Forecasting short-term freight rate cycles: do we have a more appropriate method than a normal distribution?

References and notes

• Stopford , M . 2009 . Maritime Economics, , 3rd , Oxon, , UK : Routledge .
   Google Scholar
• For example, many investment decisions made during the dot-com crisis in the late 1990s were based on the wrong assumption that internet businesses would be developed in a shorter time scale
   Google Scholar
• These are not, of course, the easy situations as when we have to predict a liquid natural gas project secured by long-term cargo contract
   Google Scholar
• Liner companies, e.g., in the 1950s and 1960s had to adapt, in order to survive, to low-cost air travel. They also faced the fact that many colonies gained independence and cargo was containerized after 1956
   Google Scholar
• Goulielmos , AM . 2002 . Business Management: Introduction for Shipping Business Executives to Nonlinear Management , Athens-Piraeus, , Greece : Stamoulis Publications . ISBN 960-351-384-9
   Google Scholar
• Onassis in 1956 achieved a profit of $80 million due to the sudden closure of Suez Canal. Onassis was wrong, however, to believe that Suez will remain closed for long; Costas Gratsos—his aide—thought the opposite, and he secretly chartered 12 tankers to Esso for 39 months. In April 1957, the canal reopened. Gratsos’ forecast was right
   Google Scholar
• Goulielmos, A. M., 2009, Risk analysis of the Aframax freight market and of its new building and second hand prices, 1976-2008 and 1984–2008. International Journal of Shipping and Transport Logistics, 1(1), 74–97
   Google Scholar
• Goulielmos , AM and Goulielmos , MA . 2009 . The problem of timing in decisions to buy or to charter a vessel . International Journal of Transport Economics , XXXVI ( 2 ) : 261 – 286 .
   Google Scholar
• The price of a 4000 TEU containership (gearless, 5 years old) between 2007 and 2010 (first quarter) lost $29.8 million, from a price of $60.3 million (data from Drewry, 2010) in 2007
   Google Scholar
• Keynes , JM . 1936/1961 . The General Theory of Employment Interest and Money , London, , UK : MacMillan & Co. Ltd .
   Google Scholar
• A frequently used model is the generalized autoregressive conditional heteroskedasticity—GARCH (1986), where the variance of observations is allowed to change over time based on their past behavior. This is a generalization of ARCH, which appeared in 1982. This is also an indication of the problems that ARCH has encountered as that was also based on normal distribution
   Google Scholar
• Soros , G . 2008 . The New Paradigm for Financial Markets , Athens, , Greece : Livanis Editions . Greek translation, ISBN 978-960-14-1835-3
   Google Scholar
• Goulielmos , AM and Psifia , M-E . 2009 . Forecasting weekly freight rates for one-year time charter 65000 dwt bulk carrier, 1989–2008, using nonlinear methods . Maritime Policy and Management , 36 ( 5 ) : 411 – 436 .
   Web of Science ®Google Scholar
• Mandelbrot , B . 1963 . The variation of certain speculative prices . Journal of Business , 36 ( 4 ) : 394 – 419 .
   Web of Science ®Google Scholar
• Mandelbrot , BB . 1972 . “ Statistical models and turbulence ” . In Lecture Notes in Physics , Edited by: Rosenblant , M. and Van Atta , C. Vol. 12 , 333 – 335 . New York : Springer .
   Google Scholar
• Mandelbrot , BB and Wallis , JR . 1969 . Some long run properties of Geophysical records . Water Resources Research , 5 : 321 – 340 .
   Web of Science ®Google Scholar
• Mandelbrot , BB and Taqqu , M . 1979 . Robust R/S Analysis of long run serial correlation . Bulletin of the International Statistical Institute , 48 : 59 – 104 .
   Google Scholar
• Mandelbrot , BB and Hudson , R . 2004 . The (Mis) Behavior of Markets: A Fractal View of Risk, Ruin and Reward , New York : Basic Books .
