Baltic Dry Index : the effects of crude oil and iron ore

(1)

1

GRADUATE THESIS INTERNATIONAL FINANCE

Baltic Dry Index: The Effects of Crude Oil and Iron Ore

Hajolt Laming

Student Number: 10880771

Thesis Supervisor: Dr. S.R. Stefan Arping

Master in International Finance

University of Amsterdam, Amsterdam Business School

(2)

2

Acknowledgements

First of all I would like to express my sincere gratitude to my thesis supervisor, Dr. S.R. Stefan Arping, for his valuable guidance and advice. I would also like to thank my parents for both their financial support and being a safe haven for anything at any moment I can imagine over the past two years. And last but not least, I want to thank my dearest friend, buddy and partner, Manon, who joined me on this MIF journey, for a 1001 reasons. She is the best example of perseverance I know, her inspiration, patience, drive and attitude are remarkable, but most of all, she helped me to carry on, in my worst moments.

(3)

3

Abstract

In my thesis, I use a time series empirical analysis to study the effect of Crude Oil and Iron Ore prices on the Baltic Dry Index. This paper examines the dynamic behavior of the Baltic Dry Index and its returns over the past five years. The analysis will focus on both these topics in a univariate time series and in relation to explanatory variables Crude Oil and Iron Ore. Are they all stand-alone macro-economic indicators or is there a correlation between them? And who is leading who? After running regressions it appears that both the Crude Oil and Iron Ore prices have impact on the Baltic Dry Index. This seems explainable as they are both major demand and supply drivers of the BDI. Fuel costs make up 60 to 70% of all costs for operating vessels and Iron Ore is the biggest dry bulk commodity that is globally being transported by sea. Decision makers have a need for tools showing them where we are in the cycle, in which direction the next imbalance will go to and in which direction the BDI sub-markets move. The result of my thesis will give macro-economic directions only and will not analyze all sub-markets separately, due to missing detail price data on the BDI supply site.

(4)

4

1. Introduction

Today’s shipping industry needs to anticipate on stagnating growth in the demand of seaborne transport. Another imbalance in supply and demand has come after a period of growing Chinese GDP figures with high demand, high freight rates and a booming

expansion of ship building capacity. Now we face an overcapacity imbalance and reaching a new balance will have substantial consequences on the operating conditions of shipping companies and transport port hub terminals. Vessels will need to be

scrapped, delivery of new building orders will have to be delayed or even cancelled and long term charter tariffs between owners and brokers will need to be renegotiated. In order to be able to support their business decisions both shipping industry investors and operators are in need of information on how to interpret the volatile nature of freight rates and their effect on the industry.

The fundamentals of global economic growth with traditional drivers such as asset inflation, debt-based consumption of middle-class people and export-oriented strategies are changing; renewable energy, environment and sustainability are important

investment drivers nowadays.

A flattening global economic growth (GDP China) and the collapse of the oil price are the major reasons for the lower growth in demand of transport for iron ore, coal and grains. These dry bulk raw materials are globally transported by dry bulk carriers. The price of this transport, the balance between supply and demand, is established at the Baltic Exchange and is known as the Baltic Dry Index (BDI). Because dry bulk transport is used for transportation of raw materials, the basis for future production, the index is often seen as an indicator for global economic activity expectations.

Global freight rates, established by the Baltic Dry Index Exchange (BDI), are considered to be an indicator for global economic growth. Other GDP growth indicators are the prices of Crude Oil, Iron Ore, Coal, Grains and Gold.

(6)

6

International trade values and volumes have exploded since the 1990s. The reasons for this substantial growth are as follows: Integration; China entering the WTO and the creation of the European internal market; the fragmentation of production systems with global comparative advantages; the increase of international transportation systems’ capacity, connectivity and reliability, and the improvement of transactional efficiency. These developments have also caused that in the last two decades the role of the Financial Industry in shipping has shifted from a traditionally funding role of mitigating the risks related to shipping that has led to the creation of insurance companies (e.g. Lloyd’s), to seeing opportunities for financial leverage. This shift is explained by De Monie, Rodrique and Notteboom (2009).

The Baltic Dry Index indicates the average costs for the transport of raw materials or dry bulk commodities, such as iron ore, coal and grains. The Index is a composite, a weighted average, of three other sub-indices of dry bulk carriers in terms of different size, volume (dwt) or routes: Capesize, Panamax and Supramax. The dynamics of the Dry Baltic sub-markets are researched by Batrinca and Cojanu (2013), and Geman and Smith (2012). The Index is established by the London Baltic Exchange (www.balticexchange.com), founded in 1744. The index is based on daily assessments of supply and demand of bulk carriers’ capacity, from a panel of shipbrokers settling the prices. Indices and

assessments are used as a settlement tool for freight derivative trades, for benchmarking physical contracts and as a general indicator of the bulk market's performance. The BDI is the world's only independent source of maritime market information for the trading and settlement of physical and derivative shipping contracts. The Baltic Exchange has over 600 member firms that encompass the majority of world shipping interests and commits to a code of business conduct. The Baltic Exchange members are responsible for a large proportion of all dry cargo and tanker fixtures, as well as the sale and purchase of

merchant vessels. Headquartered in London, with regional offices in Singapore, Shanghai and Athens, the Baltic Exchange is owned by its shareholders, most of whom are

members. It is governed by a board of directors elected by shareholders and members. Changes in the Index are driven by supply and demand in seaborne transport capacity. Typical demand drivers are: the world economy, seaborne trade or cargo volume, the

(7)

7

average haul or distance cargo is being transported, transport costs of operating shipping and random shocks such as political instability and war. Supply drivers are: ship building capacity, scrap prices, the dwt in service, inactive tonnage, the average age of vessels in service and random shocks such as changes in Safety and Environmental legislation. The Index’ movements strongly influence ship owners’ and operators’ behavior and, by doing so, they influence future supply directly. It always returns to a certain balance, though, in a process called dynamic behavior, as described by Stopford (2009) and in figure 3. There are many different decision makers, all with different interests and questions that need to be answered, making forecasting extra sensitive. However, there are three principles that can be used to find out whether or not a forecast is useful: Relevance, Rationale and Research. Figure 1 depicts a view of the global shipping industry decision makers, all having their own forecasting relationships, variables and view needed at their own moment in time. The forecast periods differ from 3 – 12 months for ship operators and 10 – 20 years for ship owners and port developers.

