Transmission of Financial Stress Shocks between the USA and the Euro Area During Different Business Cycle Phases

152 2020, XXIII, 4 Finance DOI: 10.15240/ ul/001/2020-4-010 TRANSMISSION OF FINANCIAL STRESS SHOCKS BETWEEN THE USA AND THE EURO AREA DURING DIFFERENT BUSINESS CYCLE PHASES Sil o Dajčman1, Alenka Ka kle 2, Pe e Mikek3, Dejan Romih4 1 Uni e si y o Ma ibo , Facul y o Economics and Business, Depa men o Economic Policy, Slo enia, [email p o ec ed]; 2 Uni e si y o Ma ibo , Facul y o Economics and Business, Depa men o Quan i a i e Economic Analysis, Slo enia, [email p o ec ed]; 3 Wabash College, Depa men o Economics, Uni ed S a es o Ame ica, [email p o ec ed]; 4 Uni e si y o Ma ibo , Facul y o Economics and Business, Depa men o In e na ional Economics and Business, Slo enia, ORCID: 0000-0001-9123-0183, [email p o ec ed]; Abs ac : This pape examines he ansmission o inancial s ess shocks be ween he USA and he eu o a ea o ecessiona y and non- ecessiona y egimes in he shock- ecipien economy. The in es iga ed pe iod is 1999M1–2017M11, which includes se e al episodes o ecessiona y and non- ecessiona y egimes, endogenously de e mined by he model, as well as se e al inancial s ess episodes. A e es ing o non-linea i y, we employ a i e- a iable Bayesian h eshold ec o au o eg ession model using in e na ionally compa ible da a o inancial s ess indices. Ou esul s show signi ican non-linea i ies in he inancial s ess-business cycle in e ac ions o he eu o a ea. In compa ison o he non- ecessiona y egime, he US inancial s ess shocks a e mo e de imen al o he s abili y o he Eu opean inancial sys em, ou pu g ow h, and in la ion in ecessions. US inancial s ess shocks nega i ely a ec eu o a ea unemploymen a e, bu he e ec is independen o he eu o a ea indus ial p oduc ion g ow h egime. In con as , he s abili y o he US inancial sys em is no suscep ible o he eu o a ea’s inancial s ess shocks. Howe e , due o ade ies, he inancial s ess in he eu o a ea does lead o ou pu con ac ion, while no a ec ing in la ion and unemploymen in he US. We also ound ha US indus ial p oduc ion g ow h and unemploymen a e a e suscep ible o domes ic inancial s ess shocks, mo e in he ecessiona y han non- ecessiona y episodes o he US economy. The esul s sugges a need o a ca e ul domes ic and o eign inancial s ess moni o ing and coo dina ion o mone a y au ho i ies. While his may p o i bo h economic a eas, his is ele an mo e o he Eu opean Cen al Bank han i s US coun e pa . Keywo ds: Financial s ess, in e na ional ansmission o inancial s ess, business and inancial cycles, nonlinea i y, h eshold VAR. JEL Classi ica ion: E32, E44, G01. APA S yle Ci a ion: Dajčman, S., Ka kle , A., Mikek, P., & Romih, D. (2020). T ansmission o Financial S ess Shocks be ween he USA and he Eu o A ea Du ing Di e en Business Cycle Phases. E&M Economics and Managemen , 23(4), 152–165. h ps://doi.o g/10.15240/ ul/001/2020-4-010 In oduc ion A e he global inancial c isis and he G ea Recession, a la ge and g owing body o li e a u e has examined eal business- inancial cycle linkages. To his end, Claessens e al. (2012) examined a la ge da abase o business and inancial s ess pe iods, co obo a ing ha inancial c isis pe iods a e o en longe and deepe han economic ecessions and end o ampli y and p olong he la e . Ou esea ch aims EM_4_2020.indd 152 18.11.2020 12:28:18 153 4, XXIII, 2020 Finance o con ibu e o an unde s anding o he inancial s ess-mac oeconomy nexus by s udying he spillo e s o US (eu o a ea) inancial s ess shocks and hei mac oeconomic e ec s (i.e. e ec s on indus ial p oduc ion, in la ion and unemploymen ) in o he eu o a ea (USA). This pape asks whe he hese e ec s a e con ingen on he phase o he business cycle. T adi ionally, domes ic and in e na ional inancial s ess-business cycle linkages ha e been in es iga ed wi hin he linea modelling amewo k. Recen ly, he linea amewo k has been c i icized by au ho s employing non- linea dynamic s ochas ic gene al equilib ium (DSGE) models (B unne meie & Sanniko , 2011; Mi nik & Semmle , 2012, 2013; Schlee & Semmle , 2014; Boissay e al., 2013; Chen & Semmle , 2014) o hei limi ed powe o explain he di e en ial mac oeconomic e ec s o inancial shocks in di e en egimes o he economy. This li e a u e con ends ha unde high inancial s ess o ecessiona y egimes he economy could unc ion di e en ly han in low inancial s ess o non- ecessiona y egimes. I iden i ies se e al ampli ica ion mechanisms which can swi ch he economy om one egime o ano he . Ou esea ch builds on he amewo k o his line o esea ch. To he bes o ou knowledge, only E genidis and Tsagkanos (2017) and Chen and Semmle (2018) ha e in es iga ed he egime-dependen mac oeconomic e ec s o he in e na ional ansmission o inancial s ess shocks. Bo h s udies used he inancial s ess index o he USA as a h eshold a iable and explained how he eu o a ea economy esponds o US inancial s ess shocks, condi ional on he US inancial s ess egime. The li e a u e does no show how inancial s ess shocks a e ansmi ed be ween he wo majo economies, based on he domes ic business cycle egime. We asse ha an answe o his ques ion is ele an o policymaking and aim o ill his gap in he li e a u e. Ou esea ch also ex ends he li e a u e ha uses he Composi e Indica o o Sys emic S ess (CISS), cons uc ed by Holló e al. (2012) and main ained by he Eu opean Cen al Bank (ECB). The CISS agg ega es a ious s ess indica o s o i e majo segmen s o he inancial sys em ( inancial in e media ies, money ma ke s, bond ma ke s, equi y ma ke s and o eign exchange ma ke s) based on hei ime- a ying co-mo emen s ( u he de ails a e gi en in he Appendix). As he la e a e s onge when he s ess is ele a ed in se e al segmen s o he inancial sys em, his indica o is conside ed o cap u e he ma e ializa ion o sys emic isk (Holló e al., 2012; K eme , 2016a, 2016b). As such, i has a “subs an ial and obus explana o y powe o s anda d mac oeconomic a iables” (K eme , 2016b), as demons a ed in a ious esea ch applica ions (e.g. Holló e al., 2012; K eme , 2016a, 2016b; Adam & Benecká, 2013). Fu he mo e, we complemen he exis ing li e a u e by explici ly including he labou ma ke esponse o spillo e s o a inancial s ess shock. The li e a u e on he inancial- eal sec o nexus has la gely neglec ed he e ec o a inancial s ess on unemploymen . Rela ed li e a u e on he in e na ional ansmission o unce ain y shocks, howe e , illus a es ha he ela ionship is non-linea (Mo ley & Pige , 2012; Caggiano e al., 2014, 2017). This is due o unce ain y shocks p o oking a s onge su ge in he unemploymen a e du ing ecessions han in non- ecessions. The empi ical esea ch in he pape is based on he Bayesian h eshold VAR (TVAR) modelling amewo k. Recessiona y and non- ecessiona y egimes du ing pe iod 1999M1– 2017M11 a e de e mined by he model and hen he non-linea i y o ansmission o in e na ional inancial s ess shocks o mac oeconomic a iables (indus ial p oduc ion g ow h, in la ion, and unemploymen ) is s udied by gene alized impulse esponses. The esul s show signi ican ansmission o US inancial s ess shocks o eu o a ea, and non-linea i ies in he ansmission. The emainde o he pape is s uc u ed as ollows: Sec ion 1 e iews he exis ing li e a u e, while Sec ion 2 desc ibes he me hodology. Sec ion 3 p esen s he da a and empi ical esul s, Sec ion 4 discusses he esul s and he inal sec ion summa izes he main indings. 1. Li e a u e Re iew One s and in he empi ical li e a u e on inancial s ess-business cycle linkages uses linea ec o au o eg ession (VAR) models o illus a e he des abilizing e ec s o inancial s ess shocks on he mac oeconomy (e.g. Hakkio & Kee on, 2009; Gilch is & Zak ajsek, 2012; K eme , 2016a). Ano he s and in he empi ical li e a u e also applies linea modelling o show ha inancial s ess (o s ess in some EM_4_2020.indd 153 18.11.2020 12:28:18 154 2020, XXIII, 4 Finance segmen o he inancial sys em, e.g., he c edi ma ke ) o igina ing in one coun y can spill ab oad and cause de imen al mac oeconomic de elopmen s (Helbling e al., 2011; Do e n & an Roye, 2014; Eickmeie & Ng, 2015; Ha e al., 2017). Theo e ical unde pinnings o such empi ical modelling o business and inancial cycle linkages can be ound in inancial accele a o