You are on page 1of 18

Transportation Research Part B 41 (2007) 126143 www.elsevier.

com/locate/trb

Investment timing and trading strategies in the sale and purchase market for ships
Amir H. Alizadeh *, Nikos K. Nomikos
Faculty of Finance, Cass Business School, London EC1Y 8TZ, United Kingdom Received 14 September 2005; received in revised form 13 April 2006; accepted 25 April 2006

Abstract The aim of this paper is to investigate, for the rst time, the performance of trading strategies based on the combination of technical trading rules and fundamental analysis in the sale and purchase market for dry bulk ships. Using a sample of price and charter rates over the period January 1976 to September 2004, we establish the existence of a long-run cointegrating relationship between price and earnings and use this relationship as an indicator of investment or divestment timing decisions in the dry bulk shipping sector. In order to discount the possibility of data snooping biases and to evaluate the robustness of our trading models, we also perform tests using the stationary bootstrap approach. Our results indicate that trading strategies based on earningsprice ratios signicantly out-perform buy and hold strategies in the second-hand market for ships, especially in the market for larger vessels, due to higher volatility in these markets. 2006 Elsevier Ltd. All rights reserved.
Keywords: Trading strategies; Cointegration; Shipping; Stationary bootstrap

1. Introduction Investors in shipping markets have always been faced with important and dicult decisions on investment and/or divestment timing because of the complex and volatile nature of the shipping industry. It is not surprising therefore that the dynamic behaviour of ship prices and their conditional volatilities have been the focus of many empirical studies in maritime economics literature. Traditional approaches for modelling ship prices are mainly based on general and partial equilibrium models using structural relationships between a number of variables such as orderbook, newbuilding deliveries, scrapping rates, freight rates, bunker prices, etc. (see Strandenes, 1984; Beenstock and Vergottis, 1989; Tsolakis et al., 2003, among others). More recent studies have applied real options analysis for determining ship prices; this valuation framework takes explicitly into account the operational exibility in ship management, in terms of choosing between entry and exit from the market, spot and period time-charter operations, and switching between lay-up and trading modes (see Dixit and Pindyck, 1994; Tvedt, 1997; Bendall and Stent, 2004, among others).
*

Corresponding author. Tel.: +44 207 040 0199; fax: +44 207 040 8681. E-mail addresses: a.alizadeh@city.ac.uk (A.H. Alizadeh), n.nomikos@city.ac.uk (N.K. Nomikos).

0191-2615/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.trb.2006.04.002

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

127

The price formation in the second-hand market for ships has also been examined to determine whether markets for ships are ecient and whether prices are formed rationally. For example, Kavussanos and Alizadeh (2002a), Hale and Vanags (1992) and Glen (1997), test the validity of the Ecient Market Hypothesis (EMH) in the formation of second-hand dry bulk prices. These studies argue that the failure of the EMH may either be attributed to the existence of time-varying risk premia, or reect arbitrage opportunities in the market. The latter suggests that if prices for vessels are found to deviate consistently from their rational values, then trading strategies can be adapted to exploit excess prot making opportunities.1 For example, when ship prices are lower than their fundamental values, then buying and operating these vessels may be protable since they are under-priced in comparison to their future protability (i.e. the earnings from freight operations). On the other hand, when prices are higher than their corresponding rational values, then from a shipowners point of view it may be more protable to charter in vessels, rather than buying them, since they are overpriced in comparison to their expected future protability. Despite numerous studies in the literature on ship price formation, on testing the validity of the EMH in shipping markets, and on the behaviour of ship prices and their volatilities, there has been little empirical evidence on whether sale and purchase decisions of merchant ships, based on fundamental and/or technical analysis, can be protable. For example, Adland (2000) and Adland and Koekebakker (2004) investigate the performance of technical trading rules and argue that if the market for ships is ecient, then trading strategies based on these rules should not produce wealth in excess of what can be gained through simple buy and hold strategies.2 Using both in- and out-of-sample tests, they report that, in general, trading rules do not yield excess returns that can compensate for transaction costs. Although their study seems to provide support for the EMH, given the nature of technical analysis there may be two points that could be raised. First, as they point out, their results might be dependent on the variables and set of rules used for constructing the technical trading strategies. Second, the use of technical trading rules on their own, and not in conjunction with the underlying economic theory, may not be as eective in this market. This is because the historical pattern of the underlying series alone is not enough to extract information on the future behaviour of prices, since it is widely documented that ship prices follow random walk processes. Therefore, in this study we overcome these shortcomings by developing a theoretical economic framework which links prices and earnings, and then combining such a relationship with technical rules, to extract information from the market for investment and trading purposes. In other words, we do not rely only on the past price behaviour for trading strategies, but we combine technical trading rules with fundamental analysis by using the cointegration relationship between prices and earnings. In particular, we use the priceearnings ratio as an indicator for investment or divestment timing decisions in the dry bulk shipping sector. The motivation for this stems from the importance of economic indicators and, in particular, the priceearnings (P/E) ratio (or its inverse the earningsprice, E/P, ratio) in predicting asset returns in nancial markets. For instance, P/E ratios of individual stocks or portfolios are regularly used to explain the returns in the stock market and a number of studies document the ability of P/E ratios to predict future returns of individual stocks or portfolios. For instance, Campbell and Shiller (1998) show that P/E ratios are negatively correlated with subsequent stock returns over a ten-year period. Other studies on the information content of P/E ratio in predicting stock returns include Fama and French (1992), Fuller et al. (1993), Jae et al. (1989), and Roll (1994). The spread between P/E ratios and interest rates is also used to forecast movements of broad stock market indices. For example, Lander et al. (1997) use various linear combinations of the P/E ratio and bond yields to predict returns on the S&P 500 index in a regression framework, while, Pesaran and Timmermann (1995) include both interest rates and P/E ratios as possible explanatory variables of stock market movements. In addition, a number of studies in nancial economics literature examine the performance of various strategies that may be useful in timing the market. For example, Lander et al. (1997) test their models ability to time the market, while Fuller and Kling (1990, 1994) study regression-based market timing strategies using dividend yields, and highlight the inherent diculties in nding market timing strategies.
Here by fundamental or rational value of assets we mean the discounted present value of the expected stream of income that the assets will generate over their lifetime. 2 Adland and Koekebakker (2004) use historical prices for VLCC and Aframax tankers, as well as capesize and panamax dry bulk carriers.
1

