Journal of International Money and Finance 32 (2013) 1032–1043

Contents lists available at SciVerse ScienceDirect

Journal of International Moneyand Finance

journal homepage: www.elsevier .com/locate/ j imf

Predicting severe simultaneous recessions using yieldspreads as leading indicators

Charlotte ChristiansenCREATES, Aarhus University, Department of Economics and Business, Business and Social Sciences, Fuglesangs Alle 4,8210 Aarhus V, Denmark

JEL classifications:C25E32E43F44G15

Keywords:Business cycleRecessionsYield spreadProbit model

E-mail address: CChristiansen@creates.au.dk.

0261-5606/$ – see front matter � 2012 Elsevier Lthttp://dx.doi.org/10.1016/j.jimonfin.2012.08.005

a b s t r a c t

Severe simultaneous recessions are defined to occur when at leasthalf of the countries under investigation (Australia, Canada,Germany, Japan, United Kingdom, and United States) are inrecession simultaneously. I pose two new research questions thatextend upon stylized facts for US recessions. One, are the occur-rences of simultaneous recessions predictable? Two, does the yieldspread predict future occurrences of simultaneous recessions? Iuse the indicator for severe simultaneous recessions as theexplained variable in probit models. The lagged yield spread is animportant explanatory variable, where decreasing yield spreadsare a leading indicator for severe simultaneous recessions. Both USand German yield spreads act as leading indicator for severesimultaneous recessions.

� 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Previous research considers the predictability of recessions of a single country, most often theUnited States. The yield spread (the long interest rate minus the short interest rate) is known to predictfuture single-country recessions. In this paper I consider severe simultaneous recessions instead ofsingle-country recessions as it is even more important to be able to foresee these than single-countryrecessions.

A severe simultaneous recession is defined to occur when at least half of the countries being studiedare in recession simultaneously. The countries under investigation are six large developed countries,

d. All rights reserved.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1033

namely Australia, Canada, Germany, Japan, United Kingdom, and United States. To my knowledge,simultaneous recessions have not been predicted previously.

I use the probit model to describe the indicator variable for occurrences of simultaneous recessions. Iprovide both in-sample and out-of-sample analysis at 1�12 month horizons. Future simultaneous reces-sions are predictable, and more so at long horizons than at short horizons. The yield spread has animportant influence upon the likelihood of future simultaneous recessions. Small yield spreads implyfuture simultaneous recessions.Germanyield spreads provideadditional explanatorypower for predictingfuture simultaneous recessions over and above US yield spreads. In many ways, the empirical findingsregarding severe simultaneous recessions are parallel to the findings regarding single-country recessions.

Why does the yield spread predict future recessions? The literature gives a number of differentanswers to this question, cf. the overview in Wheelock and Wohar (2009). The yield spread isa measure of the shape of the yield curve. Increasing yield spreads are a leading indicator for expan-sions and decreasing yield spreads are a leading indicator for recessions. The expectations hypothesis isoften used to explain this stylized fact. According to the expectations hypothesis the yield spread isequal to the expected future short rate and a term premium. Falling yield spreads before recessions arecaused by both factors, where the decreasing expectations to future short rates is more important, cf.(Hamilton and Kim, 2002). Another explanation why yield spreads predict future recessions is basedupon monetary policy. Tight monetary policy is used to stabilize output growth and causes the yieldspread to decrease. The power of the yield spread as a leading indicator depends on the monetaryauthority’s behavior, cf. (Estrella, 2005). Consumption smoothing across business cycles is anotherexplanation for why the yield spread is a leading indicator in Harvey (1988). When investors expectrecessions they sell short term bonds and buy long term bonds, which implies decreasing yield spreads.Yet, Harvey (1988) concerns real interest rates whereas most empirical work is done on nominalinterest rates.

Previous research shows that the yield spread is an important predictor for future output, mostoften measured by GDP growth rates. Stock and Watson (1989) show that the yield spread acts asa leading indicator for the GDP growth rate. Estrella and Hardouvelis (1991) document that a positiveyield spread predicts future increases in real economic activity. Hamilton and Kim (2002) confirm theusefulness of the yield spread for predicting future GDP growth rates. In addition, Hamilton and Kim(2002) analyze why the yield spread predicts future GDP growth rates. According to the expecta-tions hypothesis, the yield spread is the expectation of future short rates and a term premium.Hamilton and Kim (2002) show that the most important reason why the yield spread forecasts futureGDP growth rates is that low yield spreads imply falling future short rates. Interest rate volatility doesnot explain the importance of the yield spread. Estrella (2005) provides a theoretical model in whichthe yield spread explains output and inflation. He shows that the predictive ability of the yield spreaddepends on the monetary policy reaction function.