models (Be nanke & Ge le , 1989; Kiyo aki & Moo e, 1997; Ge le e al., 2007; Ge le & Kiyo aki, 2010). In inancial accele a o models, inancial s ess shocks exace ba e nega i e economic de elopmen s. The e ec s, howe e , a e mean- e e ing. As a gued by Schlee and Semmle (2014), his is because he inancial- eal sec o in e ac ions a e modelled as linea . In non- linea DSGE models he ampli ying e ec s o inancial s ess shocks a e asymme ic o egime-dependen : while hey a e s ong and mo e du able when he economy is unde a egime o high inancial s ess o ecession, hey a e small o insigni ican when he inancial sys em o economy is in a non-c isis egime (B unne meie & Sanniko , 2011; Mi nik & Semmle , 2012, 2013; Schlee & Semmle , 2014; Hub ich & Te low, 2015). A key ole in he swi ch om a no mal o a c isis egime in non-linea DSGE models is played by banks’ balance shee s and a ious ampli ica ion mechanisms. In he model o B unne meie and Sanniko (2011), upon an exogenous shock o he inancial sys em, la ge des abilizing e ec s and egime shi s (ampli ica ion) can be he esul o an endogenous esponse conce ning asse p ices due o p ecau iona y sa ings and ola ili y. O he ampli ica ion mechanisms can also be a wo k, e.g., mo emen s in isk p emia and c edi sp eads (Mi nik & Semmle , 2012, 2013), in e play be ween isk p emia, c edi cons ain s and ex ensi e dele e aging o bo owe s (Chen & Semmle , 2014), a weal h educing e ec on consump ion and in es men when asse p ices all, o a “diabolic loop” in he in e play be ween p i a e bo owe s, banks and so e eign deb (Schlee & Semmle , 2014). The empi ical li e a u e explo ing non- linea i ies in he dynamic ela ionship be ween domes ic inancial s ess and he mac oeconomy is ela i ely hin and, in gene al, suppo s he heo e ical p edic ion o non- linea i y. Classical TVAR (Mi nik & Semmle , 2013; Schlee & Semmle , 2014; Mi nik & Semmle , 2014; Chen & Semmle , 2014), Bayesian Ma ko -swi ching VAR (Abou a & an Roe, 2017; Ha mann e al., 2013), o Bayesian TVAR (Alessand i & Mum az, 2017; Cha e jee e al., 2017) modelling app oaches a e applied, while (gene alized) impulse esponses a e compu ed o in e s uc u al ela ionships be ween he a iables. Mos s udies se he inancial s ess indica o as a h eshold a iable. Howe e , Mi nik and Semmle (2012) s udy he e ec o inancial s ess shocks on economic ac i i y du ing low and high inancial s ess egimes, whe eas Cha e jee e al. (2017) ocus on ecessiona y and non- ecessiona y egimes. Al hough mos esea che s design a pa simonious bi a ia e model, some au ho s include up o h ee addi ional mac oeconomic a iables in hei model (Alessand i & Mum az, 2017; Ha mann e al., 2013; Cha e jee e al., 2017), including he in la ion a e, he mone a y policy o sho - e m in e es a e, g ow h in loans, o g ow h in he nominal e ec i e exchange a e. A a ie y o inancial s ess indices a e used: he IMF Financial S ess Index (Mi nik & Semmle , 2012, 2013; Chen & Semmle , 2014), he ZEW Financial Condi ion Index (Schlee & Semmle , 2014), he Chicago Fed Financial Condi ions Index (Alessand i & Mum az, 2017), he CISS (Ha mann e al., 2013) o sel - cons uc ed indices (Abou a & an Roye, 2017; Cha e jee e al., 2017). The e a e nume ous s udies ha explo e he non-linea esponse in indus ial p oduc ion g ow h o a inancial s ess. Ac oss a a ie y o s udied coun ies, hey ind an asymme ic esponse in indus ial p oduc ion g ow h o a posi i e shock in inancial s ess. Mi nik and Semmle (2012) s udied majo Eu opean coun ies (excluding he UK and F ance), Mi nik and Semmle (2013) looked a he USA, Ge many, I aly and F ance, Alessand i and Mum az (2017) conside ed he USA, Cha e jee e al. (2017) examined he UK, Abou a and an Roye (2017) in es iga ed F ance, Chen and Semmle (2014) add essed 15 OECD coun ies, Schlee and Semmle (2014) e iewed 10 majo eu o a ea coun ies, and Ha mann e al. (2013) app oached he eu o a ea as a whole. They o e whelmingly epo ed a signi ican , egime-dependen d op in p oduc ion g ow h in esponse o a inancial s ess shock. In gene al, he educ ion is la ge in ecessiona y o high inancial s ess egimes. Fo