Relationship between Baltic Dry Index and China economy

(1)

Relationship Between Baltic Dry Index and China

Economy

Amirahmad Bornaee

AMSTERDAM BUSINESS SCHOOL UNIVERSITY OF AMSTERDAM

(2)

Index and China Economy” is the result of my own research except as cited in the references. This thesis has not been accepted for any degree and is not

concurrently submitted in candidature of any other degree.

Student : Amirahmad Bornaee

Date : 15-09-2017

Supervisor : Dr. Simon Broda

(3)

This thesis is dedicated to my lovey wife, Zahra, who constantly supported and encouraged me during this study. This work is also dedicated to my little son,

Habib, who brought enormous excitements since he came into our life few months ago.

(4)

This study examined the relationship between the Baltic Dry Index (BDI) and China’s stock market returns. The main findings of this research suggest that the relationship between BDI and China stock market indices has changed as the fundamentals of the dry bulk market evolved over the years. In particular, BDI has been able to predict China’s Stock market over the period of 2000 till 2005 according to Granger causality tests, results of basic regression models and additional robustness tests. Therefore, we can conclude that BDI has been a significant and robust predictor of China’s market over the period of 2000-2005. However, our findings suggest this relationship does not exist anymore. In fact, the causality relation has been reveresed and China financial market indicators are Granger causing the BDI. This finding can be explained by the fact that since 2005 the new ships orders gradually entered to the fleet causing an excess capacity in the market. Moreover, the financial crisis exacerbate the situation by a significant decline on the demand. All of these resulted in a market condition where demand is generally lower than the available fleet capacity. In such circumstances, BDI price is not anymore a pure reflection of demand and instead it more reflects the supply-demand dynamics in the market. Therefore, a decline in freight rate in such situation does not neccesary reflect a lower demand or trade volume which could potentially predict a slower economic growth in near future. Instead, it reflects a market condition where global trade is not growing as fast as the supply of new fleets and therefore ships were willing to take cargoes at lower rate just to cover operating costs and reduce their unutilised capacity.

(5)

List of Figures

2.1 Main Bulk Commodities Seaborne Trade Volume 4

2.2 Demand and Supply Growth in Dry Bulk Shipping (2000-2017) 6

2.3 Baltic Dry Index (2000-2017) 7

4.1 Shanghai Stock Exchange Composite Index (SSE) (2000-2017) 12 4.2 MSCI World and MSCI Emerging Market Index (2000-2017) 13

(8)

4.1 Descriptive statistics of returns (Jan 2000 - May 2017) 13

5.1 Unit Root Test Results 16

5.2 Granger Causality Between BDI and other indices (2000-2004) 17 5.3 Regression results with BDI as the dependent variable (2000-2004) 18 5.4 Granger Causality Between BDI and other indices (2005-2017) 19 5.5 Regression results with BDI as the independent variable (2005-2017) 19 5.6 Breusch-Godfrey LM Test for Basic Model (2000-2004) 20

5.7 White Test for Basic Model (2000-2004) 20

5.8 Regression Results with rSSE_{as dependent variable (2000-2004)} ₂₁

5.9 Regression Results with rmscicn_{as dependent variable (2000-2004)} ₂₂

5.10 Robustness Analysis for mscicn (2005-2017) 24

5.11 Robustness Analysis for SSE (2005-2017) 25

5.12 Breusch-Godfrey LM Test for Basic Model (2005-2017) 25

5.13 White Test for Basic Model (2005-2017) 26

(9)

Chapter 1 Introduction

Standing at the backbone of global supply chains, maritime transport is the main driver of international trade. According to UNCATD’s review of maritime transport, world seaborne trade volumes are estimated to have accounted for over 80 per cent of total world merchandise trade.[13] Maritime industry has witnessed major growths as well as slowdowns in tandem with economic activities on key importing and exporting countries. China has played a major role in recent dynamics of maritime transport market over the last decade. While China’s economic growth was the main driver of international maritime transport, recent slowdowns in China’s economy has a significant impact on the maritime industry, especially on the dry bulk sector. According to [13], global trade in five major dry bulk commodities (iron ore, coal, gran, bauxite and alumina and phosphate rock) declined by 1.3 per cent in 2015 which was mainly due to decline in China’s steel output, which accounts for nearly half of global output, for the first time since 1981. Iron ore shipments into China increased only by 2.8 per cents, a much slower pace than 15 per cent expansion in previouss year due to reduced steel production. Similarly, for the first time in about three decades, coal seaborne shipments fell, by 6.9 per cents which was again caused by China’s slow economic growth, restrictions on low-quality coal import and newly introduced air pollution control measures in China.[13]

Baltic Dry Index (BDI) as the main index for shipping costs of dry bulk commodities, which are the raw materials for production, can be viewed as an early stage indicator of global economic growth and production and the stock market movements. Several researches have been conducted to investigate these relationships. Bakshi et al. (2011) [3], Apergis and Payne (2013) [2] and Alizadeh and Muraduglo (2014) [1] have investigated the prediction power of BDI for stock

(10)

markets in developed and undeveloped countries. Bakshi (2011)[2] showed that there is a positive and significant relationship between BDI and subsequent global stock returns, commodity returns, and industrial production growth. Futhur-more, they showed that the predictability remains in place in the presence of some alternative predictors. Apergies and Payne (2013) [2]also argued that their findings support the idea of the close relationship between the cost of shipping raw materials and the production of intermediate and final goods in that the demand for commodities and, therefore, economic activity, follows movements in the BDI. They further exhibit that this predictability does not only hold for the short-term but also persist at longer horizons. Alizadeh and Mordauglo (2014) [1] investigated the gradual diffusion of shipping freight rates information and argued that this enables us to predict the US stock returns and volatility.