128

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

However, although these studies provide empirical evidence on the performance of trading rules in nancial markets, there has been little evidence for markets that trade real assets, in particular for the transportation and shipping markets. The aim of this paper is therefore to investigate the performance of trading strategies for investment decisions in the market for second-hand ships. In doing so, the paper contributes to the literature in a number of ways. First, there has been no prior evidence on the performance of trading strategies based on signals provided by fundamental market price indicators such as the priceearnings (P/E) ratio and how eective these strategies are for investment decisions in the shipping markets. We consider ships as real capital assets which can, not only generate income through operation but also capital gain (loss) through price appreciation (depreciation). In this setting we examine whether the P/E ratio can be used to identify the optimal time to buy or sell second-hand vessels. Second, we compare the protability and riskreturn characteristics of our proposed strategies with a simple benchmark strategythe buy and hold, where one invests in the shipping market at all times. This comparison enables us to assess whether the dynamic investment strategy, in which one invests in ships most of the time but switches to risk-free investments (e.g. t-bills) when the P/E ratio is too high, is superior to static trading strategies. As a matter of fact, if the information contained in the P/E ratio is economically important, one would expect the dynamic strategies to have higher risk-adjusted returns. Third, we also compare the protability of the trading strategies across dierent vessel sizes and attribute any dierences in the results to the idiosyncratic features of each market. Finally, we also use stationary bootstrap as a technique to re-generate the underlying series and hence replicate the trading results from the dierent strategies in a simulation environment; this is done in order to discount the possibility that our results may be due to data snooping or statistical chance. Our methodology is motivated by the fact that the ratio of ship prices to operating earnings (priceearnings ratio) is a measure of whether the market for second-hand ships is under or overvalued, relative to its fundamentals. Shipowners, ship operators and charterers regularly use this ratio as an indicator of whether to buy or charter-in tonnage. The ndings of this paper also have important practical implications and can be of interest to investors in shipping markets regarding the timing of investment and divestment. In addition, recent developments in the areas of shipping investment and nance, such as the development of shipping funds and derivative contracts for ship values, may enable participants not only to invest in ships as an alternative investment but also to speculate on the future outlook of the market without incurring the costs of physically owning or operating a ship. Although the focus of the paper is in the market for ships, the same methodology can also be used for the valuation and investment analysis of other tangible assets in the transportation sector, such as the airline industry. Since airlines are often faced with the choice of whether to lease or buy aircrafts, the ratio of aircraft prices to operational earnings can also be used in the same setting to identify investment timing opportunities. The structure of this paper is as follows. Section 2 presents the theoretical background and the methodologies proposed in the asset pricing literature, which are used to relate prices and earnings for second-hand ships. The data and their properties are discussed in Section 3. Section 4 presents the empirical results and discussion on the performance of trading strategies using simulations. Finally, Section 5 concludes this paper. 2. The theoretical relationship between price and earnings Investors in the shipping industry, like investors in any other sector of the economy, are not only interested in income from the day to day operation of ships, but also interested in gains from capital appreciation in the value of the vessels. Therefore, from the investors point of view expected one period returns, EtRt+1, on shipping investments are equal to the expected one period capital gains between time t and t + 1 (EtPt+1 Pt)/Pt, plus the expected return from operation, EtPt+1/Pt, where EtPt+1 is the expected ship price at time t + 1 and EtPt+1 is the expected operating prot between period t and t + 1.3 Mathematically,   E t P t 1 P t E t P t 1 E t R t 1 1 Pt

See Section 3 of the paper for the description of operating prots and TC earnings.

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

129

Eq. (1) can be rearranged to represent the present value relationship, where the current ship price, Pt, is expressed in terms of the expected price of the vessel, expected operational prots and expected rate of return, in the following expression   E t P t 1 E t P t 1 Pt 2 1 E t R t 1 Eq. (2) is in fact a one period present value model; through recursive substitution and some algebraic manipulation, Pt can be written as the sum of the present values of the future prots plus the terminal or resale value, P sc tn of the asset. Mathematically ! ! n i n Y X Y 1 1 Pt 1 E t R t j 1 E t R t j 3 Et Pti Et P sc t n
i1 j 1 j 1

Eq. (2) can also be written in logarithmic form; however, in this case it is not possible to perform recursive substitutions to write the log of price (lnPt) in terms of the log of discounted expected earnings and log of discounted expected terminal value of the asset. Campbell and Shiller (1987) suggest a way round this by using a rst-order Taylor series expansion and linearising (1) around the geometric mean of P and P (P and P) to give ln1 Et Rt1 q lnEt P t1 1 q lnEt Pt1 ln P t k 4 where q P =P P and k = ln(q) (1 q)ln(1/q 1). Letting Etpt+1 = ln(EtPt+1), Etrt+1 = ln(1 + EtR+1) and Etpt+1 = ln(EtPt+1), Eq. (2) can be written as pt qEpt1 1 qEpt1 Ert1 k which can be solved recursively forward to yield pt
n1 X i0 n1 X i 0

qi 1 qEt pt1i

n qi Et rt1i qn Et psc tn k 1 q =1 q

Since prices and operating prot series are non-stationary, Eq. (6) should be transformed in such a way so as to derive a model with stationary variables. Following Campbell and Shiller (1987), we use the cointegration relationship between the log-price and the log-earning series for such transformation; that is the log P/E ratio.4 This is done by subtracting pt from both sides of (6) which results in p t pt or p t pt
n 1 X i0 n qi Et Dpt1i Et rt1i qn Et psc tn E t ptn k 1 q =1 q n 1 X i0

qi 1 qEt pt1i pt

n1 X i0

n qi Et rt1i qn Et psc tn k 1 q =1 q

In the above setting pt pt and psc t pt are the log P/E ratio and log resale priceearning ratio, respectively. According to Campbell and Shiller (1987), the left hand side of Eq. (8) is the actual spread, and the right hand

It has been argued that many nancial and economic time series are non-stationary. Such variables tend to have an increasing variance and do not show a tendency to revert to a long-run mean. In order to detect such behaviour in a variable one should use unit root tests such as the Phillips and Perron (1988) and Kwiatkowski et al. (1992). In general, it has been shown that correlation between nonstationary series does not accurately represent the true relationship between variables. However, there might be cases where two nonstationary variables can be related in the long-run through an equilibrium relationship, but deviate from such an equilibrium in the short run. Such a relationship is called a cointegrating relationship and implies that a linear combination of the two non-stationary series is stationary (Engle and Granger, 1987). In our case for instance, although the log of ship prices and the log of earnings are non-stationary time series, their dierence (i.e. the P/E ratio) should be stationary because ship prices and earnings are linked through the fundamental pricing relationship of Eq. (6). Thus, if the P/E ratio is too high or too low, we expect it to revert back to its long-run mean due to corrective movements in the level of earnings and ship prices.

130

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

side is the theoretical spread which is based on the expected values of earnings, discount rates and resale values of the asset. Under ecient market conditions, the two spread series should be statistically equal with similar volatility, which can be tested empirically (see Kavussanos and Alizadeh, 2002a). This model also suggests that the dierence between the actual and theoretical spreads contains very useful information for investment purposes. For example, when the actual spread is greater than the theoretical one, this implies that the actual price is above the theoretical price, which is the discounted present value of future earnings; that is, vessels are overpriced relative to their future earnings potential. Therefore, the above model suggests that the P/E ratio (spread) contains important information regarding investment timing and trading strategies in shipping markets. 2.1. Cointegration and causality An alternative but related way of explaining the information content of the P/E ratio is through the cointegrating relationship between these two variables. In order to test the existence of cointegration between second-hand prices and operational earnings, we use the Johansens (1988) reduced rank cointegration technique and estimate the following vector error correction model (VECM) Dpt Dpt
q X i1 q X i1

a i D p t i ci Dpti

q X i 1 q X i1

bi Dpti c1 pt1 hpt1 h0 e1;t 9 d i Dpti c2 pt1 hpt1 h0 e2;t

The above VECM model can be used to establish the cointegrating relationship between log-prices and log earnings which then can be used to set up a trading strategy for shipping investment. The important element of the cointegration relationship is the error correction term (ECT) which is in fact the dierence between logprices and log earnings (pt1 hpt1 h0). The constant term in the error correction term, h0, represents the long-run equilibrium relationship; it is in other words the long-run average of the P/E ratio. In order to set up a trading model then, at any month we estimate the deviation of the log P/E ratio from its long-run mean (cointegration constant). For example, when the log P/E ratio is greater than its long-run average, this indicates that earnings are low relative to ship prices or, alternatively, ship prices are overvalued relative to their earnings potential. In this case, ship prices in the market are expected to adjust in future periods by falling relative to their current levels. Similarly, when the P/E ratio is lower than its long-run average, this can be regarded as an indication that ship prices are undervalued relative to their potential earnings and, hence, it is expected that prices will increase in the next period, so that the long run earningsprice relationship is restored. The VECM model of Eq. (9) also provides a framework for testing the causal linkages between ship prices and earnings. According to the Granger Representation Theorem (Granger, 1986), if two variables are cointegrated, then at least one variable should Granger-cause the other.5 Since ship prices are determined through the discounted present value of expected earnings and the latter are determined exogenously, through the interaction between the supply and demand schedules for shipping services, we expect the causality to be unidirectional; that is, we expect earnings to Granger-cause ship prices but not the other way round. Hence, any change in earnings should aect the spread between log-prices and log earnings and result in a change in ship prices over the next period. Therefore, in this case one can argue that the log P/E ratio contains information on future changes in ship prices, which can be used for investment strategies.

