (1998), who explain the di2erential e2ects of monetary policy shocks across US regions in terms of the concentration of small &rms (taken as a measure of a “credit channel” e2ect) and the share of manufacturing in total production (accounting for the traditional “interest rate channel”). Their identi&cation of these two channels of monetary transmission rests on the presumption of “credit constrained small &rms” and “di2ering interest rate elasticities of industries” (Carlino and DeFina 1998, p. 572). As argued above, the availability of industry-level data allows us to scrutinize the determinants of the policy e2ects in a more direct way.

2 To fully exploit the advantages of disaggregated data one could, in principle, estimate the policy e2ects using &rm-level, rather than industry-level, data. This is prevented, however, by the yearly frequency and shorter time-span availability of the former. 3 See Section 4.1 for a de&nition of these variables. 4 Interestingly, the issue of whether the traditional interest rate channel alone, without &nancial frictions or credit constraints, can explain the stylized facts of the monetary transmission in the euro area is central to the work of the Eurosystem Monetary Transmission Network (see Angeloni et al., 2002).

1546

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

The paper is organized as follows. In the next section we present the methodology used to identify monetary policy shocks in the &ve countries at the aggregate level. In Section 3 the method is extended to study the industry responses to a policy shock. The linkages between the industry e2ects and some sectoral features suggested by theory are discussed in Section 4. Section 5 concludes. 2. Measuring the eects of a monetary policy shock To measure the output e2ects of exogenous monetary policy shocks, we use structural vector auto regression (SVAR). The multivariate approach of SVARs allows us to estimate exogenous monetary policy shocks while controlling for the systematic feedback between monetary policy and the main macroeconomic variables. While the impulse responses generated by the SVAR are not an estimate of the total e2ects of monetary policy, the exogeneity of the monetary policy shock makes SVAR particularly appealing to test hypotheses on the monetary transmission mechanism. Moreover, despite the ongoing debate on their usefulness, 5 their widespread use makes our results comparable to previous studies. The identi&cation method used here relies on the recursiveness assumption presented in Christiano et al. (1999). The main reason for adopting this scheme is its simplicity, which makes it a natural starting point. Obviously, simplicity also raises the question of the robustness of our &ndings. We addressed this issue in two ways. First, we estimated the benchmark VARs over a shorter subsample (1983.1–1997.3); second, we explored the robustness of our results introducing additional variables in the VAR and using alternative identifying assumptions. The results of these robustness exercise are discussed in Sections 2.1 and 3.1. In essence, the recursiveness assumption amounts to dividing the VAR variables into two sets: on one hand, those to which monetary policy reacts contemporaneously (but which respond to policy with a delay); on the other hand, those that the central bank observes with a lag (but which are immediately a2ected by policy). An appealing feature of the recursive approach is that the ordering of the variables preceding and following the monetary policy instrument does not inHuence the measurement of their responses to the monetary policy shock. Our starting point is the estimation of &ve benchmark country VARs (for France, Germany, Italy, the UK and the US) using monthly data for the 1975 –1997 period. For all countries, it is assumed that the operating instrument of monetary policy is a short-term interest rate, as is common in the literature. 6 Lag-length selection in each VAR was chosen to ensure no autocorrelation in the residuals (Lagrange multiplier test). The speci&cation adopted for every country VAR and the ordering used in the recursive identi&cation of the monetary policy shock is summarized in Table 1. 5

See e.g. the exchange between Rudebusch (1998) and Sims (1998). On the use of the short-term rate as the operating tool of the G7 central banks, see Clarida et al. (1998). We use 3-month interest rates for all European countries and the Federal Fund rate for the US; all data were taken from the OECD database “Main Economic Indicators”; sectoral data on output are from the OECD database “Indicators of Industrial Activity”. The sample period runs from January 1975 to March 1997; in a few industries, data are only available since the early 1980s. 6

France

Germany

Italy

UK

USA

Industrial production Consumer price index Commodity price index Short-term rate (3-month) Money (M3) Exchange rate AICa 3 LMb 5

Industrial production Consumer price index Commodity price index Short-term rate (3-month) Money (M3) Exchange rate 2 2

Industrial production Consumer price index Commodity price index Short-term rate (3-month) Money (M3) Exchange rate 3 5

Industrial production Consumer price index Commodity price index Short-term rate (3-month) Money (M3) Exchange rate 2 2

Industrial production Consumer price index Commodity price index Short-term rate (FF rate) Money (M1)

# lags 5

2

5

2

4

4 4

Note: Estimated on monthly data (from the OECD: “Main Economic Indicators”) over the sample period 1975.1–1997.3 and monthly dummies (all data, except the short-term rate, are in log (levels) not seasonally adjusted). Data for France begin in 1980.1. a Lags suggested by the Akaike information criteria. b Lags suggested by the Lagrange Multiplier test.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

Table 1 Country VARs

1547

1548

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

We follow Christiano et al. (1999) in the speci&cation of a parsimonious 5 variable VAR for the US, which includes industrial production, the consumer price index, a commodity price index, a short-term interest rate and a monetary aggregate. 7 In the recursive ordering, the &rst three variables enter the monetary authorities’ reaction function simultaneously (but respond to it with a lag). The monetary policy shocks thus obtained are basically equivalent to the regression residuals of the shortterm interest rate on the contemporaneous values of industrial production, the consumer and commodity price indices and the lagged values of all the VAR variables. For the US, the impulse response functions resulting from this identi&cation scheme, which appear in the &rst column of Fig. 1, show that following a monetary tightening there is a temporary reduction of industrial activity and the money stock. These patterns are consistent with theoretical a priori about the long-run neutrality of money and the short-run e2ectiveness of policy. The VAR speci&cation for the European countries also includes the exchange rate, according to the presumption that this variable is more relevant in European countries than in the US, possibly because of the greater degree of openness of the European economies. The exchange rate enters the recursive ordering after the short-term rate, thus assuming that monetary policy does not respond contemporaneously to it. 8 Fig. 1 illustrates the impulse responses of the main variables included in the VARs, along with 95 percent con&dence bands. 9 An unexpected increase of the short-term interest rate causes e2ects on the other variables that are qualitatively similar across countries and broadly in line with previous studies (e.g. Sims, 1992). The policy shock is highly persistent: in all countries, the interest rate is signi&cantly above zero in the year following the shock. Industrial production begins to decline after a few months, and bottoms after 18–24 months; about three years later it eventually returns to the level prevailing before the shock. A higher interest rate leads to a contraction in monetary aggregates in the US and Italy and to an exchange rate appreciation in France and Germany. The price level does not show clear signs of reduction, which is a common &nding in the SVAR literature and is usually interpreted as supporting the presence of nominal rigidities. 10 Quantitatively, the e2ect of a monetary policy shock on industrial 7 Unfortunately, data on the service sectors were not available. However, the literature on the monetary transmission has shown that manufacturing output is a very close proxy for GDP at the monthly frequency (see Christiano et al., 1999). 8 The inclusion of the exchange rate among the variables entering contemporaneously in the monetary authority information set (but responding with a lag) helps to deal with the so-called “price puzzle” (i.e. the fact that the price level increases after a restrictive monetary policy shock). This assumption neglects the simultaneous relation between the interest rate and the exchange rate, central to non-recursive identi&cation schemes (e.g. Sims and Zha, 1995). The monthly data used in our analysis, however, may justify the assumption of a non-simultaneous policy reaction to the exchange rate, even when some kind of exchange rate target zone is adopted. Under this premise, the monetary authority reacts to lower frequency movements of the exchange rate, as summarized by its lags in the policy equation. 9 Con&dence bands are computed with Monte Carlo simulations assuming that innovations are asymptotically normally distributed. 10 A small price response (or even a positive one) can alternatively be explained in terms of the supply side e2ects of monetary policy, studied by Christiano et al. (1997) and Barth and Ramey (2001).

