Classical business cycles for g7 and European countries

(1)

EUI

WORKING

PAPERS IN

ECONOMICS

EUI Working Paper ECO No. 95/24

Classical Business Cycles for G7

and European Countries

Michael

J.

Artis, Zenon G . Kontolemis

and

Denise

R.

Osborn

-JP

3 3 0

EUR

(2)

©

The

Author(s).

European

University

Institute.

produced

by

the

EUI

Library

in

2020.

Available

Open

Access

on

Cadmus,

European

University

Institute

Research

Repository.

(3)

EUROPEAN UNIVERSITY INSTITUTE, FLORENCE

ECONOMICS DEPARTMENT

EUI Working Paper ECO No. 95/24

Classical Business Cycles for G7

and European Countries

Mic h a e l

J.

Ar t is, Ze n o n G. Ko n t o l e m is

and

De n is e

R.

Os b o r n

UP 330

EUR

\ s V T u r 0 y

BADIA FIESOLANA, SAN DOMENICO (FI)

(4)

All rights reserved.

No part of this paper may be reproduced in any form

without permission of the authors.

© Michael J. Artis, Zenon G. Kontolemis and Denise R. Osborn

Printed in Italy in July 1995

European University Institute

Badia Fiesolana

I - 50016 San Domenico (FI)

Italy

(5)

Classical Business Cycles for G7 and European Countries

M ichael J. A rtis + , Zenon G. K ontolem is' and Denise R. O sborn'

+ European U niversity Institute and CEPR * U niversity of M anchester

January 1995

ABSTRACT

This paper proposes classical business cycle turning po in ts fo r a num ber of countries based on industrial pro du ction. The countries selected are the G7 tog ether w ith m ost m ajor European countries. This in fo rm atio n enables us to exam ine the international nature of cyclical m ovem ents. In particular, w e exam ine w h ethe r cyclical m ovem ents are sim ilar across d iffe re n t countries and consider the lead/lag relationships betw een countries at peaks and troughs.

Keywords: Classical Business Cycles, Turning Points, International Business Cycles

JEL classification: E32

Correspondence to: Zenon Kontolem is, School of Econom ic Studies, U niversity of

M anchester, M anchester M 13 9PL, UK (Z eno n.K on tole m is@ m a n.ac.uk)

A c kno w ledg em e nts: We are grateful to M ark W atson fo r pro vid ing us w ith a

GAUSS version o f th e Bry-Boschan procedure and to the OECD fo r sup plying the data. Finally, tha nks are due to W enda Zhang fo r very useful com m ents. Michael A rtis and Zenon Kontolem is acknow ledge support from the Leverhulm e T rust throu gh a grant adm inistered by the CEPR.

(6)

(7)

1 Introduction

Over the last tw o decades it has becom e evide nt th a t recessions, exhibiting substantial declines in the absolute level o f econom ic a c tiv ity , are a re a lity in m any countries. This has prom pted a revival o f in te rest in business cycle research. The m ost recent recession has played an im p ortan t role in th is revival because it was largely unpredicted by conven tion al m acroeconom ic foreca sting m odels and because m any observers have linked it to a perception th a t fina ncial liberalization and deregulation m ay have rendered the global eco no m y in he rently unstable.

Despite the recent appreciation o f the im portance of business cycle s, relative ly little is kno w n ab ou t the cha racteristics o f such cycles outside the United States. T herefore, it is unclear w h e th e r the w e a lth o f results obtained from con tem po rary business cycle research w ill be applicable to oth er countries. S im ilarly, there has been relative ly little a tte m p t to understand c ro ss-co u n try influences in business cycles, b u t it is cle arly im p o rta n t fo r policym akers to kn o w w h e th e r the occurrence and tim in g o f expansions and co n tra ctio n s across countries can be prim arily a ttribu ted to com m on shocks or to the transm ission o f shocks across cou ntrie s. In this respect, the m ost recent recession is w o rth y o f pa rticular s tu d y , since its econom ic origins are no t w ell understood. Further, although the recessions o f the 1970s and early 19 80 s m ay be a ttrib u te d to oil price shocks, d iffe re nce s in the tim ing of the ab sorp tion of these shocks across d iffe re n t econom ies are also of interest. Perhaps m ore interesting here, how ever, is w h e th e r recoveries fro m these recessions began sim u ltan eou sly in various cou ntrie s, or w h e th e r th e y were transm itte d from one or m ore lead countries.

(8)

There have been a num ber o f recent studies of internationa! business cycles, including those by Canova (1993a), Engle and Kozicki (19 93 ), Backus, Kehoe and Kydland (1 9 9 2 ), Backus and Kehoe (1 9 9 2 ), Canova and Dellas (1 9 9 2 ), fo r exam ple. H ow ever, none of these studies e x p lic itly considers tu rning po in ts, so th a t they cannot d irectly exam ine the transm ission of expansions and c o n tra c tio n s o f interest here. This is especially im p o rta n t in th a t an accum ulation o f evidence indicates asym m etry o f econom ic relationships in expansions and c o n tra c tio n s ; see Neftci (1984) and H am ilton (19 89 ) fo r classic reference.1- in the US c o n te x t, w hile A cem oglu and S c o tt (19 94 ) present som e evidence fo r the UK. It appears entirely plausible th a t d iffe re n t international m echanism s could apply at peaks and troughs.

None o f the w o rk on in ternational m echanism s hinted at above can be undertaken at the present tim e fo r the sim ple reason th a t no accepted business cycle turning points appear to be available fo r countries other than the United S tates. For the US, the National Bureau o f Econom ic Research (NBER) dates peaks and troug hs; the corresponding cycles, representing periods o f expansion and c o n tra c tio n in the level o f a c tiv ity , have becom e kno w n as "cla ssica l" business cycles. There is a related con cep t of " g ro w th " cycles, w h ich refers to cyclical m ovem ents around an underlying tre n d 1. Dates are regularly published fo r g ro w th cycle turning points in a num ber of cou ntrie s, but corresponding classical dates are not.

1 There w as a change in emphasis from classical to g ro w th cycles, particularly in Europe, after the debate in the 1960s over the possible obsolescence of the business cycle; see B ronfenbrenner (1969) fo r a conference on this them e.

(9)

It seems self-e vid ent th a t recessions, in the sense o f absolute declines in a c tiv ity , are m ore im p o rta n t th a t declines relative to trend. There are also tech nica l reasons w h y one m ay choose to analyze classical business cycles instead of g ro w th cycles. In particular, d iffe re n t detrending m ethods may yield d iffe re n t g ro w th cycle chronologies (Canova, 19 93 b), w h ic h is a problem w hen the trend is a fuzzy concept. Further, com m o nly used detrending m ethods m ay induce spurious cycles (see King and Rebelo, 19 93 , and O sborn, 1993, fo r tw o exam ples).

This paper goes some w a y to w a rd s re c tify in g the lack o f classical business cycle in fo rm atio n by proposing dates fo r a num ber of countries based on industrial production. The countries selected are the G7 tog ether w ith m ost m ajor European countries. This in fo rm atio n then enables us to exam ine the in ternational nature of cyclical m ovem ents. In particular, w e are interested in w h e th e r cyclical m ovem ents are sim ilar across d iffe re n t countries and in the lead/lag relationships betw een countries at peaks and troughs.

The paper is organized as fo llo w s . Section 2 discusses our m etho do log y fo r iden tifying classical turning points, w ith the details o f our procedure included as A ppendix 1. Results are discussed in the fo llo w in g tw o sections, w ith Section 3 considering cycles c o u n try by co u n try and Section 4 looking at in ternational aspects. C onclusions com plete the paper.