There has been little research on the predictability of BDI for the Chinese market. Given the importance of the Chinese market for the global economy, we will focus on this country by investigating the BDI prediction power specifically on Chinese stock markets and economic indices. The main research questions is to investigate whether BDI changes can predict the Chineses market indices such as MSCI China, Shanghai Stock Exchange Composite index (Research Question 1). We further check if the prediction power remain robust in the presence of other alternative predictors such as other MSCI indices as well as Oil Price (Research Question 2).

The rest of the thesis continues as follows. Section 2 describes the fun-damentals of dry bulk shipping economics. In Section 3, we will review recent literature about the Baltic Dry Index and its predicition power. Section 4 in-cludes a description of datasets and introduction to econometric models used in their research. The study results of the research are discussed in Section 5 and Section 6 contains the conclusion and recommendation for future studies in this area.

(11)

Chapter 2 Dry Bulk Shipping Economics

2.1 Seaborne Trade

Seaborne trade currently accounts for around 80% of merchandise trade. Without maritime transport, transport of raw materials and the import/export of afford-able food and final products would hardly be possible. In 2015, the volume of seaborne trade exceeded 10 billion tons with Dry Cargo commodities (excluding Containers) accounting 54% of total seaborne trade, while Tankers trade and containers account for 29% and 17% of the total respectively (see Figure 2.1). [13] The bulk carrier markets are divided based on four sizes of ships which are capesize (100,000 dwt plus), panamax (65,000-99,999 dwt), handymax/supramax (40,000-64,999 dwt) and handysize (10,000-39,999 dwt) bulk carriers.[13] Cape-size vessles, which account for about 62% of the dry bulk traffic, mainly carry coal and iron ore and are chartered for longer haul voyages. Panamax vessels can also carry coal and iron ore as well as other bulks such as cokes, fertilizers, bauxite and etc. The smaller vessels, supramax and handysize, are often chartered to transport steel products, scrap, sugar, etc. [8]

Dry bulks can be broadly divided into two categories. The five major bulks consist of Iron ore, coal, grain, phosphates and bauxite. Other bulks such as steel products, steel scrap, cement, gypsum, non-ferrous metal ores, sugar, salt, sulphur, forest products, wood chips and chemicals, etc are considered as the minor bulks.[12] Iron ore, coal and grain dominate the dry bulk shipping market with 70% of the shipping volume, with about 30%, 30% and 10% shares respectively. Major players in this market are main importers and exporters such as China, Japan, Australia, Brazil, Russia and the United States.[13]

(12)

Figure 2.1: Main Bulk Commodities Seaborne Trade Volume Source: UNCTAD 2016

2.2 Freight Rates

In this section, we discuss the fundamental drivers of the shipping freight rates. Then, we introduce Baltic Dry Index as the main indicator of dry bulk shipping costs. We will then briefly describe BDI historical trends and volatilities.

Supply Characteristics

High costs and long lead time of building new ships along with the ab-sence of storability are the main characteristics of the supply side of the freight rate market. As highlighted by Bakshi et al. (2010) [3], this makes the supply structure of the market quite predictable and inelastic to the short-term dynamics of the market. Ordering new ships requires a significant amount of investments often financed by banks which exert long-term obligations for ship owners. In addition, since building of new ships takes between 2-3 years, by the time the ships are ready to be added to the fleet, the economic situation may have already changed drastically which may result in deep supply and demand imbalances in the market. Geman and Smith (2012) [6] argued that the absence of storability makes the supply side inflexible to immediate demand shocks which results in large swings in trajectories. They explained that “Unlike many commodities, the storage of vessels to act as a buffer against shocks in supply and demand is far too expensive: Even when freight rates fell dramatically from 12,000 points in May 2008 to 670 in December 2008 (see Figure 2.2), ship-owners preferred to make their ships available at just bunker fuel costs rather than letting them idle and paying the cost of insurance. This full absence of storability is essentially shared

(13)

5 with another commodity, namely electricity”.

Demand Characteristics

While the supply structure of the freight rates is quite predictable and slow to react to immediate changes, the demand side is the main driver of the freight rates behavior. The main driver of the shipping demand is the global demand for the dry bulk commodities. Higher demand for dry bulks, as the raw materials of many industrial productions, leads to higher freight rates for shipping the bulk cargos. Several researchers have investigated relationship between the GDP and freight rates. Shen and lo (2012)[11] showed the existense of short-term and long-short-term equilibrium relationships between BDI and China’s GDP, while no significant causality were detected between BDI and the GDP of other BRIC countries. Batrinca (2014) [4] also found a significant positiove relationship between BDI and worldwide GDP.

Other factors such as bunkering costs, seasonal pressure and issues with Choke points also affect the freight rates. About 25%-33% of the operationg costs of a vessel is related to bunker prices. Therefore, a higher oil price will directly affect the freight rates. Choke points in busy shipping routes which are often exposed to accidents, terrorist attacks, piracy, millitary conflicts, and severe winters are often considered as another driving factor of BDI rates.[14] In case of any disruptions in these routes, the global supply will be adversely affected which will lead to higher freight rates.

Several literature have discussed the cyclicality of the shipping freight rates. Cufely (1972) elaborated a sequence of three key stages common to ship-ping cycles: first, a shortage of ships develops, then high freight rates stimulate over-ordering of the ships in short supply, leading finally to market collapse and recession. Simillary, Stopford (2009) [12] has characterised the shipping market cycles as a sequence of following stages: a trough, a recovery, a peak, and a col-lapse. Shipping cycles used to average eight years but became shorter recently with the big activity in Korean and Chinese shipyards. As shown in Figure 2.2, there has been high demand growth and shortage of supply in the early years of 2000s which resulted in rapid growth in shipbuilding orders. Consequently, the new ships gradually came into the fleet starting from the mid 2000s and resulted in a huge oversupply which still exist in the market. Moreover, the financial crisis in 2007-2008 resulted in a significant decline in demand for seaborne transporta-tion which resulted in a collapse in the freight rates as depicted in Figure 2.2.