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1549

Fig. 1. Responses of the main macro variables to a monetary policy shock (± 2 standard error bands).

production is somewhat larger in the continental European than in the other countries: the maximum impact, measured by the semi-elasticity of output to the interest rate shock, is about 0.9 percent in France and Italy, 0.8 in Germany, 0.7 in the USA and 0.5 in the UK (see the last row of Table 3).

2.1. Robustness The estimates of the country VAR showed that the interest rate equations display no serial correlation but, as is common in the VAR literature, the normality of the interest rate equation residuals is rejected. No parameter instability emerges—when we split the sample in two—except for Germany and the US. For the latter country it is well-known that this is related to the di2erent operating procedures adopted in the early

1550

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

eighties. 11 However, Christiano et al. (1999) show that the identi&cation scheme we adopted in this paper yields estimates of the output e2ect of monetary policy that are robust to the inclusion of other variables and across di2erent subsamples. As a matter of fact, when we re-estimated output impulse responses to a monetary policy shock for the US in a sample starting in January 1983, we found no sizeable change in the output e2ects of policy shocks. For the European countries, we carried out robustness checks along the following two dimensions. First, as for the US, we re-estimated our benchmark model over a shorter sample, beginning in January 1983. This experiment was aimed at assessing the robustness of the estimates with respect to changes in the European monetary regime (Giavazzi and Spaventa (1990) suggest that this time span includes the very similar “hard ERM” and “pre-EMU” periods). Again, the real e2ects of monetary shocks were found to be basically insensitive to this change. Second, we adopted an alternative identi&cation procedure introducing additional variables in the VAR (“augmented VAR” in what follows). For France and Italy, the German short-term rate was added “before” the domestic rate, to allow for the possibility of a contemporaneous inHuence of German monetary policy on the policy of these countries. In the VAR for the UK and Germany, we added a short-term dollar rate; &nally, a “uni&cation dummy” was used for Germany to control for the possible inHuence of the major events that unfolded at the end of 1990. We found that these changes in the identi&cation scheme have a rather limited inHuence on the absolute size of the policy e2ect on industrial output (a possible exception being France). More importantly for the purpose of this paper, they have no detectable inHuence on the sectoral e2ects, which are the core of our investigation (see the next section, particularly Section 3.1 and Table 6): the relative intensity of the e2ects of policy shocks across industries is not signi&cantly a2ected by the change in the VAR scheme. 12 As discussed more fully in Section 4, both experiments led to similar estimates of industry e2ects and almost identical conclusions about the relationships between these e2ects and their microeconomic determinants.

3. Industry eects of monetary policy In this section, we employ the recursive identi&cation scheme presented above to measure the industry e2ects of monetary policy. We estimate a VAR in which the production index of industry j in country i is added as the last variable to the VAR of country i presented before (see Table 1). A lack of data forces us to con&ne the analysis to di2erences in the output e2ects, overlooking possible di2erences in pricing behavior. The index j spans 21 manufacturing industries. These are listed in Table 2 11

Bernanke and Mihov (1998) &nd two regime switches in the sample period, one in late 1979, the other during 1982. 12 Similar results were obtained in a previous version of this paper in which the 5-lag VAR was used for all the European countries (see Dedola and Lippi, 2000). The results based on a shorter sample and on the “augmented VAR” are available from the authors upon request.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1551

Table 2 Manufacturing industries (shares of total industrial production) ISIC code

Industries

France

Germany

Italy

UK

USA

311 313 314 321 322 323 324 33 3411 342 351+352 353 36 362 371 372 381 382 383 3841 3843

Food Beverages Tobacco Textiles Wearing apparel Leather Footwear Wood and furniture Paper Printing and publishing Industrial chemicals Petroleum re&neries Non-metallic mineral Glass Iron and steel Non-ferrous metals Fabricated metal products Machinery and equipment Electrical machinery Ship building Motor vehicles

10.3 2.1 0.9 3.5 2.6 0.5 0.7 3.2 2.5 4.7 8.5 6.5 4.3 1.2 3.6 1.9 7.3 9.7 9.6 0.5 6.9

5.6 2.8 2.8 2.7 1.4 0.3 0.3 3.3 2.4 2.0 10.9 3.5 4.2 1.0 5.8 1.8 9.4 11.3 11.2 0.4 9.3

7.8 2.5 0.5 8.9 4.4 1.0 2.0 5.5 2.3 3.5 7:5∗ 0.9 7:2∗ 1.4 3.8 0.8 9.7 9.6 7.3 0:4∗ 4.9

9.6 3.0 1.1 3.7 2.2 0.3 0.7 3.0 3.2 7.1 11.3 1.5 3.8 0.7 3.6 1.3 6.0 11.8 9.1 1.2 5.5

7.8 1.4 1.5 3.0 2.3 0.2 0.3 4.7 4.2 6.4 10.0 1.7 2.8 0:8∗ 3.7 1.7 7.1 11.4 8.6 0.7 6.1

Source: OECD-STAN database; averages of annual data for the 1970 –1993 period. An asterisk indicates that monthly industrial production data are not available. That industry is thus excluded from VAR analysis of the corresponding country.

(according to a 3 or 4 digit ISIC code), which reports their percentage shares of total manufacturing output. 13 This VAR speci&cation implies that monetary policy does not respond simultaneously to industry-speci&c shocks, but it does not constrain to zero the simultaneous response of industrial output to policy shocks. It is reasonable to consider whether allowing for a simultaneous industry response is consistent with the assumption used in the identi&cation of the aggregate VAR that the contemporaneous aggregate output response is zero. A suQcient condition that there is no inconsistency is the empirical observation that the estimated simultaneous industry responses are generally not significantly di2erent from zero. Essentially identical results are obtained from an alternative identi&cation scheme that allows for a contemporaneous policy response to the industry output (see Section 4.1). The main output of our analysis is a set of 101 VARs and the associated impulse response functions, one for each of the 21 industries in each of the &ve countries (four industries lack data). For each of the &ve countries considered, Fig. 2 shows 13

The industries for which data are available account for about 90 percent of total manufacturing output in each of the countries considered. The monthly data used in the VAR are not available for all industries in some countries; these “missing” data are denoted by an asterisk in Table 2.