Figure 1 Global shipping industry decision makers

Shipping companies Cargo owners Shipbuilders Bankers Governments Port authorities Machinery manufacturers International organisations

(8)

8

In this thesis I will research the behavior of the BDI. Can the past be modelled in such a way that it is also possible to predict a forecast on the Index and how to interpret volatility? First I will do a regression on the Exchange Index returns, known as regressing to the first lag. Then I will do a regression based on the explanatory variables Crude Oil and Iron Ore prices. This is to see how they behave compared to the Baltic Dry Index. When regressing the BDI I will look at stationarity and reasons for non-stationarity such as a trend, drift or seasonal impact. What are the dynamics between the BDI and Crude Oil and Iron Ore prices?

Are Crude Oil as major cost for Dry Bulk shipping operations, and Iron Ore as major commodity transported by Dry Bulk vessels, predictors for the Baltic Dry Index?

Can investors, operators and other third parties involved use this information in making their business decisions?

Figure 3 Macroeconomic shipping industry model of Stopford

World Economy / GDP growth Commodity trade

Ship demand Supply / Demand

Ship prices _{Freight prices}

New build

market Scrap market Vessel fleet

Investor Sentiment

(9)

9

2. Literature Review

Modelling and forecasting the BDI has been done extensively in the past and will be done in the future. The financial stakes in the shipping industry are high, the balance in market dynamics changes all the time and the forecasts that are needed differ from 3 – 12 months for ship operators to 10 – 20 years for ship owners and port developers. Although it is accepted that there is no holy grail, decision makers are looking for guidance.

In my literature review I will first look how the BDI drivers are modelled and forecasted in favor of investors’ and operators’ decision making. I will look if there is a line in assumptions, using econometrics and its conclusions, and incorporate this knowledge in my research. On the one hand there are many articles on the supply side concerning shipbuilding capacity, the buying and selling of scrap ships and strategic port

development, while on the other hand the articles on the demand side concerning economic growth, cargo volume and transport distance. Each of these factors has a supply or demand role within the dynamics of BDI drivers. In other words, if the BDI can be modelled and forecasted, decision makers on the dynamic drivers of the model can anticipate their business decisions.

Secondly, I will look at articles written about Crude Oil and Iron Ore price modelling and forecasting. As fellow global trade indicators, I will look if they behave the same as the BDI and, if they are forecasted together, how they relate. Are these two variables also indicators for forecasting the BDI and BDI drivers? If so, related parties can also use this information to make their business decisions.

As econometrics methodology I will use both basic econometric literature and applied studies.

(10)

10 2.1 Theories on BDI drivers

The BDI balance is driven by demand and supply. The behavior of the index model may be called a dynamic model. The market is called efficient since there are many players and information is generally available.

Taylor (1976) demonstrates the complexity of the supply side of the market. In his model he defines sub-markets: Shipbuilding Orders and Deliveries, the Scrapping Sector, Lay-up capacity, Trade Change influencers and Freight Rates. However, as he assumes demand to be constant in time and he admits its unsophisticated representation of reality, this is the level of accuracy I find in a lot of articles. I think this is a good macro-economic model with evidence for a certain kind of structural model on the supply side. For example, if the Freight index goes down, new vessel deliveries go down as well and inactive tonnage goes up. This macro-economic process already affects many industries with many

investors and operators.

Chen (2011) analyzes the behavior of three types of sub-markets. The first type is distinguished in terms of economic variables, including the freight market, the new building market and the second-hand ship market; the second is distinguished in terms of different ship sizes, including Capesize, Panamax and Handymax vessels, and the last is distinguished as the physical market and the forward market. Most important in this work is that many hypotheses include technical ship innovations that are hard to forecast. Chen tests stationarity and unit root on the time series data and, since many economic events are non-stationary, data is transformed by logarithms. In this way, prices or rates turn to returns, which can have value to market participants. Moreover, logarithms make analysis easier. I will also apply this reasoning.

Batrinca and Cojanu (2013) show the dynamic working of the BDI sub-markets in size, cargo type and routes sailed. The BDI is calculated based on a weighted average of the three sub-markets. Empirical data analysis concludes that as of the year 2000 the BDI is more than average volatile because of worldwide economic booms and recessions. Batrinca and Cojanu conclude that the indices produced by the BDI provide useful and reliable information and improve price transparency in the market.

(11)

11

Kalouptsidi (2012) is mostly concerned with the influence of the entry and exit of vessels and the second hand market. This theory shows the interaction between uncertainty and adjustment costs; the time to build new vessels, in this case. There is a period of time between the moment of ordering a vessel and the delivery, while the demand for dwt capacity may change during this period of time. For this reason Kalouptsidi constructed a dynamic model for ship entry and exit based on observations that resale prices reflect value function. This study shows that the shorter the time to build a vessel the higher investment volatility is. Furthermore, it shows that both freight rates and inventory costs by shipping delivery lags can significantly affect trade flows.

The theory of new build order intake, delivery or delay and pricing strategy for ship builders during different moments in a time cycle is explained by Raiswell (1976). In this work he shows how pricing strategy affects the dynamic behavior and profitability of product facilities, a supply driver of BDI. Pricing Strategy is recognized as an important management tool when dealing with market imbalances and shipping life cycles. During a period of boom a shipyard should take sufficient orders in order to ensure profitable work until the next boom. Orders taken in competitive periods in between booms often need to be taken at lower prices, sometimes even without allowance for overheads or profit. Even managing production capacity utilization on delivery or delay during booms is an important management tool. The key questions here are: Am I exacerbating market volatility and impacting new imbalances? Am I maximizing my future profit?

Figure 2 Basic price quote and order intake cycle

Forecast ideal order backlog Prices quoted Order scoring rate Order backlog

(12)

12

Returns on owning Dry Bulk Cargo ships are modeled by Greenwood and Hanson (2013). Ship earnings exhibit a high degree of mean reversion, driven by industry parties’

competitive investment responses to increases in demand. Greenwood and Hanson show that high current ship earnings are associated with high second-hand ship prices and high industry investment and forecast low future returns. They conclude that pro-cyclical investor expectations, driven by a form of competition neglect, can dramatically amplify economic fluctuations in competitive industries.

All the articles described above concern modelling and forecasting the BDI, from BDI time series to dynamic models. Most of them try to explain a specific market segment, a BDI supply or demand driver. It is clear that demand is more macro-economically driven and supply is more micro-economically driven. All articles come to the conclusion that BDI market segments behavior can be explained and forecasted within a selected number of assumptions.

(13)

13 2.2 Theoretical predictors of the BDI

Research has also already been done on other major indicators of the BDI. It is generally accepted that Crude Oil, Iron Ore and the GDP all relate to global economic growth expectations.