5 A time series, pt, is said to Granger cause another time series, pt, if the present value of pt can be predicted more accurately by using past values of pt than by not doing so, considering also other relevant information including past values of pt (Granger, 1969). Therefore, the criterion for Granger causality is whether or not the variance of the predictive error of pt is reduced when past pt values are included in its prediction. In terms of the VECM of Eq. (9), pt Granger causes pt if some of the bi coecients, i = 1, 2, . . . , q are not zero and/or c1, the error correction coecient in the equation for ship prices, is signicant at conventional levels.

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

131

2.2. Trading strategies The aim of this analysis is to utilise the relationship between variables in shipping markets and devise strategies to identify the timing for sale and purchase of merchant ships. To do so, we develop a strategy which is based on the relationship between price and earnings of such vessels. As mentioned earlier, theoretically, the price of a vessel is linked to her expected operational earnings which are in turn determined by current and expected conditions in the shipping market and the world economy. This theoretical relationship between prices and earnings allows us to use the historical (empirical) spread between them to identify buy and sell opportunities in the market. In practice, the universe of potential trading rules is vast, as there are multiple combinations of relationships between variables that can produce a trading signal as well as multiple parameterizations for a given family of rules; for instance, there are dierent combinations of Moving Average (MA) rules reecting dierent time spans in the estimation of MA prices as well as dierent lter rules depending on the distance from the mean. As it is beyond the scope of this study to evaluate an exhaustive set of trading rules, we focus our eorts on two simple cases of MA rules based on the relationship between ship prices and earnings. The moving average trading strategy is mainly based on the comparison of a fast (short) and a slow (long) moving average of the PE ratio. For example, a simple MA trading strategy in the sale and purchase market for ships could be a comparison of a 12 month MA with 3 month MA of the PE ratio. This means that in a given month, a positive dierence between the 12-month MA and the 3-month MA of the PE ratios should signal a buy decision; similarly, a negative dierence signals a sell decision.6 3. Description of data For the purpose of this study, monthly prices for 5-year old ships are collected for three dierent size dry bulk carriers (capesize, panamax and handysize) from Clarksons Shipping Intelligence Network from January 1976 to September 2004. Capesize prices are for the period April 1979 to September 2004. All prices are quoted in million dollars and represent the average value of vessels traded in each category in any particular month. In shipping, operating prots can be dened as time-charter rates, or the time-charter equivalent of spot rates when a vessel is operating in the spot market, minus operating costs. In this study, we use time-charter rates as a proxy for earnings, Pt, for two reasons. First, because time-charter rates do not include voyage costs and represent the net earnings from chartering activities of the vessel. Second, since time-charter rates are hire contracts for a number of consecutive periods, they are considered to contain information about future earnings of the vessel during these periods (see Kavussanos and Alizadeh, 2002b, for a detailed discussion of timecharter rates formation). As a result, it is believed that time-charter rates (earnings) may explain price changes better than current spot rates. Monthly time-charter rates for handysize, panamax and capesize vessels over the period January 1976 (April 1979 for capesize vessels) to September 2004 are also obtained from Clarksons Shipping Intelligence Network. Finally, monthly operating expenses for each vessel size are also collected from the same source. Table 1 reports descriptive statistics of levels and logarithmic rst dierences of second-hand prices, as well as operational earnings for capesize, panamax and handysize vessels. The results indicate that mean levels of prices for larger vessels are higher than for smaller ones. Unconditional volatilities of prices (standard deviation) also follow a similar pattern; that is, prices for larger vessels uctuate more than prices for smaller vessels. Jarque and Bera (1980) tests indicate signicant departures from normality for TC earnings and price returns in all markets, while price levels for all size classes seem to be normally distributed. The Ljung and Box (1978) Q statistics for 12th-order autocorrelations in levels and logarithmic rst dierences of earnings are all signicant, indicating that serial correlation is present in all price and prot series. Finally, Engles
For the strategy implemented in this paper, a sell decision will be executed only if the investor has already bought a ship. In other words short-selling is not permitted since practically it is not possible for an investor to take a short position in a vessel. However, the development of new paper contracts on ship prices, such as the Baltic Sale and Purchase Agreement (BSPA) could allow investors to short sell the vessel values and benet from falling ship prices.
6

132

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

Table 1 Descriptive statistics of price (P) and time charter earnings (TC) for dierent size dry bulk carriers Mean Capesize Second-hand prices, P ($m) 1 year TC earnings, P ($m) Log return Dp (%) Log change, DP (%) 22.54 3.960 0.007 0.003 SD 10.11 1.183 0.071 0.101 Skew. 0.073 {0.583} 1.200 {0.000} 2.097 {0.000} 0.248 {0.079} Kurt. 0.396 {0.137} 2.709 {0.000} 17.737 {0.000} 1.753 {0.000} JB 2.557 {0.278} 167.01 {0.000} 4761 {0.000} 42.16 {0.000} Q(12) 3346 {0.000} 1950 {0.000} 58.15 {0.000} 53.25 {0.000} ARCH(12) 2867 {0.000} 1662 {0.000} 24.76 {0.016} 37.83 {0.000}

Panamax Second-hand prices, P ($m) 1 year TC earnings, P ($m) Log return Dp (%) Log change, DP (%)

15.83 3.131 0.004 0.006

6.240 1.495 0.058 0.093

0.233 {0.078} 2.034 {0.000} 0.263 {0.047} 0.467 {0.000}

0.033 {0.902} 9.674 {0.000} 3.767 {0.000} 8.995 {0.000}

3.140 {0.208} 1583 {0.000} 207.4 {0.000} 1172 {0.000}

3164 {0.000} 1901 {0.000} 57.33 {0.000} 44.29 {0.000}

2691 {0.000} 910.9 {0.000} 71.59 {0.000} 117.3 {0.000}

Handysize Second-hand prices, P ($m) 1 year TC earnings, P ($m) Log return Dp (%) Log change, DP (%)

10.54 2.300 0.002 0.005

3.996 0.896 0.052 0.056

0.066 {0.613} 1.687 {0.000} 0.011 {0.933} 0.357 {0.007}

0.693 {0.016} 6.395 {0.000} 2.571 {0.000} 2.893 {0.000}

6.138 {0.046} 751 {0.000} 94.77 {0.000} 127.3 {0.000}

3390 {0.000} 2652 {0.000} 97.56 {0.000} 94.08 {0.000}

3118 {0.000} 1904 {0.000} 54.20 {0.000} 184.2 {0.000}

Sample period is January 1976 to September 2004 for the Handysize and Panamax series and April 1979 to September 2004 for the capesize series. Figures in {} are p-values. a4 3), respectively. Their asymp Skew. and Kurt. are the estimated centralised third and fourth of the data, denoted ^ a3 and (^ p pmoments totic distributions, under the null, are T ^ a4 3 $ N 0; 24. a3 $ N 0; 6 and T ^ JB is the Jarque and Bera (1980) test statistics for normality; it is v2(2) distributed. Q(12) is the Ljung and Box (1978) Q statistic on the 12th-order sample autocorrelations of the raw series, distributed as v2(12). ARCH(12) is the Engles (1982) test for 12th-order ARCH eect; the statistic has a v2 (12) distribution.