3 © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(10)

2 Establishing Business Cycle Chronologies

Our m etho do log y fo r tu rn in g po in t dating is based on th a t em ployed by the NBER and, in particular, the w o rk of Burns and M itche ll (1946) and Bry and Boschan (1971). The la tte r are henceforth denoted as BB. BB (p 17) po in t out tha t " ...tu rn in g p o in t d e term in ation ... c a n n o t be regarded as o b je c tiv e in the sense tha t

all reasonable a n d c o n scie n tio u s in v e s tig a to rs w o u ld agree on the answers. O nly agreem ent on the a p p lica tio n o f a s p e c ific se t o f detailed, and som etim es arbitrary, p ro c e d u ra l co n v e n tio n s c o u ld b rin g a b o u t agreem ent on the cho ice o f tu rn s ." Our

classical business cycle tu rning po in t dates are based on such a set o f rules. We hope th a t oth er researchers w ill agree th a t these rules are at least reasonable. BB them selves provide a set of rules, w h ic h have been w id e ly used (see W atson,

1994, fo r a recent example) and these provide our s tarting point.

The procedure w e ad op t is detailed in the A ppendix. It consists essentially of four steps. In step I extrem e values are id entified and replaced, since w e do n o t w a n t these to unduly in fluence a procedure w h ic h is designed to look fo r broad upw ard and d o w n w a rd m ovem ents. Then step II sm ooths (or averages) values to reduce the im portance o f s ho rt-ru n erratic flu c tu a tio n s . Turning points are te n ta tiv e ly id entified in th is sm oothed series as occu rring at observations the values of w h ich are greater (less) than those fo r tw e lv e m onths on either side w h ils t peaks and troug hs are required to alternate. In step III w e return to the unsm oothed data and use sim ilar rules to id e n tify te n ta tiv e tu rning points w ith the additional requirem ents th a t the am plitude o f a phase be at least as large as one standard error of the m on th ly log changes and the duration o f a cycle (peak to peak and troug h to

(11)

trough) be at least 15 m onths. The final stage, step IV, com pares the tw o sets of te n ta tiv e tu rn in g points. W hen there is a close correspondence b e tw ee n the in dica tion s given by the tw o alternative sets of te n ta tiv e tu rn in g po in ts (and only in this case), w e co n firm the existence of a turning po in t, w ith th e fina l date being th a t id en tified in the unsm oothed (original) series.

Table 1 sho w s the tu rn in g points id entified fo r industrial p ro d u c tio n data fo r nine European cou ntrie s (Germ any, France, UK, Italy, Spain, Belgium , N etherlands, Ireland and Luxem bourg) to g e th e r w ith the USA, Canada and Japan. The data used are m o n th ly and seasonally adjusted, covering the period 1 9 6 1 :1 -1 9 9 3 :1 2 . For ease o f later inte rna tiona l com parison, tu rning po in t dates are aligned in the table in accordance w ith apparent "c o m m o n " cycles. Our (arbitrary) rule w as th a t cycles w ere aligned if the recessions either overlapped or the beginning o f recession in one co u n try w as w ith in three m onths o f the end o f recession in another. M ild recessions (defined as in volving a total decrease o f less than 5 percent) are indicated by th e use o f underlining.

A tte n tio n here is con fine d to industrial production fo r reasons o f tra c ta b ility and data a va ila bility. Nevertheless, w e recognise tha t the id e n tific a tio n o f business cycle tu rning po in ts ideally requires the analysis o f a num ber o f d iffe re n t series. H ow ever, w e believe th a t the im p ortan t characteristics w h ic h w e uncover w ill apply more generally than ju s t to industrial production. Figures 1-12 s h o w th e series tog ether w ith the id en tified turning points; the top panel o f each diagram sho w s the series sm oothed using a seven-m onth centred m oving average, w h ile the original

(12)

data are graphed in the b o tto m panel. Note th a t the sym bol " x " in the lo w e r part of a figure indicates cases w h ere a tu rning po in t w as id e n tifie d in stage III of the procedure b u t sub seq ue ntly deleted in the final stage IV since no corresponding turning po in t was id en tified close to th a t region in the sm oo th ed series.

There is no obje ctive standard w ith w h ich w e can com pare the dates w e s h o w in Table 1. W e have, ho w e ver, com pared our results w ith tho se ob tained by applying the com p ara tively m ore com plicated set o f rules em bodied in the BB procedure2. As sho w n by the com parison in A ppendix Table A 3, the BB procedure is relatively more sensitive; w ith the single exception of a cycle fo r Belgium detected by BB in 19 9 0 -9 1 , all o f our id en tified cycles correspond to the ones de tected by the BB procedure but th a t procedure also id entifies som e additional cycles. As an exam ination o f the graphs in Figures 1 to 12 indicates, the recessions in these additional cycles are n o t v e ry m arked; w h e th e r th e y represent "tru e " recessions m ust be a m atter of judgem ent. We note, how ever, th a t the additional cycle identified by BB fo r the US is n o t a business cycle according to the w idely-acce pted NBER reference chro no log y (see Table 2 below ). On balance, w e prefer the s tric te r criteria em bodied in our procedure, since the y serve to guard against small id iosyn cratic m ovem ents in industrial pro du ction being con fused w ith business cycles.

2The GAUSS procedure was w ritte n by M ark W atson and replicates th a t of the BB procedure. The BB program s are also described and used by King and Plosser,

(13)

b le 1: C la s s ic a l B u s in e s s C y c le C h ro n o lo g ie s f o r In d u s tr ia l P ro d u c ti o n CM 00 CO O 00 00 LD CM CM ' t O 00 LD £ s * " 0) ID p» O T * LD CD 2 2 P ' CM 3 CD CD O p * p * CD CD aô LD oo CD p * P ' P ' 00 TJ (M CD CM !â 2 * -£ ■«t LD CD — p * P ' O 00 co 00 CD LD CO tp CM 5 ₂ * - 5 LD CD co O P» 00 p P * 00 C-J 00 CO CD 00 E p - O CD CM CM CO 00 CM 3 *— t— r— S 2 T— o> t* LD p S O m r ' P ' CD P CD O CD CD p - P ' 00 CT) ç 00 00 CO O ni Q- i 5 T— co t* LD O p - I - ' CD CT) CD co co co CO CD CM i i *— LD p ' p^ O co CD CD p P ' p ' 00 00 CD 00 . c : 0) Cj p * r - CM CD co CD LD 00 CD LD Q)

C

T

«“ 5 5 Q . CD CM T f iD CD ''t " t O CM ID CD CD p - p * p P ' 00 00 00 CD Cf) CD co 0 co LD T— CM OO T - CM oo "«t _•Si c S

r_

«■“ *“ 2 2 ' t LD p CD CM CM O p * p - p P ' P O 00 CD 0) .$ p 00 00 O 0) ^ ) CO 1D 00 p * CM CO rË »“ 2 E CO co LD o S CD CD p - CD CM CD a P» 00 £ I c r - co r - O LD 00 LD coc : Q- *— * - S o « _LD _LD _co 55 CM 00 00 CD <u o 00 00 g) .P n> p * O co LD oo CD ' t O 00 T * CM TJ 2 2 2

T_

S 5 Ç CD LD CD O O CD CD 2 C_ï co o p * p P 00 00 CM 00 00 00 O ) p * 00 p 1 < O r - co CO P ' p CM ' t co co 2 C/5 T- *— ₅ _S ₂ r— i LD O O <D ■§ CD O CO P» 00 00 aô CM 00 CT) CD p * p» 00 “p p .5 ■p SZ -XL .C -C SI sz SI . * _c -X .£= sz - C -VC

ro

O) (TJ CT (TJ CT (TJ O ) (TJ CT) (TJ CT) (TJ CT) (TJ O) (TJ OJ (0 O ) (TJ c CD CT CD CT CD CT CD CT CD CT a) CT (D CT (D CT (D CT (U CT 0) û_ O 0 - O 0 - O 0_ O CL O CL O CL O 0- O û- O CL O 0. H f— f— H H H f— H h - H 2 © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(14)

3 Results for Individual Countries

The dates id entified in Table 1 provide a s tarting po in t fo r a closer analysis o f the cycles. In w h a t fo llo w s w e brie fly discuss the cycles in each c o u n try , highlighting any d iffic u ltie s encountered in the analysis.