Lam (2013) shows extensive econometric research on the behavior of Crude Oil, a leading commodity in the world. In this work Lam describes two feasible approaches towards building models. The first is a time series approach, using the Box-Jenkins methodology, aiming to build ARIMA-GARCH/APARCH models. The second is to seek a number of meaningful explanatory variables to build linear regression models, such as consumption, production, stock and refinery utilization, but also financial indices like NYMEX Oil contracts and the S&P 500. He concludes that the univariate approach outperforms the linear regression approach; the chosen variables are proved not to be relevant enough to accurately influence the oil prices.

Warell (2013) analyzes Iron Ore price behavior and predictability. She studies two specific changes in trading in the commodity: firstly, the introduction of spot market pricing versus dominating producers and consumers negotiating a benchmark price; secondly, whether or not a change in price regime influences the volatility. Remarkably, Warell holds that traders consider the spot market pricing less transparent than the old pricing system. Today’s market price is mainly driven by the development of the Asian market and especially by Chinese consumers who are held responsible for changing the pricing system. Once more, first the univariate Iron Ore prices behavior is researched and secondly a reduced regression model is designed, based on explanatory variables

including GDP growth in China and Freight rates. Furthermore, all have been tested for stationarity and the DF test for unit root. The tests are performed on the log of the time series. Warell concludes that transport costs are the main driver behind the change in the pricing system. The main driver of the Iron Ore price is the GDP growth in China. Chou, Yang and Chang (2011) perform an empirical research on how the Crude Oil and Steel price indices relate. In their study they apply a multivariate time series model to forecast crude oil and global steel prices and they try to determine the appropriate

(14)

14

corresponding relationship between the variables to predict future trends. In their Methodology they describe that since macro-economic variables such as oil and steel are mainly non-stationary series, they need to be unit root tested before being called

stationary and before they can be examined on their causal relationship; otherwise, they would test on co-integration. If the two are non-stationary the researchers convert the series to stationary by the finite difference method. Only then do they perform VARMA modelling. They conclude that there is a unidirectional relationship between the crude oil price and the global steel price index. This means that the price of crude oil is only impacted by its own volatility and that the global steel price index is impacted by both its own price movements and the volatility of crude oil prices. When the crude oil price increases the global steel index price follows.

Just like many other researchers, Li, Wang, Ren and Wu (2011) recognize that China is the largest iron ore consumer in the world, but that despite this there is a continuous decline in profits from the iron and steel industry. They define a theoretical

econometrics-driven equation between Iron Ore and Crude Oil firstly, and secondly between Iron Ore and the BDI, by applying AR auto regression, LS and estimation command. They recognize the drawback that their model is demand-driven and that there will be differences between the calculated equations and real price developments. Like Warell they also find that the (spot) market pricing of iron ore has not become more transparent. In the end they promote purchase cooperation.

Crude Oil modelling and forecasting theories have been tested by Behmiri and Manso (2013). They categorize modelling and forecasting in quantitative vs qualitative and within the quantitative models they distinguish Econometric, Financial and Structural models. They support the idea that time series models can predict future oil prices based on historical data because, either, data shows a systematic pattern, or there are a large number of explanatory variables with interactions having a complex structural model, or forecasting the dependent variable requires prediction of the explanatory variables. All three conditions appear to apply to oil prices. In their structural models Behmiri and Manso use oil consumption, production, OPEC behavior and inventory, but also some non-oil variables like interest rates, exchange rates and other commodity prices. They

(15)

15

conclude that in practice the time series econometrics techniques are by far the most used for modelling and forecasting crude oil applying ARIMA and ARCH/GARCH models. Frey, Manera, Markandya and Scarpa (2009) in their turn describe the relevance of the determinants of past, present and future crude oil prices in the global economy and the ever-existent urge of academic researchers and energy experts the build forecasting models. Their study discusses existing econometric literature about oil price modelling and forecasting. They show all the econometric steps taken for Oil forecasting, from the random walk to AR models.

400 800 1,200 1,600 2,000 2,400 2,800

IV I II III IV I II III IV I II III IV I II III IV I II III IV I II

2010 2011 2012 2013 2014 2015 2016

(16)

16 20 40 60 80 100 120 140 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

Crude Oil (petroleum); Dated Brent Price (US Dollars per Barrel)

The crude oil price graph shows, from left to right: the collapse of the oil price because of the financial crisis turning into an economic crisis in 2008; the return of the price level at $100 + in the years 2011 to 2014 because of almost unlimited Chinese GDP growth figures; another collapse in 2014 because of the USA shale gas boom and slowing Chinese GDP figures; Iran starting to supply oil again.

20 40 60 80 100 120 140 160 180 200 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016

(17)

17

In the Iron Ore graph you can clearly recognize the macro-economic Chinese growth figures after the financial crisis and following the new built vessel production catching up and creating an overcapacity in supply. This is a typical shipbuilding lifecycle.

Below there are the three growth figures shown together. Most remarkable is the volatility in BDI returns, but also the identical behavior of Crude Oil and Iron Ore returns.

-1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6

IV I II III IV I II III IV I II III IV I II III IV I II III IV I II

2010 2011 2012 2013 2014 2015 2016

BDI growth Oil growth iron ore growth

(18)

18

3. Methodology

3.1 Introduction

In my time series based empirical analysis on the Baltic Dry Index there are two main goals:

One is to identify if the historical nature and behavior of my dependent variable, the BDI, can be used to forecast future rates based only on its own past values. Where will the BDI go tomorrow, knowing it went down last month but up two months ago?

The other one concerns estimating the dynamic causal effects for economic forecasting of future values of the dependent BDI variable in combination with independent

variables Crude Oil and Iron Ore. Where will the BDI go tomorrow, knowing crude oil prices went up last month and the month before?

I will look at the decision how many past periods or lags to include in auto regression (AR) and auto-regressive distributed (ARD) lag models.

The assumption that the future will be like the past in time series will be tested on stationarity and non-stationarity, in which case time series have trends or breaks, and models can be adjusted once detected.

3.2 Regression models with a dependent variable and past value regressors

The most basic question is: how does the BDI behave in time? This question has the BDI as dependent variable and Time as independent variable and studies its own lagged values in time. The equation for AR(1): 𝑦𝑡 = 𝛽0 + 𝛽1𝑦𝑡−1 + 𝑢𝑡. In practice 𝛽0 and 𝛽1 are

unknown, so the forecast will be based on estimates. OLS estimators, which are constructed using historical data, can be used for estimating the coefficients.

(19)

19

The impact of a prior period on today’s value can be calculated and is called auto regression. An overview of several periods of auto regression is called a correlogram. Here you can see how many auto regression coefficients can be called significant.