1552

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

Fig. 2. Industry impact of a contractionary monetary policy shock (Note: The industry impact is measured by the percent output reduction after an unanticipated interest rate increase (1 percentage point).)

the e2ects of a 1 percentage point increase in the interest rate on aggregate industrial production and on the output of &ve large industries—food, textiles, iron, machinery and motorvehicles—which represent about half of the total manufacturing output. Most industries display a u-shaped response to the shock. The erratic behavior during the &rst

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1553

6 months is never statistically signi&cant. 14 Within each country, industry responses di2er signi&cantly both qualitatively and quantitatively. In the food industry, the impact on production is often not signi&cantly di2erent from zero and never greater than that of the aggregate industrial production. On the contrary, the heavy industries (iron, machinery and motorvehicles) show a response to policy to a markedly greater degree than other industries. To quantify the output e2ects of monetary policy across industries and countries, we construct three summary measures of impact: the industry output elasticity to a 1 percentage point interest rate increase after 24 months; the maximum elasticity recorded between 12 and 36 months after the increase; the average elasticity recorded between 18 and 24 months after the increase (so that single “peaks” have less inHuence on the impact measure). These three measures are highly correlated, suggesting that the policy effects identi&ed by our analysis do not depend crucially on the particular measure of impact that is utilized. 15 The &rst two measures are reported in Table 3 for all countries. The impact of policy on industrial output is usually negative in all of the countries and in several cases it is statistically di2erent from zero at the 95% con&dence level. Visual inspection of Table 3 reveals that the largest (negative) impacts tend to be concentrated in the lower part of the table, where the “heavy” industries are located. In the US and France, the motorvehicles industry records the largest maximum impact (respectively, −2:4 and −2:1 percent). In all of the &ve countries, the “iron and steel” and “machinery and equipment” industries record impacts (both 24-month and maximum elasticity) that are clearly larger than those recorded by the aggregate industrial production indices (see Table 3 and Figure 3). At the other extreme, the impacts on the “food”, “wearing apparel” and “footwear” industry are much smaller than what is recorded for the aggregate industrial production. To analyze the extent to which the cross-industry e2ects of monetary policy are similar across countries, we construct two measures of the uniformity of the impacts between pairs of countries: a simple linear correlation coeQcient (Table 4, panel A) and the Spearman index of rank correlation (Table 4, panel B). 16 The results reported in Table 4 are based on the 24-month elasticity, analogous results are obtained from the maximum elasticity. First, it is apparent that no two countries show an “inverse” correlation of rankings (the correlation index is never signi&cantly smaller than zero). Rather, most of the correlations are signi&cantly larger than zero, suggesting a cross-country similarity in the cross-industry pro&le of policy e2ects. We further explore this issue by using a linear regression to decompose the impact of monetary policy in industry j of country i (call it ij ) into country- and industry-speci&c components. To this end, we estimate the equation (there are 101 ij estimates obtained 14

Standard error bands are not reported here for reasons of legibility. In each country, the cross-industry correlation between the maximum and the 24 month elasticity is greater than 0.9; that between the maximum and the 18–24 month elasticity is above 0.95 and that between the 24-month and the 18–24 month elasticity is greater than 0.98. 16 The rank correlation index between country i and country j is 1 if the rankings of the elasticity of Table 3 are identical (−1 if they are reversed). Under the null hypothesis of independence, the rank correlation is distributed with zero mean and variance 1=(n − 1) (where n is the number of observations). 15

1554

Table 3 Elasticity of industrial output to a monetary policy shock Industry

France

Germany

Italy

UK

USA

Maximum elasticity

24-Month elasticity

Maximum elasticity

24-Month elasticity

Maximum elasticity

24-Month elasticity

Maximum elasticity

24-Month elasticity

Maximum elasticity

Food Beverages Tobacco Textiles Wearing apparel Leather Footwear Paper Printing and publishing Industrial chemicals Petroleum re&neries Wood and furniture Non-metallic mineral Glass Iron and steel Non-ferrous metals Fabricated metal products Machinery and equipment Electrical machinery Ship building Motorvehicles

−0.01 −0.88 0.02 −0.34 0.01 −0.82 0.37 −0.37 −1.22 −0.04 −0.12 −1.33 −1.08 −0.51 −0.99 −0.62 −1.59 −1.72 −0.53 0.41 −1.98

−0.07 −1.20 −0.11 −0.62 −0.26 −1.83 0.18 −0.50 −1.29 −0.45 −0.44 −1.37 −1.08 −0.57 −1.69 −0.65 −1.80 −1.81 −0.59 −0.34 −2.13

−0.17 −0.12 0.00 −0.02 0.15 0.19 0.96 −1.17 0.02 −1.23 −1.46 −0.89 −0.54 −1.21 −0.79 −0.96 −1.19 −0.79 −0.77 1.75 −1.24

−0.31 −0.29 −0.40 −0.43 −0.66 −0.26 0.42 −1.37 −0.53 −1.37 −1.48 −1.25 −0.81 −1.21 −1.34 −1.66 −1.21 −0.92 −0.88 1.02 −1.24

−0.33 −0.09 0.67 −0.40 −0.20 0.53 −0.17 −0.88 −1.23 n.a. −0.69 −1.18 n.a. −0.41 −0.64 −0.72 −0.89 −1.37 −0.39 n.a. −0.43

−0.64 −0.59 0.38 −0.81 −0.23 0.05 −0.17 −1.83 −1.69 n.a. −0.74 −1.28 n.a. −0.57 −1.30 −1.55 −0.91 −1.65 −0.58 n.a. −1.34

−0.19 −0.45 −0.24 −0.67 −0.42 −1.41 −0.27 −0.21 −0.45 −0.82 −0.09 −0.79 −0.50 −0.62 −1.51 −0.34 −0.14 −1.40 −1.58 0.71 −1.54

−0.20 −0.47 −0.33 −0.67 −0.43 −1.43 −0.37 −0.21 −0.56 −0.84 −0.19 −0.81 −0.50 −0.64 −1.66 −0.47 −0.23 −1.40 −1.84 0.33 −1.57

−0.20 −0.31 0.01 −0.40 0.00 0.39 0.80 −0.35 −0.14 −0.62 −0.79 −0.58 −0.86 n.a. −1.41 −1.00 −0.83 −1.63 −0.85 0.20 −1.70

−0.21 −0.33 −0.12 −0.64 −0.25 0.22 0.61 −0.44 −0.25 −0.65 −0.87 −0.73 −0.90 n.a. −1.79 −1.05 −0.85 −1.63 −0.86 0.01 −2.39

Industrial production

−0.81

−0.87

−0.75

−0.78

−0.64

−0.92

−0.49

−0.50

−0.66

−0.70

Note: The 24-month elasticity is the percentage output change registered 24 months after a 1 percentage point increase in the short-term rate. The maximum elasticity is the smallest percentage output change recorded between 12 and 36 months after a 1 percentage point increase in the short-term rate.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

24-Month elasticity

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1555

Fig. 3. 24-Month output elasticity in selected industries. Table 4 Cross-country correlation of industry e2ects FRA (A) Correlation of 24-month elasticity GER 0.50 ITA 0.51 UK 0.57 USA 0.69

GER

ITA

UK

0.50 0.41 0.53

0.02 0.54

0.53

(B) Rank correlation of 24-month elasticity GER 0.40 ITA 0.61 0.55 UK 0.51 0.15 USA 0.66 0.76

−0.01 0.59

0.45

Note: Correlation for the industries where data are available for both countries (see Table 3). The rank correlation is measured by the Spearman’s correlation index.

from the industry VARs) ij =  + i + j + ij ;

(3.1)

where i is a country index (i=1; 2; : : : ; 5) and j is an industry index (j=1; 2; : : : ; 21). The constant term  measures the average policy impact across all sectors and countries; the i coeQcients measure the simple average (across industries) deviation from  of country i; the j coeQcients measure the simple average (cross-country) deviation from