Figure 1 relates to the USA, w here our procedure identifies fiv e peaks and troughs over the relevant period. In this case w e can refer to the alte rna tive chronology provided by the NBER reference dates and Table 2 com pares these w ith our dates fo r turning points in in dustrial pro du ction. A lth o u g h the NBER dates refer to general econom ic con ditions and not sim ply to industrial p ro du ction, there is a strikin g correspondence be tw een the tw o chronologies3. In six o u t o f the ten cases the dates coincide e xa ctly, w ith generally sm all differences o th e rw ise . The only disagreem ent of note is in the dating of the 1 9 8 9 /9 0 peak, w h ere our procedure selects a date m ore tha n a year in advance o f the NBER one. In the c o n te x t of industrial p ro du ction, ho w e ver, the earlier date is m ore plausible in th a t its value exceeds the m axim um reached in 1990.

Figure 1 sho w s three cases o f an apparent turning p o in t in the original series being deleted by our procedure as it is not also id entified in the sm oothed series of the upper panel. One o f these occurs in 1967 and the oth er tw o m ore rece ntly in

3 An NBER reference troug h in 1961 is ignored in Table 2, because th is occurs to o early in the sam ple period fo r our procedure to be able to id e n tify it.

(15)

1 9 8 9 -9 0 . The exclusion o f these dates suggests th a t the m apping process, in the last stage o f the procedure, is useful in elim inating po ssibly spurious cy c le s 4.

Table 2: NBER Dates and Own Industrial Production Chronology for the US

P e a ks: NBER 6 9 M 1 2 7 3 M 1 1 8 0 M 1 8 1 M 7 9 0 M 7

In d u s tr ia l P r o d u c tio n 6 9 M 1 0 7 3 M 1 1 8 0 M 3 8 1 M 7 8 9 M 4

T r o u g h s : N BER 7 0 M 1 1 7 5 M 3 8 0 M 7 8 2 M 1 1 9 1 M 3

In d u s tr ia l P r o d u c tio n 7 0 M 1 1 7 5 M 3 8 0 M 7 8 2 M 1 2 91 M 3

Source fo r NBER dates: Survey o f Current Business, A p ril 1994.

Canada has relatively clear cycles. The only d iffic u lty encountered is th a t the last troug h in the original series is located 11 m onths aw ay fro m the corresponding troug h in the m oving average. For inclusion as a reference tu rn in g po in t, our program sets a lim it o f 5 m onths betw een the te n ta tiv e tu rn in g po in ts id e n tifie d in the original and sm oothed series. In th is case, how ever, w e override the autom atic program and include it as a reference troug h. A sim ilar situ a tio n occurs in Italy (see Figure 7) during 19 77 , w here w e set 1 9 7 7 :6 as a tro u g h despite its exclusion by the au tom atic procedure. O therw ise, despite erratic m o n th -to -m o n th m ovem ents, the tu rning points iden tified fo r Italy appear to be clear.

4lt is w o rth po in ting o u t th a t, in co n tra s t to the procedure adopted here, the original Bry-Boschan procedure com es up w ith tw o cycles during 1 9 8 9 -9 1 ; see the A ppendix. © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(16)

The case o f Japan is an in te restin g one and requires som e discussion. As show n in Figure 3, the dom inant p ro p e rty o f Japanese industrial p ro d u c tio n is o f sustained g ro w th w ith little cyclical flu c tu a tio n . Our procedure does, ho w e ver, id e n tify some cycles w ith the tw o largest recessions being in 1 9 7 4 -7 5 and the m ost recent one, 1 9 9 1 -onw ards. These tw o cycles s h o w obvious (absolute) declines in industrial pro du ction. For the recessions id en tified in 1 9 8 1 -8 2 and 1 9 8 5 -8 6 , on the other hand, industrial pro du ction is relative ly fla t, w ith declines in to ta l o f only 4 .2 and 3 .4 per cent respectively.

In con tra st, G erm any is s tra ig h tfo rw a rd and cycles are generally clear after sm oothing is undertaken. Notice th a t the value m arked w ith "E " on the lo w e r part of the diagram is classified as 'e x tre m e ' according our rules and is not regarded as a turning point. Figure 4 also reveals th a t although tw o tu rn in g points w ere identified in 1 9 8 6 in the sm oothed series, these w ere no t included in the final selection since no corresponding turning points w ere detected in the original series. In con tra st, the BB procedure w o u ld retain these.

The case o f France (Figure 5) is m ore problem atic, w ith cyclical m ovem ents during the 198 0s being fa r from clear. Our procedure selects 1980:11 and 1 9 8 1 :1 2 as a trough and peak resp ective ly, but this is a m inor flu c tu a tio n and the question is w h e th e r one should cla ssify this as a cycle at all. C ertainly in term s o f the am plitude th is is n o t ty p ic a l and o u tp u t only rises by 3.1 per cent to reach the peak in Decem ber 1981. The recession detected in 19 82 is also relative ly m inor and fu rth e r doubts m ust exist about the recession id entified in 1 9 8 4 -8 5 . For this latter

(17)

case, the points m arked "a ” in the lo w er panel im ply th a t these po tentia l turning points are excluded in accordance w ith the m inim um am plitude rule; th a t is, the change from troug h to peak is less than one standard error o f (log) changes in the series. It m ay also be noted th a t the tw o extrem e values in 19 63 and 19 68 are e ffe c tiv e ly ignored.

The UK cycles are id en tified relatively easily in Figure 6, bu t tw o features are w o rth noting. The firs t is the selection o f 1 9 7 2 :2 as a trough; th is w as classified as an extrem e observation (it is thu s adjusted fo r in the to p panel o f the diagram ) but since the procedure id en tified a troug h fo r the sm oothed series in 1 9 7 2 :1 2 , this extrem e observation (being the local m inim um in the original series) w a s selected as a troug h fo r industrial pro d u ctio n . The second po in t refers to 1 9 7 3 -7 4 . A peak is required to be larger than values 12 m onths on either side: th is rules out a sho rt cycle w ith a peak in 1 9 7 3 and the abrupt fall in industrial pro du ction early in 19 74 as a recession5.

For the case o f Belgium , sh o w n in Figure 9, the decade of the 19 80 s is again problem atic. This is especially so w ith the original data, w h ere erra tic m ovem ents are in evidence. Since, ho w e ver, our procedure requires tu rn in g po in ts to be detected in both the original and sm oothed series before th e y are accepted, the period from 1981 to 1 9 9 0 is classed as an expansion fo r Belgium . The troug h dated in 1991 m ay also be controversial due to the erratic nature o f the

5 This fall relates to the "thre e da y" w o rking w e ek en forced to econom ise on stocks of fuel during a coal m ining strike, and hence m ay be a ttrib u te d to special factors. © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(18)

flu c tu a tio n s . Industrial pro du ction fo r the N etherlands (Figure 10) raises sim ilar problem s due to its noisy m ovem ents. T w o pa rticular cases to note are the recessions id entified from 1976 to 19 78 and from 19 87 to 1988, during w h ich industrial pro du ction declined by 5 .4 and 8 .8 per cen t resp ective ly. The la tte r one, at least, appears to be reasonably clear in the graph after the values are sm oothed.

The data fo r Luxem bourg, sho w n in Figure 12, are noisy but e x h ib it quite marked cyclical flu c tu a tio n s . Indeed, w ith seven peaks and six troughs, Luxem bourg has experienced m ore business cycles than any other co u n try considered. As the sm allest c o u n try in the sam ple, it may be m ore susceptible to such flu ctu a tio n s. The principal d iffic u lty in dating relates to the 1976 recession w h ic h our procedure de tects, although this does not seem to be an im p o rta n t cyclical flu c tu a tio n in the graph o f the original series.

