The Relationship with Economic Growth - Science Direct

Available online at www.sciencedirect.com

ScienceDirect Procedia - Social and Behavioral Sciences 210 (2015) 416 – 424

4th International Conference on Leadership, Technology, Innovation and Business Management

Baltic Dry Index as a Major Economic Policy Indicator: The relationship with Economic Growth Melike E. Bildiricia *, Fazıl Kayıkçıb , Işıl Şahin Onat†, a, b

Yıldız Technical University, Social Science Institute, Istanbul, 34349,Turkey.

Abstract Since its establishment, the Baltic Dry Index has become one of the foremost indicators on the cost of shipping and an important barometer on the volume of worldwide trade and manufacturing activity.Global factors also play important role in supply and demand of BDI index. BDI and global markets have common economical and financial movement due to market supply and demand which is as a result of turmoil’s and crisis. After economic recessions and during economic growth, demand of raw materials increase as production and investments are also increase, as a result transportation volume grows accordingly. On the other hand, during economic slowdowns, demand of raw material decreases which creates utilized capacity. In this paper, MSIAH(3)-VAR(4) model is selected to analyze the relationship between BDI and economic growth for the United States. BDI and GDP are cointegrated for the United States. The crisis regime tends to last 3.13 years on the average, while the Regime 2 is comparatively more persistent with 3.11 years. Finally, Regime 3 which corresponds to the high growth tends to last 2.55 years on the average. Crisis regime of economy is the most persistent regime in the US. Thus, BDI can be used for an indicator of a crisis in GDP growth for the United States. Keywords: BDI, Markov Switching VAR, Markov Switching Granger Causality ,Economic Growth

1. Introduction The Baltic Exchange has a long history going back to 1744. In 1985, the Baltic Exchange developed the Baltic Dry Index (BDI) as a general indicator, consisting mainly of raw commodities such as grain, coal, iron ore, copper and other primary materials. Since its establishment, the BDI has become

*

Corresponding author. Tel. + 90-212-383-2527 E-mail address: [email protected]

fax. +90-212-259-4202

1877-0428 © 2015 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). Peer-review under responsibility of the International Conference on Leadership, Technology, Innovation and Business Management doi:10.1016/j.sbspro.2015.11.389

Melike E. Bildirici et al. / Procedia - Social and Behavioral Sciences 210 (2015) 416 – 424

417

one of the foremost indicators on the cost of shipping and an important barometer on the volume of worldwide trade and manufacturing activity (Faqin Lina Nicholas C.S. Sim, 2013; 59:1-18. ) Investors are always looking for practical economic indicators that they can use to help them make informed investing decisions. Recently, Baltic Dry Index can be sources of economic indicator on a global scale. In addition to that the BDI depends on volatile of crude oil prices and port and docking fees which makes BDI to be sensitive for global demand and manufactured goods (Economic SYNOPSES, Federal reserve banks of St. Louis). Oomen (2012) mentioned that Baltic Dry Index (BDI) which is a source of measurement to determine cost of raw materials around the world such as iron, coal, cement, grain. Average of price of 23 different shipping routes around the world compiles daily to form the Baltic Dry Index. Economic indicators such as unemployment rate, inflation and oil prices that can be manipulated or influenced by governments and speculators, however, Baltic Dry Index is difficult to manipulate because it is driven by clear forces of supply and demand. One of the reasons for BDI to be difficult to manipulate and influence is number of ships around the world is limited with up to a certain extend therefore in order to manipulate and increase the supply, more ships need to be built which will be very costly. After economic recessions and during economic growth, demand of raw materials increase as production and investments are also increase, as a result transportation volume grows accordingly. On the other hand, during economic slowdowns, demand of raw material decreases which creates utilized capacity. Global factors also play important role in supply and demand of BDI index. BDI and global markets have common economical and financial movement due to market supply and demand which is as a result of turmoil’s and crisis. Iron ore, coal, phosphate, grain and alumina are main goods of dry bulk transportation. These goods are mostly dynamics of construction and energy sector. Moreover, freight rate is determined by raw material demand as transportation need continues to remain the same. In this sense, our work is related to, among others, Korajczyk and Viallet (1989), Cutler, Poterba, and Summers (1991), Harvey (1991, 1995), Bekaert and Hodrick (1992), Campbell and Hamao (1992), Ferson and Harvey (1993), Heston and Rowenhorst (1994), Bekaert and Harvey (1995), Dumas and Solnik (1995), De Santis and Gerard (1997), Fama and French (1998), Griffin and Karolyi (1998), Rowenhorst (1998), Bossaerts and Hillion (1999), Jorion and Goetzmann (1999), Rangvid (2006), Guidolin and Timmermann (2008), Bekaert, Hodrick, and Zhang (2009), Pakthuanthong and Roll (2009), Rapach, Strauss, and Zhou (2009), Hjalmarsson (2010), and Henkel, Martin, and Nardari (2010). The volatility of the bulk shipping market has gained wide attention, and much research regarding this volatility has been undertaken. In the past decades, econometric and statistical methods, such as VAR, GARCH and VECM models, have been widely used in shipping market analysis and forecasting. For example, Kavussanos and Alizadeh-M (2001) analysed seasonal volatility considering ship type, lease term, market environment, etc. Veenstra and Franses (1997) found that cointegration relations exist between several freight rate time series. Duru and Yoshida (2011) studied the lag and price elasticity of the bulk shipping market through the long-term freight index. The results indicate that the log-linear model is not a good method for bulk shipping market forecasting because of the spurious regression. Byoung-wook (2011) decomposed the bulk shipping market freight time series into a longterm trend component and a temporary particular component with a random model. Bashi (2011) investigated the importance of the BDI growth rate as a predictor that stems from two findings. First, the BDI growth rate exhibits a positive and statistically significant relation to subsequent global stock returns, commodity returns, and industrial production growth. Second, the predictability is corroborated in statistical terms, in-sample and out-of-sample, as well as through metrics of economic significance, and in the presence of some alternative predictors. Movements in the BDI growth rate, thus, capture variation across the real and financial sectors, and the association appears stable across a multitude of economies. (Bashi et.all:2011). Studies that focus on BDI have mostly used VAR-VECM models. However, MS-VAR model have been used in this study ecause of the nonlinear structure of the economic time series, especially GDP