The yield spread is also an important predictor for future US recessions. This is not surprising asrecessions are naturally related to GDP growth rates. US recessions are dated by the NBER BusinessCycle Dating Committee. Estrella and Mishkin (1998) investigate the predictability of future USrecessions using probit models. They show that the yield spread is the most promising explanatoryvariable. Estrella and Hardouvelis (1991) use the yield spread to predict future US recessions withina probit model. Dueker (1997) extends upon this by considering a dynamic probit model where thelagged recession variable is included as an explanatory variable. Estrella and Trubin (2006) containsome practical guidelines about using the yield spread as a leading indicator. This is evidence of thepopularity of the yield spread as a leading indicator for US recessions. Wright (2006) shows that usingthe federal funds rate in addition to the yield spread improves the predictability of future recessions.He considers the likelihood of a recession occurring during successive quarters instead of duringa specific quarter as is usual. Rudebusch and Williams (2009) show that the yield spread is better ableto forecast future recessions than professional forecasters are. Kauppi and Saikkonen (2008) suggesta dynamic autoregressive probit model to estimate US recessions from yield spreads, where the laggedrecession indicator and lagged recession probability are used as an explanatory variable. They alsoprovide improved multi-period forecasts.

Although most research in this area concerns US recessions, there are several international studies.The international studies consider predicting recessions in each country separately. Stock and Watson

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–10431034

(2003) forecast GDP growth rates for seven developed countries (Canada, France, Germany, Italy, Japan,UK, and US) using various economic explanatory variables including the yield spread. The yield spreadis a good predictor, but there are variations across countries and across periods. Moneta (2005)considers the euro area collectively and uses the yield spread to forecast future euro area recessions.The findings for the euro area are similar to those for the US. Chinn and Kucko (2010) investigate thepredictive ability of the yield spread for economic activity and recessions in different countries(Canada, France, Germany, Italy, Japan, the Netherlands, Sweden, UK, and US). The predictive power ofthe yield spread varies across countries and is declining over time. Schrimpf andWang (2010) considerfour large economies (Canada, Germany, UK, and US) and investigate the ability of the yield spread topredict output growth. They find evidence of structural breaks. Again, the power of the yield spread isdeclining over time. Nyberg (2010) investigates recessions in the US and Germany. He uses the dynamicautoregressive probit model of Kauppi and Saikkonen (2008). The yield spread is an importantexplanatory variable for both countries, but other variables provide additional explanatory power suchas stock returns, interest rate differential between US and Germany, and the other country’s yieldspread. IMF (2002) considers the synchronization of business cycles, where a recession in a country issynchronized if at least half of the other countries are in recession at the same time. IMF (2002)describes the occurrences of synchronized recessions in the past. Unlike other studies of recessions,IMF (2002) does not consider predictions. Plosser and Rouwenhorst (1994) and Bernard and Gerlach(1998) find that not only domestic yield spreads but also foreign yield spreads are leading indicatorsfor future output and recessions for a given country, respectively.

Wheelock and Wohar (2009) review the literature on the ability of the yield spread to predictoutput growth and recessions. In general, they consider the literature in favor of the yield spread as animportant leading indicator.

The remaining part of the paper is structured as follows. First, the data are introduced in Section 2.Second, the econometric framework is laid out in Section 3. Third, the in-sample results are discussedin Section 4, which is followed by the out-of-sample evidence in Section 5. International yield spreadsare considered in Section 5. Finally, Section 7 concludes.

2. Data

I use monthly data covering the period 1953M04 to 2010M12. The starting point is determined bythe availability of the US 10-year interest rate and the recession data for Japan.

2.1. Recession data

I use the NBER business cycles to date the US recessions. These cycle dates are publicly available andhave been used in all previous studies of US recessions that I am aware of.1 I use the same definition ofrecessions as in Estrella and Trubin (2006): The month after the “peak” month defines the first monthin recession and the month of the “through”month defines the last month in recession. Thus, the mostrecent recession is 2008M01 through 2009M06. I introduce a dummy variable that equals 1 when theUS is in recession and 0 otherwise; USt.