(14)

The market is still facing a signficinat excess capacity (oversupply) and given the slow economic growth in China and other emerging markets, it is expected to take a few more years for the market to recover to a more balanced situation.

Figure 2.2: Demand and Supply Growth in Dry Bulk Shipping (2000-2017) Source: Crucial Perspective, Bloomberg and Clarkson

2.3 Baltic Dry Index (BDI)

The Baltic Dry Index (BDI) is considered as the main indicator of the shipping cost for dry bulk commodities, including coal, grain, iron ore, finished steel and other similar materials. Started in 1985, the Baltic Freight Index (BFI) was de-rived based on the weighted average of shipping cost of 13 trade routes: grain (five routes), coal (three routes), iron ore (one route) and general charter (four routes). Since November 1991, BDI came into operation and replaced BFI. The Baltic Exchange modifies the routes and weightings to reflect changes in trade volumes. In October 2001, the routes in BDI expanded to cover 26 shipping routes and four vessel sizes: Handysize, Supramax, Panamax and Capesize. [9] BDI is often used by industry analysts as a market barometer and leading indi-cator of world economic situations. The index components are used as a physical chartering benchmark for floating rate contracts as well as for the settlement of shipping derivative contracts (FFAs). The panelists of the Baltic Exchange, the international shipbroking firms submit their views about the current shipping cost for representative cargoes and routes on a daily basis. The rate assessments are then weighted and aggregated to form the overall index for BDI and its com-ponents which are four size specific indices: the Baltic Capesize Index (BCI), the Baltic Panamax Index (BPI), the Baltic Supramax Index (BSI) and the Baltic Handysize Index (BHSI). Each index is based on calculating a time-charter av-erage (TCA) which is expressed in US dollars per voyage day. TCA is used to assess the daily average revenue of a given vessel by subtracting expenses (e.g. port costs, etc) from voyage revenue and dividing the adjusted number by voyage

(15)

7 days. TCA for an entire class is derived by aggregating the TCA for all vessels within the same class. [9]Since 1st of July 2009, the index has been a compos-ite of the Dry Bulk Timecharter Averages and is calculated using the following formula:

BDI =CapesizeT CA + P anamaxT CA + SupramaxT CA + HandysizeT CA

4 ∗ 0.10907849

(16)

Literature Review

In this chapter, we will review recent literature about the role of the Baltic Dry Index as an economic indicator. In particular we will focus on the study results regarding BDI predictive content for China’s economy. The prediction power of the Baltic Dry Index has been investigated in several studies. Bakshi (2011) [3] showed that there is a positive and significant relationship between BDI and sub-sequent stock returns, commodity returns, and industrial production growth in several economies. Futhurmore, they showed that the predictability remains in place in the presence of some alternative predictors. They investigated whether BDI growth rate is able to predict stock returns across the world in an in-sample as as well as out-of-sample analysis and if it survives the addition of alterna-tive predictors. They identified that the three-month BDI growth rate consti-tutes a common predictor of dollar denominated stock returns in a multitude of economies. They further investigated the economic significance of this pre-dictability and showed that the certainty equivalent returns and Sharpe ratios of a BDI-based portfolio strategy outperforms those obtained with a strategy that assumes i.i.d. returns. Their study was on a broad range of economies by using MSCI and Industrial production indices of each country as the dependent variable. However, the study results do not exhibit any significant BDI effect for China’s economy. Alizadeh and Muraduglo (2011) [1] show that there is a significant relationship in a way that changes in shipping freight rates can predict stock returns. Their results show that changes in BDI rates predict US size and sector indices as well as stock index returns worldwide. They exhibit that such predictability is not associated with time-varying risk premia, but it is consis-tent with the gradual information-diffusion hypothesis and the information that originates in international shipping freight reaches investors in the stock market with a time lag. In addition, they compare the predictive power of the BDI and

(17)

9 oil prices and show that the shipping freight rates have better predictive ability with respect to US stock-market prices across different size and sector indices, as well as worldwide country and sector indices. However, while the study covered a range of developed and developing countries, it did not include China. Therefore, the study results and conlcusion cannot be extended to China’s economy without further investigations. Apergies and Payne (2013) [2] analyze the predictive con-tent of the BDI with respect to financial markets as well as industrial production of G7 countries. Their findings support the idea of a close relationship between the cost of shipping raw materials and the production of intermediate and final goods. They further exhibit that this predictability does not only hold for the short-term but also persist at longer horizons by perofrming a causalitiy analysis which shows that the BDI appears to Granger-cause both the financial asset prices and industrial production in both the short-run and long-run and therefore, the causal dynamics reveal that the BDI contains properties of a leading indicator. They further investigated the robustness of BDI in the presence of other indica-tores such as MSCI world Index and Oil prices. The study was mainly focused on G7 countries and therefore their findings need to be validated with regards to developing economies such as China. In line with previous researches, Oome (2012) [10] investigated the BDI predictive content by dividing the dataset into four sub-periods. The empirical results supports the conjecture that a higher BDI return will result in a higher stock market return in the next month. However, the results are only significant for 2001-2007 period. They further explained that this might be due to the economic boom effect in this period or due to the the index modification in 1999, which made the index a weighted average of some sub indices, and only from that moment reflected the bulk trading volumes ac-curately. They also showed that the Panamax Index is the main driver of BDI predictive power. Their results also shows that BDI can predict China’s MSCI Index in the period of 2001-2007 and its prediction power survives the additon of alternative predictors such Oil price or MSCI World into the regression model.