   Google Scholar
• Einstein, A., 1905, Uber die von der molekularkinetischen Theorie der Warme geforderte Bewegung von in rehenden Flussigkeiten suspendierten Teilchen, Annala of Physics 322
   Google Scholar
• A random object runs a distance at the square root of its time. This was the mathematical proof of the Brownian motion or random walk going back to 1827. It was Yule [21] who invented autoregressive (AR) models, which developed into GARCH and its many other variations
   Google Scholar
• Yule , GU . 1927 . On a method of investigating periodicities in disturbed series, with special reference to Wolfer’s sunspot numbers . Philosophical Transactions of the Royal Society , 226 : 267 – 298 .
   Google Scholar
• This is, in the simple language of Mandelbrot and Hudson [18], a measure to compare the extent to which real data fits the bell curve. The bell curve has three units kurtosis. The fat tail, which is common in real life, shows kurtosis over three. Schoutens [23] found for the Standard and Poor’s 500 1970–2001 kurtosis of 43.36; 7.17 for the Dow Jones on Black Monday; 5.78 for Nasdaq and 4.63 for the French CAC-40
   Google Scholar
• Schoutens , W . 2003 . Levy Process in Finance: Pricing Financial Derivatives , Chichester, , UK : John Wiley and Sons, Inc. .
   Google Scholar
• As shown by Fama [25], large variations greater than five standard deviations from the mean occurred 2000 times more frequently than expected [18]. This, for a normal distribution, occurs once in 7000 years. In reality, it occurs once every 3–4 years
   Google Scholar
• Fama , EF . 1965 . The behavior of stock-market prices . Journal of Business , 38 ( 1 ) : 34 – 105 .
   Web of Science ®Google Scholar
• Name made up by the first letters of the surnames of the economists who invented this test, i.e., Brock–Dechert–Scheinkman [27]. With the BDS-statistic the difference in the dispersion of the observations is tested, in spaces for a dimension from 2 to n, compared with white noise (iid)
   Google Scholar
• Brock, W., Dechert, W. and Scheinkman, J., 1986, A Test of Independence Based on the Correlation Dimension. SSRI Report No 8702. University of Houston, Wisconsin and Chicago, Department of Economics
   Google Scholar
• McConville , J and Rickaby , G . 1995 . Shipping Business and Maritime Economics: An Annotated International Bibliography , London, , UK : Mansell .
   Google Scholar
• If one excludes works by organizations like Drewry, Lloyd’s Shipping Economist, Terminal Operators, and UNCTAD, the percentage is 0.5. In this percentage, we had an MSc in 1983 and a PhD in 1990
   Google Scholar
• Hearth , J . 1970 . Forecasting Shipping Demand, October , London, , UK : Shipbuilding and Shipping Record .
   Google Scholar
• Seland, J., 1960, On Forecasts and Forecasters: Pitfalls in Predicting Tonnage Demand, January (London, UK: The Shipping World)
   Google Scholar
• Nielsen, R. E., 1959, Oil Tanker Economics (Bremen, W. Germany: Institute of Shipping Economics)
   Google Scholar
• Goulielmos, A. M., 1974, A vintage model approach to some problems of shipping economics. Unpublished PhD thesis, University of Brunel, Uxbridge, UK
   Google Scholar
• Koopmans , TC . 1939 . Tanker Freight Rates and Tankship Building: An Analysis of Cyclical Fluctuations , Haarlem, , the Netherlands : De Erven Bohn; London: PS King and Staples .
   Google Scholar
• Koopmans [34] and Einarsen [36] were criticized by Zannetos [37] that ship retirements and ship replacements were not related, as Koopmans wrongly said that: “a new investment means also replacement”
   Google Scholar
• Einarsen, J., 1938, Reinvestment Cycles and their Manifestation in the Norwegian Shipping Industry (Bergen, Norway: University Institute of Economics)
   Google Scholar
• Zannetos , Z . 1966 . The Theory of Oil Tankship Rates , Boston, MA : MIT Press .