A special AR model is the ARMA model, the Auto-regressive Moving Average ARMA (p,q) is defined as follows:

𝑌_𝑡 = 𝜙0 + ∑𝑝𝑖=1𝜙𝑖𝑌𝑡−𝑖 + ∑𝑞𝑗=1𝜃𝑗𝜖𝑡−𝑗 + 𝜖𝑡

1. p refers to the number of auto-regressive terms 2. q refers to the number of lagged error terms

3. 𝜙 refers to the coefficients of the auto-regressive terms and the constant 4. 𝜃 refers to the coefficients of the moving average terms

The 𝜖𝑡 defined above is a white noise process, meaning that it is a sequence of variables

which have mean zero, variance 𝜎2, and zero correlation in time. They are also independent and identically distributed.

3.3 Time Series Regressions with multiple Predictors

Adding independent variables and lags gives an extension to an auto regression model gives and auto-regressive distribution model (ADL). Multiple predictors and their lags lead to double subscripting of the regression coefficients and regressors. The concepts of a model with two lags and two predictors included are:

𝑦_𝑡=𝛽0+𝛽1𝑦𝑡−1+𝛽2𝑦𝑡−2+..+𝛽𝑝𝑦𝑡−𝑝+𝛿1𝑥𝑡−1+𝛿2𝑥𝑡−2+..+𝛿𝑝𝑥1−𝑝+𝜃1𝑧𝑡−1𝜃2𝑧𝑡−2+..+𝜃𝑝𝑧1−𝑝+𝜇𝑡

With E(𝜇𝑡|𝑦𝑡−1, 𝑦𝑡−2, … , 𝑥𝑡−1, 𝑥𝑡−2, … , 𝑧𝑡−1, 𝑧𝑡−2) = 0

The random variable has stationary distribution and become independent when 𝜌 becomes large

(20)

20

As many economic time series are non-stationary the model need tested if there are trends or breaks when applied, the assumption might not hold.

To see if model and assumptions hold we can do the Granger Causality Test. For

forecasting in these time series, the F-statistics test is useful. It tests if lags of regressors are predictive. In the Descriptive Statistics chapter both the model results and the Granger test outcome is included.

Conducting a t-statistic test, a random function of the data with a known distribution under 𝐻0, and a decision rule. Usually the rule is to reject if the test statistic, or its

absolute value exceeds a critical value.

3.4 Stationarity - Non Stationarity

A time series 𝑦𝑡 is stationary if its probability distribution does not change over Time.

That is, if the joint distribution of (𝑦𝑠+1, 𝑦𝑠+2, … , 𝑦𝑠+𝑡) does not depend on s regardless

of the value of T; otherwise, 𝑦𝑡 is said to be non-stationary. A pair of time series, 𝑦𝑡 and

𝑥_𝑡, are said to be jointly stationary if the joint distribution of (𝑦𝑠+1, 𝑥𝑠+1, 𝑦𝑠+2, 𝑥𝑠+2, ….,

𝑦_𝑠+𝑇, 𝑥𝑠+𝑇) does not depend on s, regardless of the value of T. Stationarity requires the

future to be like the past, at least in a probabilistic sense. In case of non-stationarity a time series can strongly influence its behavior and properties; basic hypothesis tests, confidence intervals and forecasts can be unreliable.

I will test my time series data on two important types of non-stationarity: trend and breaks.

A trend is a persistent long-term movement of a variable in time. Especially in economic time series we consider a trend stochastic instead of deterministic, meaning that the variable moves randomly over time and is not linear and change unpredictable. The simplest version of such a model would be the random walk model, where the changes in the BDI are independent and identically distributed with a zero mean. The equation of this process can be written as follows: 𝑦𝑡+1 = 𝑦𝑡 + 𝜖𝑡+1. Here 𝜖𝑡+1 represents a white

noise term with a mean of zero, constant variance and zero auto-correlation. Changes in 𝑦𝑡 are independent from each other. Due to a trend, seasonality or changing variance,

(21)

21

the distribution of the variable is not constant and needs to be tested on stationarity. If a series has an obvious tendency the forecast should include an adjustment for this drift. It is called a random walk with drift and the equation looks like: 𝑦𝑡+1 = 𝛼 + 𝑦𝑡 + 𝜖𝑡+1. The

random walk model is a special case of the AR(1) model in which 𝛽1 = 1. So if Yt follows

an AR(1) with 𝛽1 = 1, then 𝑦𝑡 contains a stochastic trend and is non-stationary. If,

however, |𝛽1| < 1 and 𝑢𝑡 is stationary, then the joint distribution of 𝑦𝑡 and its lags does

not depend on t, so 𝑦𝑡 is stationary. To find trends we use the Dickey-Fuller test in the

AR(1) model, also known as OLS t-statistic testing.

A process is stationary if it is mean and the variance and auto-covariance of 𝛾𝑡 do not

change over time: 𝐸(𝑦𝑡) = 𝜇

𝑣𝑎𝑟(𝑦𝑡) = E[(𝑦𝑡 − 𝜇)2] = Υ0 = 𝜎2

Cov(𝑦𝑡,𝑦𝑡−𝑘) = E[(𝑦𝑡 - 𝜇)(𝑦𝑘−1 - 𝜇)] = Υ𝑘

I will apply the Dickey-Fuller test on the simple AR(1) process: 𝑦𝑡 = 𝜌𝑦𝑡−1 + 𝑥𝑡 𝛿 + 𝜖𝑡

Breaks arise when the population regression function changes due to macro-economic or political changes and can lead to misleading inference and forecasting. A problem is that the OLS regression estimates an average relationship over the full sample period and leads to poor forecasting. The Gregory Chow test can be used for testing for a break on a known date and the sub-Wald statistic if the date is unknown.

3.5 The order or number of lags

The number of lags, both in a time series auto-correlation model and a multiple variable model, can be determined. For both models the rule applies that too little lags means loss of forecast accuracy but too many increases uncertainty.

An information criterion that can be used is the Bayes Information Criterion (BIC), also known as the Schwarz Information Criterion, and is calculated as follows:

(22)

22

𝐵𝐼𝐶_𝑝 = ln [𝑆𝑆𝑅𝑝_𝑇 ] + (P+1)𝑙𝑛_𝑇𝑡 where 𝑆𝑆𝑅𝑝 is the sum of squared residuals of the estimated

𝐴𝑅_𝑝.