1556

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

 of industry j. 17 The i measure the response heterogeneity that can be attributed to country-speci&c factors, after controlling for industry di2erences (identical across countries). Similarly, the j measures the response heterogeneity related to industry-speci&c factors, after controlling for country e2ects (identical across industries). The estimates of Eq. (3.1) are reported in Table 5, where both the 24-month and the maximum elasticity are used as impact measures. The estimated constant  from the 24-month elasticity equation (&rst column) indicates that an unexpected interest rate increase of 1 percentage point reduces industrial activity by 0.5 percent, in the average industry of the average country (0.8 percent using the maximum elasticity). Interestingly, none of the country &xed e2ect ( i ) is signi&cantly di2erent from zero in either regression. A simple Wald test on the coeQcients of these variables cannot reject the joint null hypothesis that these coeQcients are all equal to zero. This &nding suggests the absence of signi&cant country-speci&c di2erences in the transmission mechanism of monetary policy in the &ve countries considered. 18 Marked di2erences appear instead across industries: the heavy industries experience a large and signi&cant decline in activity (bottom part of Table 5), while industries producing non-durables (e.g. food, wearing apparel and footwear) record signi&cant smaller than average response to the monetary policy shock (i.e. a positive and often signi&cant j coeQcient). The results show that the cross-industry variability swamps the (hardly detectable) cross-country variability. Indeed, di2erences across industries are as large as 2 percentage points, more than 10 times greater than the maximum di2erence recorded across countries. 3.1. Robustness To verify the robustness of the measures of industry impacts presented in Table 3, we explored two main modi&cations to the above procedure for both the 24-month and the maximum elasticity. 19 First, using the same country VAR that was used before, we changed the identi&cation scheme by placing the industry output variable within the information set of monetary policy. Moving the industry output from “block B” (i.e. the set of variables that respond contemporaneously to policy) to “block A” (i.e. the variables that do not respond contemporaneously to policy), as done in this experiment, is useful to ensure the consistency requirement that neither the individual industry output nor its aggregate counterpart respond contemporaneously to a policy shock. As a matter of fact, given the quasi-zero contemporaneous responses to policy of industry output recorded under our benchmark identi&cation scheme (which places industry output in block B), it turns out that this modi&cation of the identi&cation scheme leads to basically identical values of the industry elasticities: the correlation coeQcient between the 24-month elasticity 17 Obviously, the and the coeQcients cannot be estimated independently because, being deviations, i j

both the industry e2ects and the country e2ects sum to zero (i.e. i i = j j = 0). Therefore, Eq. (3.1)

is estimated under the constraints: 5 = − i=5 i and 21 = − j=21 j . 18 Previous estimates derived from a common VAR scheme, reported in Dedola and Lippi (2000), delivered similar results. 19 We are grateful to the referees of this Journal for raising these issues.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1557

Table 5 Decomposition of industry responses by country and industry e2ects Dependent variable

Constant Dummies Country France Germany Italy UK US Industry Food Beverages Tobacco Textiles Wearing apparel Leather Footwear Paper Printing and publishing Industrial chemicals Petroleum re&neries Wood and furniture Non-metallic mineral Glass Iron and steel Non-ferrous metals Fabricated metal products Machinery and equipment Electrical machinery Ship building Motorvehicles

24-Month elasticity

Maximum elasticity

CoeQcient

CoeQcient

Standard error

Standard error

−0:53∗∗∗

0.05

−0:76∗∗∗

0.05

−0.10 0.08 0.10 −0.08 0.01

0.09 0.10 0.10 0.10 0.08

−0.12 −0.01 −0.05 0.07 0.11

0.10 0.09 0.10 0.09 0.08

0:35∗∗∗ 0.16 0:62∗∗∗ 0.17 0:44∗∗∗ 0.31 0:87∗∗∗ −0.42 −0.06 −0.07 −0.12 −0.10 −0.19 −0.15 −0:54∗∗∗ −0.20 −0.40 −0:85∗∗∗ −0.29 1:32∗∗∗ −0:85∗∗∗

0.10 0.11 0.13 0.10 0.10 0.34 0.24 0.14 0.22 0.26 0.27 0.29 0.13 0.20 0.15 0.15 0.25 0.15 0.19 0.31 −0.23

0:48∗∗∗ 0.19 0:65∗∗∗ 0.13 0:40∗∗∗ 0.11 0:90∗∗∗ −0.33 −0.11 −0.10 −0.08 0.02 −0.07 0.04 −0:79∗∗∗ −0.31 −0.24 −0:72∗∗∗ −0.19 1:01∗∗∗ −0:97∗∗∗

0.11 0.14 0.16 0.08 0.11 0.39 0.18 0.10 0.29 0.24 0.21 0.22 0.11 0.17 0.13 0.24 0.23 0.15 0.26 0.27 −0.24

No. of observations: 101 Note: Pooled (cross-section cross-country) least squares; White Heteroskedasticity-Consistent standard errors (in italics); *, ** and *** indicate rejection of the null hp of zero coeQcients at the 10, 5 and 1 percent level, respectively.

of Table 3 and the one calculated placing the industry output in block A is 0.998 (see the &rst cell of Table 6). Second, we modi&ed the country VARs of France, Germany and Italy by introducing a foreign policy rate in the information set of monetary policy (the “augmented VAR” described in Section 2.1). Moreover, a dummy variable was added to the VAR for Germany to control for reuni&cation event. As for the case of our benchmark VAR, 24-month elasticity measures were constructed using both a block-A and a block-B

1558

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

Table 6 Cross-correlations of alternative elasticity measure

Benchmark–Block A Augmented–Block B Augmented–Block A

Benchmark

Benchmark–Block A

Augmented–Block B

0.998 0.957 0.945

0.955 0.946

0.989

identi&cation scheme. The correlations between the di2erent elasticity measures are reported in Table 6. The exercise indicates that the correlation between the elasticity obtained from the augmented VAR and the benchmark measures of Table 3 is high, equal to 0.957. As before, changing the identi&cation scheme (i.e. moving the industry variable from block B to block A) appears to have a limited e2ect. 4. Industry features and monetary policy eects The heterogeneity of industry responses in all the countries raises the natural question of whether such di2erences are systematically related to sectoral features. Theoretical models of the monetary transmission help identifying such characteristics. The purpose of this section is to build proxies for some of these features and to relate them to the di2erent industry impacts documented above. 4.1. Measuring the industry determinants of monetary policy responses To construct these proxies, we use information drawn from Amadeus, a &rm-level database that contains yearly balance sheet information for major public and private European companies from the manufacturing industries considered earlier (ISIC 3/4 digits) for the period 1993–1997. 20 The &rms considered are markedly di2erent in terms of size (value added, number of employees) and access to capital markets (both listed and unlisted companies are included). Comparable data for the US were constructed using the Quarterly Financial Reports and the Employment Size of Firms. 21 To measure the industry demand sensitivity to the traditional interest-rate channel, we use a durability dummy, which identi&es industries producing durable goods. 22 The 20 The data in Amadeus provide information on the entire distribution of the industry features considered, such as mean and median. For the 21 industries of the four European countries studied here, the database has observations on about 42,000 &rms. The data are likely to be biased towards medium to large-sized &rms, because the companies surveyed in Amadeus must comply with at least one of the following criteria: (a) turnover greater than 12 million USD; (b) more than 150 employees; (c) total assets greater than 12 million USD. 21 Respectively from the US Bureau of Census and the OQce of Advocacy—US Small Business Administration. The indicators we construct from these database are population averages. 22 The industries are selected on the basis of the economic destination of production used in the national accounts statistics. According to this criterion, the industries producing “durable” output are denoted by the ISIC codes beginning with digits: 33, 36, 37, 38 (cf. Table 2). An alternative measure, which includes industries 34 and 35 (paper and chemicals) among the durable output producers, does not change the results.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1559