Spain and Ireland (Figures 8 and 11) are fascin atin g since the y exh ib it very fe w classical cycles. A lth o u g h industrial pro du ction in Spain fell sub stantia lly in 1974- 75 and during the la te st recession, Ireland displays no cycles exce pt fo r those resulting from tw o m ild recessions (1 9 7 4 -7 5 and 19 79 -80).

Tables A 4 -A 1 5 in the A pp en dix record the sta tis tic a l cha racteristics, c o u n try by c o u n try, o f each id en tified business cycle. Rather than e x p lic itly discussing these here, w e m ove on to our principal interest, w h ic h is relationships across countries.

(19)

4 International Business Cycles

The firs t question th a t needs to be investigated is w h e th e r cycle s d iffe r in term s of their tim in g, th a t is w h e th e r th e y are synchronised w ith each oth er. In ad dition, it is also o f in te rest w h e th e r cycles d iffe r in term s o f th e ir in te n s ity and duration.

Table 3 provides com parative sum m ary in fo rm a tio n on regim e-specific cha racteristics. The figu res in the table are derived by calcula ting the averages across all com pleted expansions and con tra ction s in the sam ple; the average m on th ly decline (rise) in Industrial P roduction m ultiplied by the average du ration of co n tra ctio n s (expansions) gives the to ta l change, also sh o w n in the table.

Table 3: Classical Business Cycle Characteristics

R e c e s s io n s E x p a n s io n s M o n t h ly c h a n g e T o ta l c h a n g e D u ra tio n M o n t h ly c h a n g e T o ta l c h a n g e D u r a tio n U S A - 0 . 7 6 3 - 8 . 6 7 2 1 5 0 . 4 9 3 2 2 . 5 1 2 4 6 C A N - 0 . 6 5 0 - 9 . 1 9 5 1 4 0 . 6 2 5 2 0 . 3 3 0 3 5 J A P - 0 . 7 4 3 - 1 0 . 0 9 5 1 3 0 . 5 3 5 2 8 . 8 7 8 5 6 GER - 0 . 5 1 9 - 1 1 . 3 7 3 2 4 0 . 4 5 6 3 3 . 7 0 8 7 7 FRA - 0 . 7 9 5 - 7 . 8 6 2 11 0 .4 5 1 1 2 . 3 5 0 4 2 UK - 0 . 7 7 8 - 9 . 8 1 0 1 4 0 . 4 4 8 1 8 . 4 4 5 4 5 IT A - 1 . 4 6 2 - 1 3 . 7 4 2 1 5 0 . 6 7 3 3 5 . 4 9 1 6 3 S P A - 1 . 3 0 8 1 6 . 5 2 9 1 3 1 . 1 4 6 2 6 . 5 3 5 9 0 BEL - 0 . 8 4 2 - 1 1 . 4 1 7 1 4 0 . 8 7 7 1 7 . 3 2 3 3 9 NET - 0 . 5 1 5 - 9 . 2 4 7 2 0 0 . 6 0 0 1 5 . 4 0 8 3 0 IRE - 0 . 7 3 9 - 1 0 . 6 4 1 1 5 0 . 7 4 0 3 9 . 2 1 9 5 3 L U X - 1 . 8 7 9 - 2 1 . 8 2 1 1 6 1 .1 7 1 3 0 . 3 6 2 3 9 A v g - 0 . 9 1 6 - 1 1 . 7 0 0 3 1 5 0 . 6 8 4 2 5 . 0 4 6 51

Notes: ta) Changes are expressed in terms o f logarithms, white durations refer to m onths. Ibl AH figures are averages over com pleted recessions o r contractions.

(20)

The firs t th in g to note is the asym m etry of the cycles across all countries; o u tput falls m uch m ore a b ru p tly tha n it rises. Sim ply averaged across countries, industrial production falls by 0 .9 1 6 per cen t a m onth in recession, com pared to a 0 .6 8 4 per cent rise during expansions. Indeed, only in the individual cases of Belgium and the Netherlands is the average m o n th ly change in recessions sm aller in m agnitude than the average g ro w th in expansions. Since industrial pro du ction has g ro w n over the po st-w a r period, it is n o t surprising to find th a t econom ies sta y, on average, approxim ately three tim es as long in the expansion phase as th e y do in recession. W hen the y occu r, recessions in G erm any appear to be severe; the y last longer than in other countries, w ith the three recessions detected here having an average duration of tw o years and th e y result in a to ta l decline o f in dustrial pro du ction of over 10 percent. A t the oth er extrem e, France has experienced fo u r recessions at an average duration o f less than a year and w ith the sm allest average decline of under 8 percent. In expansions, Spain and Ireland sho w , resp ective ly, the longest average duration and the largest to ta l change. W hilst G erm any stays in recessions fo r longer periods she also g ro w s fo r longer in the recovery phase; the ratio of the duration of the up turn to th a t o f the d o w n tu rn is close to the overall average.

Table 4 sho w s the du ratio ns o f com plete cycles, th a t is from peak-to-peak and tro u g h -to -tro u g h . G erm any experiences long cycles com pared to oth er countries w ith the average duration fro m peak-to-peak being 101 m on th s in co n tra s t to 58 .5 m onths fo r the USA and 6 9 .3 m onths fo r Japan. In this respect, the group of European countries o f France, Belgium, the Netherlands and Luxem bourg appear to exh ib it shorter cycles than Germ any, Japan or, less m arkedly, the USA. H ow ever,

(21)

it m ust be borne in m ind, both here and elsewhere, th a t these results should be interpreted w ith care since fe w cycles occu r w ith in our sam ple period6.

Table 4: A verage Durations o f Cycles

C o u n try P e a k -to -P e a k T r o u g h -to -T roug h

U S A 5 8 . 5 6 3 . 5 C A N A D A 4 7 . 4 4 8 . 8 J A P A N 6 9 . 3 6 8 . 5 G E R M A N Y 1 0 1 . 0 9 3 . 0 F R A N C E 5 3 . 0 2 9 . 0 U K 5 7 . 4 6 1 . 2 IT A L Y 7 7 . 7 5 7 5 . 0 S P A IN 1 8 5 . 0 1 8 7 . 0 B E L G IU M 5 3 . 0 6 4 . 3 N E T H E R L A N D S 4 9 . 5 5 0 . 6 IR E L A N D 6 7 . 0 6 8 . 0 L U X E M B O U R G 5 4 . 1 6 4 8 . 0 A v e r a g e 7 2 . 7 5 7 1 . 4

From the results in the appendix it can be seen th a t the m ost severe recession (during the period under investigation) w as the one associated w ith the firs t oil shock in 1 9 7 3 -7 5 . In the USA, fo r exam ple, industrial pro du ction declined by approxim ately 16 per cen t from peak to trough. The s h o rte st recession occurred during 1 9 8 0 in the USA (see also Z arno w itz and M oore, 1 9 81 ), lasting only 4 m onths and im plying an average fall in the index o f industrial p ro du ction o f nearly 1.5 per cent an m onth.

6The big diffe re nce betw een the peak-to-peak and tro u g h -to -tro u g h durations fo r France is the result o f the exclusion of the last tro u g h -to -tro u g h cycle w h ich has not ye t been com pleted.

(22)

An interesting observation can be made by com paring the figu res fo r USA and Canada. C ontrary to the conven tion al w isdom , our dates suggest th a t the average duration of US expansions is large com pared to Canada7. F urtherm ore Canada experienced an additional recession in the m id -198 0s w h ic h , alth ou gh lasting only 7 m onths, involved a substantial fall in industrial p ro du ction. M oreover, w h ils t the firs t tw o co n tra ctio n s are m ilder fo r the case o f Canada, the three cycle s occuring after 1979 are deeper and longer lasting. For exam ple, during the last recession, industrial production in Canada declined by nearly 10 per cen t from peak to trough com pared w ith only 4 .6 per cen t fo r the USA.