Another method is the Akaike Information Criterion: 𝐴𝐼𝐶𝑝 = ln [ 𝑆𝑆𝑅𝑝

𝑇 ] + (P+1) 2 𝑇

(23)

23

4. Data and Descriptive Statistics

All my statistical calculations, modelling and forecasting have been done using EViews 9.5 software.

The raw data of the Baltic Dry Index can be downloaded from the Baltic Exchange website. Crude Oil prices and Iron Ore raw data prices come from the commodity prices of the Index-mundi website. In paragraph 2.2 of the literature review I already showed and commented on the three different price graphs.

4.1 The Baltic Dry Index

Below the ARMA(1) model for the Baltic Dry Index is described. The model as such has no economic or financial value, but the Inverted AR and MA roots are indicators for (Non) stationarity. In this case both are > 1 in absolute terms, so the model can be called stationary.

Dependent Variable: BALTIC_DRY_INDEX_ACTUALS Method: ARMA Maxim um Likelihood (OPG - BHHH) Date: 08/30/16 Tim e: 13:37

Sam ple: 2010M07 2016M04 Included obs ervations : 70

Convergence achieved after 21 iterations

Coefficient covariance com puted us ing outer product of gradients

Variable Coefficient Std. Error t-Statis tic Prob.

C 1177.650 256.6069 4.589315 0.0000

AR(1) 0.846421 0.091288 9.271968 0.0000

MA(1) -0.070700 0.179581 -0.393692 0.6951

SIGMASQ 99856.01 12769.28 7.820018 0.0000

R-s quared 0.664880 Mean dependent var 1153.214

Adjus ted R-s quared 0.649647 S.D. dependent var 549.8083

S.E. of regres s ion 325.4349 Akaike info criterion 14.47996

Sum s quared res id 6989921. Schwarz criterion 14.60845

Log likelihood -502.7986 Hannan-Quinn criter. 14.53100

F-s tatis tic 43.64808 Durbin-Wats on s tat 1.944512

Prob(F-s tatis tic) 0.000000

Inverted AR Roots .85

(24)

24

The t-statistic with three degrees of freedom and 5% probability is 3,18. Our observed values for C and AR(1) are bigger than 3,18 so we say that the null hypothesis Η0: β = 0 is

rejected at the 5% level.

The F-distribution with three degrees of freedom and 5% probability is 8,57. Here we observe 43,65 also here we say that the null hypothesis Η0: β = 0 is rejected at the 5%

level.

Dependent Variable: Baltic Dry Index

Method: ARMA Maximum Likelihood (OPG - BHHH) Sample: 2010M07 2016M04

Included observations: 70

Coefficient covariance computed using outer product of gradients

Variable Coefficient Std. Error t-Statistic Prob.

C 1176.183 259.4285 4.533746 0.0000 AR(1) 0.147024 0.809187 0.181694 0.8564 AR(2) 0.602054 0.634696 0.948571 0.3464 MA(1) 0.662975 0.834812 0.794161 0.4300 MA(2) -0.114646 0.166929 -0.686792 0.4947 SIGMASQ 98671.95 13049.58 7.561315 0.0000

R-squared 0.668854 Mean dependent var 1153.214 Adjusted R-squared 0.642983 S.D. dependent var 549.8083 S.E. of regression 328.5155 Akaike info criterion 14.52591 Sum squared resid 6907037. Schwarz criterion 14.71864

F-statistic 25.85359

Prob(F-statistic) 0.000000

Inverted AR Roots .85 -.71 Inverted MA Roots .14 -.81

(25)

25

Dependent Variable: BALTIC_DRY_INDEX

Method: ARMA Maximum Likelihood (OPG - BHHH) Sample: 2010M07 2016M04

Included observations: 70

Convergence not achieved after 500 iterations

Coefficient covariance computed using outer product of gradients

C 1161.631 179.8698 6.458175 0.0000 AR(1) 0.621068 0.219183 2.833555 0.0062 AR(2) -0.478752 0.283013 -1.691626 0.0957 AR(3) 0.322385 0.210891 1.528680 0.1314 MA(1) 0.194139 0.328186 0.591551 0.5563 MA(2) 0.868771 1.184015 0.733750 0.4659 MA(3) 0.463101 0.539334 0.858654 0.3938 SIGMASQ 79115.61 89608.03 0.882908 0.3807

R-squared 0.734485 Mean dependent var 1153.214 Adjusted R-squared 0.704508 S.D. dependent var 549.8083 S.E. of regression 298.8713 Akaike info criterion 14.43136 Sum squared resid 5538093. Schwarz criterion 14.68833 F-statistic 24.50124 Durbin-Watson stat 2.012745 Prob(F-statistic) 0.000000

Inverted AR Roots .65 -.01-.70i -.01+.70i Inverted MA Roots .14+.99i .14-.99i -.47

(26)

26

Lags: The number of past periods impacting today’s forecast can be calculated via the auto-correlation coefficient:

1,96 X _𝑇_1/21 = 0,234. This means that the auto-correlation is significant for four periods, meaning that for forecasting AR(4) is the most efficient method.

The forecast based on the above for four periods (months) is given above.

Date: 08/30/16 Time: 10:11

Sample: 2010M07 2016M04

Included observations: 70

Autocorrelation

Partial Correlation

AC

PAC

Q-Stat Prob

1 0.797 0.797 46.400 0.000

2 0.618 -0.046 74.744 0.000

3 0.464 -0.042 90.912 0.000

4 0.255 -0.250 95.865 0.000

5 0.184 0.229 98.504 0.000

6 0.166 0.083 100.67 0.000

7 0.139 -0.006 102.21 0.000

8 0.142 -0.049 103.84 0.000

9 0.138 0.029 105.41 0.000

10 0.178 0.210 108.08 0.000

11 0.220 0.047 112.21 0.000

12 0.199 -0.153 115.64 0.000

-500 0 500 1,000 1,500 2,000 2,500 2016m1 2016m2 2016m3 2016m4 BALTIC_DRYF ± 2 S.E. Forecast: BALTIC_DRYF Actual: BALTIC_DRY_INDEX_ACTUALS Forecast sample: 2016M01 2016M04 Included observations: 4

Root Mean Squared Error 307.5844 Mean Absolute Error 303.7525 Mean Abs. Percent Error 77.23591 Theil Inequality Coefficient 0.254503 Bias Proportion 0.975240 Variance Proportion 0.002874 Covariance Proportion 0.021887 Theil U2 Coefficient 2.579886 Symmetric MAPE 54.44755

(27)

27

Null Hypothesis: BALTIC_DRY_INDEX_ACTUALS has a unit root Exogenous: Constant

Lag Length: 0 (Automatic - based on SIC, maxlag=10)

t-Statistic Prob.*

Augmented Dickey-Fuller test statistic -2.763494 0.0689

Test critical values: 1% level -3.528515

5% level -2.904198

10% level -2.589562

*MacKinnon (1996) one-sided p-values.