presumption is that this dummy captures the industry demand sensitivity to changes in interest rates, either because expenditures for non-durables is expected to Huctuate less or because it does not need to be &nanced over several periods. Therefore, we expect the output e2ect of monetary policy to be stronger in the industries producing durables. 23 Two additional sets of indicators are constructed to capture factors that a2ect industry heterogeneity via a supply side channel. The former includes measures of short-term debt (industry mean and median ratio of short-term debt to total debt) and two different proxies of the industry’s &nancing requirements (industry mean and median of &rms’ working capital and the industry index of external nancial dependence computed by Rajan and Zingales, 1998). These indicators relate to a supply channel of monetary policy: short-term debt and &nancial structure requirements may a2ect &rms’ responsiveness to policy shocks by a2ecting their production decisions impinging on the user cost of capital, as suggested not only by a classical interest rate channel but also quantitative limited participation models (e.g., Christiano, 1991; Fuerst, 1992). For instance, short-term debt is proportional to the impact of changes in the interest rate on the user cost. Working capital is constructed from the balance sheet data in Amadeus as the di2erence between current liabilities and current assets (divided by total liabilities), while the nancial dependence index is a measure of “the technological reason why some industries depend more on external &nance than others (Rajan and Zingales, 1998).” 24 Following these authors, we assume that these technological di2erences are common across countries, so that the US index is used for other countries too (also in interaction with a country index of &nancial development). Since all above variables measure the extent to which production decisions depend on interest rates, we expect the output e2ect of monetary policy to be stronger in industries with larger values for each of the above variables (i.e. a negative partial correlation coeQcient). The second class of indicators includes: rm size (mean and median number of employees per &rm in each sector), &nancial leverage (industry mean and median ratio of total debt to shareholders’ capital) and a measure of the incidence of interest rate expenditures on cash Hows, called the interest burden (industry mean and median ratio of interest rate payments to operating pro&ts). We interpret the &rst two variables as proxies for the borrowing capacity of &rms. Their use is consistent with quantitative general equilibrium models of &nancial frictions. For instance, Fisher (1999) shows that &rms that are credit constrained, due to asymmetric information problems, in the long run have a lower leverage ratio, are considerably smaller and employ less workers than non-credit-constrained &rms. We will thus interpret smaller &rm size and leverage

23 In the case of an interest rate increase, a larger output reduction was observed. Therefore, the expected sign of the partial correlation coeQcient between the estimated elasticities (see Table 3) and the durability dummy is negative. 24 The variable is de&ned as capital expenditures minus cash How from operations divided by capital expenditures and is constructed with data from large, publicly traded &rms (Rajan and Zingales, 1998, pp. 563–564). As argued by these authors, this is an advantage, since for this kind of &rms it is likely that the actual amount of external funds raised equals their desired amount.

1560

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

variables as indicators of more stringent credit constraints. 25 Hence, we expect an inverse relationship (i.e. a positive partial correlation coeQcient) between the e2ectiveness of monetary policy and the &rm size and leverage indicator (for a given level of interest burden). 26 Finally, the interest burden indicator, as suggested by Barth and Ramey (2001), is consistent with a cost-channel e2ect, whereby industries with higher interest expense are more likely to increase their relative price in response to a monetary tightening. However, it may also relate to &rms’ responsiveness to shocks occurring via a deterioration of their creditworthiness, as suggested by Bernanke and Gertler (1989) and Carlstrom and Fuerst (1997). Both considerations led us to expect that a higher interest rate burden increases the impact of monetary policy. The sources and de&nitions of all the variables used in the analysis are detailed in Appendix A. 4.2. Regression analysis This section reports the results of a regression analysis in which the output e2ects of monetary policy, measured by the elasticities of Table 3, are related to the mean industry characteristics presented above. Since these elasticities are averages of the industry behavior over the estimation period, the explanatory variables are also measured as averages over the available period. The use of averages reduces the possibility that the results depend on a particular outcome of the data in any given year. 27 All estimates, reported in Table 7, include country &xed e2ects to control for unobserved (industry invariant) country features. 28 The coeQcients of the industry explanatory variables are assumed to be common across country (this assumption is not rejected by the data). Regressions 1, 2, and 3, based on alternative elasticity measures, illustrate our main &ndings. Industries producing durable output and with higher short-term &nancing requirements, as measured by the working capital variable, show a signi&cantly greater output responsiveness to monetary policy shocks. Note also that while the working capital variable is highly signi&cant, the &nancial dependence measure taken from Rajan 25 This is also consistent with Giannetti (2000) who &nds that “the cost of debt is lower for more levered &rms” and that “more levered &rms are the ones with a higher share of long-term debt to total debt” (using information from Amadeus). Both &ndings lead her to conclude that high leverage is a signal of the ability to get loans at better terms. 26 Conversely, for a given level of leverage, we expect the e2ects to be greater if the interest burden (cost channel) is higher. In our sample, the correlation coeQcient between leverage and the interest rate burden is 0.5. 27 Due to limited data availability, the industry characteristics derived from Amadeus are measured toward the end of the sample period over which the VARs are estimated. Clearly, it would have been better to have them measured as averages at various points over the sample period. However, we have no evidence that these features have evolved over time in a signi&cant way. For instance, Barth and Ramey (2001) relate their industry price responses, based on VARs as well, with a single datapoint for interest expense, observed in a particular quarter. 28 Standard errors are calculated using the White heteroskedasticity consistent estimator (Greene, 2000, p. 463) that allows us to take account of the non-spherical disturbances typical of cross-section data without taking a stand on the source of this feature.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1561

Table 7 Industry determinants of monetary policy e2ects Explanatory variable

Dependent variable 24-Month elasticity

Maximum elasticity

24-Month elasticity (Augm. VAR)

24-Month elasticity

Eq. (1)

Eq. (2)

Eq. (3)

Eq. (4)

Eq. (5) (IV estimation)

−0:47∗∗∗ (0.14) −2:18∗∗∗ (0.85) −0.16 (0.27) −0.11 (0.51) 0:09∗∗∗ (0.04) 0:26∗∗∗ (0.11) −0.13 (0.09)

−0:49∗∗∗ (0.14) −2:11∗∗∗ (0.86) −0.06 0.27) −0.02 (0.52) 0:07∗∗ (0.04) 0:23∗∗ (0.11) −0.10 (0.09)

−0:37∗∗ (0.16) −2:05∗∗ (0.98) −0.16 (0.31) −0.24 (0.59) 0:10∗∗∗ (0.04) 0:26∗∗ (0.12) −0.10 (0.11)