In aligning cycles across countries in Table 1, w e a tte m p t to present furthe r inform ation on the synch ron isatio n o f cycles. It is, indeed, s trik in g h o w fe w recessions in th is period are confined to one co u n try alone. O nly those experienced by Italy in 1 9 6 4 and the UK tw e n ty years later are, by our alignm ent conven tion , unique.

Compared to the subsequent tw o decades, the 196 0s sho w s fe w classical business cycle recessions. The firs t tru ly in ternational recession here is th a t o f 19 73 -74. A lthough the s e ve rity o f th is w as undo ub te dly associated w ith the oil price rises,

7One reason fo r the big diffe re nce betw een the average duration rates betw een the tw o countries is the additional cycle experienced by Canada in the m id -198 0s. A lso the BB dates, reported in A pp e n d ix 1, suggest th a t a second cycle exists in the 19 8 9 -9 2 period, th a t is a 'dou ble d ip' recession did occu r (this is not consistent w ith the NBER dates). This small diffe re nce in the turning points reduces the average duration fo r the USA cycle (peak-to-peak) to 5 0 .6 m onths com pared to 4 8 .2 fo r Canada, thus elim inating the apparent gap betw een the tw o countries. © The Author(s). European University Institute. produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(23)

the peaks in G erm any and the US pre-date the oil price increase o f 1 9 7 4 s. A ltho ug h avoided by som e cou ntrie s, industrial p ro d u c tio n in a num ber o f countries also declined in 1 9 7 6 -7 7 . Indeed, the evidence o f Table 1 is th a t th is recession was a European phenom enon. Then, betw een 1979 and 1981 all cou ntrie s w ith the single exception o f Spain experienced one, or som etim es tw o , recessions. Of course, oil price rises again played a role here. A p a rt fro m the U K 's id io syn cra tic decline in in dustrial p ro d u c tio n in 1984, fo u r oth er cou ntrie s sh o w e d declines in o u tp u t in the m id -1 9 8 0 s . O f these, how ever, those o f Canada and Japan were relatively slight. Finally, another international business cycle recession w as in evidence around the turn o f the decade.

In sum m ary, three recessions are com m on across alm ost all c ou ntrie s, tho se being the ones com m encing in 1 9 7 3 -7 5 , 1 9 7 9 -8 0 and 1 9 8 9 -9 1 . These are associated w ith large falls in o u tp u t and are exam ined in fu rth e r detail below .

Table 5 com pares the con tra c tio n s w h ich occurred in 1 9 7 3 -7 4 and w h ic h are associated w ith the firs t oil price shock. It show s the dates o f the tu rn in g points fo r the tw e lv e c ou ntrie s analyzed to g e th e r w ith the cum u la tive (percentage) decline in industrial p ro d u ctio n , the du ration (in m onths) and in fo rm a tio n con cern ing the tim ing o f the tu rn in g points relative to those in the US and G erm any. 8

80 il prices rose gradually from O ctober 1973, w h en the oil em bargo w as im posed, to January 19 74 . The biggest rise occurred be tw een Decem ber and January w hen the price o f oil rose from approxim ately $5 a barrel to $ 1 2 in January. © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(24)

Industrial p ro d u ctio n declined s ig n ific a n tly during th a t period w ith the sm allest c o u n try, Luxem bourg, experiencing the largest fall. A m ong the rem aining countries Japan and Spain experienced the biggest declines, the N etherlands the sm allest. In term s o f d u ratio n G erm any had th e longest (23 m onths) and France the sho rte st (9 m onths! recession w h ile the rest varied from 10 to 16 m onths.

Table 5: The 1 9 7 3 -7 5 Recession

C o u n try P e a k /T ro u g h D e p th (% ) D u ratio n L e a d (-)/L a g ( + ) R elativ e to U S A L e a d (-)/L a g ( + ) R e la tiv e to GER U S A 7 3 M 1 1 / 7 5 M 3 1 6 . 0 2 3 1 6 + 3 /- 4 C A N A D A 7 4 M 3 / 7 5 M 5 1 2 . 1 2 8 1 4 -i 4 / + 2 +11-2 J A P A N 7 4 M 1 / 7 5 M 3 2 2 . 6 5 3 1 4 + 2 / 0 + 5 /- 4 G E R M A N Y 7 3 M 8 / 7 5 M 7 1 2 .9 2 1 2 3 - 3 / + 4 F R A N C E 7 4 M 8 / 7 5 M 5 1 5 . 9 8 9 9 + 9 / + 2 + 1 21-2 U K 7 4 M 6 / 7 5 M 8 1 2 . 6 2 3 1 4 + 7 / + 5 + 1 0 / + 1 IT A L Y 7 4 M 6 / 7 5 M 4 1 7 . 4 1 5 1 0 + 7 / + 1 + 1 0 / - 3 S P A IN 7 4 M 8 / 7 5 M 8 2 1 . 3 9 5 1 2 + 9 / + 5 + 1 2 / + 1 B E L G IU M 7 4 M 4 / 7 5 M 7 1 3 .1 4 1 1 5 + 5 / + 4 + 8 / 0 N E TH E R . 7 4 M 8 / 7 5 M 8 9 . 3 3 3 1 2 + 9 / + 5 + 1 2 / + 1 IR E L A N D 7 4 M 2 / 7 5 M 4 1 1 . 0 5 6 1 4 + 3 / + 1 + 6 / - 3 L U X E M . 7 4 M 8 / 7 5 M 8 5 3 . 5 6 4 1 2 + 5 / + 5 + 8 / + 1 A v e ra g e 1 8 . 1 9 1 3 . 7 5 + 5 . 2 / + 2 . 2 + 8 . 5 / - 1 . 3

Finally, the table sho w s th a t, w ith the exception o f Germ any, all the countries w e n t into the recession a fte r the US. On exit, the US preceded all cou ntrie s except Japan, w ith w h ic h it w as contem poraneous. Perhaps the m ost rem arkable thing about the u p turn, ho w e ver, is th a t despite the w id e ly varying durations o f the recession, on ly fiv e m on th s separated the troughs across these 12 countries. This, w ith the contem poraneous dating of those US and Japan, indicates th a t a sim ultaneous sho ck m ay have been responsible.

(25)

C o u n try P e a k /T ro u g h D e p th ‘ % ) D u ratio n Lead(-)/L ag < + )

R e la tiv e to U S A L e a d ()/L a g ( + 1 R elativ e to GER U S A 8 0 M 3 / 8 0 M 7 5 . 9 8 0 4 + 3 /n a C A N A D A 7 9 M 8 / 8 0 M 6 7 . 6 0 7 1 0 -7 /-1 -4 /n a J A P A N n a /n a G E R M A N Y 7 9 M 1 2 / n a 1 2 . 1 3 9 ' 3 5 ' - 3 /n a F R A N C E 7 9 M 8 / 8 0 M 1 1 6 . 2 8 6 1 5 - 7 / + 4 -4 /n a U K 7 9 M 6 / 8 1 M 5 1 5 . 8 0 6 2 3 - 9 / + 1 0 -6 /n a IT A L Y 8 0 M 3 /'n a 1 6 . 3 3 8 ' 3 9 ' 0 /n a + 3 /n a S P A IN n a /n a B E L G IU M 7 9 M 1 2 / 8 0 M 1 2 1 5 . 2 0 9 1 2 - 3 / + 5 0 /n a N E TH E R . 8 0 M 3 / n a 1 3 . 3 5 3 ' 3 2 ' 0 /n a + 3 /n a IR E L A N D 7 9 M 9 / 8 0 M 1 2 1 0 . 2 2 6 1 5 - 6 / + 5 -3 /n a L U X E M . 7 9 M 1 2 / 8 1 M 4 2 1 . 5 0 7 1 6 - 3 / + 9 0 /n a Notes: (a) * (b)

indicates one recession experienced over the pe riod 1979 to 1982. na indicates tha t no peak (trough) is dated fo r this recession