example, Mi nik EM_4_2020.indd 154 18.11.2020 12:28:19 155 4, XXIII, 2020 Finance and Semmle (2013) es ima ed he esponse o be 2.5 imes s onge and Alessand i and Mum az (2017) e en epo ed a six old s onge esponse in he ecession/high s ess egime in he USA. Addi ionally, Ha mann e al. (2013) ind a nega i e esponse in he in la ion a e, g ow h in loans and in e es a e o he eu o a ea in a high s ess egime. Beyond he li e a u e on dynamic ela ionship be ween domes ic inancial s ess and he b oade economy, se e al channels o in e na ional inancial s ess ansmission ha e been iden i ied, including he ade channel ( inancial s ess shocks a e mo e likely o be ansmi ed and a ec he economic ac i i y o coun ies ha mu ually ade mo e), inancial ma ke in eg a ion ( he capi al ma ke and global ope a ion o banks), global shocks, and con agion (see Apos olakis & Papadopoulos, 2014; Do e n & an Roye, 2014, and e e ences he ein). As al eady no ed, only E genidis and Tsagkanos (2017) and Chen and Semmle (2018) in es iga e he egime-dependen e ec s o he in e na ional ansmission o inancial s ess shocks. Chen and Semmle (2018) apply a bi a ia e ( he IMF’s Financial S abili y Index and indus ial p oduc ion g ow h) global VAR (GVAR) model o in es iga e he e ec s o inancial s ess shocks o igina ing in he USA (as a case o a la ge coun y) and Belgium (as a small coun y) on inancial s ess and indus ial p oduc ion g ow h in 15 OECD coun ies. The impulse esponses show ha , in a high inancial s ess egime, a inancial s ess shock in he USA leads o a s a is ically signi ican inc ease in he domes ic inancial s ess in i e coun ies. This con as s wi h only one coun y o Belgium. Unde a high s ess egime a home, indus ial p oduc ion dec eases signi ican ly only o one coun y due o a US inancial s ess shock and none in he case o Belgium. In con as , unde a low s ess egime, a inancial s ess shock in he USA and Belgium signi ican ly inc eases domes ic inancial s ess indices in all obse ed coun ies. Howe e , he esponses o domes ic indus ial p oduc ion g ow h a e no signi ican ac oss all coun ies. E genidis and Tsagkanos (2017) analysed he in e na ional ansmission o inancial s ess shocks om he USA o he eu o a ea by speci ying a model wi h se e al endogenous a iables (including indus ial p oduc ion, capi al lows, sho - e m in e es a e and inancial s ess index) and a se o exogenous a iables (including p oxies o unce ain y in he USA and he eu o a ea, gold p ices, s ock ma ke p ices, commodi y p ices, and expec a ions abou u u e in la ion). Thei esul s indica e ha , unde a high inancial s ess egime in he USA, a posi i e shock (i.e. inc ease) o US inancial s ess is de imen al o he eu o a ea’s economic ac i i y, in e es a e and inancial s ess. The esponses a e no signi ican unde a low s ess egime. Besides he di e ences no ed in he in oduc ion, ou pape di e s om he s udies o Chen and Semmle (2018) and E genidis and Tsagkanos (2017) in se e al o he impo an ways. Fi s , bo h s udies use a inancial s ess index as he h eshold a iable. The economies in es iga ed in hese s udies a e hus ei he in a high o in a low inancial s ess egime. Following Cha e jee e al. (2018) and Chen and Semmle (2014), he h eshold a iable in ou pape is g ow h in domes ic indus ial p oduc ion. Condi ional on he h eshold alue, he conside ed economy is hus in ei he a ecessiona y (low g ow h) o a non- ecessiona y egime. We, hus, can answe di e en ques ions o hose in he e e enced s udies. Second, in Chen and Semmle (2018), a bi a ia e TVAR is applied, while E genidis and Tsagkanos (2017) conside se e al endogenous and exogenous a iables. We es ima e a i e- a iable TVAR model in which, as commonly ound in he li e a u e, all a iables a e endogenous. Thi d, E genidis and Tsagkanos (2017) use di e en inancial s ess indica o s o economies: he S . Louis Fed Financial S ess Index o he USA and he CISS o he eu o a ea. As he e a e ce ain di e ences in he cons uc ion o di e en inancial s ess indica o s ( o a e iew, see Kliesen e al., 2012), we ind i impo an o use indica o s cons uc ed by employing he same me