418


which has been used as the measure of economic performance fluctuates as the business cycles. In these models in perspectives of business cycles, the parameters are assumed to be constant over the sample period which means the relationship between GDP and BDI is stable. But the world has experienced many significant crises during the past decades. For this reason, the relationship between BDI and economic growth must be analysed in perspectives of the business cycles because countries and the world experienced many significant crisis. If in time series analysis, phase of the business cycle must be taken into account, the estimated parameters would be incorrect and misleading. One way to overcome these problems is to divide the sample into sub-samples, based on the structural breaks; however, in most cases the exact date of these changes are not known and the researcher must determine it endogenously based on the data. But there is no guarantee that the relationship between GDP and BDI in the same date as the break dates of the variables itself (Falahi:2011;4165-4170; Bildirici:2012;179-205). In this paper, MS-VAR model is selected to analyze the relationship between BDI and economic growth for the United States. Although this study can be defined as complementary to the previous empirical papers, it differs from the existing literature for some aspects. Firstly, as being distinguished from the previous works, it employs Markov Switching VAR method. Secondly, it is used Markov Switching Granger Causality analysis. MS-Granger causality approach allow analysis of the Granger Causality in different regimes of a business cycle. 2. Data In this study, the relationship between BDI(BDI=ln(bdit/bdit-1) and economic growth (Y=ln(GDPt/GDPt-1) is investigated by MS-VAR method. Quarterly data covers the period of 1986(1)– 2014(1) for the United States. The data are taken from Bloomberg. BDI is the value of the index and GDP is in terms of current U.S. dollars. 3. Methodology 3.1. MS-VAR Method Hamilton (1989) proposed a simple nonlinear framework for modeling economic time series with a permanent component and a cyclical component as an alternative to a stationary linear autoregressive model. Clements and Krolzig (2002) and Holmes and Wang (2003), Cologni and Manera(2006) and Bildirici, Alp and Bakırtaş (2011) used to MS-AR and MS-VAR model to test impact on GDP of oil shock. Falahi (2011) and Bildirici (2012 a, b) used MS-VAR model for analysis of the relationship between energy consumption and economic growth. q

MSI(.)-VAR(.) model is

yt

P st ¦ Ai ( st ) yt 1 ut ,

(1)

k 0

ut st ~ NID 0, ¦ ( st ) . and Ai (.) shows the coefficients of the lagged values of the variable in different

regimes, and

¦

shows the variance of the residuals in each regime.