C o u n try P e a k /T ro u g h D e p th (% ) D u ratio n L e a d (-)/L a g ( + ) R elativ e to U S A L e a d (-)/L a g ( + ) R elativ e to GER U S A 81 M 7 / 8 2 M 1 2 9 . 4 2 3 1 7 n a / + 1 C A N A D A 81 M 4 / 8 2 M 1 0 1 7 . 0 2 7 1 8 - 3 /- 2 n a /-1 J A P A N 8 1 M 1 1 / 8 2 M 1 0 4 . 2 1 6 11 + 4 / - 2 na/-1 G E R M A N Y n a / 8 2 M 1 1 1 2 . 1 3 9 ' 3 5 ' n a /-1 F R A N C E 8 1 M 1 2 / 8 2 M 8 4 . 3 8 6 8 + 5 /- 4 n a /-3 U K n a /n a IT A L Y n a / 8 3 M 6 1 6 . 3 3 8 ' 3 9 ' n a / + 6 n a / + 7 S P A IN n a /n a B E L G IU M n a /n a N E TH E R . n a / 8 2 M 1 1 1 3 . 3 5 3 ' 3 2 ' n a /-1 n a / 0 IR E L A N D n a /n a L U X E M . n a /n a

Notes: (a) * indicates one recession experienced over the pe riod 1979 to 1982. (b) na indicates th a t no peak ttrough) is dated fo r this recession

(26)

Notice also th a t the US lead tim e fo r the peak seems to be longer com pared to th a t fo r the troug h (5 .2 and 2 .2 m onths respectively). Com pared w ith the US, G erm any seems to have gone early into the recession, but came out late.

Tables 6 and 7 com pare the co n tra ctio n s of the 1 9 7 9 -8 2 period. Since a num ber o f countries experienced tw o cycles at the beginning o f the 80s w e present the results fo r the tw o cycles separately, w h ile indicating w here on ly one cycle occurs. The US and Canada each experienced tw o recessions in the 1 9 7 9 -8 2 period. In both countries the second w as the more severe and lasted longer. Further, Japan w e n t throu gh one (mild) recession at about the same tim e as the second US/Canada cycle . France w as the only European c o u n try to experience both recessions, b u t here the second w as relatively m ild9. European cou ntrie s, notably Germany, Italy and the Netherlands, experienced a single prolonged recession spanning both periods. Indeed, Europe led the 19 79 d o w n tu rn , w ith all the European c ountries going in to recession before the USA, in co n tra s t to the 19 7 3 -7 4 cycle. It may be germ ane to note th a t the European countries decided at the same tim e to prioritize c o u n te r-in fla tio n a ry policies, and began the ir pa rticip a tio n in the ERM.

In recovery fro m the fir s t recession, North A m erica once again led the w a y. For the US and Canada the second recession w as, how ever, the m ore severe. Flere the US, German and Japanese econom ies all emerged w ith in tw o m onths of each other,

9One can argue th a t the extra (minor) cycle experienced by France was associated w ith the fam ous 'M itte rra n d U -turn'(see Sachs and W yplosz, 1986, for example). © The Author(s). European University Institute. produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(27)

again raising the qu estion o f the synch ron ous nature o f the process w h ic h m ig ht have led to this.

Table 8: The Latest Recession

C o u n try P e a k /T ro u g h D e p th (% ) D u ra tio n L e a d {-)/L a g ( + ) R elativ e to U S A L e a d (-)/L a g { + ) R elativ e to GER U S A 8 9 M 4 / 9 1 M 3 4 . 6 0 3 2 3 - 2 6 / C A N A D A 8 9 M 4 / 9 1 M 2 9 . 9 2 8 2 2 0 /-1 - 2 6 / J A P A N 9 1 M 5 / + 2 5 / - 1 / G E R M A N Y 9 1 M 6 / + 2 6 / F R A N C E 9 2 M 4 / + 3 6 / + 1 0 / UK 9 0 M 6 / 9 2 M 5 8 .0 9 1 2 3 + 1 4 / + 1 4 - 1 2 / IT A L Y 8 9 M 1 2 / + 8 / - 1 8 / S P A IN ' 9 0 M 1 / + 9 / - 1 7 / B E L G IU M ' 9 0 M 3 / + 1 1 / - 1 5 / N E TH E R . 9 1 M 2 / + 1 0 / - 4 / IR E L A N D L U X E M . 9 2 M 5 / + 3 7 / + 1 1 /

Note: * indicates that, according to our dates, tw o dow nturns were experienced during this period. The earlier peak is used here.

Finally table 8 com pares the latest con tra ction s. A lth o u g h w e have id e n tifie d the trough fo r only three cou ntrie s (USA, Canada and the UK) the in fo rm a tio n in the table reveals th a t, as w ith the 1 9 7 3 -7 4 co n tra ctio n , the US and Canada w ere the firs t to go into recession. In addition, the tim e lag is s ub stantia lly longer and varies from 8 m onths fo r Italy to 36 m onths fo r the case of France. Finally, from the table we can also see th a t G erm any w e n t late in to recession; th is m ay re fle c t the id iosyn cratic 'u n ific a tio n s h o c k ' suffered by tha t c o u n try.

(28)

5. Conclusions

In this paper w e have exam ined the existence and properties o f classical cycles in industrial p ro d u c tio n for a num ber o f countries. The countries selected w ere the G7 tog ether w ith m ost m ajor European cou ntrie s, enabling us to exam ine the international nature o f cyclical m ovem ents. Our dates w ere derived using m echanical rules w h ic h e ffe c tiv e ly reproduce the NBER reference dates fo r the US. Therefore, w e are c o n fid e n t th a t our procedure is reasonable. For countries other than the US, w e k n o w of no oth er dating exercise com parable to th a t undertaken here.

A lthough a tte n tio n was confined to industrial pro du ction, w e believe th a t we have uncovered features o f in te rest in the international transm ission o f business cycles. A lthough our d ating exercise looks at each c o u n try in isolation, w e have established th a t very fe w business cycles are confined to a single c o u n try . This indicates that m acroeconom ists should be alert to tu rning points (both peaks and troughs) resulting fro m external fa cto rs, rather than alw ays being the consequence of dom estic events. Further, this indicates the potential value of in fo rm atio n from other cou ntrie s in the co n s tru c tio n o f leading indicators and the prediction of turning points.

Our s tu d y also find s th a t m any countries started the ir recoveries from the 19 73 74 recession alm ost sim ultaneously, w h ile the same is true fo r the recoveries o f the US, Japan and G erm any from the 19 8 1 -8 2 recession. This raises the question of w h a t gives rise to such an in ternational effe c t. On the oth er hand, how ever, the

(29)

m ost recent recession seems to be d iffe re nt, since by the end o f 1 9 9 3 , o f these three do m in ant econom ies, only the US appeared to be in an expansion phase. W ith the possible exception once again of this last recession, there appears to be greater correspondence in the dates of troug hs across countries than in th a t o f peaks.

APPENDIX 1: PROCEDURES FOR DATING CLASSICAL BUSINESS CYCLES The Brv-Boschan procedure

The approach em ployed by BB is related to the process o f tu rn in g point determ in ation practised by the NBER. it involves, in the firs t instance, the detection o f extrem e values and the ir s u b stitu tio n by in te rpo lated values. Turning points are then detected fo r a sm oothed series and these are sub seq ue ntly used as a basis fo r the id e n tific a tio n o f cycles.

The idea is th a t sm oo th in g (accom plished by applying a m oving average filte r) w ill sim p lify the id e n tific a tio n of expansions and con tra ction s and the selectio n o f the general neighbourhoods o f "p o te n tia l" peaks and troug hs. The procedure continues and the neighbourhood o f po tential turns is redefined by id e n tify in g peaks and troug hs corresponding to those of a series tha t is only s lig h tly sm oo th ed by a shorter m oving average. As a result o f the second filte r one com es closer to the eventual lo catio n o f the tu rning point. Once the im m ediate neighbourhood of potential tu rn s is established the analysis shifts back to the original data.