hodology, especially i he esea ch aims o examine egime-dependen in e na ional s ess spillo e s (shock in o eign inancial s ess spilling o e in o domes ic inancial s ess) and he ansmission o inancial s ess shocks. Fou h, while he e e enced s udies apply a classical TVAR, we eso o Bayesian me hods in TVAR es ima ion and in e ence, which is ad an ageous o e classical maximum likelihood me hods in es ima ion and in e ence (impulse esponse analysis) (see Koop & Po e , 1999; Kwon, 2003). EM_4_2020.indd 155 18.11.2020 12:28:19 156 2020, XXIII, 4 Finance 2. Me hodology To explo e he USA-eu o a ea inancial s ess spillo e s and hei egime-con ingen mac oeconomic e ec s, we employ a Bayesian TVAR model. As no ed by Alessand i and Mum az (2017), a TVAR is capable o cap u ing a s uc u al b eak associa ed wi h inancial o economic c ises. Ano he use ul ea u e o he model is ha he h eshold a iable is one o he endogenous a iables in he model and ha he h eshold alue which swi ches he economy om one egime o ano he is de e mined by he da a and no subjec i ely by he esea che . To s udy he inancial s ess spillo e s om he USA in ecessiona y and non- ecessiona y egimes in he eu o a ea, we apply he ollowing wo- egime TVAR model (see, e.g., Chen & Lee, 1995): (1a) whe e =1, …,T, X = [y EA, u EA, π EA, ciss US, ciss EA]T is a ec o o endogenous a iables consis ing o se e al domes ic (eu o a ea) a iables and one o eign (US) a iable. In his we ollow se e al ecen s udies on he e ec s o in e na ional spillo e s o unce ain y shocks on mac oeconomy (e.g. Caggiano e al., 2017; Fon aine e al., 2017; Huang e al., 2018). The eu o a ea a iables included a e he annual, i.e., yea -on-yea , g ow h in indus ial p oduc ion (see, e.g., Ha mann e al., 2013; K eme , 2016a) (y EA), unemploymen a e (u EA), annual in la ion (π EA), and (sys emic) inancial s ess le el in he eu o a ea (ciss EA). The US a iable is he US inancial s ess (ciss US). z –d is a h eshold a iable; d is h eshold lag; z* is he unknown h eshold alue; c1 (c2, espec i ely) is a egime-speci ic 5 × 1 ec o o cons an s; p is he lag leng h o he TVAR; A1,i (A2,i, espec i ely) is he egime-speci ic ma ix o coe icien s o lag i. e1 (e2 , espec i ely) is a 5 × 1 ec o o i.i.d. egime-speci ic e o s (see Chen & Lee, 1995). The h eshold a iable in (1a), z –d, is he d- h lag o y EA. Financial s ess spillo e s om he eu o a ea in o he USA and hei mac oeconomic e ec s in ecessiona y and non- ecessiona y egimes o he US economy a e s udied by an iden ical model wi h a sligh ly di e en endogenous ec o , i.e., X : (1b) whe e X = [y US, u US, π US, ciss US, ciss EA]T; and y US, u US and π US e e o annual g ow h in indus ial p oduc ion, unemploymen a e and in la ion in he USA. The h eshold a iable in (1b), z –d, is he d- h lag o y US. Depending on z –d, he economy desc ibed by (1a) and (1b) is hus endogenously labelled as being in a ecessiona y egime when z –d ≤ z*, o in a non- ecessiona y egime when (z –d > z*). No e ha his de ini ion o ecessions da ing may di e om hose, e.g., o he NBER and he CEPR, which obse e qua e ly eal GDP g ow h and se e al o he economic a iables in o de o iden i y business cycle u ning poin s. The model is piece-wise (wi hin- egime) linea . Each ow o (1a) and (1b) hus desc ibes a egime-speci ic VAR model. The model, howe e , is non-linea in ime (Chen & Lee, 1995). Following he commonly used app oach, he lag, p, o Models (1a) and (1b) is de e mined by applying he Akaike in o ma ion c i e ia o he linea VAR model (limi ing he maximum lag (p) o 12): , (2) whe e X is as de ined abo e o (1a) and (1b), espec i ely; c is a 5 × 1 ec o o eg ession cons an s; Ai is a ma ix o coe icien s o lag i; and e is a 5 × 1 ec o o i.i.d. e o s. A e de e mining he lag o de , Models (1a) and (1b) a e es ima ed by Bayesian me hods, applying he app oach o Alessand i and Mum az (2017) who ollow Sims and Zha (1998) and Banbu a e al. (2010). Mo e speci ically, we use he same no mal in e se Wisha p io o es ima ing he egime-speci ic VAR pa ame e s. We closely ollow Alessand i and Mum az (2017) and use he AR(1) es ima es o he indi