P st define the dependence of the mean vairable st.

P of the K – dimensional time series vector on the regime

419


In

an

MS-VAR

model,

st

is

governed

by

f f Pr ª st ^st 1`i 1 , ^ yt 1`i 1 º Pr ^st st 1; U` , ¬ ¼

a

Markov

chain

and

(2)

where p includes the probability parameters. That is, the state in period t would depend only on the state in period t-1. On the other hand, the conditional probability distribution of yt is independent of st-1, that is,

P( yt Yt 1 ,st 1 ) Pr ( yt Yt 1 ) .

(3)

It is assumed that s follows an irreducible ergodic M state Markov process with the transition matrix defined as,

P

ª p11 «p « 21 « M « ¬ pM 1

p12 p22

L L

M

L L

pMM

p1M º p2 M »» M» » pMM ¼

(4)

The Markov chain is ergodic and irreducible; a two-state Markov chain with transition probabilities pij has unconditional distribution given by:

Pr(st=1)=(1-p22)/Pr(st-1=2) , Pr(st=2)=(1-p11)/Pr(st-1=1)

(5)

Pr(st=1)=(1-p) To estimate the MS models, there are available different ways as the maximum likelihood estimate(MLE) and the expectation maximization (EM) suggested by Hamilton. The EM algorithm has been designed to estimate the parameters of a model where the observed time series depends on an unobserved or a hidden stochastic variable. To make inference, it was used probability

[it 1

Pr ¬ª st 1

iterative method for t= 1,2,..T, while taking the previous value of this

i :t 1;T º¼

as input.

3.2. Markov Switching Granger Causality Warne (2000), Psaradakis et al. (2005) proposed a different approach to causality based on the Granger causality. Falahi (2011) utilized Short-run or weak Granger causalities for MSIA(.)-VAR(.) model by following the Granger causality in the concept of Markov Switching. Based on the coefficients of the lagged values of Yt and BDIt in the equations it is determined the existence of causalities between these two variables. In the equation vector where the dependent variable is BDIt, if any of the coefficients of lagged variables of Yt are statistically different from zero, the obtained test process will result in the acceptance of the causality. In any of the regimes, based on the coefficients of the lagged values of Yt and BDIt in the equation for BDIt and Yt, we could determine the existence of nonlinear causality between these two variables. In the equation for BDIt, if any of the coefficients of Yt be significantly different from zero in any of the regimes,

Yt

m

n

i 1

i 1

D1 ¦ EiYt i ¦Ii BDIt i H t

(6)

420


BDI t

m

n

i 1

i 1

D 2 ¦ J i BDIt i ¦ OiYt i et

(7)

It is concluded that LYt (LBDIt) is a Granger cause of LECt (LYt) in that regime. Granger

0 causalities are detected by testing H 0 :I 12 and H 0 :I 21 0 . The methodology requires the estimation of either an MSIA(.)-VAR(.) or a MSIAH(.)-VAR (.) model. (k )

(k )

4 . Empirical Results The integration order of the Y and BDI was determined by using the the test of Ng and Perron (2001). The results of unit root tests were given in Table 1. The results indicate that the Y and LBDI appear to be stationary. After unit root test, Johansen procedure was used to determine the possible existence of cointegration between BDI and Y. Johansen Cointegration result in Table 1 determined that the null hypothesis of no cointegration was rejected. If the variables are cointegrated, they can be used for test of MS- Granger Causality. Table 1: Unit Root Test Results

MZa MZt -30.200 -3.718 -48.907 -4.929 -13.800 -2.580 -8.100 -1.980 -5.700 -1.620 Johansen Cointegration Test Rresult r=0 0.097 rd1 0.002