−0:43∗∗∗ (0.09) −1:74∗ (1.07)

−0:49∗∗∗ (0.12) −1:98∗∗ (0.94)

(0.01 (0.27) 0:07∗∗ (0.03) 0:18∗∗ (0.09) −0.11 (0.08)

0.02 (0.06) 0:21∗ (0.14) −0.10 (0.12)

Industry dummies

−1:20∗∗ (0.53) −1:58∗∗∗ (0.50) −1:25∗∗ (0.64) −1:14∗∗ (0.51) −1:09∗∗∗ (0.28) —

−1:48∗∗∗ (0.53) −1:83∗∗∗ (0.50) −1:63∗∗∗ (0.65) −1:23∗∗∗ (0.51) −1:20∗∗∗ (0.28) —

−0.52 (0.61) −1:53∗∗∗ (0.58) −1:20∗ (0.75) −1:11∗∗ (0.59) −1:13∗∗∗ (0.32) —

−1:10∗∗∗ (0.43) −1:32∗∗∗ (0.49) −0.98 (0.53) −1:06∗∗∗ (0.39) −0:84∗∗∗ (0.25) Yes

−1:11∗∗∗ (0.36) −1:16∗∗ (0.49) −1:15∗∗ (0.56) −1:00∗∗∗ (0.32) −0:97∗∗∗ (0.30)

No. of observations: U 2 —Adj: R

98 0.23

98 0.19

98 0.23

98 0.52

98 0.20

Durability dummy Working capitala Financial dependenceb Short-term debta Firm sizec (100 employees per &rm) Leveragec Interest burdena Country &xed e2ect France Germany Italy UK US

Note: Pooled (cross-section cross-country) least squares; White Heteroskedasticity-Consistent standard errors (in parentheses). *, ** and *** indicate rejection of the null hypothesis of zero coeQcients at the 10.5 and 1 percent level, respectively. a Industry average. b Rajan and Zingales (1998) measure of external &nance dependence (equivalent results are obtained when this variable is interacted with an index of the country’s &nancial development). c Industry median (average for the US). Source data are described in Appendix A.

and Zingales is not statistically related to the output e2ects of monetary policy. 29 Similarly, no signi&cant systematic link appears between the policy e2ects and the industry measure of short-term debt. 29 This is true also for the case, not reported in Table 7, in which this variable is interacted with country measures of &nancial developments (total capitalization or domestic credit over GDP).

1562

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

Two of the variables that proxy &rms’ borrowing capacity, namely leverage and &rm size, are signi&cant and have the expected (positive) sign. 30 The &nding of a positive relationship between the output e2ects of monetary policy and the &rm size variable is often criticized as being weak evidence in favor of the credit channel hypothesis, since it might simply reHect other underlying reasons for the size distribution of &rms (e.g. Eichenbaum, 1994). Note that our approach, based on disaggregated data, allows us to tackle this critique directly, by controlling for several observed and unobserved (e.g. regression 4 in Table 7) industry e2ects. This is a key advantage of using micro data for studying the monetary transmission mechanism. The correlations detected by our regressions are consistent with the hypothesis that greater borrowing capacity by the &rms of an industry reduces the output e2ects of monetary policy. Finally, the interest-burden indicator is found not to be signi&cantly related to the monetary policy e2ects. Analogous results are obtained when the alternative measures of the impact of monetary policy discussed in Section 3.1 are used. As an illustration of this case, regression 3 reports the results that are obtained when the 24-month elasticity is obtained from the “augmented VAR” using the “Block B” identi&cation scheme. As reported in Table 6, the correlation between this elasticity measure and the one used in regression 1 is 0.957. The estimates of Eq. (3) reveal no major di2erences with respect to the estimates of Eq. (1). That similar results are obtained when elasticity measures obtained from di2erent VARs are used indicates that the estimates are relatively robust. Overall, these &ndings are consistent with the view that the output e2ects of monetary policy reHect both a traditional demand channel (as captured by the durability dummy) and a cost channel, namely the supply e2ect impinging on &rms’ production decisions (captured by the working capital measure and the broad credit channel variables). Even though lack of industry data on wages and prices prevents us from separately identifying the demand and cost e2ects of monetary policy, as in Barth and Ramey (2001), our results corroborate their hypothesis that the factors leading to sectoral di2erences in the transmission mechanism should be systematically related to di2erences in working capital requirements, industry demand features and the fraction of credit constrained &rms. Moreover, the estimates are consistent with these authors’ &nding that the output e2ect of monetary policy is larger when both channels are active. Quantitatively, the economic signi&cance of the variables related to the &rms’ borrowing capacity (size and leverage) appears relevant. The estimated marginal e2ect of increasing the typical &rm size by 250 employees is suQciently large to half the negative e2ect experienced by the “durable” industries. Considering that the range of variation of the (median) &rm size in our sample ranges from 50 to 500 employees (this interval accounts for approximately 90 percent of the observations), the size variable appears related to di2erential impacts of about 0.5 percentage points. Similarly, the 90 percent of the observations on leverage range from 1 to 4. The estimates of Table 7 suggest that this is related to a variation in the impacts of monetary policy as large as 1 percentage point. These magnitudes are large if judged in comparison with 30 This result is obtained using the median indicators for the four European countries. No signi&cant coeQcient is obtained when the average measure is used.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1563

Table 8 “Synthetic” 24-month elasticity of industrial production

Synthetic elasticity

France

Germany

Italy

UK

USA

−0:74 (0.05)

−0:81 (0.06)

−0:70 (0.05)

−0:73 (0.06)

−0:77 (0.06)

the range of variation of the policy impacts (Table 3) and, overall, are consistent with a non-negligible quantitative role of credit frictions in the transmission mechanism of monetary policy. 31 4.3. On cross-country di?erences in the e?ect of monetary policy None of the regressions of Table 7 detects a signi&cant cross-country di2erence in the e2ects of monetary policy that is imputable to a country-speci&c component. The joint hypothesis of identical country &xed e2ects cannot be rejected at the 10 percent con&dence level in regressions 1, 2, 4, and 5. This result is analogous to the one presented in Table 5. These &ndings, and the fact that the null hypothesis of equal (across country) coeQcients is not rejected in most of the regressions in Tables 5 and 7, strongly suggest that the (limited) cross-country di2erences detected with aggregate data are not likely to result from unobserved country-speci&c factors. Di2erences across industries are as large as 2 percentage points while the largest cross-country di2erential, about 0.4 percentage points, is recorded between the maximum elasticity of Italy and the UK (see Table 3). Excluding the UK, cross-country di2erences are smaller than 0.2 percentage points. This evidence, however, might not be suQcient to dismiss the view that country e2ects are important. An advocate of the latter view might argue that the small cross-country di2erences presented in Tables 3 and 5 mask o2setting composition and country e2ects. We addressed this issue by re-estimating the regression in Table 5 setting the country dummies to zero. We then computed estimates of country e2ects from the weighted average of the estimated industry e2ects using the industrial production shares of Table 2. 32 Table 8 reports these estimated “synthetic” e2ects along with their standard errors (in parenthesis) computed with the delta method. Two observations are in order. First, the synthetic elasticities are very similar across countries and their di2erences even smaller than those reported in Table 5. Second, they are strikingly close to those directly estimated from the country VARs in Section 2 and reported in the bottom line of Table 3, with the only exception of the UK. For this latter case, the aggregate VAR elasticity, equal to −0:49, does not fall within two standard deviations from the synthetic estimate. Therefore, this may be interpreted as evidence of a UK country-speci&c e2ect that somehow attenuates the impact of 31 These &ndings have a theoretical counterparts in the results in Bernanke et al. (1999): in their quantitative general equilibrium model of the &nancial accelerator, the response of output to a given monetary impulse is about 50% greater than in an economy without credit market frictions. 32 We thank a referee for suggesting this exercise.