idual endogenous a iables o ob ain he p io s, means and scaling ac o s. The o e all p io is τ = 0.1, while he p io o he cons an is se a 0.0001, and he igh ness o he p io s in he sum o coe icien s is λ = 10τ. The p io o he h eshold lag is se a a maximum o 4, while he p io o he h eshold alue is no mally dis ibu ed, i.e., z* ~ N(z, s), whe e z is he sample mean o he h eshold a iable and s is he a iance in he p io o a iable z . EM_4_2020.indd 156 18.11.2020 12:28:19 157 4, XXIII, 2020 Finance Condi ional on he ini ial z* and d, he condi ional pos e io dis ibu ion o he egime-speci ic VAR coe icien is no mal and he condi ional pos e io o he a iance-co a iance ma ix is in e se Wisha dis ibu ed. The Gibbs sample o Chen and Lee (1995) wi h 15,000 i e a ions and 5,000 bu n-ins is used o d aw he pos e io dis ibu ions. The h eshold alue is d awn om he Me opolis Has ings s ep. Fo a mo e comp ehensi e desc ip ion o he se ings, e e o Alessand i and Mum az (2017). Dynamic esponses om mac oeconomic a iables o inancial s ess shocks a e explo ed using he gene alized impulse esponses, i.e., non-linea impulse esponse analysis, sugges ed by Koop e al. (1996). The gene alized impulse esponses o a egime R(GIRFR ), whe e R = ecessiona y, non- ecessiona y, a e de ined as ollows (Koop e al., 1996; Alessand i & Mum az, 2017): , (3) whe e E deno es expec a ion; h is he ime ho izon; Φ a e all he pa ame e s o Models (1a) and (1b), espec i ely; XR –1 is he egime-speci ic his o y; and η is a shock. Gi en he pa ame e s o Models (1a) and (1b), espec i ely, and he egime-speci ic his o y, he i s e m on he igh side o (3) is he condi ional expec a ion o he endogenous a iable unde a shock (a one s anda d de ia ion shock conce ning he a iable o in e es ), while he second e m is he condi ional expec a ion o he endogenous a iable wi h no shock. Empi ically, he condi ional expec a ions o a speci ic egime R a e es ima ed by simula ing he model (100 eplica ions) o all possible s a ing alues in ha egime. Impulse esponses (3) a e hen compu ed (by 500 Gibbs i e a ions) o all his o ies in he egime. Mean esponses and he 68% con idence in e al a e inally compu ed. The compu a ion is pe o med by he MATLAB codes o Alessand i and Mum az (2017). We wish o hank he au ho s o making hei code a ailable. S uc u al iden i ica ion o inancial s ess shocks is achie ed h ough he Cholesky iden i ica ion scheme, as is s anda d in he TVAR li e a u e. Simila ly, ollowing he e e enced li e a u e, we place he eal economic a iables (y EA, u EA) and in la ion (π EA) i s , ollowed by he inancial a iables (ciss US, ciss EA). This implies ha he inancial a iables a e con empo aneously a ec ed by eal economic a iables, bu no he e e se. We place US inancial s ess be o e ha o he eu o a ea, as in Chen and Semmle (2018) and se e al o he s udies on he in e na ional ansmission o shocks om he USA (e.g. Do e n & an Roye, 2014; Caggiano e al., 2017; Huang e al., 2018). This is in line wi h he no ion o he la ge global impac o he US economy and i s inancial ma ke s as compa ed o he eu o a ea. This o de is also suppo ed g aphically (see Fig. 1 in con inua ion) due o he ac ha majo inancial s ess e en s du ing he sample pe iod, which o igina ed in he USA, no only ele a ed inancial s ess in he USA, bu also con empo aneously in he eu o a ea, whe eas, du ing a majo s ess e en in he eu o a ea ( he eu o deb c isis), his was no he case. By placing ciss US in he penul ima e place, we assume ha sys emic inancial s ess in he USA is con empo aneously a ec ed by all eu o a ea “ eal” mac oeconomic a iables (indus ial p oduc ion, unemploymen , in la ion), albei only wi h a lag ega ding he eu o a ea inancial s ess. By placing ciss EA las , we assume ha all o he a iables in he model con empo aneously a ec inancial s ess in he eu o a ea, whe eas he eu o a ea’s inancial s ess a ec s o he a iables albei only wi h a lag. 3. Da a and Empi ical Resul s The da a consis o mon hly obse a ions o he pe iod 1999M1–2017M11, limi ed a he s a by he bi h o he eu o a ea and a he end by he a ailabili y o da a o he a iable ciss US. A