(30)

Table A 1: B rv -Boschan P rocedures

I. D ete rm ina tion o f e xtrem e s a n d s u b s titu tio n o f values

II. D ete rm ina tion o f cycle s in 12-m onth m o v in g average (extrem es replaced). A. Id e n tific a tio n o f p o in ts higher (or lo w e r) than G m o n th s on eith er side.

B. E nfo rcem en t o f a lte rn a tio n o f turns b y se le ctin g h ig h e s t o f m ultiple pe aks (or lo w e s t o f m u ltip le troughs)

III. D ete rm ina tion o f corre spo nding turns in Spencer curve (extrem es replaced). A. Id e n tific a tio n o f h ig h e s t (or lo w e s t) value w ith in ± 5 m onths. B. E nfo rcem en t o f m in im um cycle d u ratio n o f 15 m o n th s b y e lim inating lo w e r peaks a n d hig h e r troug hs o f s h o rte r cycles.

TV. D e te rm ina tion o f corre spo nding turns in s h o rt-te rm m o v in g average o f 3 to 6 m onths, de pending on M C D (m onths o f c y c lic a l dom inance).

A . Id e n tific a tio n o f h igh est (or lo w e s t) value w ith in ± 5 m o n th s o f se le cte d turn in Spencer curve.

V. D ete rm ina tion o f tu rn in g p o in ts in u n sm oo th ed series.

A . Id e n tific a tio n o f h ig h e st (or lo w e s t) value w ith in ± 4 m o n th s, o r M CD term , w h ic h e v e r is larger, o f selected turn in s h o rt-te rm m o v in g average. B. E lim ination o f tu rn s w ith in 6 m on th s o f be ginning an d e n d o f series. C. E lim ination o f peaks (or troughs) a t b o th ends o f series w h ich are lo w e r (or higher) than values closer to end.

D. E lim ination o f cycle s w hose du ration is less than 15 m onths. E. E lim ination o f phases w hose duration is less than 5 m onths. VI. S tatem e nt o f fin a l tu rn in g p oints.

Table A1 sho w s the procedure fo llo w e d by Bry-Boschan (1 9 7 1 ). The procedure in itially identifies, po tentia l tu rn in g points on the 12-m onth m oving average, then uses the Spencer curve to id e n tify the highest (low est) value w ith in ± 5 m onths of the selected turns in the 12 -m on th m oving average. W hen the set of te n ta tive turning points fo r the Spencer curve is established, a m inim um cycle duration requirem ent is im posed. The ne xt step is to map, once m ore, the po tential turning points id entified on the Spencer curve on to a shorter m oving average (3 to 6

(31)

m onths). S p e cifica lly the procedure identifies the highest (low est) value, fo r this sho rt m oving average, w ith in ± 5 m onths o f the selected turns in the Spencer curve. In the final stage the procedure repeats the exploration fo r tu rn s using the original, unsm oothed series, based on the record o f te n ta tiv e tu rning po in ts from the previous stage, th a t is, the sh o rt-te rm m oving average. In this stage, how ever, the program id entifies the highest (low est) value w ith in ± 4 m onths (or MCD term) of the selected tu rn s in the sh o rt-te rm m oving average. In ad dition oth er rules are enforced, including m inim um phase and cycle durations of 5 and 15 m onths respectively.

The procedure adopted

A sim plified version o f the BB procedure is fo llo w e d in this paper; Table A2 provides an o v e rv ie w o f this procedure. The results o f our procedure and those of BB are com pared in Table A3.

As in the BB procedure, the process begins w ith a search fo r "e x tre m e " values or outliers. This is necessary fo r the second stage in the id e n tific a tio n process since these extrem e values have to be adjusted before the sm oothing of the data. In this paper extrem e values are defined as those values w hose change com pared w ith both adjacent m onths is greater than 3 .5 standard errors of the (log) differenced series. Jum ps in the index o f industrial pro du ction, th a t is perm anent increases or decreases in the series, are the re fore not classified as extrem e observations. Ideally one w o u ld have liked to use the standard error of a corrected series as a

(32)

basis fo r the id e n tific a tio n o f extrem e values but th is is not practical. For such extrem e values, the series are adjusted prior to stage II10.

In the second stage the procedure determ ines cycles in a 7-m onth (centred) m oving ave ra g e '1. Values id en tified as local m inim a or m axim a are chosen as te n ta tiv e tu rning points and these are subjected to a sim ple te s t fo r the proper alternation of peaks and troughs.

The location o f locally extrem e po in ts is determ ined by the id e n tific a tio n o f points higher (lower) than 12 m onths on either side. We experim ented w ith 9 and 15 m onths but the 12 -m on th criterion proved to give the m ost s a tis fa c to ry balance betw een the elim ination o f too m any flu c tu a tio n s and the reten tion of shallow flu c tu a tio n s 12. As far as the en forcem e nt of the proper alternation o f peaks and troug hs in concerned, the procedure chooses, o f the tw o or m ore adjacent peaks (troughs), the highest (low est) one. A t the end of this process a list o f te n ta tive tu rning points is obtained.

10The value w h ic h is id entified as "e x tre m e " , say x,, is s u b s titu te d using the average o f the tw o adjacent observations, th a t is x," = (x ,, + x t+1)/2.

” BB use a long (12 m onth) and a sho rt (3 or 6 m onths) m oving average in their procedure. Since w e only ad op t a tw o stage procedure and use only one sm oothed series, a 7 -m o nth m oving average proved to be the best choice between shorter and longer m oving averages.

12The results of applying the 9, 12 and 15-m onth local m in im um /m axim um rule are available upon request.

(33)

Tahtp A ? : P rocedure fo r pro gra m m e d id e n tific a tio n o f tu rn in g p o in ts

I. D e te rm ina tion o f extrem e values

II. D e te rm ina tion o f Cycles in 7-m onth m oving average

A. Id e n tific a tio n o f p o in ts hig h e r flow er) than 12 m o n th s on eith e r side. B. E n fo rce m e n t o f a lte rn a tio n o f turns b y se le ctin g the h ig h e s t o f m u ltip le p e a ks (lo w e s t o f m u ltip le troughs).

III. D e te rm ina tion o f tu rn in g p o in ts on u n sm oo th ed series

A . Id e n tific a tio n o f p o in ts hig h e r (lower) than 12 m o n th s o n e ith e r side. B. E nfo rcem en t o f a lte rn a tio n o f turns b y s e le c tin g the h ig h e s t o f m u ltip le pe aks Ilo w e s t o f m u ltip le troughs).

C. Id e n tific a tio n o f fla t segm ents.

D. Id e n tific a tio n an d exclusion o f "o u tlie rs " fro m "p o s s ib le " tu rn in g po in ts.

E. E nfo rcem en t o f a lte rn a tio n o f turns b y s e le c tin g the h ig h e s t o f m u ltip le pe aks (lo w e s t o f m u ltip le troughs).

F. Id e n tific a tio n o f "sh o rt c y c le s " (less than 15 m o n th s fro m peak to peak o r troug h to trough).

G. M in im u m am p litu de rule req uiring the am p litu de o f a phase (peak to troug h o r tro u g h to peak) be a t least as b ig as 1 s ta n d a rd e rro r o f lo g changes.

IV. Com parison o f te n ta tiv e tu rn in g p o in ts selected fo r s m o o th e d a n d origin al series

A . Exclusion o f "p o s s ib le " tu rn in g p o in ts o f u n s m o o th e d series th a t do n o t co rre sp o n d to sim ila r turns ( ± 5 m onths) o f the m o v in g average.

