Business Cycles around the Globe: A Regime-switching Approach

Sumru Altuğ and Melike Bildirici

Koç University and CEPR;Yıldız Technical University

6th Colloquium on Modern Tools for Business Cycle AnalysisLuxembourg, September 24-27

Introduction• How do business cycles in developing and

emerging market economies differ from those industrialized countries?

• Köse, Otrok and Whiteman (2003) and Köse, Otrok, and Prasad (2008) examine the sources of macroeconomic fluctuations in samples that include both developed, emerging and developing economies using a dynamic factor framework.

• Benczur and Ratfai (2009) use the “Real Business Cycle” approach to examine the business cycle characteristics of 62 countries.

Introduction•We examine 27 countries as a way of

uncovering the sources of business cycle fluctuations.

•We consider a representative set of developed and emerging market economies.

•We employ a simple nonlinear regime switching approach to describe stylized facts of business cycles in these countries.

•We compare our approach for dating business cycles to the nonparametric Harding-Pagan approach.

Studies of business cycles

•Backus and Kehoe (1992) analyze the

properties of historical business cycles for 10 developed countries.

•Stock and Watson (2000) use data on 71 variables to characterize U.S business cycle phenomena over the period 1953-1996

•Artis and Zhang (1997) or Artis, Kontolemis, and Osborn (1997) study the European business cycle

Modeling business cycles• Dynamic factor model introduced by Sargent and

Sims (1977). Altug (1989) estimates a version of the Kydland-Prescott model using maximum likelihood by treating the economy-wide technology shock as an unobserved factor

• Parametric and nonparametric approaches to dating euro area business cycles [Artis, Marcellino, and Proietti (2003) ]

• Markov processes to describe the underlying state of the economy. [Neftçi (1982) and Hamilton (1989)]. Markov switching model with the dynamic factor framework [Diebold and Rudebusch (1996) and Chauvet (1998)].

Applications of MS Model•Applications to developed and developing

economies ▫Many applications for the US and G7

economies▫Euro area economies [Artis, Krolzig and Toro

(2004), Krolzig and Toro (2005)]▫Emerging and developing economies

Non-parametric Bry-Boschan method [Rand and Tarp (2002)]

Taylor, Sheperd and Duncan (2005) for Australia Girardin (2005) for 10 East Asian countries Moolman (2004) Taştan and Yıldırım (2008) for Turkey

Harding-Pagan Approach

•Harding and Pagan (2002a, 2002b, 2005): A business cycle is defined as a pattern in the level of aggregate activity.▫This contrasts with much recent work which

identifies business cycles in terms of the cyclical time series behavior of the main macroeconomic variables and their co-movement with cyclical output (RBC approach)

• Identifies classical business cycles in the manner of Burns and Mitchell.

Modeling asymmetries in business cycles

• Several ways of capturing asymmetry:▫ Recessions are due to permanent negative shocks [Hamilton

(1989)]▫ ``Plucking model'‘ of business cycles, where recessions occur as

temporary deviations from the long-run level of GDP [Kim and Nelson (1999a)]

• Evidence in favor of either forms of asymmetry is mixed. ▫ Koop and Potter (1999) find evidence in favor of the Hamilton

(1989) model ; Sichel (1993, 1994) and Beaudry and Koop (1999) for the peak-reverting model.

• How to proceed?▫ Plucking model of cyclical fluctuations relevant for developed

countries. ▫ Crises and sharp recessions in emerging market economies

suggest that negative shocks may have permanent effects on the level of real output.

A Markov switching model

•Suppose nt depends on an unoberseved Markov state variable denoted st nt= μ(st)+nt-1

•y*t=nt+z*t where z*t follows an ARIMA(r,1,0) process.

•Differencing yields:

yt=μ(st)+zt

where yt = y*t - y*t-1 and zt is a stationary AR(p) process in (log) differences

A Markov switching model •Assuming A(L) zt =εt where {εt} is an i.i.d.

N(0,σ2) process, and applying A(L) to both sides of the above equation yields

1

( ) ( ( ( ))p

t t j t j t j t j tj

y s a y s s

• In this case the unobserved state is

defined as st=1,2, where st=1 denotes a “contraction” and “st=2” an “expansion”

A Markov switching model• More generally, suppose st=1,…,m. For

example, there might exist situations where a third regime is appropriate.