   Google Scholar
• “Maritime” covers ships, ports, and yards Goulielmos [39]
   Google Scholar
• Goulielmos, A. M., 2001, Matters of Definition: Maritime Economics Revisited, scientific yearbook honorary in favor of late Prof. D. Kodosakis (Greece: University of Piraeus), pp. 211–28
   Google Scholar
• Tinbergen, J., 1959, A shipbuilding cycle?, Selected readings, first appeared in 1931. 1934, Scheepsbouwruimte en vrachten, De Nederlandse Conjunctuur, March 23–35
   Google Scholar
• Nobel laureate in 1969 for Econometrics
   Google Scholar
• Goulielmos and Kaselimi [43] reviews four to five works that tried to forecast port traffic. This paper will appear in the International Journal of Shipping and Transport Logistics, Vol. 3
   Google Scholar
• Goulielmos, A. M. and Kaselimi, V., 2011, A non-linear forecasting of container traffic: the case-study of the Port of Piraeus, 1973–2008. International Journal of Shipping and Transport Logistics, 3(1), 72–99
   Google Scholar
• Strandenes, S. P., 1984, Price determination in the time-charter and second-hand markets, WP 6, Centre of Applied Research, Norwegian School of Economics and Business Administration
   Google Scholar
• Kavussanos and Alizadeh [46] criticized her for not taking into account the stochastic properties of her variables and for not taking 20 years data, at least, to examine a full shipping cycle. These authors assume a single 20 years cycle, which has ignored short-term cycles for 4 years or even long waves of 54 years (Kondratieff waves)
   Google Scholar
• Kavussanos , MG and Alizadeh , AH . 2002 . Efficient pricing of ships in the dry bulk sector of the shipping industry . Maritime Policy and Management , 29 ( 3 ) : 303 – 330 .
   Google Scholar
• Goulielmos , AM and Psifia , M-E . 2009 . Is history repeated? Cycles and recessions in shipping markets, 1929 and 2008 . International Journal of Shipping and Transport Logistics , 1 ( 4 ) : 329 – 360 .
   Web of Science ®Google Scholar
• Beenstock , M . 1985 . A theory of ship prices . Maritime Policy and Management , 12 : 215 – 225 .
   Google Scholar
• He was criticized too by [46] for short data and for ignoring stochasticity
   Google Scholar
• Vergottis, A., 1988, An econometric model of world shipping. PhD thesis, City Univeristy Business School, London, UK
   Google Scholar
• He also did not take into account the stochastic properties of his variables (as argued by [46])
   Google Scholar
• Hale , C and Vanags , A . 1992 . The market for second hand ships: some results on efficiency using cointegration . Maritime Policy and Management , 19 : 131 – 140 .
   Google Scholar
• This means information available for historical prices only
   Google Scholar
• The time was rather short (<20 years). Their formulation was inappropriate to test EMH, as co-integration (due to Engle and Granger [55]) is not a sufficient, but a necessary, condition, as argued by [46]
   Google Scholar
• Engle , RF and Granger , CWJ . 1987 . Cointegration and error correction representation, estimation, and testing . Econometrica , 55 : 251 – 276 .
   Web of Science ®Google Scholar
• Glen , D . 1997 . The market for second hand ships: further results on efficiency using cointegration analysis . Maritime Policy and Management , 24 : 245 – 260 .
   Google Scholar
• Johansen , S . 1988 . Statistical analysis of cointegration vectors . Journal of Economic Dynamics and Control , 12 : 231 – 254 .
   Web of Science ®Google Scholar
• They came to mixed results, due to same mistakes found also in Hale and Vanags, as mentioned by [46]
   Google Scholar
• Li , J and Parsons , MG . 1997 . Forecasting tanker freight rate using neural networks . Maritime Policy and Management , 24 ( 1 ) : 9 – 30 .
   Google Scholar
• These terms and acronyms are defined below
   Google Scholar
• A computer-based simulation, considered as imitating the human nervous system that can learn from experience
   Google Scholar
• Campbell , JY and Shiller , RF . 1987 . Cointegration and test of present value models . Journal of Political Economy , 95 : 1062 – 1088 .
   Web of Science ®Google Scholar
• Campbell , JY and Shiller , RF . 1988 . Stock prices, earnings, and expected dividends . Journal of Finance , 43 : 661 – 676 .
   Web of Science ®Google Scholar
• Fama , EF . 1991 . Efficient capital markets II . Journal of Finance , 46 ( 5 ) : 1575 – 1617 .
   Web of Science ®Google Scholar
• Jarque , CM and Bera , AK . 1981 . Efficient tests for normality, homoscedasticity and serial independence of regression residuals: Monte Carlo evidence . Economics Letters , 7 ( 4 ) : 313 – 318 .