The studies mentioned were conducted on a broad range of countries and did not focus on China’s economy. Furthermore, they did not include any Granger causality test as a formal way to test whether a causal relationship ex-ists between BDI and the economic index of the study. One of the few studies with special focus on China was Shen and Lo (2010) [11]. They investigated the causality relationships between the BDI and the gross domestic product (GDP) of the BRIC (Brazil, Russia, India and China) countries from the perspectives of equilibrium analysis and Granger’s causality test. They applied equilibrium anal-ysis to explore the long-term and short-term relationships between the BDI and

(18)

BRIC’s GDP and the results indicated that there is a long-term and short-term equilibrium relationship between the BDI and China’s GDP while no significant relationship are found for other BRIC countries. They further discussed that there is bidirectional causality between the BDI and China’s GDP which high-lights the important position of China in the international freight rate market. However, in order to avoid inference bias they excluded data for 2008 and 2009 due to significant fluctuation of BDI and economic growth. Therefore, the time horizon of their study is limited to the years between 1996 to 2007. Their study can be further extended to shed more light into BDI effect on China’s financial market. Chang and Lin (2009) [5] studied the Granger causality relationship be-tween BDI and major market indices for emerging markets as well as the United States. They found that there is a causal relationship between China’s stock market and the BDI, while other stock markets did not have any impact on the BDI. They examined the Granger causality relationship between Shanghai Stock Exchange Index (SSE) and BDI and find that SSE was Granger causing BDI in the period 2003-2008. In line with the researches mentioned above, this study focuses on causality relationship between Baltic Dry Index and China’s stock market.

(19)

Chapter 4 Description of Data and Econometric Model

4.1 Data Description

In this chapter I will describe the data sets and explain the econometric models used for the research.

The Baltic Dry Index (BDI) is a freight rate indicator issued daily by the London-based Baltic Exchange. Started in May 1985, the index was intially called BFI (Baltic Freight Index) and included 13 shipping routes covering a variery of dry cargoes ranging from 14,000 metric tons of fertilizer to up to 120,000 mt of coal. [1] The index has been evolving over the years by redefining the shipping routes and cargoes to better reflect the activities in the market. In November 1999, a major revision of the index took place where BDI replaced the BFI and it became an equally weighted index of its sub-indices which currently covers Handysize , Supramax, Panamax, and Capesize dry bulk ships carrying a range of commodities including coal, iron ore and grain over 23 shipping routes. The index is published daily and is available via DataStream or the Baltic Exchange website. The monthly logarithmic returns of the BDI are defined as follows:

rBDI_t−i = ln(BDIt−i) − ln(BDIt−i−1)

where BDIt−i is the BDI at begining of the month t − i. While BDI will be

the primarly indicators of bulk freight rates in this study, we will also investiagte the robusness of our results by usings Baltic Capesize index (BCI) and Baltic Panamax index (BPI) as the alternative freight rates indicators.

(20)

The Shanghai Stock Exchange Composite Index (SSE) has been collected via Datastream. Started in December 19, 1990 with a base value of 100, SSE is a capitalization-weighted index which tracks the daily price performance of all A-shares and B-shares listed on the Shanghai Stock Exchange. In addition, the Morgan Stanley Capital International (MSCI) Index for China has been collected. MSCI China captures large and mid-cap representation across China H shares, B shares, Red chips, P chips and foreign listings. With 150 constituents, the index covers about 85% of China’s equity universe. We will use SSE and MSCI China as representative indices of China’s financial market in our research.

Figure 4.1: Shanghai Stock Exchange Composite Index (SSE) (2000-2017)

In addition to Indices particularly selected for China, we have selected MSCI World and MSCI Emerging markets and Dated Brent Oil Price as alter-native predictors in order to investigate the robustness of our results. The MSCI World Index is a free float-adjusted market capitalization weighted index that is designed to measure the equity market performance of developed markets. The MSCI World Index consists of the following 23 developed market country indexes. MSCI Emerging Markets (MSCI EM) was launched on Janauary 2001 but the data prior to that has been back-casted. MSCI EM is designed to measure the equity market performance of the emerging markets. In addition, Dated Brent Oil has been collected from DataStream. The logarithmic returns of all indices are calcuatlated as follows:

r_tX = ln(Xt) − ln(Xt−1)

where Xt is the index or price at month t. The descriptive statistics of

the beginning-of-the-month logarithmic returns are presented below in Table 4.1. As shown in the table, the average returns of BDI are relatively lower and show

(21)

13 higher volatility than other indices. In addition, high kurtosis of BDI returns highlights its heavy-tailed distribution and not conforming to normal disribution characteristics. Figures for China indices (SSE and MSCI China) are relatively simillar to each other with SSE having a slightly higher volatility and lower skewness. Among MSCI indices, MSCI Emerging Market shows higher returns on average as well as higher volatiliy than MSCI World as it only includes the emerging economies.

Figure 4.2: MSCI World and MSCI Emerging Market Index (2000-2017)

rBDI rSSE rM SCI−CN rM SCI−W orld rM SCI−EM rOil

Mean -0.001 0.004 0.003 0.002 0.004 0.003 Median 0.012 0.009 0.012 0.008 0.008 0.013 Maximum 0.712 0.256 0.198 0.136 0.201 0.289 Minimum -1.297 -0.288 -0.300 -0.208 -0.297 -0.422 Std. Dev. 0.234 0.085 0.081 0.048 0.067 0.108 Skewness -1.088 -0.399 -0.612 -0.831 -0.576 -0.532 Kurtosis 8.440 4.395 4.147 5.145 4.827 3.878 Jarque-Bera 298.938 22.492 24.481 64.152 40.642 16.565 Probability 0.000 0.000 0.000 0.000 0.000 0.000 Observations 209 209 209 209 209 209

Table 4.1: Descriptive statistics of returns (Jan 2000 - May 2017)

4.2 Granger Causality

The Granger causality test will be used to formally check whether one time series would be useful in forecasting another one. The concept was proposed by Granger (1969) [7] to investigate how much of Y can be explained by its own past values as well as lagged values of X. For two-variable models this can be formulated as followings:

(22)