For Australia, Canada, Germany, Japan, and the United Kingdom I use the Economic Cycle ResearchInstitute (ECRI) business cycle dates.2 The ECRI business cycle data appear to be the standard datasource for non-US recessions, cf. (Sensier et al., 2004; Nyberg, 2010; Chinn and Kucko, 2010, andSchrimpf and Wang, 2010). The ECRI is not only the standard business cycle data source, but also theonly data source for non-US recessions.3 Still, the ECRI business cycle data is reliable because the ECRIuses the same methodology to date business cycles as the NBER. In fact, for the US the ECRI and NBER

1 The NBER business cycle dates are available from www.nber.org/cycles/.2 The ECRI business cycle dates are available at www.businesscycle.com/resources/cycles.3 Earlier research such as Bernard and Gerlach (1998) and Ahrens (2002) use the business cycle dates estimated by Artis et al.

(1997) based on the industrial production variable. Now, the ECRI business cycle data have taken over because it is based onseveral macroeconomic variables, e.g. (Sensier et al., 2004).

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1035

business cycle dates are identical. The recession indicators for Australia, Canada, Germany, Japan, andthe United Kingdom are denoted AUt, CAt, GEt, JPt, and UKt and are defined similarly to USt.

Fig. 1 shows the time series evolution of the individual recession indicators. Table 1 tabulates thefrequency of the recession indicators. Germany is most often in recession (22% of the sample period)and Australia is the least often in recession (8% of the sample period). On average, each country is inrecession 15% of the sample period. The recession periods of the six countries are obviously notidentical.

The variable CRt counts the number of simultaneous recessions.

CRt ¼ AUt þ CAt þ GEt þ JPt þ UKt þ USt (1)

The larger CRt is, the more severe is the overall crisis of the world economy. Table 1 tabulates CRt.Most of the time, no countries are in recession (395 out of 693 months). 19% of the time only onecountry is in recession and 10% of the time two countries are in recession. In 94 months (14%) there aresimultaneous recessions in at least three countries. Of these instances, exactly three countries inrecession is most common (54 months), then four (26), and then five (14). All six countries are never inrecession simultaneously.

I construct a dummy variable for severe simultaneous recessions in the six countries. The variableDRt is an indicator for the occurrence of simultaneous recessions at time twhich equals 1 when at leastthree countries are in recession at time t and 0 otherwise. Thus, I define a severe simultaneousrecession period when at least half of the countries under investigation are in recession at the sametime.

DRt ¼�

0 if CRt � 21 if CRt � 3 (2)

Table 1 also tabulates DRt. Most often DRt is 0 (in 599 out of 693 months) indicating that the worldeconomy is not in severe simultaneous recession. In 94 months (14%) there are simultaneous reces-sions in at least three countries (i.e. whereDRt¼ 1). Thus, the amount of time that theworld economy isin severe simultaneous recession is comparable to the amount of time that each country is in recession.

Fig. 2 shows the number of countries in recession (CRt) as well as the recession indicator (DRt). Theperiods of simultaneous recessions are detailed in Table 2. There are six severe simultaneous recessionsin the sample period. On average, a severe simultaneous recession period lasts for 16 months. The two

0

1

1960 1970 1980 1990 2000 2010

Australia

0

1

1960 1970 1980 1990 2000 2010

Canada

0

1

1960 1970 1980 1990 2000 2010

Germany

0

1

1960 1970 1980 1990 2000 2010

Japan

0

1

1960 1970 1980 1990 2000 2010

United Kingdom

0

1

1960 1970 1980 1990 2000 2010

United States

Fig. 1. Single-country recessions.

Table 1Frequency tabulation of recession variables.

Value AU CA GE JP UK US DR CR

0 640 92% 603 87% 544 79% 562 81% 617 89% 582 84% 599 86% 395 57%1 53 8% 90 13% 149 22% 131 19% 76 11% 111 16% 94s 14% 134 19%2 70 10%3 54 8%4 26 4%5 14 2%6 0 0%

The table shows the distribution of the recession variables; AU, CA, GE, JP, UK, and US are the recession indicator variables forAustralia, Canada, Germany, Japan, UK, and US. DR is an indicator for at least three countries in simultaneous recession. CRcounts the number of simultaneous recessions.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–10431036

most recent severe simultaneous recessions are the dot-com bubble in 2001 and the financial crisis in2008–2009.

2.2. Yield spread

The symbol for the yield spread is YSt. The term structure data are available from the FRED databaseat the Federal Reserve Bank of St. Louis. The yield spread is the difference between the 10-year rate andthe 3-month rate. The 10-year rate is the 10-year constant maturity rate (FRED symbol is G10). The 3-month rate is the 3-month Treasury bill secondary market rate (TB3MS). The choices for constructingthe yield spread are similar to previous studies including (Estrella and Mishkin, 1998) and (Wright,2006). Below, I consider international yield spreads as well.