Augmented Dickey-Fuller Test Equation

Dependent Variable: D(BALTIC_DRY_INDEX_ACTUALS) Method: Least Squares

Date: 08/30/16 Time: 10:35

Sample (adjusted): 2010M08 2016M04 Included observations: 69 after adjustments

BALTIC_DRY_INDEX_ACTUALS(-1) -0.193513 0.070025 -2.763494 0.0074

C 205.7082 89.81144 2.290446 0.0251

R-squared 0.102321 Mean dependent var -18.81159

Adjusted R-squared 0.088922 S.D. dependent var 333.1340

S.E. of regression 317.9777 Akaike info criterion 14.39040

Sum squared resid 6774357. Schwarz criterion 14.45515

F-statistic 7.636899 Durbin-Watson stat 2.050042

(28)

28 4.2 Baltic Dry Index including independent variables Crude Oil and Iron

Ore

In the correlation matrix it is clear that there is not a clear colleration between the BDI and Oil and Iron Ore. But there is between Oil and Iron Ore both major commodities.

Baltic Dry Index Crude Oil Iron Ore

Mean 1153.214 91.88729 115.6740 Median 1044.000 107.1950 127.5650 Maximum 2713.000 124.9300 187.1800 Minimum 317.0000 30.80000 39.60000 Std. Dev. 549.8083 26.88847 42.34786 Skewness 0.982900 -0.914315 -0.282626 Kurtosis 3.461761 2.385521 1.914088 Jarque-Bera 11.89298 10.85429 4.371248 Probability 0.002615 0.004396 0.112408 Sum 80725.00 6432.110 8097.180 Sum Sq. Dev. 20857950 49886.30 123740.6 Observations 70 70 70

Baltic Dry Index Crude Oil Iron Ore

Baltic Dry Index 1.000000 0.335166 0.583220

Crude Oil 0.335166 1.000000 0.798252

(29)

29 Dependent Variable: BDI returns

Method: ARMA Maxim um Likelihood (OPG - BHHH) Sam ple: 2010M08 2016M03

Included obs ervations : 68

C -0.025601 0.032275 -0.793224 0.4306

OILRETURNS(1) -0.128329 0.384320 -0.333910 0.7396

IRONORERETURNS(1) -0.100437 0.432755 -0.232088 0.8172

AR(1) -0.086418 0.182014 -0.474785 0.6366

SIGMASQ 0.067028 0.010439 6.421099 0.0000

R-s quared 0.010520 Mean dependent var -0.022394

Adjus ted R-s quared -0.052304 S.D. dependent var 0.262206

Sum s quared res id 4.557938 Schwarz criterion 0.445608

Inverted AR Roots -.09

Dependent Variable: BALTIC_DRY_INDEX

Method: ARMA Maxim um Likelihood (OPG - BHHH) Date: 08/31/16 Tim e: 13:44

Sam ple: 2010M07 2016M02 Included obs ervations : 68

C 675.7313 570.4955 1.184464 0.2409 CRUDE_OIL(1) 2.497817 10.88392 0.229496 0.8193 CRUDE_OIL(2) -10.07562 11.37095 -0.886084 0.3791 IRON_ORE(1) 1.635709 5.144703 0.317940 0.7516 IRON_ORE(2) 8.659393 4.625882 1.871944 0.0661 AR(1) 0.713824 0.138750 5.144670 0.0000 AR(2) -0.027992 0.155805 -0.179664 0.8580 SIGMASQ 94691.29 15989.71 5.922013 0.0000

Sum s quared res id 6439008. Schwarz criterion 14.80237

(30)

30

The t-statistic with three degrees of freedom and 5% probability is 3,18. Our observed values are all but one smaller than 3,18 so we say that the null hypothesis Η_0: β = 0 cannot be rejected at the 5% level.

The F-distribution with three degrees of freedom and 5% probability is 8,57. Here we observe 18,16 so here we say that the null hypothesis Η_0: β = 0 is rejected at the 5% level.

Dependent Variable: BDI returns

Method: ARMA Maxim um Likelihood (OPG - BHHH) Sam ple: 2010M08 2016M01

Included obs ervations : 66

C -0.038290 0.034332 -1.115311 0.2696 OILRETURNS(1) -0.045101 0.465975 -0.096789 0.9232 OILRETURNS(2) -0.731759 0.498753 -1.467176 0.1480 OILRETURNS(3) -0.306473 0.461438 -0.664170 0.5094 IRONORERETURNS(1) -0.162715 0.635489 -0.256047 0.7989 IRONORERETURNS(2) 0.514954 0.561931 0.916401 0.3635 IRONORERETURNS(3) -0.239071 0.580012 -0.412183 0.6818 AR(1) -0.099208 0.198411 -0.500010 0.6191 AR(2) -0.079831 0.169219 -0.471762 0.6390 AR(3) 0.136205 0.131943 1.032298 0.3065 SIGMASQ 0.060277 0.009788 6.158082 0.0000

R-s quared 0.119495 Mean dependent var -0.027657 Adjus ted R-s quared -0.040597 S.D. dependent var 0.263649 S.E. of regres s ion 0.268947 Akaike info criterion 0.363696 Sum s quared res id 3.978297 Schwarz criterion 0.728638 F-s tatis tic 0.746412 Durbin-Wats on s tat 1.896953 Prob(F-s tatis tic) 0.678104

Inverted AR Roots .44 -.27-.49i -.27+.49i

Pairwise Granger Causality Tests Date: 08/31/16 Time: 12:25 Sample: 2010M07 2016M04 Lags: 2

Null Hypothesis: Obs F-Statistic Prob.