Xt= γ1+ m X i=1 aiXt−i+ m X i=1 biYt−i+ ε1t Yt= γ2+ m X i=1 ciXt−i+ m X i=1 diYt−i+ ε2t

where a, b, c and d are regressions coefficients. ε1and ε2 are error terms

and mis the optimal lags of X and Y. The null hypothesis is that Y does not Granger-cause X in the first regression:

H0: b1 = b2 = ... = bm = 0

and, in order to test the reverse relation, the null hypothesis would be that X does not Granger-cause Y in the second regression:

H0: c1 = c2 = ... = cm = 0

4.3 Econometric model

In order to investigate the role of BDI in predicting China’s financial market (research question 1), we use the following predictive regression model:

rt = α0+ α1rt−1BDI + α2rBDIt−2 + α3rt−3BDI + εt

where rt is the beginning of the month logarithmic return of China’s

index (SSE or MSCI China) and r_t−iBDI are the first, second and third lags of BDI’s monthly returns. Moreover, α0 is the constant and ε0 is the error term.

We further investigate the robustness of our results (research question 2) by extending the model to include alternative predictors such as MSCI world, MSCI Emerging markets and Oil Price. The extended regression model is as follows:

(23)

15

rt= α0+ α1rt−1BDI + α2rt−2BDI + α3rBDIt−3 + β1rXt−1+ β2rt−2X. + β3rt−3X

where rX

t−i are the are the first, second and third lags of the alternative

predictor. Ordinary least squares (OLS) method is used to estimate the linear regression models. Futhermore, the t-values of all the predictive regressions are based on heteroscedasticity standard errors.

(24)

Study Results

In this section, we present the empirical results of the study and try to address the research questions formulated in the first chapter. We first conduct unit root tests to examine the stationarity of the data series. Second, by dividing the time horizon into two sub-periods, we then analyze the causality relationship between BDI and the Shanghai Exchange Composite Index (SSE) and the MSCI China by performing Granger causality tests. Third, based on the findings of the Granger causality test, we estimate and analyze the regression model. Finally, we will investigate separate robustness tests for each sub-period.

5.1 Unit Root Test

Unit root tests are done on monthly data of all series. As shown in Table 5.1, all the level data are non-stationary as their t-values are not significant enough to reject the null hypothesis of non-stationary process at the 1% significance level. In contrast, the null hypotheisis are rejected for the first-difference of all data series. Hence, we can continue our studies using the returns of variables.

ADF test statistics ADF test statistics

Variables Levels Prob 1st Diff. Prob.

BDI -2.271 0.448 -6.996 0.00

SSI -3.915 0.013 -5.223 0.00

MSCI CN -3.032 0.126 -12.264 0.00

MSCI World -2.607 0.228 -5.756 0.00

MSCI EM -2.237 0.466 -11.965 0.00

Table 5.1: Unit Root Test Results

(25)

17

5.2 Granger Causality

As discussed in Section 2.2, high demand and shortage of supply in the early years of the 2000s resulted in high growth in shipbuilding orders which gradually came into the fleet in the second half of the 2000s and resulted in the huge over supply which still exists. In addition, the financial crisis in 2007-2008 resulted in a significant decline in demand for seaborne transportation. Therefore, due to significant changes in market fundamentals, we divided the time horizon of the study into two sub-periods: (i) From January 2000 until end of 2004 and (ii) from January 2005 until May 2017. We perform the Granger causality tests and further steps of the study based on these two subperiods.

5.2.1 Causality Relationships for the Sub-Period 2000-2004

In this section, we investigate the causality relationship between BDI and MSCI China and SSE in the period 2000-2004. As shown in Table 5.2, there is unidirec-tional Granger causality between BDI and SSE and MSCI China. BDI Granger causes in the period between 2000 to 2004. This confirms our hypothesis that BDI can predict China’s financial market as an early indicator of economic ac-tivities in short term. This finding is also in line with finding of other researches. Since the dry bulk market used to be very demand driven in that period, changes in BDI were clearly reflecting the demand for raw materials of prodcutions and consquently growth of China’s economy. We also estimated a basic regression model with BDI as the dependent variable and the results confirms the Granger causality results that in this period MSCI China and SSE could not predict the BDI developments as shown in Table 5.3.

Null Hypothesis Obs F-Statistic Prob.

MSCI CN does not Granger Cause BDI 58 0.625 0.539 BDI does not Granger Cause MSCI CN 58 3.903 0.026

SSE does not Granger Cause BDI 59 0.0106 0.919

BDI does not Granger Cause SSE 59 2.880 0.095

(26)

Dependent variable rBDI t rBDIt r_t−1SSE(α1) -0.119 (-0.566) rSSE t−2 (α2) -0.009 (-0.054) rSSE t−3 (α3) -0.363 (-1.094) rmscichina t−1 (α1) 0.137 (1.068) rmscichina t−2 (α2) 0.182 (1.163) rmscichina t−3 (α3) 0.111 (0.657) R-squared 0.039 0.042 Adj R-squared -0.015 -0.013 Durbin-Watson stat 1.450 1.400 Wald F-statistic 0.521 1.098

Table 5.3: Regression results with BDI as the dependent variable (2000-2004)

5.2.2 Causality Relationships for Sub-Period 2005-2017

In this section, we investigate the causality relationship between BDI and MSCI China and SSE in the period 2005-2017. As shown in Table 5.4, in the period after 2005, BDI no longer granger causes the MSCI China and SSE indices. The same conclusion is achieved while testing causality relationships between BDI and some global indices such as MSCI World and MSCI Emerging market. In the period prior to 2005, the Dry Bulk market used to be in a demand-driven situation where dry bulk fleet size (supply) were typically lower than demand and therefore changes in demand were directly reflected in BDI rates. However, since 2004-2005 the new ship orders entered to the market leading to a generally over-supplied market. The financial crises in 2007-8 exacerbated the situation by a significant decline in demand. These all resulted in a market where the demand side was generally lower than the available fleet size. Therefore BDI price is not anymore a direct reflection of demand, instead it more reflects the supply-demand dynamics in a market where ships were willing to take cargos with lower rate just to cover operating costs. Therefore, what changes in BDI reflect is no longer a clear representation of global demand and future economic activities. Another highlight is that the causality relationship is reversed as MSCI China and SSE are granger causing the BDI which again stresses the important role of China’s

(27)

19 economy on freight rate markets. We also estimated a basic regression model with BDI as the dependent variable and the results overall confirms the granger causality results that in this period MSCI China and SSE could not predict the BDI developments as shown in Table 5.5.