de ailed desc ip ion o all a iables wi h s a is ical summa y is con ained in Tab. 1. We can obse e ha he mean annual g ow h in indus ial p oduc ion in eu o a ea was sligh ly (0.02 pe cen age poin s) highe bu also mo e ola ile han in he USA. Mean unemploymen a e in eu o a ea was 3.42 pe cen age poin s highe han in he USA, bu less ola ile, while mean in la ion a e in eu o a ea was 0.43 pe cen age poin s lowe and less ola ile han in he USA. CISS in eu o a ea was on a e age highe and mo e ola ile han in he USA. The dynamics o and and he ecessiona y egimes, as de e mined by (1a) and (1b), a e p esen ed in Fig. 1. Clea y, he inancial s ess in he eu o a ea and in he USA du ing he EM_4_2020.indd 157 18.11.2020 12:28:19 158 2020, XXIII, 4 Finance Va iable no a ion Mean Max Min S . de . Desc ip ion y EA 0.98 9.34 −21.32 4.87 Annual g ow h in indus ial p oduc ion (pe cen age change compa ed o he same mon h in he p e ious yea ) o o al indus y, excluding cons uc ion; mon hly equency; calcula ed om he mon hly seasonally adjus ed index (2015 = 100). Da a sou ce is OECD (2019a). y US 0.96 8.47 −15.34 4.18 u EA 9.46 12.1 7.3 1.30 Unemploymen a e, seasonally adjus ed, exp essed as a pe cen age o he ac i e popula ion; mon hly equency. Da a sou ce is Eu os a (2019). u US 6.02 10.0 3.8 1.76 π EA 1.75 4.12 −0.62 0.96 In la ion a e (pe cen age change compa ed o he same mon h in he p e ious yea ); mon hly equency; calcula ed om he consume p ice index (2015 = 100). Da a sou ce is OECD (2019b). π US 2.18 5.6 −2.1 1.28 ciss EA 0.19 0.78 0.03 0.16 The CISS o he eu o a ea. Mon hly alues a e calcula ed om weekly alues acco ding o he a i hme ic a e age. The CISS is cons uc ed by Holló e al. (2012), main ained by he ECB and compu ed in se e al s eps. Fi s , h ee di e en inancial s ess measu es o each o i e segmen s o he inancial sys em ( inancial in e media ies, money ma ke s, bond ma ke s, equi y ma ke s and o eign exchange ma ke s) a e collec ed. In he second s ep, he da a on indi idual s ess measu es a e a anged in ascending o de , hen ans o med in o hei empi ical cumula i e dis ibu ion unc ion. A e his ans o ma ion, 15 indi idual s ess ac o s a e dis ibu ed in he in e al (0.1). Nex , he s ess ac o s o each o he i e segmen s a e agg ega ed acco ding o he a i hme ic a e age o ob ain i e subindices. The inal s ep consis s o agg ega ing he subindices o a composi e indica o o inancial s ess (CISS), based on he subindices’ weigh s ( he mos weigh is ca ied by inancial in e media ies and equi y ma ke subindices) and ime- a ying c oss-co ela ions be ween he subindices. The ob ained CISS indica o is dis ibu ed o e he in e al (0.1) and measu es he “ex-pos sys emic s ess, i.e. isk ha has ma e ialised al eady” (Holló e al., 2012). Da a sou ce is ECB (2019). ciss US 0.14 0.73 0.02 0.13 The CISS o he USA is measu ed by he same me hodology as . Whe eas as well as a simila indica o o indi idual eu o a ea coun ies is main ained by he ECB, he CISS o he USA was calcula ed by K eme (2016b). An ex ended ime se ies o wi h a weekly equency un il No embe 10, 2017 was calcula ed and kindly sha ed by Man ed K eme . Da a sou ce is K eme (2016b). Sou ce: own Tab. 1: Da a summa y and desc ip ion EM_4_2020.indd 158 18.11.2020 12:28:19 159 4, XXIII, 2020 Finance obse ed pe iod is s ongly connec ed ( he co ela ion coe icien is 0.872). The CISS o he eu o a ea and he USA aces ou he majo e en s ha caused u moil in he espec i e inancial ma ke s. The la ges spikes in he CISS o bo h a eas a e associa ed wi h he global inancial c isis and he eu o a ea so e eign deb c isis. O he impo an e en s in he p e-global inancial c isis pe iod coincide wi h he do -com c isis, he 9/11 e o a acks in he USA, and he En on and Wo ldCom bank up cies. Clea ly, he decoupling o he CISS o he wo a eas is obse able du ing he eu o a ea’s so e eign deb c isis, implying ha he sys emic c isis was limi ed o he eu o a ea. The inancial s ess in he eu o a ea dec eased in 2013. In he pe iod om 2014 un il mid-2016, a endency o inc easing s ess is no iceable in bo h a eas. As he ECB (2016) no es, his is ela ed o