Fig. 3 shows the time series evolution of the yield spread and the indicator for simultaneousrecessions. The yield spread is strongly variable during the sample period. The yield spread is typicallypositive but it is negative in some shorter periods. In general, the yield spread is falling and it is oftennegative in the period up to a simultaneous recession. This gives an early signal that the yield spreadcould be an important explanatory variable for simultaneous recessions.

3. Probit model for simultaneous recessions

The indicator variable for simultaneous recessions (DRt) is a binary choice variable. Thus, DRt iscomparable to the single country recession indicator, say USt that is previously described by the probitmodel, cf. (Estrella and Mishkin, 1998; Dueker, 1997; Wright, 2006; Estrella and Trubin, 2006;Rudebusch and Williams, 2009; Kauppi and Saikkonen, 2008 and Schrimpf and Wang, 2010). Inconsequence, the probit model is used to describe DRt as well.

0

1

2

3

4

5

6

1960 1970 1980 1990 2000 2010

Number of Simultaneous Recessions

0

1

1960 1970 1980 1990 2000 2010

Indicator for Simultaneous Recessions

Fig. 2. Simultaneous recessions.

Table 2Simultaneous recession periods.

No From Through Duration

1 1953M08 1954M05 102 1973M12 1975M03 163 1980M02 1980M07 23

1981M05 1981M05 –

1981M08 1982M11 –

4 1990M07 1992M03 215 2001M04 2001M11 86 2008M03 2009M06 16

The table shows the beginning and ending (both months included) of each of the periods of simultaneous recessions andtheir duration (in months).

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1037

Let xt denote the explanatory variables at time t and b be the parameter vector, then the probabilityof a simultaneous recession depends on xt as follows:

PrðDRt ¼ 1jxt ; bÞ ¼ F�x0t�kb

�DRt ¼ F

�x0t�kb

�þ εt

(3)

where Fð$Þ is the cumulative distribution function of the standard normal distribution and εt is theerror term. The probit model is applied to both in-sample and out-of-sample analysis. Horizonsbetween 1 and 12 months are considered; k ¼ {1,2,.,12}.

The explanatory variables in the full model are a constant, the lagged indicator for simultaneousrecessions, and the lagged yield spread. I assume a publication lag for the simultaneous recessionindicator variable of 9 months, which is similar to the publication lag considered in Nyberg (2010) forthe single-country recession indicator variable. Therefore, the indicator for simultaneous recessions islagged another 9 months compared to the forecast horizon.

xt�k ¼ ð c DRt�k�9 YSt�k Þ0: (4)

-3

-2

-1

0

1

2

3

4

5

1960 1970 1980 1990 2000 2010

Indicator for Simultaneous RecessionsUS Yield Spread

Fig. 3. US yield spread and indicator for simultaneous recessions.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–10431038

Using the lagged explained variable itself as an explanatory variables is similar to (Dueker, 1997)and it is what he terms a dynamic probit model. In addition, I consider the effect of each of theexplanatory variables individually using model (a)

xt�k ¼ ð c DRt�k Þ0 (5)

and model (b)

xt�k ¼ ð c YSt�k�9 Þ0: (6)

4. In-sample results

Table 3 shows the results from estimating the probit model for the entire sample period for horizonsk¼ 3,6,9,12.4 The explanatory power is measured by the pseudo R-squared. It is calculated as in Estrella(1998) and it compares the log-likelihood of the estimated probit model (lU) to the log-likelihood ofa probit model including only a constant (lC):

pseudo R2 ¼ 1��lUlC

��ð2=TÞlC: (7)

The explanatory power is increasing, the longer the forecast horizon. Thus, the yield spread is mosthelpful in predicting simultaneous recessions further into the future rather than in the immediatefuture. At a 3-month horizon, the R-squared is 0.04, whereas it is much larger at the 12-month horizonwhere it is 0.12.

The coefficient to DRt�k�9 is positive, but it is only significant at relatively short horizons, namely upto a 6-month horizon. This entails that current simultaneous recessions imply a stronger likelihood offuture simultaneous recessions k þ 9 periods ahead where k � 6. This simply says that recessions tendto persist for a several months but not years.

The coefficient to YSt�k is negative across all horizons. This means that smaller yield spreads todayimply a stronger likelihood of simultaneous recessions in the future. The coefficient to YSt�k issignificant which implies that the yield spread is an important leading indicator.

The estimated probit models for the simultaneous recession indicator DRt are similar to the probitmodels for the US recession indicator, USt. For USt it is also the case that there is positive dependenceupon the lagged recession indicator and negative dependence upon the lagged yield spread. Similar tosingle-country recessions the predictability of future recessions is increasing with the forecast horizon,cf. (Dueker, 1997).