Null Hypothesis Obs F-Statistic Prob.

MSCI CN does not Granger Cause BDI 149 3.614 0.008

BDI does not Granger Cause MSCI CN 149 0.418 0.796

SSE does not Granger Cause BDI 149 2.767 0.030

BDI does not Granger Cause SSE 149 1.644 0.167

Table 5.4: Granger Causality Between BDI and other indices (2005-2017)

Dependent variable rSSE t rM SCIchina rBDI t−1 (α1) -0.022 (-7.800) 0.000 (-0.014) r_t−2BDI(α2) 0.0423 (1.980) 0.017 (0.766) rBDI t−3 (α3) -0.047 (-1.837) 0.010 (0.347) R-squared 0.0282 0.005 Adj R-squared 0.008 -0.016 Durbin-Watson stat 1.764 1.699 Wald F-statistic 1.846 0.330

Table 5.5: Regression results with BDI as the independent variable (2005-2017)

5.3 Regression Results

Following the causality relationships found in the previous section, we will proceed with estimating the regression models for each sub-periods.

5.3.1 Results for Sub-Period 2000-2004

In this section, the basic regression model is estimated for the period from begin-ning of 2000 until the end of 2004. Based on the Granger causaility results, the model can be formulated as shown below with MSCI China and Shanghai Stock Exchange Composite Index (SSE) returns being the dependent variable (rt) and

BDI returns as the independent variable (rBDI). The results of basic regressions

are shown in the second column of Tables 5.8 and 5.9 . It is seen that the lagged BDI returns have positive effect on China’s stock market returns. In particular,

(28)

at the 10 % significant level, the one-month lagged BDI returns have a signif-icant relation with MSCI China return. Moreover, the two-month lagged BDI returns can significantly predict the SSE. Simillar findings have been achieved when we sub-indices of BDI, namely BCI and BPI, are used as the independent variables (Models D and E). Breusch-Godfrey and White hetorskedasticity tests have been conducted to check for residual autocorrelations and hetorskedasticity on the regression models. As shown in 5.6 and 5.7 the results suggest that the null hypothesis of no autocorrelation and homoscedasticity can neither be rejected for the regression models with SSE and MSCI China as the dependent variables.

rt= α0+ α1rt−1BDI + α2rBDIt−2 + α3rBDIt−3

Dependent Variable: rSSE rmscicn

F-statistic 0.617 0.788 Prob. F(3,53) 0.607 0.506 Obs*R-squared 2.025 2.560 Prob. Chi-Square(3) 0.567 0.465

Table 5.6: Breusch-Godfrey LM Test for Basic Model (2000-2004)

Dependent Variable: rSSE _rmscicn

F-statistic 0.409 1.134 Prob. F(3,56) 0.747 0.343 Obs*R-squared 1.285 3.436 Prob. Chi-Square(3) 0.733 0.329 Scaled explained SS 1.522 2.760 Prob. Chi-Square(3) 0.677 0.430 Table 5.7: White Test for Basic Model (2000-2004)

As the next step, we include some common predictors to the model in order to test the robustness of basic regression results. We use MSCI World, MSCI Emerging and Oil Date Brent as three predictors in our robustness analysis. The extended regression model can be formulated as shown in the equation below. Regression coefficients and t-values as well as other details of each model are presented in Tables 5.8 and 5.9. The results of alternative models (A, B and C) show that the lagged BDI returns still remain positive and significant predictors of both SSE and MSCI China in the presence of all the common predictors.

(29)

21

Dependent Var:rSSE _Basic

Model

Alternative models Sub − indice M odels

A B C D E rBDI_(α 1) 0.083 (1.629) 0.111 (2.153) 0.115 (1.953) 0.085 (1.766) rBDIα2) 0.119 (1.817) 0.131 (1.873) 0.110 (1.510) 0.131 (1.967) rDBI_(α 3) -0.008 (-0.136) 0.015 (0.228) 0.018 (0.287) -0.012 (-0.169) rmsciW orld t (β1) -0.206 (-1.502) rmsciW orld t (β2) -0.136 (-0.666) rmsciW orld t (β3) 0.163 (0.762) rmsciEM t (β1) -0.145 (-1.776) rmsciEM t (β2) -0.098 (-0.667) rmsciEM t (β3) 0.166 (1.361) roil t (β1) 0.017 (0.272) r_toil(β2) 0.060 (1.402) roil t (β3) -0.038 (-0.510) rBCI_(α 1) 0.098 (2.237) rBCI(α2) 0.043 (0.785) rBCI_(α 3) 0.021 (0.441) rBP I_(α 1) 0.018 (0.401) rBP I(α2) 0.186 (4.004) rBP I_(α 3) -0.022 (-0.409) R-squared 0.092 0.168 0.185 0.146 0.080 0.142 Adj R-squared 0.043 0.068 0.087 0.043 0.031 0.096 Durbin-Watson stat 2.061 2.175 2.190 2.065 2.015 2.030 Wald F-statistic 2.998 3.280 3.406 2.734 2.448 8.569

(30)