1564

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

monetary policy shocks. However, it is clear from the relevant column of Fig. 1 that the UK output impulse response is imprecisely estimated and thus it may be that the above discrepancy is nevertheless within the limits sample uncertainty. 33 Overall, this exercise indicates that the small cross-country di2erences detected by the country VARs are consistent with the heterogeneity in the industry impacts reported in Table 3. This is mainly due to the relatively similar sectoral shares of the countries in our sample (Table 2). Besides con&rming the usefulness of disaggregated data for the study of monetary policy, these &ndings may also have a bearing on the policy debate. In the Euro area, for instance, there is interest about the possible heterogenous e2ects that the ECB policy might have on di2erent countries. Our inability to &nd signi&cant cross-country di2erences imputable to a country-speci&c component suggests that the asymmetric e2ects of a common monetary policy on France, Germany and Italy, which together add up to nearly two third of the Euro area GDP, are unlikely to be large. Note, moreover, that much larger di2erences have historically been recorded across the industry of each of these countries. Some heterogeneity in the e2ects of monetary policy seems to be a natural phenomenon in an economy composed by sectors with di2erent structural characteristics (durability, working capital, credit worthiness). 4.4. Sensitivity The results are reasonably robust. The use of alternative impact measures (as from regressions 2 and 3) somewhat weakens the signi&cance of the size and the leverage variable (see regression 2 or 5), with no major consequences for the other variables. This also holds when industry dummies are included to control for the e2ect of unobserved (country invariant) industry characteristics (regression 4) and instrumental variable estimation is performed (regression 5), in order to control for possible biases related to measurement errors in the indicators. 34 Overall, the point estimates of all the variables, with the exception of the size variable in regression 5, remain signi&cant and their values do not change much across regressions. The coeQcients of the durability, &rm size and working capital are almost identical across regressions. These results are essentially analogous to those obtained in a previous version of the paper (Dedola and Lippi, 2000), where the industry impacts were measured using the same 5-lag structural VAR for all countries. 35 The main di2erence between the previous estimates and the current ones concerns the cross-country heterogeneity of the 33 Mojon and Peersman (2001), using VAR techniques to the study of the aggregate e2ects of a monetary shock on 10 countries members of the euro area, conclude that given the large con&dence bands, one cannot &nd conclusive evidence of asymmetric cross-country e2ects of monetary policy. 34 We instrument the leverage, size, working capital and interest burden variables of Eq. (4) with their ranks across countries and industries (e.g. ranking all the industries in all the countries according to &rm size and using the rank as an instrument for that variable). 35 In that paper, we also used two additional measures of interest rate sensitivity measuring the industry degree of openness to trade (average ratio of export to value added) and the investment intensity of the production process (average ratio of investment to value added). No signi&cant e2ect of the openness measure emerged, while the investment ratio was signi&cant. The latter &nding, con&rmed by Peersman and Smets (2002) using a di2erent methodology, brings additional evidence to bear on the importance of the traditional interest rate channel.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1565

country &xed e2ects that appeared slightly more pronounced, but still much smaller than the cross-industry di2erences. However, the ranking of the industry e2ects and, most importantly for our purposes, its relation to the structural industry features are not a2ected by the choice of the identi&cation scheme: the industry e2ects of monetary policy measured by these alternative schemes remain signi&cantly related to the durability, working capital, leverage, &rm size and interest-burden variables. 5. Concluding remarks This paper investigated the output e2ects of monetary policy using disaggregated data at the industry level from &ve OECD countries. The &rst step of the analysis documented two empirical regularities: (i) a signi&cant cross-industry heterogeneity of policy e2ects; (ii) a similarity across countries of the cross-industry distribution of policy e2ects: some industries, e.g. motorvehicles (food), show a systematic above (below) average response to monetary policy shocks. A decomposition of the sectoral e2ects into industry-speci&c and country-speci&c components indicates the absence of signi&cant cross-country di2erences in the transmission mechanism of monetary policy. The second step of the analysis related these regularities to industry characteristics suggested by monetary transmission theories. 36 The analysis reveals that the impact of monetary policy is stronger in industries that produce durable goods, have greater &nancing requirements (working capital) and a smaller borrowing capacity (i.e. smaller &rm size and leverage ratio). The economic signi&cance of variables related to &rms’ borrowing capacity (size and leverage) indicates a non-negligible quantitative role of credit frictions, con&rming the predictions obtained from quantitative general equilibrium models, e.g. Bernanke et al. (1999). Overall, disaggregated data contain information that is useful to understand the monetary transmission mechanism. This route is also being followed by other economists; it is one building block of the Eurosystem Monetary Transmission Network, an important project that coordinates economists from the European Central Bank and the other National central banks to improve the understanding of the workings of monetary policy in the euro area (see the contributions in Angeloni et al. (2003)). Acknowledgements We thank the Editor, Jordi Gali, and three anonymous referees for several helpful suggestions. We also bene&ted from the comments of Juan Dolado, Luigi Guiso, Bernd Hayo, Anil Kashyap, Frank Smets and seminar participants at Tel Aviv University, the Bank of Italy, the Monetary Transmission Network of the European Central Bank and the World Econometric Society 2000. We thank Miria Rocchelli and Patrizia Passiglia 36 To measure these industry characteristics for each of the &ve countries considered, we used an original &rm-level database containing information from a large number of listed and unlisted &rms. The resulting summary statistics, based on information from the balance sheets of about 42,000 &rms over a 5-year period, is available from the authors upon request.

1566

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

for their expert research assistance. The views herein are personal and do not involve the responsibility of the institutions with which we are aQliated. Appendix A. Data sources and de#nitions The following data are used in the VARs estimates; the sample runs from 1975:1 to 1997:4 for all countries, but for France, for which the starting date is 1980:1. • Industrial production: monthly indices from OECD Main Economic Indicators. • Industrial sectors production (ISIC 3/4 digits): monthly indices from OECD Indicators of Industrial Activity. • CPI: monthly data from OECD Main Economic Indicators. • Interest rates: monthly averages of the Federal Fund Rate (US) and the 3 month interbank rate for all other countries; from BIS Data Bank. For Italy, 3 month interbank rate from the domestic screen-based market (MID). • Exchange rates: monthly averages of the real e2ective (trade weighted) exchange rate from IFS (“rec” line). • Money stock: M1 and M3 monetary aggregates, national de&nitions, monthly data from BIS Data Bank. A synopsis of de&nitions and sources of the variables used in the regressions of Table 6 appears in Table A1. The left column lists the countries and industries upon which the analysis is based. The right column lists the variables that are used in the regressions. The dependent variables appear in the upper panel of this column, they are given by the (semi)elasticity of industrial output to an interest rate structural innovation, 24 months after the shock, and at its maximum after 12–36 months. Explanatory variables used in the regressions are listed in this column below the dependent variables. The &rst variable is a durability dummy, which is 1 if the industry produces durable goods. The economic destination of production is from the national accounts statistics: according to this criterion, the “durable” output industries are denoted by the ISIC codes beginning with 33, 36, 37 or 38. An alternative measure, which includes industries 34 and 35 (paper and chemicals) among the durable output producers, does not a2ect the results in Table 6. The other explanatory variables are constructed from yearly balance sheet data of individual &rms contained in Amadeus. First, the variable of interest is constructed at the &rm level for a given year. The average &rm value for this variable is then calculated over the available period (1993–1997). Finally, the latter measures are used to compute the industry mean and median value of each variable. Comparable industry averages for the US were constructed using the Quarterly Financial Reports and the Employment Size of Firms, respectively, from the US Bureau of Census and the OQce of Advocacy—US Small Business Administration. The variables are de&ned as follows: • working capital (ratio): the di2erence between the balance sheet item “current liabilities” and the item “current assets” divided by “total assets” • short-term debt (ratio): (short-term debt)/(total debt)