(1982) ARCH tests for 12th-order ARCH eects indicate the existence of autoregressive conditional heteroscedasticity in all series. Phillips and Perron (1988) (PP), unit root tests are performed on the log-levels and log-dierences of second-hand prices and time-charter rates (earnings), for the three size dry bulk carriers. Results from these tests suggest that log-levels of all price and earnings series are non-stationary, while their rst dierences are stationary, indicating that variables are integrated of order one, I(1). Also, PP unit root tests on the spread between logs of second-hand prices and time-charter rates for dierent size vessels indicate that all spread series are stationary. Studies in the literature argue that PP tests may have low power in rejecting the unit root null hypothesis in favour of the alternative of stationarity (see Harris, 1995; Maddala and Kim, 1998). Lee et al. (2000) suggest that one way of overcoming this problem is by conducting unit root tests which test the null of stationarity against the alternative of a unit root, such as the test developed by Kwiatkowski et al. (1992), henceforth KPSS test. In the KPSS test, the null hypothesis of stationarity is rejected in favour

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

133

1976- 01

1985-12

1993- 01

1977-06

1980-04

1981-09

1983-02

1984-07

1987-05

1990-03

1991-08

1994-06

1995-11

40 35 30 25 20 15 10 5 0

40. 0 35. 0 30. 0 25. 0 20. 0 15. 0 10. 0 5.0 0.0 1997-04 1998-09 2000-02 2001-07 2002-12 2004-05

TC rate 000'$

Price m$

1978-11

PANAMAX Price

1988-10

PANAMAX TC rate

Fig. 1. Historical prices and time-charter rates for panamax dry bulk carriers.

of the unit root alternative, if the calculated test statistic exceeds the corresponding critical values. KPSS test results also conrm that the log-price and time-charter earnings series are non-stationary, I(1), while the spreads between prices and time-charter earnings are in fact stationary.7 These results also provide early evidence that log prices and time-charter earnings are cointegrated, and render support for the use of the VECM specication for modelling ship price changes. Finally, Fig. 1 plots the second-hand prices along with one year time-charter rates for a panamax dry bulk carrier over the sample period. It can be seen that while prices and time-charter rates move together in the long-run, they tend to vary over time and under dierent market conditions. For example, it can be observed that just before any shipping market recovery, the spread between TC earnings and prices tends to narrow (e.g. in 1978, 19871988, and 20022003), while the spread between TC earnings and prices tends to widen during market downturns (e.g. in 1980, 1990 and 1997) which is another indication of the importance of priceearning relationship in investment timing in shipping markets. Graphs for the capesize and handysize TC earnings and prices, not presented here, indicate a similar pattern. In addition, comparison of behaviour of prices across dierent vessel sizes reveals that prices for all three categories of dry bulk carriers tend to move close together over the long-run while their short run behaviour seems to be dierent and show idiosyncratic stochastic behaviour over time. Dierent short-term dynamics of dierent size ship prices might be related to dierences in the supply and demand for each type of vessel and the prevailing conditions in the shipping industry. 4. Empirical results Having identied that ship prices and earnings are I(1) variables, cointegration techniques are used next to examine the existence of a long-run relationship between these series. The lag length (q = 1) in the VECM of Eq. (9) is chosen on the basis of the Schwarz Bayesian Information Criterion (SBIC) (Schwarz, 1978). LR tests indicate that an intercept term should be included in the long-run relationship.8 Johansens (1988) reduced rank cointegration method is then used to establish the cointegration relationship between ship prices and earnings. This method involves assessing the rank of the long-run coecients matrix, C, through the kmax and ktrace statistics.9 The rank of C in turn determines the number of cointegrating relationships; for instance,
7 8

Unit root results are not presented here but are available from the authors. Johansen (1991) proposes the following statistic to test for the appropriateness of including an intercept term in the cointegrating P 2 ^ ^ ^ vector against the alternative that there are linear trends in the level of the series; T n ir1 ln1 ki ln1 ki $ v n r where ki ^ and ki represent the i smallest eigenvalues of the model that includes an intercept term in the cointegrating vector and an intercept term in the short run model, respectively. Acceptance of the null hypothesis indicates that the VECM in Eq. (9) should be estimated with an intercept term in the cointegrating vector. These results are not presented here and are available from the authors. It should also be noted that the inclusion of an intercept term is also justied on the basis that the intercept term reects the mean P valueQ of the P/E ratio. 9 C is the coecient of xt 1, in the matrix representation of VECM of Eq. (9), where xt pt pt 0 ; Dxt k i1 i Dxti Cxt1 et .

134

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

Table 2 Result of Johansens reduced rank cointegration test of log-prices (p) and log time-charter (p) q q X X ai Dpti bi Dpti c1 pt1 hpt1 h0 e1;t Dpt
i1 i1

Dpt

q X i1

ci Dpti Lags

q X i1

d i Dpti c2 pt1 hpt1 h0 e2;t kmax HA rP1 r=2 rP1 r=2 rP1 r=2 17.22 2.94 37.29 3.41 17.40 2.98 15.67 9.24 15.67 9.24 15.67 9.24 kmax 90% CVs ktrace H0 r=0 r61 r=0 r61 r=0 r61 HA r=1 r=2 r=1 r=2 r=1 r=2 20.18 2.94 40.71 3.41 20.37 2.98 19.96 9.24 19.96 9.24 19.96 9.24 ktrace ktrace 90% CVs Normalised coint. vector [1 h h0] [1 1.455 1.095]

Pair of variables

kmax H0

Handysize ln P and ln TC (p and p) Panamax ln P and ln TC (p and p) Capesize ln P and ln TC (p and p)

q=1

r=0 r61 r=0 r61 r=0 r61

q=1

[1 1.237 1.348]

q=1

[1 1.134 1.707]

Sample period is January 1976 to September 2004 for the Handysize and Panamax series and April 1979 to September 2004 for the capesize series. Johansens (1988) reduced rank cointegration tests for each pair are estimated using a model with a constant in the cointegrating vector and no trend. The appropriate number of lags in each case is chosen by minimising SBIC. kr1 tests the null hypothesis of r cointegrating vectors against the alternative of r + 1. kmax r; r 1 T ln1 ^ P ^ ktrace T n ir1 ln1 ki tests the null that there are at most r cointegrating vectors against the alternative that the number of cointegrating vectors is greater than r, where n is the number of variables in the system (n = 2 in this case). CVs represent critical values from Osterwald-Lenum (1992).