Dependent Var:rmscicn _Basic

Model

Alternative M odels Sub − indice M odels

A B C D E rBDI_(α 1) 0.224 (1.969) 0.209 (1.797) 0.247 (1.995) 0.235 (2.007) rBDI(α2) 0.116 (1.476) 0.082 (1.031) 0.088 (1.094) 0.098 (1.337) rBDI_(α 3) -0.017 (-0.224) -0.028 (-0.339) 0.007 (0.083) 0.014 (0.156) rmsciW orld t (β1) 0.263 (1.628) rmsciW orld t (β2) -0.036 (-0.132) rmsciW orld t (β3) 0.101 (0.407) rmsciEM t (β1) 0.091 (0.566) rmsciEM t (β2) -0.251 (-1.147) rmsciEM t (β3) 0.109 (0.643) roil t (β1) -0.123 (-1.156) r_toil(β2) -0.022 (-0.274) roil t (β3) -0.105 (-0.875) rBCI_(α 1) 0.179 (2.169) rBCI(α2) 0.120 (2.126) rBCI_(α 3) -0.022 (-0.375) rBP I_(α 1) 0.226 (2.203) rBP I(α2) 0.037 (0.483) rBP I_(α 3) 0.020 (0.264) R-squared 0.135 0.155 0.171 0.171 0.154 0.117 Adj R-squared 0.088 0.053 0.072 0.077 0.109 0.070 Durbin-Watson stat 1.827 2.017 1.956 1.834 1.840 1.783 Wald F-statistic 5.400 3.992 2.748 2.701 6.729 2.984

(31)

23

5.3.2 Results for Sub-Period 2005-2017

In this section, the basic regression is estimated for the period beginning from 2005 until the end of 2017. As discussed in the previous section, the causality relationship is reversed in this period, meaning that other predictors, namely MSCI China and SEE, are Granger causing the BDI returns. Hence, the BDI return is used as the dependent variable (rBDI

t ) and MSCI China and SSE returns

as the independent variables (rchina

t−1 ) as shown in the regression models below. As

shown in Tables 5.10 and 5.11, the lagged MSCI China and SSE returns both have positive effect on BDI returns. At the 10% significance level, the one-month and two-one-month lagged MSCI China returns have a significant effect on BDI return. Simillarly, the one-month lagged SSE returns have a positive effect on BDI returns. Simillar findings have been achieved when we sub-indices of BDI, namely BCI and BPI, are used as the independent variables (Models D and E). Breusch-Godfrey and White hetorskedasticity tests have been conducted to check for residual autocorrelations and hetorskedasticity on the regression models. As shown in Tables 5.12 and 5.13 the results suggest that the null hypothesis of no autocorrelation and homoscedasticity can neither be rejected for the regression model with SSE as the dependent variables. For the regression model with MSCI China as the dependent variable, the null hypothesis of no autocorrelation cannot be rejected but the White test rejects the null hypothesis of homoscedasticity. However, this would not affect our conclusions as we are using heteroskedasticity and autocorrelation consistent (HAC) standard errors in the regression models.

(32)

Basic Model

A B C D E rmscicn_(α 1) 0.743 (1.986) 0.915 (1.963) 0.910 (1.454) 0.541 (1.388) 1.247 (2.932) 0.513 (1.318) rmscicn_α 2) 0.446 (1.840) 0.161 (0.528) -0.839 (-1.595) 0.344 (1.209) 0.048 (0.135) 0.786 (2.934) rmscicn_(α 3) -0.448 (-1.509) 0.017 (0.038) 0.265 (0.415) -0.468 (-1.530) -0.465 (-1.047) -0.456 (-1.466) rmsciW orld t (β1) -0.336 (-0.717) rmsciW orld t (β2) -0.620 (1.220) rmsciW orld t (β3) -1.208 (-1.432) rmsciEM t (β1) -0.107 (-0.184) rmsciEM t (β2) 1.684 (2.735) rmsciEM t (β3) -1.106 (-1.336) roil t (β1) 0.477 (2.468) r_toil(β2) -0.010 (-0.040) roil t (β3) -0.115 (-0.604) R-squared 0.085 0.119 0.142 0.115 0.068 0.079 Adj R-squared 0.066 0.082 0.106 0.078 0.049 0.060 Durbin-Watson stat 1.783 1.761 1.780 1.857 2.096 1.876 Wald F-statistic 3.043 2.006 2.451 2.467 3.110 4.379

Table 5.10: Robustness Analysis for mscicn (2005-2017)

We extended the regression model by adding some common predictors of the stock market such as MSCI World, MSCI Emerging Market and Oil Dated Brent. The extended model can be formulated as shown below with r_tchina being returns of MSCI China or SSE. Regression coefficients and t-values as well other details are presented in Table 5.10 and 5.11. The results of alternative models (A, B and C) in Table 5.10 show that MSCI China remain a valid predictor in the presence of MSCI World, but it did not remain as a significant predictor when having MSCI EM or Oil in the model. However SSE remains a robust predictor in the presence of all indicators, namely MSCI World, MSCI EM, and Oil as shown in Table 5.11.

(33)

25

r_tBDI = α0+ α1rt−1china+ α2rchinat−2 + α3rchinat−3 + β1rXt−1+ β2rt−2X. + β3rt−3X

Basic Model

A B C D E rSSE_(α 1) 0.602 (2.261) 0.509 (1.993) 0.437 (1.766) 0.476 (1,917) 1.021 (2.980) 0.462 (1.593) rSSE_α 2) 0.356 (1.408) 0.255 (0.952) 0.094 (0.348) 0.350 (1.426) 0.099 (0.289) 0.583 (2.332) rSSE(α3) -0.101 (-0.435) 0.101 (0.426) 0.140 (0.587) -0.149 (-0.632) -0.368 (-1.044) 0.020 (0.080) rmsciW orld t (β1) 0.258 (0.560) rmsciW orld t (β2) 0.509 (1.254) rmsciW orld t (β3) -1.215 (-2.043) rmsciEM t (β1) 0.443 (1.175) rmsciEM t (β2) 0.731 (2.430) rmsciEM t (β3) -0.876 (-2.121) r_toil(β1) 0.557 (2.812) roil t (β2) -0.055 (-0.232) r_toil(β3) -0.179 (-1.009) R-squared 0.062 0.116 0.133 0.109 0.060 0.060 Adj R-squared 0.042 0.078 0.096 0.071 0.040 0.041 Durbin-Watson stat 1.796 1.789 1.836 1.894 2.022 1.937 Wald F-statistic 2.033 2.384 2.761 2.360 3.479 1.997