   Web of Science ®Google Scholar
• Bendall , H and Stent , AF . 2003 . Investment strategies in market uncertainty . Maritime Policy and Management , 30 ( 4 ) : 293 – 303 .
   Google Scholar
• Kavussanos , M . 2003 . Time varying risks among segments of the tanker freight markets . Maritime Economics and Logistics , 5 ( 3 ) : 227 – 250 .
   Google Scholar
• Lyridis , DV , Zacharioudakis , P , Mitrou , P and Mylonas , A . 2004 . Forecasting tanker market using artificial neural networks . Maritime Economics and Logistics , 6 : 93 – 108 .
   Google Scholar
• Engelen , S , Meersman , H and Van de Voorde , E . 2006 . Using system dynamics in maritime economics: an endogenous decision model for ship-owners in the dry bulk sector . Maritime Policy and Management , 33 ( 2 ) : 141 – 158 .
   Google Scholar
• The model started as a business game in 2005 between Antwerp and Delft Technical University and was presented at the 2005 IAME conference
   Google Scholar
• Sterman , JD . 2000 . Business Dynamics: Systems Thinking and Modeling for a Complex World , New York : McGraw-Hill .
   Google Scholar
• Goulielmos , AM and Psifia. , M-E . 2006 . Variations in charter rates for a time series between 1971 & 2002: can we model them as an effective tool in shipping finance? . International Journal of Transport Economics , XXXIII ( 2 ) : 257 – 278 .
   Google Scholar
• Goulielmos , AM and Psifia , M-E . 2006 . Shipping finance: time to follow a new track? . Maritime Policy and Management , 33 ( 3 ) : 301 – 320 .
   Google Scholar
• Goulielmos , AM and Psifia , M-E . 2007 . A study of trip and time charter freight rate indices: 1968–2003 . Maritime Policy and Management , 34 ( 1 ) : 55 – 67 .
   Google Scholar
• Jing , L , Marlow , PB and Hui , W . 2008 . An analysis of freight rate volatility in dry bulk shipping markets . Maritime Policy and Management , 35 ( 3 ) : 237 – 251 .
   Google Scholar
• GARCH is not appropriate if time series have a memory
   Google Scholar
• Exponential GARCH
   Google Scholar
• Engelen, S., Norouzzadeh, P., Dullaert, W. and Rahmani, B., 2011, Multifractal features of spot rates in the Liquid Petroleum Gas shipping market. Energy Economics, 33(1), 88–98
   Google Scholar
• Thalassinos, E. I., Hanias, M. P., Curtis, P. G. and Thalassinos, Y. E., 2009, Chaos theory: forecasting the freight rate of an oil tanker. International Journal of Computational Economics and Econometrics, 1(1), 76–88
   Google Scholar
• Hurst , HE . 1951 . The long-term storage capacity of reservoirs . Transactions of the American Society of Civil Engineers , 116 : 770 – 799 .
   Google Scholar
• Greene , MT and Fielitz , BD . 1977 . Long term dependence in common stock returns . Journal of Financial Economics , 4 : 339 – 349 .
   Web of Science ®Google Scholar
• Peters, E. E., 1994, Fractal Market Analysis: Applying Chaos Theory to Investment and Economics (New York: Wiley Finance Edition)
   Google Scholar
• The bias is created by investors in capital markets as they react in the current economic conditions until new information comes in and modifies the previous bias
   Google Scholar
• This can be used to reveal periodic cycles by plotting log R/S to log n [82] (pp. 91–92). Vn , however, provides an easier way and gives a better estimate of the cycle length and is also more precise, even when noise is present. Hurst [80] used Vn to test the stability [82] (p. 92)
   Google Scholar
• This means to divide the adjusted range by the local standard deviation. This was considered a master stroke [82] (p. 57). In simple language, rescaled range (R/S) is a name short for “range (R) divided by standard deviation (S)” [18] (p. 202). Dividing by standard deviation is described as normalizing, i.e., “volatility is normalized.” This gives another name of this method—the normalized volatility method
   Google Scholar
• Between (R/S) n and there is a linear relationship of the classical form: Y = a + bX
   Google Scholar
• Adjusted to a mean of zero and this is always non-negative
   Google Scholar
• n is treated as a distance index or time index of time series
   Google Scholar
• If , then observations are not identically and independently distributed. In the case of non-independent and non-identical observations, the present influences the future. Mathematically, this last relationship is expressed by: (*), where C is a correlation measure. If the system follows a random walk, then the correlation (C ) between present and future is zero, as there is no dependence between observations and the system is random. The method helps to see if an apparently random time series can be characterized by long-term memory. In fact, it is based on the range, which is rescaled by removing any trend (i.e., it is de-trended). (*) This measure is a correlation of the increments; it has been defined by [82] (pp. 184–185). It expresses the correlation of the changes in position of a process over time t with all increments of time t that precede and follow it
   Google Scholar
• If the time series was increasing in the previous time period, it will keep increasing in the present and vice versa
   Google Scholar
• Steeb , W-H . 2008 . The Nonlinear Workbook, , 4th , Singapore : World Scientific .