if rank(C) = 1 then there is a single cointegrating vector describing the long-run equilibrium relationship between the variables. In this case, C can be factored as C = ch 0 , where c and h are 2 1 vectors.10 Using this factorisation, h 0 represents the vector of cointegrating parameters and c is the vector of error correction coefcients measuring the speed of convergence to the long-run steady state. Results from these tests are reported in Table 2. The kmax and ktrace statistics indicate the existence of one cointegrating vector between ship prices and TC earnings in each market. This means that log-prices and TC earnings are linked through a unique long-run relationship and any deviation from this equilibrium is restored through the short-term adjustment of these variables. The estimated cointegrating vectors, i.e. [1 h h0] from Eq. (9), are also presented in the same table. These unrestricted cointegrating vectors are then used in the estimation of the VECM model; estimation results for the models are presented in Table 3. Residual diagnostics indicate that autocorrelation and heteroscedasticity are present in the residuals of all the regressions. Consequently a Newey and West (1987) correction for serial correlation and heteroscedasticity is applied to the standard errors of the regressions. Examination of the vector of error correction coecients, c, provides insight into the adjustment process of the dierent variables towards equilibrium. Consider rst, the system of equations for the capesize market. The cointegrating vector, h, is signicant in both equations, and the signs of the speed of adjustment coecient (negative for ship prices and positive for TC earnings) are consistent with convergence of ship prices and TC earnings towards their long-run relationship. For instance, in response to a positive deviation from their long-run relationship at period t 1, i.e. pt1 hpt1 h0 > 0, ship prices the following period will decrease and earnings will

Similarly, if rank(C) = 0, C is a 2 2 null matrix and the VECM is reduced to a VAR model in rst dierences. Finally, if rank(C) = 2, then all variables in Xt1 are I(0) and a VAR model in levels is appropriate.

10

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143 Table 3 Result of VECM for three size dry bulk carriers q q X X ai Dpti bi Dpti c1 pt1 hpt1 h0 e1;t Dp t
i1 i1

135

Dpt

q X i1

ci Dpti

q X i1

d i Dpti c2 pt1 hpt1 h0 e2;t

Estimated model for: Capesize Dpt ci i = 1, 2 0.030 (0.012) [2.541] 0.203 (0.057) [3.579] 0.142 (0.039) [3.646] 0.153 Statistics 13.29 0.685 p-Value {0.000} {0.408} Dpt 0.043 (0.018) [2.421] 0.071 (0.085) [0.828] 0.358 (0.059) [6.092] 0.129 DF 2 2 Panamax Dpt 0.028 0.011 [2.402] 0.145 0.056 [ 2.579] 0.144 0.036 [4.056] 0.123 Statistics 16.45 3.534 Dpt 0.078 0.018 [4.289] 0.166 0.088 [1.880] 0.319 0.056 [5.697] 0.140 p-Value {0.000} {0.060} DF 2 2 Handysize Dpt 0.027 (0.010) [2.868] 0.241 (0.052) [4.608] 0.208 (0.049) [4.230] 0.189 Statistics 18.16 4.028 Dpt 0.023 (0.010) [2.214] 0.123 (0.056) [2.181] 0.416 (0.053) [7.875] 0.196 p-Value {0.000} {0.045} DF 2 2

Dpt1

Dpt1

R2 Causality test Dpt ! Dpt Dpt ! Dpt

Sample period is January 1976 to September 2004 for the Handysize and Panamax series and April 1979 to September 2004 for the capesize series. Standard errors, in (), are corrected for serial correlation and/or heteroscedasticity using the Newey and West (1987) method. Numbers in [] are t-statistics; DF are the degrees of freedom for the causality tests.

increase, thus restoring equilibrium in the market. The same pattern is evident in the panamax and handysize markets. More rigorous investigation of the interactions between the variables can be obtained by performing Granger causality tests, which are presented in the same table. According to the Granger (1986) representation theorem, if two price series are cointegrated, then causality must exist in at least one direction. Theoretically, we expect operational earnings to Granger-cause ship prices. We test such causality between the variables by imposing the appropriate restrictions on the VECM model. Tests for the joint signicance of the lagged cross-market returns and error correction coecients, conrm the conjecture that TC earnings Granger-cause ship prices. On the other hand, ship prices cause TC earnings only in the handysize market at the 5% level, and there is no evidence of causality at the 1% level. 4.1. Protability of trading rules There can be unlimited number of ways to set up trading strategies based on MA or lter rules, depending on factors such as the variable on which the rule is applied, the length of MA series considered, and the distance from the mean in the case of lter rules. However, we choose to apply two simple MA based rules to illustrate the importance of the priceearnings relationship (ratio) in determining ship prices and consequently market timing in the sale and purchase market for ships. Our trading strategy is based on the deviation of the log P/E ratio from its long-run mean. In order to determine the timing of sale and purchase, we devise two MA series using the deviation of log P/E ratio from its long-run mean, one slow [e.g. MA(12) or MA(6)] and one fast [MA(1)], as shown in Fig. 2 for the panamax market. The dierence between the two constructed MA series is then used as an indicator for buy and sell

136

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

0.8 0.6 0.4 0.2 0 -0.2 -0.4 -0.6 -0.8 -1

1976-01

1977-06

1978-11

1980-04

1981-09

1983-02

1984-07

1985-12

1987-05

1988-10

1990-03

1991-08

1993-01

1994-06

1995-11

1997-04

1998-09

2000-02

2001-07

2002-12

MA12

MA1

Fig. 2. Plot of moving average 12 (MA12) and moving average 1 (MA1) of historical log-priceearning ratio in panamax market.

signals in the second-hand market. A positive dierence between the slow and the fast MA series signals a sell decision, while a negative dierence signals a buy decision. If investors do not hold a position in the shipping market then we assume that they invest their funds in treasury-bills. In assessing the performance of our strategies, we also consider transaction costs, the income from operating the vessels in the charter markets as well as the depreciation in the value of the vessel. More specically: transaction costs are incurred every time a buy or sell decision is implemented; they typically come in the form of brokerage commission for the sale and purchase shipbrokers who arrange the deals. Operating prots are calculated as the dierence between monthly charter earnings and operating expenses for each month the vessel is in our portfolio.11 Finally, depreciation represents the reduction in the value of the vessel due to wear and tear each month the investor holds the vessel in his/her portfolio. This is estimated as the average decline in the value between a 5-year old and a 10-year old vessel and is 0.5% per month for each type of vessel. It should also be noted that the proposed strategy is structured and implemented in a way that is entirely forward looking. In other words, investment decisions at any point in time are decided on the basis of information that is available to investors at that specic point in time. This way we provide a more realistic and accurate representation regarding the performance of the trading strategies. We then apply the outlined MA trading model to each sub-market within the dry bulk shipping sector. For comparison purposes we also consider the performance of a benchmark buy-and-hold strategy. This strategy mimics the behaviour of a ship operating company which owns a ship and operates her in the charter market throughout her economic life. Hence, the performance of this strategy reects primarily the income from the operation of the vessel in the charter market. Since in the buy and hold strategy one is always an investor in the market, it is an appropriate benchmark for the proposed dynamic trading strategy which suggests investing in the shipping market only when the timing is right. Table 4 presents the annualized mean returns, annualized standard deviations of returns and Sharpe ratios, which scale the mean returns by their standard deviations, for the dierent strategies. It can be noted that both the MA(6, 1) and MA(12, 1) strategies out-perform the buy and hold strategy as indicated by the Sharpe ratios across all markets. For example, when MA(12, 1) trading rule is applied, the Sharpe ratios for handysize, panamax and capesize investments increased to 0.667, 0.982 and 0.798, respectively, reecting the joint eect of increase in mean returns and reduction in standard deviations of return on the investment in each market. It can also be seen that the gain through such investment strategies and trading rules is greater in the markets for larger vessels due to higher volatility in these markets compared to the market for handysize vessels and better or more frequent trading opportunities arising due to such variation in prices.