Y BDI 1%* 5%* 10%*

*

MSB 0.123 0.101 0.174 0.233 0.275 r=0 rd1

MPT 1.328 0.541 1.780 3.170 4.450 11.146 0.183

Asymptotic critical Values

MSIAH(3)- VAR(4) model was selected based on the Akaike Information Criteria (AIC) and LR test. In selected models, in order to determine the number of regimes, first of all, a linear VAR is tested against a MSVAR with 2 regimes, and the H 0 hypothesis, which hypothesizes linearity, was rejected by using the LR test statistics. Since it was observed that two regime models overruling the linear model are insufficient in explaining the relationships between the mentioned variables, and models with 3 regime are considered. Therefore, secondly a MSVAR model with 2 regimes is tested against a MSVAR model with 3 regimes; H 0 hypothesis, which specifies that there are 2 regimes, was rejected and MSVAR with 3 regimes was accepted as the optimal model because of the LR statistic was greater than the 5%

critical value of F . The transition probability matrix is ergodic and cannot be irreducible. Ergodic transition probability matrix confirms stationarity of the regime. As the details can be found at Hamilton (1994) and Gallager (1996), ergodic transition probabilities matrix is always covariance-stationary. The MSIAH(3)-VAR(4) model was estimated for the United States and the results were given in Table 2. Regime 1 is recession or crisis regime. The moderate growth regime is regime 2 and high growth regime is regime 3. The model tracks fairly well the crisis of 1990-1991, 2001, 2008 and recent slow down. In the estimated MS-VAR model, the total time length of expansion period (Regime 2 and Regime 3) is longer than the total time length for recession (Regime 1) as expected. The results are 2

421


signified the presence of significant level of asymmetries for the business cycles experienced by the United States. The first regime tends to last 3.13 years on the average, while the Regime 2 is comparatively more persistent with 3.11 years. Finally, Regime 3 which corresponds to the high growth tends to last 2.55 years on the average. Crisis regime of economy is the most persistent regime in the US. As the calculated regime probabilities are Prob(st=1|st−1=1)=0.681, Prob(st=2|st−1=2)=0.679 and Prob(st=3|st−1=3)= 0.607, the persistence of each regime is significantly high. By moving from the conditions described above, the presence of important asymmetry in the business cycle in US is accepted. The computed probability (i.e. Prob(st = 3|st−1=1) = 0.107) reflects the chance that a recession is followed by a period of high growth and the computed probability (i.e. Prob(st = 2|st−1 = 1) = 0.211) reflects the possibility of entering the crisis regime from moderate regime of economy is higher than the possibility of entering the crisis regime of high growth phase. Table 2: MSIAH(3)-VAR(4) Model Eestimates for USA

Estimation sample: 1965 – 2010 Regime 1 Variables: BDIt Yt Variables: Regime –specific Intercept Constant Constant -14.52 0.19 Regime –specific autoregressive coefficients Yt-1 Yt-1 -16.37 0.42 Yt-2 Yt-2 9.89 -0.21 Yt-3 Yt-3 28.73 -0.02 Yt-4 Yt-4 -6.52 0.63 BDIt-1 BDIt-1 0.26 0.01 BDIt-2 BDIt-2 -0.22 -0.01 BDIt-3 BDIt-3 0.29 0.01 BDIt-4 BDIt-4 -0.79 -0.01 Regime- specific standart error 0.116 0.004 SE Regime properties: Duration and Probablities of regimes

Regime 1 Regime 2 Regime 3

Regime 2 BDIt

Yt

Regime 3 Variables: BDIt

Yt

Constant -31.4

0.92 Yt-1 Yt-2 Yt-3 Yt-4 BDIt-1 BDIt-2 BDIt-3 BDIt-4

3.11 23.67 -6.40 10.85 0.12 0.30 0.44 -0.05

0.02 0.03 0.04 0.01 -0.01 -0.001 -0.001 0.001

0.105

0.003

Transition

probabilities

Prob.

Duration

Transition p.

0.240 0.415 0.345

3.13 3.11 2.55

Regime 1 Regime 2 Regime 3

log-likelihood : 468.73 linear system : 400.24 ; AIC criterion Chi(42) =[0.0000] ** Chi(48)=[0.0000] ** DAVIES=[0.0000] **

Regime 1

0.681 0.057 0.154

: -7.40 linear system :

Regime 2

0.212 0.679 0.239

47.82

0.27

-5.49 1.61 -28.45 -13.20 -0.12 -0.47 -0.35 0.53

0.21 0.51 -0.27 0.26 0.06 0.01 0.02 0.01

0.143

0.004

Regime 3

0.107 0265 0.607

-7.02 LR linearity test: 136.97

4 . Conclusion In the MSIAH(3)-VAR(4) model, the dependent variable in the first equation is the innovation of GDP. The estimated coefficients of BDI in equation 6 are statistically significant at the conventional level but not substantial in all regimes and parameter estimates in Regime 3 are positive which means that when the economy in a high growth regime, BDI improvements effects economic growth positively. The