From this po in t on w a rds the process of turning po in t selection d iffe rs fro m the BB m ethod. As w e po inted out earlier the process n o w m oves back to the original series (w ith o u t ad ju stm e nt fo r extrem e values) w hereas the BB procedure em ploys tw o fu rth e r interm ediate stages, nam ely the use o f the Spencer curve and a s h o rt term m oving average. In ad dition, our procedure iden tifies te n ta tiv e tu rn in g points for the original series independently of the turning po in ts id en tified in the sm oothed series. That is, the ne w process of iden tification beginning in stage III is n o t based

(34)

on any results obtained in the earlier stage II13. Those turning points selected in the original series, w h ic h do n o t correspond to turns in the 7 -m o nth m oving average, are excluded from the final list o f tu rn in g points.

In the stage III a num ber of additional rules are also em ployed, as listed in table A2. S pecifically, w here tu rning points are selected w ith in a sequence o f tw o or more adjacent and equal values, the procedure selects the last one o f these as the turning point. In ad dition, since the original series includes extrem e observations w h ich are n o t adjusted, it is possible th a t these are chosen as tu rn in g points.

Furtherm ore a m inim um duration of 15 m onths fo r cycles (peak to peak and trough to trough) is im posed. If sho rte r cycles are id entified the peak (trough) w ith the lo w est (greatest) value is excluded. Finally, in c o n tra st to BB, w e do set a m inim um am plitude rule: our procedure requires the am plitude from peak to trough and troug h to peak to be at least as large as one standard error of the (log) differenced series.

,3This is in co n tra s t to the BB procedure fo r w h ic h search in stages II, III and

(35)

T he dates com pared

Table A 3 com pares the dates derived using the Bry-Boschan procedure w ith the one derived after applying the alternative sim pler m etho do log y described here. There are tw o thin gs to note: firs tly , in the case o f a sequence o f equal values w e choose the last one as the cyclical turn in accordance w ith Bry and Boschan (19 71 ). This is not, ho w e ver, the case w ith the BB program w e are using, w h ich chooses the firs t one as the tu rn 14. Secondly, as noted in the te x t o f Section 2, the BB m ethod iden tifies m ore cycles than our procedure, w ith the differences relating to relative ly m ild m ovem ents.

,4BB assert th a t "In the case o f equal values the rule is to choose the la s t one

as the c y c lic a l turn, i.e ., the m o n th before the reve rsal o f the c y c lic a l process be gins", p.1 2. © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(36)

A 3 : A C o m p a ri s o n o f C la s s ic a l B u s in e s s C y c le C h ro n o lo g ie s f o r In d u s tr ia l P ro d u c ti o n © The Author(s). European University Institute. produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(37)

APPENDIX 2 DETAILS OF BUSINESS CYCLE CHARACTERISTICS

Tables A 4 -A 1 6 s h o w the cha racteristics o f classical business cycles experienced by the USA, Canada, Japan, G erm any, France, UK, Italy, Spain, Belgium, Netheriands, Ireland and Luxem bourg w ith in our sam ple period. Here in fo rm atio n is tabulated fo r each recession and expansion id en tified in Table I o f th e te x t. In each table, the firs t fo u r colum ns include the dates of the recessions and the average and c um u la tive (percentage) changes of in dustrial p ro d u ctio n over the this period and fin a lly the d u ratio n o f th a t phase (in m onths). The n e x t fo u r colum ns display sim ilar ch a ra cte ristics fo r the expansions.

Dates fo r each expansion (con tra ction ) refer to the period beginning the m o n th after the turning point. T h a t is an expansion (contraction) phase begins the firs t m onth after the troug h (peak). T h ° dates in the tables therefore refer to the firs t m onth of each phase rather than to the tu rning point itself.

(38)

CO O CO CO co co «- o IO CM CO IO co o O 3 h- O IO >— o o lo »- r* «-co co q q co cri ò cm cm co CN IO CM CM 0) -* O) co c £ CO Ü . cu -C I ? o o CO CO i“ CM CO ^ co n r' io t— cm lo co Ò - - C O CO (N CO CM CM « -o co co ò CN (U co c CU CD co CD CDco IO LO CD CO (J O O O O 2 CO CD <3-6 c V o> o a> Û. < co D < 03 -Q (0 c co CM o T— ₅ CO 'io o CD (U a X LO CO co CO CN i LO co co 5 i CO co o o r> CO C o co co r-* co LO CM tu 3 Q tu x z CN co o co CM CM c o CO CN 00 CM o tu o CO o CD q C D C D 5 >7 co IO ai oó T S 8 o -* |_ tu r ~ Q) CL CJ 3 "O o CL (0 in 3 ■o c 0) co c (0 -C o « co <c 0) CL (P T3 (0 c (0 O IO < _0J o « c cu 2 o tu 3 O co LO CO CM LO LO T— CN co o CM o 00 r> co IO ò ò o ò ò ò co co r -CD r— CO <D r^. CO co 00 CO ai »— 2 s IO o co o f"* CO CN co oo LO o co CN CN co co CO LO co CM o CM r— CM CD co r - C D q r ~ q «— CM ai ai r— CN CO _ _LO _o _co co o CD LO o co LO q LO CN r** O _T v o o o tu co c tu (J _ CO _ co Is*- o CD CO co CO LO io CM CO CD LO co O O Ò O o o o c CO Is* CN co o r— t— '(/> LO o o r^ co CN q o _CM _'t 00 _LO C D *— «— 5 àò LO CO O 00 2 S ^ 2 LO o -2 CD N CO IN CO ^ 05 5 CN CN CD LO CU CO tu 0) > < © The Author(s). European University Institute. produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.

(39)

T a b le A 6 : J a p a n - P e rc e n ta g e c h a n g e in In d u s tr ia l P ro d u c ti o n c o C' CO co in CO CO r*' o 00 CO en n n* n o to (0 3 3 o O 05 3) T3 r-' CD ,_ CO a> -x O «3 CN O CO 00 C 05 CO o CD C 05 CO CD CO o CO Û- o r— q CO 03 Q- 05 q t— 6 2 05 05 CO CO 5 2 n i in CO CO CO CN CN CN CO CO 5 ci> 2 H i O 3 O 3 h- o h- o f- H <u O) CO CD 05 in 0505 CN CO CN CD CO 05 CO C n- r-' CN in CÔ 'tf- CO n (0 in CO _c (J 6 d 6 d co .c (J d d d O c c CO o «3 0) 3 05 ■o o 2 a l c in m £ C CO CN CD O T - 3 o :> *“ IE to _in ■o t/5 _CO Cl 0 0 ₅ c j~ 05 05 a ra 05 c a CD n * _CN X LU i CD 0> o> LU 2n -CO 5 i u o cn 0 0 c CD in CN 0 0 (0 r '' 0 0 -C o a) C o in CO o><0 Co CO in T— T-- *— r — »— CN CO CN (0 © (0 3 o 3 Q « a CL a> n O) O) CO CD i ' ' in > . 05 _Ccn o) o T_ 05 CO C 3 LO 05 c C 3 CD CN CO n-(0 o CD CN o «0 (0 o q q *— CO 6 h ni CO d E 6 H • 05 n i n i « 2 CN « o ™ 2 o -* o -* l— CO ^ Q) CL r*i < l— Ct3 ^ 05 Q_ 0) -Q (0 a> o> 0 0 CO CO CO »- 05 O) r** CN n - 05 c 0 0 CN C CO (0 CD CO CN n- <0 CO in CO in £ (J «7 O O o _(Jsz o O O O . c c CO (0 © 05 5 c CO o 00 05 c n n * 05 o ₅ ₅ 03 (0 g 5 5 »— O)re CO in «E CD »(0 05 O r * ' in IE 0) o n* ni CN 00 00_CD a) > < CD n -05 CN Ç0 05 > < a: 2 CN 2 cn 2 2

T—

'tf

T—

in CD CO E 00 CD n - o 00 00 © The Author(s). European University Institute. version produced by the EUI Library in 2020. Available Open Access on Cadmus, European University Institute Research Repository.