• In this case “low growth”, “normal growth” and “high growth” states.

• Also capture secular changes in growth rates over the sample period by allowing for trends in such growth rates.

0

1

2

( ) ( ) ( ) ( ) ( ) ,

where { } is an i.i.d. process such that | (0, ( ) ).t

p

t t t j t j t j t tj

t t t

y v s s t a s y s

s N s

Sample of countries

Anglophone countries plus Japan

EU countries

Emerging economies

Data• Let yi,t=ln(Yi,t) where (Yi,t) denotes the real GDP

of country i in quarter t.• Quarterly GDP at constant prices measured in

units of national curency• We take the annual quarter-to-quarter growth

rate of GDP for country i as Δyi,t= ln(Yi,t)-ln(Yi,t-4).• For seasonally unadjusted data, this

transformation tends to eliminate any seasonal effects that might exist at the quarterly frequency.

• Following Stock and Watson (2005) we smoothed out high frequency movements in the different series by taking four-quarter averages of the annual quarter-to-quarter growth rates.

Results•The business cycle characteristics of the

developed countries are similar in terms of▫ the expected growth rates of real output in

the different phases▫the durations of the phases.

•Recessions are milder in the EU•Business cycle in Europe tends to lag the

busines cycle in the U.S.•Synchronization of business cycles in

developed countries varies by period – the “Great Moderation.”

Results•For emerging countries:

▫Distinct groups of emerging economies by severity and duration of crises as well as strength and length of expansions – East Asian, Latin American, and other emerging economies

▫Differences even among groups of geographical and historical proximity – Chile vs Argentina, Malaysia vs Singapore

▫The role of policy and institutional factors, global conditions, membership in regional blocs – Turkey vs Mexico, S. Africa, Israel

Anglophone countries plus Japan• 3-regime models fit best for Australia, Canada

and Japan whereas 2-regime models are adequate to describe the business dynamics of (de-trended) output growth for the UK and the US.

• The magnitude of change in expected growth during a recessions varies across the different countries. - Australia ,Canada, UK and the US display

negative expected growth during recessions - Japan tends to grow less during recessions.

• Aside from the UK, expected growth rates of output during expansions tend to be similar.

Anglophone countries plus Japan• The model is successful in identifying the major

recessions of 1973-1975, 1980-1982,1990, as well as 2008.

• Growth slowdowns based on GP growth for Australia, Canada, the UK and US during 2001. Some evidence for recession in the US.

• The duration of recessions for Australia and the US are around 3 quarters .

• Longer recessions for Canada, Japan and the UK. • Severe and lengthy recession for Canada in 1990. • A long period of low growth and stagnation for Japan

in the 1990’s and early 2000’s.

The EU countries• “Core” EU countries such as Austria, France, Germany, Italy and the

Netherlands as well as countries in the “periphery” such as Greece and Spain.

• 2-regime models are selected for all EU countries except Italy and Spain.

• Less evidence for real output declines during a contraction for most of the core EU countries such as Austria, France, Germany, Italy and the Netherlands compared to the Anglophone countries .

• There are expected output declines for Italy and Spain but these effects are not significantly different from zero. By contrast, Finland, Greece and Sweden experience significant absolute output declines during recessions.

• For Austria, France, Germany, Italy, and the Netherlands, the duration of recessions and expansions are similar to those for the Anglophone countries plus Japan. By contrast, the duration of recessions for the remaining EU countries is longer.

The EU countries• The worldwide recessions associated with the oil shocks of 1973-1975 and 1980-

1982 and the 1982 recession register for the EU countries as does the effects of the financial crisis of 2007-2008.

• Three recessions identifed by CEPR Business Cycle Dating Committee: 1974:3-1975:1, 1980:1-1982:3, and 1992:1-1993:3, which are also identified by our chronology.

• However, the countries are not uniform in their response to such events as oil shocks. ▫ France experiences a double-dipped recession during the 1980-1982 period

as in the US. Recessions in Italy and the Netherlands are spread out over the entire 1980-1983 or 1984 period.