Table 4 concerns the in-sample predictability represented by pseudo R-squared. In addition, theroot mean square error (RMSE) and the mean absolute error (MAE) are shown. The RMSE and the MAEare both more applicable for linear models, but they still give an indication of the appropriateness ofthe models. For horizon k these are calculated as follows

RMSEk ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1T

XTt¼1

�bpt;tþk � DRtþk

�2vuut

MAEk ¼ 1T

XTt¼1

����bpt;tþk � DRtþk

����(8)

where T is the number of observations. bpt;tþk is the fitted probability of a recession at time t þ k;bpk;tþk ¼ Fðx0tþkbbÞwhere bb is the vector of estimated coefficients for the probit model given in Equation

(3).When bpt;tþk � 0:5 themodel predicts that the economy is in recession at time tþ k. The table showsthe results for the full model and for models (a) and (b).

4 The estimation is done partly in EViews and partly in GAUSS.

Table 3Probit model for simultaneous recessions.

Horizon (k) 3 6 9 12

Constant �0.99 (0.09)*** �0.83 (0.09)*** �0.70 (0.08)*** �0.61 (0.08)***DR (-k-9) 0.80 (0.18)*** 0.47 (0.20)** 0.27 (0.21) �0.18 (0.20)YS (-k) �0.22 (0.06)*** �0.34 (0.06)*** �0.45 (0.07)*** �0.52 (0.07)***Pseudo R-squared 0.04 0.06 0.10 0.12

Probit models for DR (indicator for at least three countries in simultaneous recession) where the explanatory variables area constant, DR lagged, and the yield spread (YS) lagged for horizons of k ¼ (3,6,9,12) months. Robust standard errors inparenthesis. */**/*** indicate that the parameter is significant at the 10%/5%/1% level of significance.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1039

At very short horizons (up to 2 months), the in-sample predictability is mainly caused by the laggedsimultaneous recession indicator DRt�k�9, which is seen by the fact that the pseudo R-squared is notmuch smaller for the full model than for model (a). From around a 4-month horizon the yield spread isthe most important explanatory variable, seen by comparing the pseudo R-squared values for the threemodels. At long horizons (from around a 7-month horizon) the predictability is almost exclusivelycaused by the yield spread.

Fig. 4 shows the in-sample predictability at a 12-month horizon graphically. The figure shows thefitted probability of a simultaneous recession, bpt;tþ12 for each model. The fitted probability of simul-taneous recessions is almost identical for the full model andmodel (b), whereasmodel (a) is not helpfulfor in-sample predictions. The predicted probability of a recession is increasing a lot before all reces-sions. The model identify the two first simultaneous recessions correctly. For the last three recessions,the predicted probability does increase but not above the 0.50 threshold. Thus, the ability of the yieldspread to predict future recessions is decreasing over the sample period. This is in accordance with theresults in Schrimpf and Wang (2010) regarding individual-country recessions.

5. Out-of-sample results

The out-of-sample predictability is investigated. I use a rolling window of 360 observations toestimate the probit model. The out-of-sample period is 1984M12 to 2010M12, giving 318 observa-tions. The out-of-sample period covers the last three recessions. Again, I use horizons of k ¼ 1,2,.,12months.

Using the first 360 observations I estimate the probit model andmake one out-of-sample forecast ofthe simultaneous recession probability k months ahead, bp1;1þk. Then the probit model is re-estimatedfor the updated sample where the oldest observation is discarded and one more recent observation is

Table 4In-sample predictability.

Horison (k) 1 2 3 4 5 6 7 8 9 10 11 12

Full modelPseudo R-squared 0.049 0.042 0.044 0.050 0.053 0.059 0.066 0.081 0.095 0.101 0.108 0.122RMSE 0.325 0.325 0.323 0.319 0.317 0.315 0.314 0.310 0.307 0.308 0.308 0.308MAE 0.211 0.211 0.209 0.206 0.203 0.201 0.200 0.196 0.193 0.193 0.192 0.190Model (a)Pseudo R-squared 0.046 0.030 0.017 0.010 0.004 0.001 0.000 0.000 0.001 0.002 0.004 0.006RMSE 0.325 0.327 0.329 0.329 0.328 0.329 0.330 0.330 0.330 0.330 0.330 0.330MAE 0.212 0.214 0.216 0.216 0.216 0.217 0.217 0.218 0.218 0.218 0.218 0.218Model (b)Pseudo R-squared 0.000 0.003 0.014 0.028 0.039 0.052 0.065 0.080 0.094 0.101 0.108 0.121RMSE 0.343 0.342 0.340 0.337 0.333 0.329 0.324 0.319 0.315 0.313 0.311 0.309MAE 0.235 0.234 0.232 0.228 0.224 0.218 0.213 0.207 0.202 0.199 0.196 0.192