Monthly time-charter earnings are estimated on the assumption that the vessel will be on-hire 29 days per month or 348 days per year. The remaining 17 days per year represent time o-hire for repairs and maintenance.

11

2004-05

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143 Table 4 Result of empirical simulation of trading strategies Handysize MA12/MA1 on P/E ratio Mean return St. Dev. Sharpe ratio MA6/MA1 on P/E ratio Mean return St. Dev. Sharpe ratio Buy and hold Mean return St. Dev. Sharpe Ratio 0.08611 0.12907 0.66713 0.07332 0.13346 0.54936 0.08487 0.18406 0.46109 Panamax 0.14814 0.15083 0.98216 0.14017 0.15353 0.91296 0.11459 0.20967 0.54651 Capesize 0.16127 0.20209 0.79800 0.15751 0.20197 0.77983 0.08097 0.25288 0.32019

137

Mean return and St. Dev. are the annualized mean returns (monthly mean return 12) and standard deviation of returns p (monthly standard deviation 12), respectively, for the dierent trading strategies. Sharpe ratio is the ratio of mean returns over the standard deviation of returns. Sample period is January 1976 to September 2004 for handysize and panamax series and April 1979 to September 2004 for the capesize series.

The cumulative returns on the MA(12, 1) trading rule and buy and hold investment strategy in the handysize, panamax and capesize markets are shown in Figs. 35, respectively. The signicant increase in cumulative returns when the active MA(12, 1) trading rule is employed, compared to buy and hold strategy, is evident for the larger vessels (panamax and capesizes). In fact, it is also interesting to note that the proposed trading model correctly identies the buy signal during the lucrative shipping markets of 20032004 when earnings increased sharply compared to ship prices. 4.2. Data snooping and the stationary bootstrap The results in the previous section are encouraging regarding the performance of our proposed trading strategies. However, an important issue which arises when evaluating technical trading rules, is that of data snooping. According to Sullivan et al. (1999) and White (2000) data snooping occurs when a dataset is used more than once for data selection and inference purposes. In other words, using the same dataset frequently for testing trading strategies, may increase the probability of having satisfactory results purely due to chance or due to the use of posterior information rather than the superior ability of the trading strategies. The method most commonly used in the literature to assess the performance of trading strategies and test for data snooping is bootstrap. The bootstrap, introduced by Efron (1979), is a resampling method that uses the empirical distribution of the statistic of interest, rather than the theoretical distribution implied by
1200 1000 800 600 400 200 0
Ja n77 Ja n79 Ja n81 Ja n83 Ja n85 Ja n87 Ja n89 Ja n91 Ja n93 Ja n95 Ja n97 Ja n99 Ja n01 Ja n03

MACum Ret

BHCum Ret

Fig. 3. Cumulative return on MA trading strategy for handysize bulk carrier.

138

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

4500 4000 3500 3000 2500 2000 1500 1000 500 0


Ja n81 Ja n83 Ja n85 Ja n87 Ja n77 Ja n91 Ja n93 Ja n95 Ja n79 Ja n89 Ja n97 Ja n03 Ja n99 Ja n01

MACum Ret

BHCum Ret

Fig. 4. Cumulative return on MA trading strategy for panamax bulk carrier.

4000 3500 3000 2500 2000 1500 1000 500 0


Ap rAp 79 rAp 80 rAp 81 r-8 Ap 2 rAp 83 r-8 Ap 4 rAp 85 r-8 Ap 6 rAp 87 r-8 Ap 8 rAp 89 rAp 90 r-9 Ap 1 rAp 92 r-9 Ap 3 rAp 94 r-9 Ap 5 rAp 96 r-9 Ap 7 rAp 98 rAp 99 r-0 Ap 0 rAp 01 r-0 Ap 2 rAp 03 r-0 4

MACum Ret

BHCum Ret

Fig. 5. Cumulative return on MA trading strategy for capesize bulk carrier.

statistical theory, to conduct statistical inference. The main advantage of bootstrap is that it can approximate the properties of the sampling distribution of the underlying statistic even when such a distribution is not parametrically dened, or the underlying statistic is complex and not easy to obtain. Bootstrap techniques have also been used by Brock et al. (1992), who test whether trading results from some trading rules can be explained by time-series models, as well as Sullivan et al. (1999) who use bootstrap to test the joint performance of several technical rules. However, the ordinary bootstrap method is only valid in the case of iid observations. When ordinary bootstrap techniques are applied to serially dependent observations, as is the case with ship-prices and earnings, the resampled series will not retain the statistical properties of the original dataset and yield inconsistent results and statistical inference (see Ruiz and Pacual, 2002). In view of that, several non-parametric methods for dealing with serially dependent data have been developed. One such method is the stationary bootstrap method of Politis and Romano (1994). This procedure is based on re-sampling blocks of random length, where the length of each block follows a geometric distribution. This way it generates random samples which preserve the serial dependence property of the original series and are also stationary. This is important since our proposed trading strategy relies on the premise that the P/E ratio is stationary; see Appendix A for technical details. Therefore, in order to statistically assess the performance of our trading strategies, we use the stationary bootstrap technique to regenerate random paths that ship prices and earnings may have possibly followed over the sample period, whilst maintaining the distributional properties of the original series. We then implement the proposed trading strategies using the simulated price and earnings series which, in turn, generate a distribution of trading statistics under the dierent trading rules. Therefore, our approach in using bootstrap is dierent from the previous literature in the sense that we bootstrap to generate paths of price and earning series and assess the protability of PE based trading strategies. We start by bootstrapping the log-dierence series. Then these bootstrapped series are transformed back into levels to construct P/E ratios which are used to trigger buy and sell decisions based on the MA trading

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

139

strategies. In implementing these strategies, we consider transaction costs, the depreciation in the value of the vessel and operating prot, along the lines described in the previous section of the paper. As a benchmark model, we also consider the buy and hold strategy in which one is always long in the market and hence benets from a constant stream of income, arising from the operation of the vessel in the charter market, irrespective of the level of the P/E ratio. Both the MA and the buy and hold strategies are implemented for each one of the 1000 bootstrapped series, thus generating a series of empirical distributions of mean returns and Sharpe ratios. Under the null hypothesis that a dynamic trading strategy is no better than a buy and hold strategy or, equivalently, that there is no information or signals in the original P/E ratio, the prot from the MA strategies should be no better than the prot from a buy and hold strategy. The results of the bootstrap simulations, are reported in Table 5. The table contains the mean annual return (obtained as the mean return from the trading strategies implemented on the 1000 bootstrapped series), the standard deviation of the mean returns, as well as the average Sharpe ratio across the bootstrapped series. Several points merit discussion here. First, it can be observed that the mean returns and Sharpe ratios from the dierent strategies are similar to those observed in the empirical series under the same trading rule. Furthermore, their comparative performance is also similar; in other words, in the simulated series, capesize returns are higher than panamax returns and panamax returns are higher than handysize returns. The trading rules based on log-priceearnings ratio seem to out-perform the static investment tactics both in terms of
Table 5 Result of stationary bootstrap simulation of trading strategies Handysize Panel A: mean returns and Sharpe ratios MA12/MA1 on P/E ratio Mean return St. Dev. Sharpe ratio MA6/MA1 on P/E ratio Mean return St. Dev. Sharpe ratio Buy and hold return Mean return St. Dev. Sharpe ratio Panamax Capesize