CRUDE_OIL does not Granger Cause BALTIC_DRY_INDEX 68 1.88246 0.1607

BALTIC_DRY_INDEX does not Granger Cause CRUDE_OIL 1.02156 0.3659

IRON_ORE does not Granger Cause BALTIC_DRY_INDEX 68 2.98316 0.0578

BALTIC_DRY_INDEX does not Granger Cause IRON_ORE 1.01076 0.3698

IRON_ORE does not Granger Cause CRUDE_OIL 68 4.97157 0.0099

(31)

31

Date: 08/31/16 Time: 12:34 Sample: 2010M07 2016M04 Included observations: 70

Correlations are asymptotically consistent approximations

BALTIC_DRY_INDEX,C... BALTIC_DRY_INDEX,C... i lag lead 0 0.3352 0.3352 1 0.3280 0.3310 2 0.3193 0.3280 3 0.2851 0.3511 4 0.2529 0.3841 5 0.2171 0.4082 6 0.1777 0.4209 7 0.1401 0.4173 8 0.0974 0.4150 9 0.0563 0.3903 10 0.0294 0.3575 11 0.0095 0.3223 12 -0.0046 0.2508

(32)

32

Vector Autoregression Estimates Date: 08/31/16 Time: 12:28

Sample (adjusted): 2010M09 2016M04 Included observations: 68 after adjustments Standard errors in ( ) & t-statistics in [ ]

BALTIC_DR... CRUDE_OIL IRON_ORE

BALTIC_DRY_INDEX(-1) 0.654328 -0.002465 -0.001867 (0.11771) (0.00208) (0.00347) [ 5.55862] [-1.18280] [-0.53736] BALTIC_DRY_INDEX(-2) 0.013047 0.000991 0.004950 (0.11806) (0.00209) (0.00349) [ 0.11051] [ 0.47433] [ 1.42045] CRUDE_OIL(-1) 3.057280 1.213116 0.234913 (6.87381) (0.12169) (0.20291) [ 0.44477] [ 9.96921] [ 1.15774] CRUDE_OIL(-2) -2.755226 -0.344600 -0.197892 (6.43621) (0.11394) (0.18999) [-0.42808] [-3.02441] [-1.04159] IRON_ORE(-1) 4.532198 0.091721 1.188282 (4.05495) (0.07178) (0.11970) [ 1.11770] [ 1.27773] [ 9.92737] IRON_ORE(-2) -2.477277 0.004848 -0.259810 (4.33203) (0.07669) (0.12788) [-0.57185] [ 0.06322] [-2.03172] C 88.52267 2.314686 0.344254 (145.358) (2.57325) (4.29080) [ 0.60900] [ 0.89952] [ 0.08023] R-squared 0.698287 0.966122 0.962125 Adj. R-squared 0.668610 0.962790 0.958400 Sum sq. resids 5334270. 1671.718 4648.092 S.E. equation 295.7145 5.234998 8.729160 F-statistic 23.52977 289.9340 258.2611 Log likelihood -479.6731 -205.3592 -240.1278 Akaike AIC 14.31391 6.245859 7.268464 Schwarz SC 14.54239 6.474337 7.496943 Mean dependent 1118.309 92.36294 115.0806 S.D. dependent 513.6919 27.13865 42.79809

Determinant resid covariance (dof adj.) 1.76E+08

Determinant resid covariance 1.27E+08

Log likelihood -923.9544

Akaike information criterion 27.79278

(33)

33

5. Results

Taking the BDI as dependent variable and Crude Oil as independent variable I find a coefficient of about 6,8, as shown by the EViews equation estimation results below. This means that any change in Crude Oil forecasts a change of 6,8 in the BDI. This can

economically be explained by the fact that more than 50% of the costs of operating ships are fuel costs. In Chapter 6 I will show that it does not work the other way around. The t-statistic with three degrees of freedom and 5% probability is 3,18. Our observed values for C and crude oil are smaller than 3,18 so we say that the null hypothesis Η0: β =

0 cannot be rejected at the 5% level.

The F-distribution with three degrees of freedom and 5% probability is 8,57. Here we observe 8,6 also here we say that the null hypothesis Η0: β = 0 cannot be rejected at the

5% level.

Dependent Variable: BALTIC_DRY_INDEX_ACTUALS Method: Least Squares

C 523.4748 223.5457 2.341691 0.0221

CRUDE_OIL__PETROLEUM___D 6.853390 2.336226 2.933530 0.0046 R-squared 0.112336 Mean dependent var 1153.214 Adjusted R-squared 0.099283 S.D. dependent var 549.8083 S.E. of regression 521.8018 Akaike info criterion 15.38061 Sum squared resid 18514842 Schwarz criterion 15.44485 Log likelihood -536.3213 Hannan-Quinn criter. 15.40613 F-statistic 8.605600 Durbin-Watson stat 0.413879 Prob(F-statistic) 0.004565

(34)

34

The results below show that price changes in independent variable Iron Ore, predict a price adjustment in the dependent variable BDI.

In this overview the general overall information is show about the three variables returns. They go all down over 5 years. The BDI clearly has highest maximum, minimum and standard

deviation. This volatility we also saw in the index prices.

Dependent Variable: BALTIC_DRY_INDEX_ACTUALS Method: Leas t Squares

Date: 08/30/16 Tim e: 13:26 Sam ple: 2010M07 2016M04 Included obs ervations : 70

C 277.3272 157.4091 1.761825 0.0826

IRON_ORE_PRICE__US_DOLLA 7.572031 1.278937 5.920564 0.0000

Sum s quared res id 13763205 Schwarz criterion 15.14828

BDI return OIL returns IRON ORE returns

Mean -0.015630 -0.008267 -0.010896 Median -0.003955 0.004366 -0.012123 Maxim um 0.570660 0.179322 0.184197 Minim um -0.938398 -0.249799 -0.190219 Std. Dev. 0.266267 0.081731 0.084608 Skewnes s -0.654592 -0.761247 0.094685 Kurtos is 4.269341 4.105247 2.542194 Jarque-Bera 9.559914 10.17622 0.705662 Probability 0.008396 0.006170 0.702696 Sum -1.078481 -0.570411 -0.751815 Sum Sq. Dev. 4.821069 0.454232 0.486781 Obs ervations 69 69 69

(35)

35

6. Robustness Checks

Additionally, I looked whether or not the results also work the other way. Is the BDI a predictor of Crude Oil? The results below show negative.

We can say that there is a unidirectional relationship between the BDI and Crude Oil prices: the BDI is impacted by its own volatility and the Crude Oil price; the Crude Oil price only by its own volatility.

Below you can see the unidirectional relationship between the BDI and Iron Ore prices: the BDI is impacted by its own volatility and the Iron Ore price; the Iron Ore price only by its own volatility.