422


dependent variable of the second equation is BDI, the estimated coefficients of Y in equation 7 are both statistically significant at the conventional level and substantial in all regimes but parameter estimates in threee different regimes varies. The MSIAH(3)-VAR(4) model was estimated for the United States and the results were given in Table 2. Regime 1 is recession or crisis regime. The moderate growth regime is regime 2 and high growth regime is regime 3. The model tracks fairly well the crisis of 1990-1991, 2001, 2008 and recent slow down. In the estimated MS-VAR model, the total time length of expansion period (Regime 2 and Regime 3) is longer than the total time length for recession (Regime 1) as expected. The results are signified the presence of significant level of asymmetries for the business cycles experienced by the United States.

References Alziadeh, A. and Talley, W. (2010) Microeconomic determinants of dry bulk shipping freight rates and contract times, Transportation, vol. 38, pp. 561-579. AndersWarne, (2000). "Inference in Cointegrated VAR Systems," The Review of Economics and Statistics, MIT Press, vol. 79(3), pages 508-511, August. Bakshi, G. Panayotov, G. and Skoulakis, G. (2010), The Baltic Dry Index as a Predictor of Global Stock Returns, Commodity Returns, and Global Economic Activity. Bakshi, Gurdip, Panayotov, Georgeand Skoulakis, Georgios(2011), The Baltic Dry Index as a Predictor of Global StockReturns, CommodityReturns, and Global Economic Activity, February. Baltic Exchange, About Us, last accessed on 30 June 2014, from http://www.balticexchange.com Baltic Exchange, last accessed on 30 June 2014, from http://www.learningmarkets.com/understanding-the-baltic-dry-index/ Bildirici, M., Aykaç Alp andE.,Bakırtaş, T.,( 2011). “The Great Recession and The Effects of Oil Price Shocks and The U.S. Recession: A Markov-Switchingand TAR-VEC Analysis” Journal of Energyand Development, 35(2). Bildirici, Melike (2012),Economic growth and electricity consumption in Africa and ASIA:MS-VAR and MS-GRANGER causality analysis; MPRA Paper No. 41796,179-205. Byoung-wook,KO (2011),Dynamics of Dry Bulk Freight Market:Through the Lens of a Common Stochastic Trend Model,The Asian Journal of Shipping and Logistics,Vol:27,3,pages 387-404,December. Clements, M.C. &Krolzig, H.-M., 2002. "Modelling Business CycleFeatures Using SwitchingRegimeModels," Economics Series WorkingPapers 9958, University of Oxford, Department of Economics,1-17. Cologni, Alessandro&Manera, Matteo, (2008). "Oilprices, inflation and interest rates in a structural cointegrated VAR model forthe G-7 countries," EnergyEconomics, Elsevier, vol. 30(3), pages 856-888, May. Duru, O. (2010) A fuzzy integrated logical forecasting model for dry bulk shipping index forecasting: An improved fuzzy time series approach, Expert Systems with Applications, vol. 37 (7), pp. 5372-5380. Duru, Okan. Bulut, Emrah and Yoshida, Shigeru (2010),Bivariate Long Term Fuzzy Time Series Forecasting of Dry Cargo Freight Rates,The Asian Journal of Shipping and Logistics,Vol:26,2,pages 205-223,December. Falahi F. 2011. Causal relationship between energy consumption (EC) and GDP: A Markov-switching (MS) causality. Energy. 36(7): 4165–4170. Gallagher,N.B. and Wise, B.M. (1996), “The Process Chemometrics Approach to Chemical Process Fault Detection and Supervision,” J. of Process Control, 6(6), 329-348. Geman, H. and Smith, W. (2012) Shipping Markets and Freight Rates: An Analysis of the Baltic Dry Index, The Journal of Alternative Investments, vol.1, pp. 98-109. Hamilton, J.D. (1989). A new approach to the economic analysis of nonstationary time series and the business cycle, Econometrica, 57, 357-384. Hamilton, James D.(1994).”Time Series Analysis”,Princeton Unıversity Press, page 257-291. Holmes,Mark and Wang,Ping(2003). "Oil Price Shocks and the Asymmetric Adjustment of UK Output: A Markov-switching approach," International Review of Applied Economics, Taylor & Francis Journals, vol. 17(2), pages 181-192. Kavussanos, Manolis G. &Alizadeh-M, Amir H., 2001. "Seasonality patterns in dry bulk shipping spot and time charter freight rates," Transportation Research Part E: Logistics and Transportation Review, Elsevier, vol. 37(6), pages 443-467, December. Ng, Serena and Perron, Pierre (2001),”LAG Length Selection and the Construction of Unit Root Tests with Good Size and Power” Econometrica, Volume 69, Issue 6, pages 1519–1554, November.