0.05676 0.02569 0.50964 0.04892 0.02623 0.42321 0.04404 0.04435 0.24268

0.11113 0.02407 0.94118 0.11109 0.02306 0.87162 0.07399 0.03922 0.38178

0.12820 0.03004 0.77713 0.12996 0.02725 0.77702 0.06266 0.03711 0.27708

Panel B: 90% Empirical Condence Intervals and p-values Excess returns MA12 relative to buy and hold (0.0418, 0.0604) {0.329} MA6 relative to buy and hold (0.0485, 0.0516) {0.411} Sharpe ratios MA12 relative to buy and hold MA6 relative to buy and hold (0.0555, 0.5825) {0.084} (0.1367, 0.4845) {0.182}

(0.0241, 0.0853) {0.133} (0.0152, 0.0781) {0.108} (0.1719, 0.9911)** {0.008} (0.1838, 0.7810)** {0.003}

(0.0076, 0.1150)** {0.024} (0.0169, 0.1100)** {0.008} (0.2300, 0.8150)** {0.000} (0.2483, 0.7642)** {0.000}

Results are based on 1000 realisations of the trading strategies based on the stationary bootstrap of Politis and Romano (1994). Mean return and Sharpe ratio are the average mean returns and Sharpe ratios across 1000 simulations. St. Dev. is the standard deviation of mean returns across 1000 simulations. p-Values are in {} and measure the signicance level for which we can reject a one-tail test on the null that mean returns or Sharpe ratios are not dierent between the MA and the buy and hold strategies. 90% Empirical Condence Intervals are in brackets (). Numbers in [] are t-statistics; DF are the degrees of freedom for the causality tests. Sample period is January 1976 to September 2004 for handysize and panamax series and April 1979 to September 2004 for the capesize series.

140

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

increasing average returns and in terms of the Sharpe ratios. For instance, a comparison of Sharpe ratios in Table 5 reveals a threefold increase in capesize and panamax markets and a twofold increase in handysize vessels, compared to the buy and hold strategy, when the MA(12, 1) trading rule is used. More formal statistical tests are conducted by considering the empirical condence intervals for excess returns and Sharpe ratios. More specically, for each simulated series we estimate the excess return of the MA trading strategy relative to the buy and hold strategy as well as the excess performance of the Sharpe ratio relative to the Sharpe ratio in the buy and hold strategy. We then construct 90% empirical condence intervals for the excess returns based on the bootstrap simulations to test whether excess returns are signicantly different from zero (or in other words, whether the P/E based trading strategies provide signicantly higher returns compared to the static strategy). These are constructed as the 5% and 95% percentiles of the ordered excess returns series. If the value of zero is not contained within the condence interval then returns under the MA strategy are signicantly better, at the 10% level, compared to the buy and hold strategy. For comparison purposes, we also construct the empirical p-values for the tests. These are simply calculated as the ratio of frequency of occurrence of negative excess returns over the total number of simulations (1000 replications) and reect the signicance level, for which the null hypothesis that there is not signicant dierence between the returns can be rejected, using a one-tail test. Overall, these results indicate that the MA strategies provide signicant increases in Sharpe ratios compared to the ordinary buy and hold strategies for the panamax and capesize markets. Figs. 68 plot the distributions of simulated returns of MA(12, 1), MA(6, 1) and static trading strategies for handysize, panamax and capsize markets, respectively. These graphs clearly illustrate the benets of using trading signals derived from ship prices and earnings in shipping investment, as the distribution of simulated returns based on MA rules show signicant shifts to the right with relatively lower dispersion. Overall, our simulation analysis conrms that the relationship between price and earnings in shipping markets contains important information about the future behaviour of ship prices. This relationship reects and responds to changes in market fundamentals and deviations from this relationship can be used to identify investment and divestment timing in shipping markets and, hence, exploit signicant returns in the market. Our results also reveal that the combination of technical trading and fundamental analyses could be more
16 14 12 Density 10 8 6 4 2 0 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20 0.25

HOLD_STRATEGY MA6_STRATEGY MA12_STRATEGY

Fig. 6. Comparison of simulated return distribution of active and passive trading strategies in the handysize market.

17.5 15.0 12.5 Density 10.0 7.5 5.0 2.5 0.0 -0.05

HOLD_STRATEGY MA6_STRATEGY MA12_STRATEGY

0.00

0.05

0.10

0.15

0.20

0.25

Fig. 7. Comparison of simulated return distribution of active and passive trading strategies in the panamax market.

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143


16 14 12 Density 10 8 6 4 2 0 -0.05 0.00 0.05 0.10 0.15 0.20 0.25

141

HOLD_STRATEGY MA6_STRATEGY MA12_STRATEGY

Fig. 8. Comparison of simulated return distribution of active and passive trading strategies in the capesize market.

eective in the market for larger vessels (panamax and capesize) due to higher volatility and price uctuations in these markets compared to the market for handysize ships. This can be attributed to the fact that the markets for smaller ships are more ecient in the sense that there are more participants and these ships are more exible to operate in terms of the cargo that they can carry and the routes in which they operate. It can also be argued that while larger ships could be more suitable for the purpose of asset play in the shipping markets, the timing of investment is of crucial importance as potential gains and losses are higher in these sectors of the shipping industry. 5. Summary and conclusions In this study we propose a new approach for timing investment and divestment decisions in shipping markets. In particular, we utilise the relationship between variables in shipping markets and devise strategies to identify the timing for sale and purchase of merchant ships. The theoretical relationship between ship prices and TC earnings, based on the discounted present value model, is discussed in detail and a cointegration relationship is established between ship prices and earnings. Based on this cointegration relationship, we develop a trading strategy which measures the deviation of the P/E ratio, which is also the cointegrating vector, from its long-run equilibrium and signals sale and purchase opportunities using moving average trading rules. Such strategies are then applied to historical series which reveal promising results when compared with static buy and hold strategies. To ensure the consistency of our model, we also perform simulations which conrm the superiority of the MA trading rules further. Overall, our results show that the relationship between price and earnings in shipping markets contains important information about future behaviour of ship prices, which can be used for investment timing in shipping markets. We also show that investors in shipping markets can benet from applying technical trading rules when making sale and purchase decisions. Acknowledgements We would like to thank Professor Ken Small and three anonymous referees for their extremely helpful comments. This paper has also largely beneted from the comments of participants at the 2004 Shipping Investment Seminar at Cass Business School, at the Hong Kong Shipowners Association seminar and at the 2005 International Association of Maritime Economists Conference, Limassol, Cyprus where it was awarded the Most Innovative Paper of the Conference prize. Appendix A Here we present the algorithm that is used to implement the stationary bootstrap resampling technique of Politis and Romano (1994). The description of the algorithm here follows from Appendix C of Sullivan et al. (1999).