he G eek bailou e e endum in July 2015, u moil in he Chinese s ock ma ke u moil, and he unce ain y spu ed by he B exi e e endum. In 2017, he CISS in bo h a eas e u ned o ela i ely low le els, indica ing ela i e inancial s abili y. Fig. 1 also p esen s ecessiona y egimes in bo h economies. Fo he eu o a ea, ecessions de e mined by Model (1a) (i.e. pe iods o which z –d ≤ z*) success ully ace ou he ecession pe iods, as de ined by he Cen e o Economic Policy (CEPR). The model also e eals some sho e pe iods o indus ial p oduc ion con ac ions a he end o 2001 and 2003 as ecessiona y. Fo he USA, Model (1b) aces ou he 2001 ecession and he G ea Recession as de ined by he Na ional Bu eau o Economic Resea ch (NBER). Addi ionally, a slack in annual indus ial p oduc ion g ow h a he end o 2015 and he s a o 2016 is also iden i ied as ecessiona y by Model (1b). Fig. 1: CISS o he eu o a ea and he USA, inancial s ess e en s and ecessiona y egimes in he eu o a ea and he USA Sou ce: own No e: Bo h plo s show he inancial s ess indices o he eu o a ea (ciss EA) and he USA (ciss US) du ing he pe iod 1999M1–2017M11. The ecessiona y egimes, de e mined by (1), a e ep esen ed by he shaded a eas. Majo inancial s ess e en s in he eu o a ea and he USA a e deno ed in bo h plo s by e ical lines: bu s o he do -com bubble in Ma ch 2000, 9/11 e o a acks (Sep embe 2001), En on bank up cy (Decembe 2001), Wo ldCom bank up cy (July 2002), BNP Pa ibas (in es men und edemp ion suspension, Augus 2007), Bea S e ns collapse (Ma ch 2008), Le- hman B o he s bank up cy (Sep embe 2008), G eece bailou (May 2010), eu o a ea deb c isis e ex (in Augus 2011, ea o deb c isis sp eading om pe iphe y o he co e eu o a ea), G eek bailou e e endum (July 2015) and B exi e e endum (June 2016). EM_4_2020.indd 159 18.11.2020 12:28:20 160 2020, XXIII, 4 Finance Al hough he p esen ed heo y sugges s non- linea modelling, es ing o (non-)linea i y is well ad ised (see Hub ich & Te äs i a, 2013). By pe o ming he Lag ange mul iplie es o non- linea i y as de eloped by Te äs i a and Yang (2014), he linea model (2) was ejec ed in a ou o a non-linea VAR model. The null o linea i y (Model (2)) agains he non-linea speci ica ion in he o m o a smoo h ansi ion VAR (STVAR) model was es ed. The STVAR model, wi h he same endogenous se o a iables, lag s uc u e and ansi ion a iable, was es ima ed. The null o he linea VAR was ejec ed agains he non- linea speci ica ion ( -s a is ic = 146.1761 and p- alue = 0.0018002 o he eu o a ea model (1a); -s a is ic = 105.0075 and p- alue = 8.7848e-06 o he US model (1b)). We wish o hank Caggiano e al. (2017) o hei MATLAB code o he es . We nex p esen he GIRFs o analyse he ela ionship be ween US inancial s ess (ciss US) shocks and he mac oeconomic esponse in he eu o a ea o e a ho izon o 60 mon hs om he shock. The GIRFs a e compu ed by (3), based on he esul s o Model (1a). Model (1a) was es ima ed on ou lags, as indica ed by Akaike in o ma ion c i e ia o Model (2). The esul s in Fig. 2 a e indica i e o ine ia in inancial s ess: once he USA is hi by a inancial s ess shock, he inancial s ess le el inc eases a e wa ds o se e al mon hs. They also co obo a e some ea lie indings in he li e a u e e ealing ha shocks o US inancial ma ke s abili y spill o e in o he eu o a ea inancial sys em: a posi i e shock in he US inancial s ess esul s in a signi ican inc ease in he eu o a ea’s inancial s ess, wi h only sligh di e ences be ween he egimes. While an unexpec ed inc ease in he US inancial s ess is de imen al o he eu o a ea economy unde a non- ecessiona y egime, i Fig. 2: Response o he eu o a ea o a one s anda d de ia ion shock in ciss US Sou ce: own No e: Gene alized impulse esponses o a one s anda d de ia ion inc ease in ciss US hi ing he eu o a ea in ecessiona y and non- ecessiona y pe iods o e a 60-mon h ho izon. Median esponses and he 68% con idence in e al a e d awn (do ed lines o a non- ecessiona y egime and shaded a eas o a ecessiona y egime), as is ypical in Bayesian in e- ence analysis, compu ed om (3), based on he esul s om (1a). EM_4_2020.indd 160 18.11.2020 12:28:21