Table 5.11: Robustness Analysis for SSE (2005-2017)

Dependent Variable: rSSE rmscicn

F-statistic 1.433 1.047 Prob. F(3,142) 0.236 0.374 Obs*R-squared 4.377 3.225 Prob. Chi-Square(3) 0.224 0.358

(34)

Dependent Variable: rSSE rmscicn F-statistic 0.549 3.066 Prob. F(3,145) 0.650 0.030 Obs*R-squared 1.674 8.888 Prob. Chi-Square(3) 0.643 0.031 Scaled explained SS 3.706 17.185 Prob. Chi-Square(3) 0.295 0.001

(35)

Chapter 6 Conclusion and Recommendations

This study examined the relationship between the Baltic Dry Index (BDI) and China’s stock market returns which has been represented by the MSCI China Index and the Shanghai Stock Exchange Composite index (SSE). We investigated the predictive power of BDI by conducting Granger causality tests followed by predictive regression models. The main findings of this research suggest that the relationship between BDI and China stock market indices has changed as the fundamentals of the dry bulk market evolved over the years. In particular, BDI has been able to predict China’s Stock market over the period of 2000 till 2005 according to Granger causality test and results of basic regression models. This has been further confirmed by showing that BDI still remains a significant predictor when we added other alternative market predictors such as MSCI World, MSCI Emerging market and the Oil price into the regression models. Therefore, we can conclude that BDI has been a significant and robust predictor of China’s market over the period of 2000-2005. However, the findings are quite different for the period after 2005 . Our findings suggest that BDI can no longer predict China’s stock market as the results of Granger causality and the regression models show. In fact, the causality relation has been reveresed and SSE and MSCI China are Granger causing the BDI. The outcome of basic regression models with BDI as the dependent variable and SSE and MSCI China as independent variables confirms this relationship. The results have been shown to be robust as they remain valid even when we add other market indicators such as MSCI World, MSCI Emerging market and Brent Oil price. This finding can be explained by the fact that since 2005 the new ships orders gradually entered to the fleet causing an excess capacity in the market. Moreover, the financial crisis in 2007-2008 exacerbate the situation by a significant decline on the demand. All of these resulted in a market condition where demand is generally lower than the available fleet capacity. In such circumstances, BDI price is not anymore a pure reflection of

(36)

demand and instead it more reflects the supply-demand dynamics in the market. Therefore, a decline in freight rate in such situation does not neccesary reflect a lower demand or trade volume which could potentially predict a slower economic growth in near future. Instead, it reflects a market condition where global trade is not growing as fast as the supply of new fleets and therefore ships were willing to take cargoes at lower rate just to cover operating costs and reduce their unutilised capacity.

This study can be further extended in multiple ways. We recommend future researches to incorporate other indicators such as trade volume, import and export data of dry bulk commodities and investigate whether they can be used as a better predictors than the freight rates in recent years. Since trade volume is more linked to the market activities than the freight rates, it may lead to a better prediction especially in a current market where changes in BDI does not necessarily move in tandem with changes in demand side (trade volume). There has been little empirical studies on the supply side of the shipping market. Therefore, another recommendation is to study the impact of fleet supply growth and provide projections about the freigh rates based on the global demand growth scenarios.

(37)

References

[1] Amir H. Alizadeh and Gulnur Muradoglu. Stock market efficiency and in-ternational shipping-market information. Journal of Inin-ternational Financial

Markets, Institutions & Money, 2014.

[2] N. Apergis and J.E. Payne. New evidence on the information and predictive content of the baltic dry index. International Journal of Financial Studies, 2013.

[3] G. Bakshi, G. Panayotov, and G. Skoulakis. The baltic dry index as a predictor of global stock returns, commodity returns, and global economic activity. SSRN, 2011.

[4] Ghiorghe I. Batrinca and Gianina S. Cojanu. The determining factors of the dry bulk market freight rates. In 2014 International Conference on

Economics, Management and Development, 2014.

[5] Chang and Lin. Causality analysis between baltic dry index and stock mar-kets in brics. Master’s thesis, 2009.

[6] H. Geman and W.O. Smith. Shipping markets and freight rates: An analysis of the baltic dry index. The journal of alternative investments, 2012.

[7] C.W.J. Granger. Investigating causal relations by econometric models and cross-spectral methods. Econometrica, 1969.

[8] Bingqian Luo. Do the sub-indices of the bdi have better predictability for stock market returns than the bdi? Master’s thesis, Tilburg University, 2015.

[9] Payton Odom. Shipping indexes signal global economic trends. Technical report, Globalization and Monetary Policy Institute 2010 Annual Report, 2010.

[10] J.G.M Oome. The baltic dry index: A predictor of stock market returns? Master’s thesis, Tilburg University, 2012.

(38)

[11] Chien-Wen Shen and Chih-Hsin Lo. Causality analysis of dry bulk cargo freight and bric economic growth. Technical report, National Science Council of the Republic of China, Taiwan, 2010.

[12] Martin Stopford. Maritime Economics. Routledge, 2009.

[13] UNCTAD. Review of maritime transport 2016. Technical report, UNCTAD, 2016.

[14] Dino Roberto Zuccollo. The baltic dry index: A leading economic: Indicator and its use in a south african context. In Proceedings of the 2013 SAAA

(39)

(40)

Relationship between Baltic Dry Index and China economy