Dependent Variable: CRUDE_OIL__PETROLEUM___D Method: Least Squares

C 72.98453 7.129104 10.23755 0.0000

BALTIC_DRY_INDEX_ACTUALS 0.016391 0.005588 2.933530 0.0046 R-squared 0.112336 Mean dependent var 91.88729 Adjusted R-squared 0.099283 S.D. dependent var 26.88847 S.E. of regression 25.51881 Akaike info criterion 9.344864 Sum squared resid 44282.25 Schwarz criterion 9.409107 Log likelihood -325.0702 Hannan-Quinn criter. 9.370382 F-statistic 8.605600 Durbin-Watson stat 0.099070 Prob(F-statistic) 0.004565

(36)

36

Dependent Variable: IRON_ORE_PRICE__US_DOLLA Method: Least Squares

C 63.87006 9.680552 6.597770 0.0000

BALTIC_DRY_INDEX_ACTUALS 0.044921 0.007587 5.920564 0.0000 R-squared 0.340146 Mean dependent var 115.6740 Adjusted R-squared 0.330442 S.D. dependent var 42.34786 S.E. of regression 34.65178 Akaike info criterion 9.956731 Sum squared resid 81650.72 Schwarz criterion 10.02097 Log likelihood -346.4856 Hannan-Quinn criter. 9.982249 F-statistic 35.05308 Durbin-Watson stat 0.253304 Prob(F-statistic) 0.000000

(37)

37

7. Conclusion

The main question of my thesis is: are Crude Oil and Iron Ore prices / indices predictors of the BDI? The answer is yes.

Decision makers need guidance in their strategic and operational planning. Bulk shipping, Crude Oil and Iron Ore are indicators of global economic growth.

In their own submarkets, decision makers will use these macro-economic developments to win market share or just survive another crisis or lifecycle dip.

Imbalances in the supply and demand of the Bulk Shipping Industry have global economic and even financial impact. At this moment we are in a supply overcapacity market imbalance.

The new ship building industry in Asia is employing tens of thousands of workers, many of whom are laid off because of a structural overcapacity of active vessels. Because the backlog for new orders is low, deliveries are delayed or even cancelled.

During the ship building boom financial institutions stepped into this market and as a result many ship building companies are heavily leveraged by debt and have difficulties repaying loans and interest. This situation also negatively impacts Asian banks and national governments.

The vessel scrap industry is booming. Today the Indian and Bangladeshi beaches are packed with bulk carriers and tankers waiting to be decommissioned under miserable and dangerous working conditions.

With today’s new environmental and sustainability regulations in place, ship designers and owners must adopt these new situations in their fleet composition growth strategy. Port authorities must encourage oil companies to build LNG supply stations for this new sustainability balance.

Vessel and port regulations are globally incomparable. For example, in the North Sea and the Gulf of Mexico stricter safety and environment regulations to shipping operations apply, than in the South Chinese Sea.

(38)

38

Moreover, the demand side of the market is not yet positive. Chinese GDP figures still go down and other upcoming markets like Brazil and Russia are facing economic and

political instabilities.

In my data analysis I have found that the Baltic Dry Index times-series is non-stationary, just like Crude Oil and Iron Ore.

The BDI and sub-market indices do not behave randomly. To a certain degree they react on the market prices of yesterday.

(39)

39 REFERENCES

Barberakis, J., Schinas, O. and Stefanakos, Ch. N., 2011, Regression and probability analysis of dry bulk indices, Conference: IMAM (International Maritime Association of the Mediterranean).

Batrinca, Ghiorghe and Cojanu, Gianina, 2013, The Dynamics of the Dry Bulk Sub-Markets, Scientific

Papers, Journal of Knowledge Management, Economics and Information Technology, December

2013, 13-23.

Batrinca, Ghiorghe and Cojanu, Gianina, 2014, The determining factors of the dry bulk market freight rates, 2014 International Conference on Economics, Management and Development, 109-112. Behmiri, Niaz Bashiri and Manso, Jose R. Pires, 2013, Crude oil price forecasting techniques: a comprehensive review of literature, SSRN Electronic Journal , January 2013, 30-48

Chen, Shun, 2011, Modelling and Forecasting in the Dry Bulk Shipping Market.

Chang, Su-Chiung, Chou, Ming-Tao and Yang, Ya-Ling, 2011, A study of the Dynamic Relationship between Crude Oil Price and the Steel Price Index, Review of Economics & Finance, November 2011,

Article ID: 1923-7529-2012-02-30-13.

De Monie, Gustaaf, Notteboom, Theo and Rodrique, Jean-Paul, 2009, Economic Cycles in Maritime

Shipping and Ports: The path to the Crisis of 2008.

Franses, Philip Hans B.F. and Veenstra, Albert Willem, 1996, A co-integration approach to forecasting freight rates in the dry bulk shipping sector.

Frey, Giliola, Manera, Matteo, Markandya, Anil and Scarpa, Elisa, 2009, Econometric models for oil price forecasting: A critical survey, CESifo Forum 1/2009,29-44

Geman, Hlyette and Smith, William O., 2012, Shipping markets and freight rates: an analysis of the Baltic Dry Index, Journal of Alternative Investments 15 (1), 98-109.

Greenwood, Robin and Hanson, Samual G., 2013, Waves in Ship Prices and Investment, The quarterly

Journal of Economics 2015, volume 130, issue 1, 55-109.

International Maritime Organisation, 2012, International Shipping Facts and Figures – Information on Resources on Trade, Safety, Security, Environment, Maritime Knowledge Centre June 2012.

(40)

40 Ko, Byoung-wook, 2010, A mixed-Regime Model for Dry Bulk Freight Market, The Asian Journal of

Shipping and Logistics, volume 26, number 2, December 2010, 185-204

Kunst, Robert M., 2009, Applied Time Series Analysis – Part 1.

Lam, Derek, 2013, Time Series Modelling of Monthly WTI Crude Oil returns.

Li, Hua, Ren, Erlin, Wang, Bin and Wu, Caiying, 2011, Empirical Analysis of the Influencing Factors on Iron ore Prices, Artificial Intelligence, Management Science and Electronic Commerce (AIMSEC), 2011

2nd International Conference, 3004-3008.

Lun, Y.H. Venus, and Quaddus, Mohammed A., 2009, An empirical model of the bulk shipping market,

International journal of Shipping and Transport Logistics, Vol. 1, No. 1, 2009, 37-54.

Kalouptsidi, Myrto, 2012, Time to Build and Fluctuations in Bulk Shipping, American Economic Review

vol. 104, no. 2, February 2014, 564-608.

Raiswell, J.E., The Problem of Pricing Strategy in Shipbuilding, Dynamica Digital Archive volume 2,

part 2, Spring 1976.

Stopford, Martin, 1988, Maritime Economics, 3rd Edition 2009.

Taylor, A.J., 1976, The Dynamics of Supply and Demand in Shipping, Dynamica Digital Archive volume

2, part 3, Summer 1976.

Baltic Dry Index : the effects of crude oil and iron ore

GRADUATE THESIS INTERNATIONAL FINANCE