• The main recession in the 1990’s for the EU countries is the one associated with the ERM crisis of 1992.▫ Austria, France, Germany, Italy, and Spain as well as the UK suffer declines

in growth during the period 1991-1993.• Idiosyncracies in experiences

▫ Italy and Spain experience episodes of “high” growth in the late 1960’s, the late 1980’s and also in the period between 1995-2000 and even later for Italy.

▫ Nordic banking crisis for Finland and Sweden in the 1990’s

The East Asian countries

• We select 3-regime models for Hong Kong and Singapore but 2-regime models are adequate to describe the business cycle dynamics of de-trended real output growth for S.Korea,Malaysia Taiwan.

• All East Asian countries except Hong Kong, S. Korea and Taiwan display positive growth in the “low” growth state as noted by Girardin (2005).

• The average duration of recessions for the developed East Asian countries is 3.74 quarters while the average duration of the “normal” growth regime around 18 quarters, which are comparable to those for the developed countries. The East Asian economies display episodes of “high” growth averaging 6.5 quarters.

The East Asian countries

• The 1997 East Asian crisis registers as a recession for all four developed E. Asian countries.

• Hong Kong, Singapore and Taiwan, three small open economies with strong trade and financial linkages to the rest of the world, also experience recessions and output declines during 2000-2001 due to slowdown in US and regional economic growth.

• All of the East Asian countries are affected by the financial crisis that erupted in the US in 2007-2008.

• Differences in cyclical dynamics depending on the degree and nature of openness. ▫ S. Korea and Taiwan display more stable growth

interrupted by some major recessions over the sample period

▫ Hong Kong and Singapore experience growth slowdowns and episodes of “high” growth that oftentimes alternate with each other.

Other emerging economies• 2-regime models for Turkey and S.Africa and a 3-regime model for Israel.• Strong role of policy and institutional factors for all three countries.• S. Africa: a small decline in output during recessions but low growth during

expansions. These due to the regime of trade sanctions against the S. African state until the dismantling of apartheid in 1994.

• Turkey: short-lived recessions amid short-lived expansions. These reflect the severe financial and banking crises of 1994-1995, 1999-2000 and 2001-2002 as well as the effects of the First Gulf War in 1991 and the Marmara earthquake in 1999.

• Israel▫ absolute output declines during the “recessionary” regime. This reflects the demand-

driven recession in Israel during 2000-2001,which arose from a worsening security situation due to the intifada and reflected the impact of global economic conditions.

▫ Second, a “high growth” regime which is also characterized by high volatility. This to the period following the end of Israel’s hyperinflationary episode in the mid 1980’s to the end of the 1990’s.

▫ Since 2003, a regime of “normal” growth in Israel that is characterized by positive growth and low volatility.

The Latin American Countries•3- regime models can be selected for Chile

and Uruguay.•2-regime models are appropriate for

Argentina, Brazil and Mexico.•The expected growth rates of output in the

bad regime are estimated to be significantly negative for all of Latin American countries, with the largest declines for Argentina and Brazil.

•The duration of recessions averages nearly 8 quarters and that of expansions 11 quarters.

The Latin American Countries

•Chile and Uruguay tend to display short episodes of “high” growth.

•Regional crises and idiosyncratic experiences:▫1980’s debt crisis▫1994-1995 Tequila crisis▫1998-2002: Argentina’s sovereign debt

default ▫1999:collapse of the Real Plan▫2000-2003: recessions in Brazil and Chile

A comparison with the Harding-Pagan Approach

• Harding and Pagan (2002a,b) have advocated an alternative approach to characterizing business cycles that has closer parallels with the Burns-Mitchell methodology.

• They argue that the approach based on the Markov switching model may produce different business cycle characteristics relative to linear models depending on the assumed features such as conditional heteroskedasticity, persistence and non-normality of the process.

• Harding and Pagan (2002b) proposed a modification to the Bry-Boschan algorithm. (BBQ algorithm)

BBQ algorithm

•The results obtained using the BBQ method are broadly consistent with the results based on the MS model.

•The BBQ dating underestimates the duration of recessions for Japan because it underestimates the recession that occured in the early 2000’s.