Pseudo R-squared, root mean squared error (RMSE), and mean absolute error (MAE) for in-sample predictability for probitmodels for DR at horizons k ¼ 1,2,..,12 months. Explanatory variables at full model: constant, DR(-k-9), and YS(-k). Explanatoryvariables at model (a): constant and DR(-k-9). Explanatory variables at model (b): constant and YS(-k).

0.0

0.2

0.4

0.6

0.8

1.0

1960 1970 1980 1990 2000 2010

Indicator for Simultaneous RecessionsIS ForecastIS Forecast (DR(-21) only)IS Forecast (YS(-12) only)

Fig. 4. In-sample predictability at 12-month horizon.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–10431040

included. A new out-of-sample estimate is then calculated. The out-of-sample estimation continuesthis way.

Table 5 shows the out-of-sample predictability as measured by the pseudo R-squared, the RMSE andMAE calculated similar to the in-sample counterparts. T is now the number of out-of-sample forecasts.

The properties of the out-of-sample predictability are in every sense similar to the in-samplepredictability. This also applies to the size of the pseudo R-squared values. The out-of-samplepredictability is stronger the longer the forecast horizon. At short horizons, the autoregressivecomponent (DRt�k�9) is most important and at long horizons the yield spread (YSt�k) is most important.At, say the 12-month horizon, model (b) that only relies on the lagged yield spread has about the samepredictive power as the full model that also relies on the lagged simultaneous recession indicator itself,whereas model (a) that relies only on lagged simultaneous recession indicator has very low predictivepower: the pseudo R-squared is 0.15 for the full model and model (b), whereas it is 0.01 for model (a).

Table 5Out-of-sample predictability.

Horison (k) 1 2 3 4 5 6 7 8 9 10 11 12

Full modelPseudo R-squared 0.055 0.037 0.018 0.008 0.009 0.017 0.038 0.057 0.075 0.101 0.125 0.147RMSE 0.342 0.346 0.350 0.352 0.352 0.351 0.346 0.344 0.341 0.338 0.335 0.332MAE 0.223 0.223 0.221 0.218 0.216 0.213 0.209 0.205 0.202 0.201 0.198 0.195Model (a)Pseudo R-squared 0.047 0.030 0.018 0.007 0.001 �0.003 0.002 0.001 �0.002 0.005 0.009 0.012RMSE 0.343 0.347 0.350 0.352 0.353 0.353 0.352 0.353 0.353 0.352 0.351 0.351MAE 0.235 0.241 0.246 0.249 0.251 0.252 0.252 0.251 0.251 0.251 0.250 0.249Model (b)Pseudo R-squared �0.001 �0.010 �0.022 �0.023 �0.010 0.008 0.033 0.057 0.080 0.105 0.129 0.150RMSE 0.353 0.356 0.358 0.358 0.355 0.352 0.347 0.344 0.340 0.337 0.334 0.331MAE 0.247 0.241 0.234 0.228 0.222 0.217 0.211 0.207 0.203 0.201 0.198 0.194

Pseudo R-squared, root mean squared error (RMSE), and mean absolute error (MAE) for out-of-sample predictability of probitmodel for DR at horizons k ¼ 1,2,..,12. Rolling window estimation, window length of 360 months. Explanatory variables at fullmodel: constant, DR(-k-9), and YS(-k). Explanatory variables at model (a): constant and DR(-k-9). Explanatory variables at model(b): constant and YS(-k).

0.0

0.2

0.4

0.6

0.8

1.0

1990 1995 2000 2005 2010

Indicator for Simultaneous RecessionsOOS ForecastOOS Forecast (DR(-21) only)OOS Forecast (YS(-12 only)

Fig. 5. Out-of-sample predictability at 12-month horizon.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1041

Fig. 5 shows the out-of-sample predictability at the 12-month horizon. Model (a) cannot predictany of the simultaneous recessions and it is evident that at long horizons the autoregressivecomponent is not useful. The full model and model (b) have almost identical forecasts. They correctlyidentify the 2001 (second last) simultaneous recession. For the 1990 (third last) and the 2008–2009(last) simultaneous recessions, the predicted probability of future simultaneous recessions doesincrease up to the recessions, but not above the 0.50 threshold, although it is very close for thelast one.6. International yield spreads

Bernard and Gerlach (1998) predict individual-country recessions. They find that in addition todomestic yield spreads, US and Germanyield spreads are leading indicators for recessions. Sensier et al.(2004) also provide evidence that international factors have predictive power for individual-countryrecessions.