142

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

The stationary bootstrap is calculated as follows: Given the original sample of T observations, X(t), t = {1, . . . , T}, we start by selecting a smoothing parameter, q = qT, 0 < qT 6 1, TqT ! 1 as T ! 1, and then form the bootstrapped series, X(t)*, as follows: 1. At t = 1, select X(1)* at random, independently and uniformly from {X(1), . . . , X(T)}. Say for instance that X(1)* is selected to be the Jth observation in the original series, X(1)* = X(J) where 1 6 J 6 T. 2. Increment t by 1. If t > T, then stop. Otherwise draw a standard uniform random variable U independently of all other random variables (a) if U < q, then select X(2)* at random, independently and uniformly from {X(1), . . . , X(T)}, (b) if U > q, then expand the block by setting X(2)* = X(J + 1), so that the X(2)* is the next observation in the original series following X(J). If J + 1 > T, then reset J + 1 to 1, so that the block continues from the rst observation in the sample. 3. Repeat step 2 until we reach X(T)*. 4. Repeat steps 13, 1000 times. Therefore, the stationary bootstrap re-samples blocks of varying length from the original data, where the block length follows a geometric distribution, with mean block length 1/q. In general, given that X(t)* is determined by the Jth observation X(J) in the original series, then X(t + 1)* will be equal to the next observation in the block X(J + 1) with probability 1-q and picked at random from the original observations with probability q. Regarding the choice of q, a large value of q is appropriate for data with little dependence, and a smaller value of q is appropriate for data that exhibit more serial dependence. The value of q chosen in our experiments is 0.1, corresponding to a mean block length of 10. This follows other studies in the literature, most notably Sullivan et al. (1999). Furthermore, we also perform sensitivity tests with dierent values of q, and nd that the results presented in this section are not sensitive to the choice of q. References
Adland, R.O., 2000. Technical trading rule performance in the second-hand asset markets in bulk shipping. Foundation for Research in Economics and Business Administration, Bergen, Norway, Working Paper No. 04/2000. Adland, R.O., Koekebakker, S., 2004. Market eciency in the second-hand market for bulk ships. Maritime Economics and Logistics 6 (1), 115. Beenstock, M., Vergottis, A., 1989. An econometric model of the world market for dry cargo freight and shipping. Applied Economics 21, 339359. Bendall, H.B., Stent, A., 2004. Ship investment under uncertainty: a real option approach. Working Paper, University of Technology, Sydney, Australia. Brock, W., Lakonishok, J., LeBaron, B., 1992. Simple technical trading rules and the stochastic properties of stock returns. Journal of Finance 47 (5), 17311764. Campbell, J.Y., Shiller, R.J., 1987. Cointegration and test of present value models. Journal of Political Economy 95, 10621088. Campbell, J.Y., Shiller, R.J., 1998. Valuation ratios and the long-run stock market outlook. The Journal of Portfolio Management 24 (2), 1126. Dixit, A.K., Pindyck, R.S., 1994. Investment Under Uncertainty. Princeton University Press. Efron, B., 1979. Bootstrap methods: another look at the jackknife. Annals of Statistics 7, 126. Engle, R.F., 1982. Autoregressive conditional heteroskedasticity with estimates of the variance of United Kingdom ination. Econometrica 50 (4), 9871008. Engle, R.F., Granger, C.W., 1987. Cointegration and error correction: representation, estimation, and testing. Econometrica 55 (2), 251 276. Fama, E., French, K., 1992. The cross-section of expected stock returns. Journal of Finance 46 (2), 427466. Fuller, R.J., Kling, J.L., 1990. Is the stock market predictable? The Journal of Portfolio Management 15 (4), 2836. Fuller, R.J., Kling, J.L., 1994. Can regression-based models predict stock and bond returns. The Journal of Portfolio Management 19 (3), 5663. Fuller, R.J., Huberts, L.C., Levinson, M.J., 1993. Returns to E/P strategies, HiggledyPiggledy growth, analysts forecast errors, and omitted risk factors. The Journal of Portfolio Management 19 (2), 1324. Glen, D.R., 1997. The market for second-hand ships: further results on eciency using cointegration analysis. Maritime Policy and Management 24 (3), 245260. Granger, C., 1969. Investigating causal relations by econometric models and cross-spectral methods. Econometrica 37 (3), 424438. Granger, C., 1986. Developments in the study of cointegrated variables. Oxford Bulletin of Economics and Statistics 48 (3), 213227.

A.H. Alizadeh, N.K. Nomikos / Transportation Research Part B 41 (2007) 126143

143

Hale, C., Vanags, A., 1992. The market for second-hand ships: some results on eciency using cointegration. Maritime Policy and Management 19 (1), 3139. Harris, R., 1995. Using Cointegration Analysis in Econometric Modelling. Prentice Hall/Harvester Wheatsheaf, UK. Jae, J., Keim, D., Westereld, R., 1989. Earnings yields, market values, and stock returns. Journal of Finance 44 (1), 135148. Jarque, C.M., Bera, A.K., 1980. Ecient test for normality, homoscedasticity and serial dependence of regression residuals. Economics Letters 6, 255259. Johansen, S., 1988. Statistical analysis of cointegration vectors. Journal of Economic Dynamics and Control 12 (2/3), 231254. Johansen, S., 1991. Estimation and hypothesis testing of cointegration vectors in Gaussian vector autoregressive models. Econometrica 59 (6), 15511580. Kavussanos, M.G., Alizadeh, A.H., 2002a. Ecient pricing of ships in the dry bulk sector of the shipping industry. Maritime Policy and Management 29 (3), 303330. Kavussanos, M.G., Alizadeh, A., 2002b. The expectations hypothesis of the term structure and risk premia in dry bulk shipping freight markets; An EGARCH-M approach. Journal of Transport Economics and Policy 36 (2), 267304. Kwiatkowski, D.P., Phillips, C.B., Schmidt, P., Shin, Y., 1992. Testing the null hypothesis of stationarity against the alternative of a unit root. Journal of Econometrics 54 (1/2/3), 159178. Lander, J., Orphanides, A., Douvogiannis, M., 1997. Earnings forecasts and the predictability of stock returns: evidence from trading the S&P. The Journal of Portfolio Management 23 (4), 2435. Lee, C.I., Gleason, K.C., Mathur, I., 2000. Eciency tests in the French derivatives market. Journal of Banking and Finance 24 (5), 787807. Ljung, G.M., Box, G.E.P., 1978. On a measure of lack of t in time series models. Biometrika 65, 297303. Maddala, G.S., Kim, I., 1998. Unit Roots, Cointegration and Structural Change. Cambridge University Press, UK. Newey, W.K., West, K.D., 1987. A simple positive denite heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55 (3), 703708. Osterwald-Lenum, M., 1992. A note with the quantiles of the asymptotic distribution of the ML cointegration rank test statistics. Oxford Bulletin of Economics and Statistics 54 (3), 461472. Pesaran, M.H., Timmermann, A., 1995. Predictability of stock returns: robustness and economic signicance. Journal of Finance 50 (4), 12011228. Phillips, P.C.B., Perron, P., 1988. Testing for a unit root in time series regressions. Biometrica 75, 335346. Politis, D.N., Romano, J.P., 1994. The stationary bootstrap. Journal of the American Statistical Association 89 (428), 13031314. Roll, R., 1994. Style return dierentials: illusions, risk Premia, or investment opportunities? Working Paper, University of California, Los Angeles. Ruiz, E., Pacual, L., 2002. Bootstrapping nancial time series. Journal of Economic Surveys 16 (3), 271300. Schwarz, G., 1978. Estimating the dimension of a model. Annals of Statistics 6, 461464. Strandenes, S.P., 1984. Price determination in the time-charter and second-hand markets. Working Paper No. 06, Centre for Applied Research, Norwegian School of Economics and Business Administration. Sullivan, R., Timmermann, A., White, H., 1999. Data-snooping, technical trading rule performance, and the bootstrap. Journal of Finance 54 (5), 16471691. Tsolakis, S.D., Cridland, C., Haralambides, H.E., 2003. Econometric modelling of second-hand ship prices. Maritime Economics and Logistics 5 (4), 347377. Tvedt, J., 1997. Valuation of VLCCs under income uncertainty. Maritime Policy and Management 24 (2), 159174. White, H., 2000. A reality check for data snooping. Econometrica 68 (5), 10971126.

You might also like