•Our estimate of average duration of expansions is 19 quarters versus 30 quarters according to the BBQ algorithm

BBQ algorithm•According to BBQ algorithm, the

percentage decline in output during recessions ranges between 1% for the US to close to 3% for Canada.

•The BBQ algorithm predicts shorter recessions and somewhat longer expansions than the MS-AR approach for EU countries.

•Both approachs tend to agree on the point that recessions tend to be milder in the euro area countries.

•The results for the emerging economies are also similar.

Business cycle dating•Business cycle dating properties broadly

consistent with the results of NBER, CEPR and ECRI type dating.

•Average of pairwise correlations of recession probabilities used to capture worldwide recessions▫1974-1975: 0.7993 1979-1983: 0.8311▫1991: 0.7579 2001: 0.4501▫ 2007-2008: 0.9548

•1994-2002: 8 years of emerging market crises or one 8-year crisis? John Taylor

•Issues of financial contagion and policy actions to avert them

Business cycle synchronization• Canova, Ciccarelli, and Ortega (2007) show

that business cycles tend to become more synchronized during recessions than expansions.▫ According to their results, expansions tend to

have large idiosyncratic components whereas declines in economic activity have common timing and dynamics, both within and across countries.

• We examine correlation of pairwise recession probabilities across 1970-2009 and 1990-2009 for developed countries and for 1990-2009 for the emerging economies.

Developed countries, 1970-2009•Australian recessions uncorrelated except

with those of Canada: 0.21 correlation•Canadian economy primarily linked to US

economy: 0.41 correlation• Japan correlated with UK and US and less

so with other European countries•UK highly correlated with US and all

European countries except Germany•European countries correlated with each

other

Developed countries, 1990-2009•Australia uncorrelated with any economy

in the sample (rise of China?)•Canada increases correlation with US and

the UK but more detached from EU countries

•Japan essentially on its own [Stock and Watson (2005)]

•UK more correlated with Spain (mutual strong growth in 2000’s?)

•EU economies more correlated amongst each other

A European business cycle?• Artis and Zhang, 1997 or Artis, Kontolemis, and Osborn, 1997).• Canova, Ciccarelli, and Ortega (2007) use a panel VAR setting

with a time-varying index structure on the underlying VAR coefficients to uncover the factors underlying cyclical fluctuations in the G-7 countries.▫ In contrast to other work, they find no evidence for the

independent effect of a European cycle driving the behavior of a key set of aggregate variables for France, Germany and Italy

• Our study:▫ Business cycle characteristics similar for developed

economies▫ US and EU countries highly correlated, especially over

period 1990-2009 period

Emerging economies•Two distinct groups

▫East Asian countries comprised of Hong Kong,Malaysia, Singapore, and S. Korea

▫Argentina, Mexico, and Turkey: crises during the 1990's and 2000's.

•More to it than national or regional cycles, though.▫Chile, Mexico, and Singapore show at least

as strong if not stronger cross-correlations with the U.S., as do Argentina, Brazil, and Malaysia.

A world business cycle?• Lumsdaine and Prasad (2003): construct a common

component by weighing output growth in 17 OECD countries using estimates of time-varying conditional volatility.

• Our approach: examine confluence of turning points• The existence of large common disturbances that are

associated with worldwide recessions: oil shocks of the 1970's and 1980's and other global factors surrounding them as well as the financial crisis of 2007-2008.▫ IMF World Economic Outlook (2009), “From Recession to

Recovery”: “Developments in the US often play a pivotal role in highly synchronized recessions, [which also tend to be more deeper and longer].”

• Policy matters! As do institutional, historical, political factors, endowments and trade patterns.▫ Diamond and Rajan (2009) suggest that the lessons of previous

crises helped emerging market economies in avoiding the worst effects of the 2007-2008 crisis.

Conclusion• We used information on business cycle behavior

for 27 individual countries. We showed that:▫ Characteristics of the developed countries differ

from developing ones.▫ Emerging economies tend to exhibit quite disperate

behavior relative to each other.• Can we find a “world business cycle”? • Can we find a European business cycle?• Tendency for less correlated behavior among

some groups of developed countries during 1990-2009: absence of large shocks?

• A note on methodology: Markov switching model versatile enough to examine both developed and emerging economy experiences.