In this spirit, I consider international yield spreads as additional predictors. Unfortunately, they areavailable for a shorter time period than US yield spreads.5 In particular, I introduce the German yieldspread as predictor of simultaneous recessions. YSGE denotes the German yield spread and theexplanatory variables at horizon k in the international model are given as:

xt�k ¼ ð c DRt�k�9 YSt�k YSGEt�k Þ0: (9)

The sample period runs from 1977M05–2010M12, which gives 404 observations out of which 68 aresimultaneous recessions.

5 I use the following yields (DataStream code in parenthesis) to construct the yield spreads: Australia bond yield 10-year(ABND10Y), Australia dealer bill 90-days middle rate (ADBR090), Canada government bond 10-yearþ (CN13867), Canadatreasury bill 3-month (CN13884), German benchmark bond 10-year bond yield (BDBRYLD), German 3-month rate (ECWGM3M),Japan benchmark bond yield 10 year (JPBRYLD), Japan interbank 3-month rate (BBJPY3M), United Kingdom benchmark bond10-year yield (UKMBRYD), and United Kingdom treasury bill 3-month yield (UKTBTND). The relevant international yieldspreads are available as follows: Australia from 1976M02, Canada from 1980M02, Germany from 1977M05), Japan from1986M07, and United Kingdom from 1986M06.

Table 6International probit model for simultaneous recessions.

Horizon (k) 3 6 9 12

Constant �0.56 (0.13)*** �0.21 (0.14) �0.01 (0.16) 0.11 (0.16)DR(-k-9) 0.10 (0.27) �0.47 (0.32) �0.53 (0.29)* �1.01 (0.26)***YS(-k) �0.07 (0.06) �0.24 (0.07) *** �0.39 (0.08)*** �0.48 (0.08)***YSGE(-k) �0.62 (0.07)*** �0.85 (0.08)*** �0.74 (0.09)*** �0.61 (0.09)***Pseudo R-squared 0.26 0.36 0.36 0.35

Probit models for DR (indicator for at least three countries in simultaneous recession) where the explanatory variables area constant, DR lagged, the US yield spread (YS) lagged, and the German yield spread (YSGE) lagged for horizons of k ¼ (3,6,9,12)months. The sample is 1977–2010. Robust standard errors in parenthesis. */**/*** indicate that the parameter is significant at the10%/5%/1% level of significance.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–10431042

The results for the in-sample probit model are shown in Table 6. The effect of yield spreads uponfuture simultaneous recessions is negative both for US and German yield spreads. This implies that thesmaller any of the twoyield spreads are, the stronger is the likelihood of future simultaneous recessions.At short horizons, only the German yield spread is significant, whereas at long horizons both US andGerman yields are significant in explaining future simultaneous recessions. The predictive power ofyield spreads is increasing as the forecast horizon increases. This is similar to the full model that onlyuses US yield spreads and that covers a longer sample period. The explanatory power for the interna-tional model is much stronger than for the full model. At the 12-month horizon the pseudo R-squared is0.35 in the international model whereas it is 0.20 for the full model (for the equivalent sample period).

Fig. 6 shows the in-sample predictability for the international model at the 12-month horizongraphically. The international model identifies all the recessions correctly. Thus, it appears that topredict recent recessions it is advantageous to consider international yield spreads in addition to USyield spreads.

Finally, I use the yield spreads from all six countries as predictors at the same time (the sample isreduced even further and begins in 1986 (results not tabulated)). Only the US, German, and Japaneseyield spreads are significant when predicting future simultaneous recessions. I also consider a some-what longer sample period (from 1980) and only use US, German, Australian, and Canadian yieldspreads (results not tabulated). Here the Australian yield spread is insignificant at all horizons. Theseresults provide further evidence of the importance of the German yield spread in addition to the USyield spread.

0.0

0.2

0.4

0.6

0.8

1.0

1980 1985 1990 1995 2000 2005 2010

Indicator for Simultaneous RecessionsIS Forecast International Model

Fig. 6. In-sample predictability of international model at 12-month horizon.

C. Christiansen / Journal of International Money and Finance 32 (2013) 1032–1043 1043

Overall, international yield spreads provide additional explanatory power for predicting futuresimultaneous recessions over and above US yield spreads.