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1567

Data Appendix Variables Country France

Elasticity

Germany

maximum 24 month

Italy

D E P E N D E N T

United Kingdom United States Durability Dummy (1 if ISIC code equals 33, 36, 37, 38)

Industry Working capital (ratio to total assets)

ISIC

311 313 314 321 322 323 324 341.1 342 351 353 330 360 362 371 372 381 382 383 384.1 384.3

Food Beverages Tobacco Textiles Wearing apparel Leather Footwear Paper Printing and publishing Industrial chemicals Petroleum refineries Wood and furniture Non-metallic mineral Glass Iron and steel Non ferrous metals Fabricated metal product Machinery and equipment Electrical machinery Shipbuilding Motorvehicles

mean median

Short term debt: (ratio to total debt) mean median

Employees per firm mean median

Leverage: (total debt) / (own capital) mean median

Interest burden: (i-payments)/profit mean median

E X P L A N A T O R Y

1568

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

• &rm size: number of employees per &rm (in units) • leverage (ratio): (total debt)/(shareholders’ funds) • interest burden (ratio): (interest rate payments)/(operating pro&ts).

References Angeloni, I., Kashyap, A., Mojon, B., 2003. Monetary Policy Transmission in the Euro Area. Cambridge University Press, Cambridge. Angeloni, I., Kashyap, A., Mojon, B., Terlizzese, D., 2002. Monetary transmission in the Euro area: Where do we stand? European Central Bank, Working paper no. 114. Barth, M.J., Ramey, V.A., 2001. The cost channel of monetary transmission. In: Bernanke, B., Rogo2, K. (Eds.), NBER Macroeconomic Annuals 2001. MIT Press, Cambridge, MA, pp. 199–240. Bernanke, B., 1993. How important is the credit channel in the transmission of monetary policy? A comment. Carnegie Rochester Conference Series on Public Policy 39, 47–52. Bernanke, B., Gertler, M., 1989. Agency costs, net worth, and business Huctuations. American Economic Review 79, 14–31. Bernanke, B., Gertler, M., 1995. Inside the black box: the credit channel of monetary policy transmission. Journal of Economic Perspectives 9, 27–48. Bernanke, B., Mihov, I., 1998. Measuring monetary policy. Quarterly Journal of Economics 113, 869–902. Bernanke, B., Gertler, M., Gilchrist, S., 1999. The &nancial accelerator in a quantitative business cycle framework. In: Taylor, J.B., Woodford, M. (Eds.), Handbook of Macroeconomics, Vol. 1-C. North-Holland, Amsterdam, pp. 1341–1393. Carlino, G., DeFina, R., 1998. The di2erential regional e2ects of monetary policy. The Review of Economics and Statistics 80, 572–587. Carlstrom, C., Fuerst, T., 1997. Agency costs, net worth, and business Huctuations: A computable general equilibrium approach. American Economic Review 87, 893–910. Christiano, L.J., 1991. Modeling the liquidity e2ect of a money shock. Quarterly Review, Federal Reserve Bank of Minneapolis 15 (1), 3–34. Christiano, L.J., Eichenbaum, M., Evans, C., 1997. Sticky price and limited participation models of money: A comparison. European Economic Review 41, 1201–1249. Christiano, L.J., Eichenbaum, M., Evans, C., 1999. Monetary policy shocks: What have we learned and to what end? In: Taylor, J.B., Woodford, M. (Eds.), Handbook of Macroeconomics, Vol. 1A. North-Holland, Amsterdam, pp. 65 –148. Clarida, R., Gali, J., Gertler, M., 1998. Monetary policy rule in practice: Some international evidence. European Economic Review 42, 1033–1067. Dedola, L., Lippi, F., 2000. The monetary transmission mechanism: Evidence from the industries of &ve OECD countries. CEPR Discussion paper no. 2508, July. Eichenbaum, M., 1994. Comment in monetary policy. In: Mankiw, N.G. (Ed.), Studies in Business Cycles, Vol. 29. University of Chicago Press, Chicago. pp. 256 –261. Fisher, J., 1999. Credit market imperfections and the heterogeneous response of &rms to monetary shocks. Journal of Money, Credit and Banking 31, 187–211. Friedman, M., Schwartz, A., 1963. A Monetary History of the United States. Princeton University Press, Princeton, NJ. Fuerst, T., 1992. Liquidity, loanable funds and real activity. Journal of Monetary Economics 29, 3–24. Gertler, M., Gilchrist, S., 1994. Monetary policy, business cycles, and the behavior of small manufacturing &rms. Quarterly Journal of Economics 59, 309–340. Giannetti, M., 2000. Risk sharing and the growth of &rms: Theory and international evidence. Mimeo., Bank of Italy. Giavazzi, F., Spaventa, L., 1990. The New EMS. In: De Grauwe, P., Papademos, L. (Eds.), The European Monetary System in the 1990’s. Longman, London. Greene, W., 2000. Econometric Analysis, 4th Edition. Prentice-Hall, Englewood Cli2s, NJ.

L. Dedola, F. Lippi / European Economic Review 49 (2005) 1543 – 1569

1569

Kashyap, A., Stein, J., 1994. Monetary policy and bank lending. In: Mankiw, G.N. (Ed.), Monetary Policy, Studies in Business Cycles, Vol. 29. University of Chicago Press, Chicago, pp. 221–256. Mojon, B., Peersman, G., 2001. A VAR description of the e2ects of monetary policy in the individual countries of the Euro area. European Central Bank, Working paper no. 92. Peersman, G., Smets, F., 2002. The industry e2ects of monetary policy in the Euro area. European Central Bank, Working paper no. 165. Rajan, R., Zingales, L., 1998. Financial dependence and growth. American Economic Review 88, 559–586. Rudebusch, G., 1998. Do measures of monetary policy in a VAR make sense? International Economic Review 39, 907–931. Sims, C., 1992. Interpreting the macroeconomic time series facts. European Economic Review 36, 975–1111. Sims, C., 1998. Comment on Rudebusch: ‘Do measures of monetary policy in a VAR make sense?’ International Economic Review 39, 933–941. Sims, C., Zha, T., 1995. Does monetary policy generate recessions? Manuscript, Yale University.