423


Nicholas, C.S. Sim, andLin, Faqin (2013).Trade, Income and the Baltic Dry Index, European Economic Review, 59, 1-18. Oomen, J. (2012,) The Baltic Dry Index: A predictor of stock market returns? Master Thesis, Tilburg University, Tilburg. Psaradakis Z.,Morten O.R. and Sola M. (2005), Markov-Switching causality and the Money-Output Relationship, Journal of Applied Econometrics, 665-83. Sengupta, Rajdeep and Tam, Yu Man (2009). “Recent Movements in the Baltic Dry Index”, Economic SYNOPSES, Federal reserve banks of St. Louis, No:12. Veenstra, A.W. andFranses, P.H.( 1997).A co-integration approach to forecasting freight rates in the dry bulk shipping sector,Transportation Research Part A, Vol. 31,6, pp. 447–458. Zuccollo, Dino Roberto (2014).“The Baltic Dry Index: A Leading Economic Indicator and its use in a South African context”, Master Thesis, University of the Witwatersrand.

Appendix A. A.1. The Figure of Dynamics of MSIAH (3) – VAR(4) Model Correlogram: Standard resids

1

ACF-BDID

P ACF-BDID

0

0.2

Spectral density: Standard resids BDID

Density: Standard resids

0.50

0.1

BDID

N(s=0.923)

QQ Plot: Standard resids BDID u normal

2.5 0.0

0.25

-2.5 1 1

5 9 13 Correlogram: Prediction errors ACF-BDID

P ACF-BDID

0

0.0 0.5 1.0 Spectral density: Prediction errors 0.2 BDID 2 0.1

1 1

5 9 13 Correlogram: Standard resids ACF-USAGDP

P ACF-USAGDP

0

0.0 0.2

-2.5 0.0 2.5 Density: Prediction errors BDID

N(s=0.258)

1

-2 0 2 QQ Plot: Prediction errors

0.0

0.5 1.0 -1 0 1 Spectral density: Standard resids Density: Standard resids USAGDP

0.4

0.1

BDID u normal

2.5

USAGDP

N(s=0.969)

-2 0 2 QQ Plot: Standard resids USAGDP u normal

2 0

0.2

-2 1 1

5 9 13 Correlogram: Prediction errors ACF-USAGDP

P ACF-USAGDP

0

0.0 0.2 0.1

1

5

9

13

0.0

0.5 1.0 Spectral density: Prediction errors 100 USAGDP 75 50 25 0.5

1.0

-2.5 0.0 2.5 Density: Prediction errors USAGDP

N(s=0.00506)

2.5

-2 0 2 QQ Plot: Prediction errors USAGDP u normal

0.0 -2.5 -0.02 -0.01 0.00 0.01 0.02

-2

0

2

424


A.2. The Figure of Predicted h-step probabilities of stability Predicted h-step probabilities when st = 1

1.0

Regime 1 Regime 3

Regime 2

0.5

Predicted h-step probabilities when st = 2

1.0

Regime 1 Regime 3

Regime 2

0.5

0 0.4

20 40 Probability of duration = h Regime 1 Regime 3

Regime 2

Predicted h-step probabilities when st = 3

1.0

Regime 1 Regime 3

Regime 2

0.5

0 1.0

20 40 Probability of duration dh Regime 1 Regime 3

Regime 2

0 0.75

20 40 Probability of staying in the same regime h p Regime 1 Regime 3

Regime 2

0.50 0.5

0.2

0.25 0.00

0

20

40

0

20

40

0

20

40

A.3. The Figure of Probabilities of regimes for BDI and Y MSIAH(3)-VAR(4), 1987 (2) - 2014 (1) 2

BDID

USAGDP

1

1.0

1990 Probabilities of Regime 1 filtered predicted

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

1995

2000

2005

2010

2015

smoothed

0.5

1.0


smoothed

0.5

1.0


smoothed

0.5

1990