7. Conclusion

I provide the first analysis of severe worldwide recessions as measured by the occurrence ofsimultaneous recessions in a number of individual countries. I consider recessions in Australia, Canada,Germany, Japan, United Kingdom, and US. A severe simultaneous recession is when at least half of thecountries are in recession at the same time. The indicator for severe simultaneous recessions followsa probit model where the explanatory variables are the lagged indicator for simultaneous recessionsand the lagged yield spread.

The future simultaneous recessions are predictable in-sample and to some extend also out-of-sample. The future simultaneous recessions are more predictably at longer horizons than at shorterhorizons. Both German and US yield spreads are leading indicators for simultaneous recessions.

Acknowledgments

I thank an anonymous reviewer and seminar participants at Universitat Rovira i Virgili, Spain forhelpful comments and suggestions. I acknowledge financial support from CREATES and the DanishCouncil for Independent Research, Social Sciences. CREATES (Center for Research in EconometricAnalysis of Time Series) is funded by the Danish National Research Foundation.

References

Ahrens, R., 2002. Predicting recessions with interest rate spreads: a multicountry regime-switching analysis. Journal of Inter-national Money and Finance 21, 519–537.

Artis, M., Kontolemis, Z.G., Osborn, D.R., 1997. Business cycles for G7 and European countries. Journal of Business 70, 249–279.Bernard, H., Gerlach, S., 1998. Does the term structure predict recessions? The international evidence. International Journal of

Finance and Economics 3, 195–215.Chinn, M.D., Kucko, K.J., 2010. The Predictive Power of the Yield Curve across Countries and Time. Working Paper 16398. NBER.Dueker, M.J., 1997. Strengthening the case for the yield curve as a predictor of U.S. recessions. Federal Reserve Bank of St. Louis

Review, 41–51.Estrella, A., 1998. A new measure of fit for equations with dichotomous dependent variables. Journal of Business and Economic

Statistics 16, 198–205.Estrella, A., 2005. Why does the yield curve predict output and inflation. Economic Journal 115, 722–744.Estrella, A., Hardouvelis, G.A., 1991. The term structure as a predictor of real economic activity. Journal of Finance 46, 555–576.Estrella, A., Mishkin, F.S., 1998. Predicting U.S. recessions: financial variables as leading indicators. Review of Economics and

Statistics 80, 45–61.Estrella, A., Trubin, M.R., 2006. The yield curve as a leading indicator: some practical Issues. Current issues in economics and

finance. Federal Reserve Bank of New York 12, 1–7.Hamilton, J.D., Kim, D.H., 2002. A reexamination of the predictability of economic activity using the yield spread. Journal of

Money, Credit, and Banking 34, 340–360.Harvey, C.R., 1988. The real term structure and consumption growth. Journal of Financial Economics 22, 305–333.IMF, 2002. Chapter 3: Recessions and Recoveries. World Economic Outlook.Kauppi, H., Saikkonen, P., 2008. Predicting U.S. recessions with dynamic binary response models. Review of Economics and

Statistics 90, 777–791.Moneta, F., 2005. Does the yield spread predict recessions in the euro area? International Finance 8, 263–301.Nyberg, H., 2010. Dynamic probit models and financial variables in recessions forecasting. Journal of Forecasting 29, 2115–2230.Plosser, C.I., Rouwenhorst, K.G., 1994. International term structures and real economic growth. International Journal of

Monetary Economics 33, 133–155.Rudebusch, G.D., Williams, J.C., 2009. Forecasting recessions: the puzzle of the enduring power of the yield curve. Journal of

Business and Economics Statistics 27, 492–503.Schrimpf, A., Wang, Q., 2010. A reappraisal of the leading indicator properties of the yield curve under structural instability.

International Journal of Forecasting 26, 836–857.Sensier, M., Artis, M., Osborn, D.R., Birchenhall, C., 2004. Domestic and international influences on business cycle regimes in

Europe. International Journal of Forecasting 20, 343–357.Stock, J., Watson, M., 1989. New indexes of coincident and leading indicators. NBER Macroeconomic Annual 4, 351–394.Stock, J.H., Watson, M.W., 2003. Forecasting output and inflation: the role of asset prices. Journal of Economic Literature 41,

788–829.Wheelock, D.C., Wohar, M.E., 2009. Can the term spread predict output growth and recessions? A Survey of the literature.

Federal Reserve Bank of St. Louis Review, 419–440.Wright, J.H., 2006. The Yield Curve and Predicting Recessions. Working Paper. Federal Reserve Board.