   Google Scholar
• It is called an object in which the parts are in some way related to the whole, or the whole has self-similar components. If objects are solid and continuous, as in Euclidean plane geometry, they have integer dimension. A probability density function that is statistically self-similar is a fractal distribution. This means that in different increments of time, the statistical characteristics remain the same. The fractal market hypothesis is valid when many investors have different investment horizons and the information set, important to each investment horizon, is different. The market becomes unstable if all investors have the same information set [82] (p. 309)
   Google Scholar
• It is such a system if it has a fractal dimension and exhibits sensitive dependence on initial conditions
   Google Scholar
• Lo , AW . 1991 . Long-term memory in stock market prices . Econometrica , 59 : 1279 – 1313 .
   Web of Science ®Google Scholar
• It shows the probability that two points are within a certain distance from one another
   Google Scholar
• Siriopoulos , C . 1998 . Analysis and Tests of Univariable Financial Time Series, in Greek , Athens, , Greece : G Dardanos editions, Tipothyto .
   Google Scholar
• Small , M . 2005 . Applied Nonlinear Time Series Analysis: Applications In Physics, Physiology And Finance, Series A , Vol. 52 , Singapore : World Scientific .
   Google Scholar
• (1) AR 1927, autoregressive process, a stationary stochastic process where the current value of a time series is related to the past p values (p = any integer). An AR (1) process has an infinite memory. (2) ARMA, autoregressive moving average process, a stationary stochastic process that can be a mixed model of AR and MA. MA, a stationary stochastic process in which the observed time series is the result of the MA of an unobserved random time series. (3) ARIMA, autoregressive integrated moving average process, a non-stationary stochastic process related to ARMA. (4) ARCH 1982, an AR with conditional heteroskedasticity, a non-linear stochastic process where the variance is time varying and conditional upon the past variance, with high peaks at the mean and fat tails. These are models for the analysis of the predictability of variance introduced by Engle, Bollerslev, Nelson, and others. (5) The GARCH 1986, as mentioned, means the generalized ARCH. It refers to a set of statistical tools to model data whose variability changes with time. The changes in variability are controlled by the data’s own past behavior and it has been generalized toaccommodate more circumstances as a further development of initial ARCH. (6) EGARCH 1991, exponential GARCH
   Google Scholar
• This initially means that the error terms are not normal and independently distributed with expected mean 0 and variance equal to σ 2. We may symbolize this by: This is called “normality assumption” in the classical regression model [100] (p. 23). A second variant is This means that errors are not only iid, including the assumptions of (7), but also cover the further assumption that all characteristics of the error distributions are identical, including all moments (and not only Mean and Variance as in (7)). Random variables by satisfying (7) satisfy also (8). Independence here means zero covariance, but the reverse is not true, unless there is a multivariate normality. Two random variables with zero covariance are uncorrelated
   Google Scholar
• Stewart , J and Gill , L . 1998 . Econometrics, , 2nd , Europe : Prentice-Hall .
   Google Scholar
• Details of the concepts of kurtosis and skewness are in the Appendix
   Google Scholar
• de Vries, C. G., 2001, Fat tails and the history of the guilder, Tinbergen Magazine, Fall(4), 3–6
   Google Scholar
• Bretton Woods: A New Hampshire resort in the US where UN Monetary and Financial Conference was convened in 1944
   Google Scholar
• Pareto, V., 1896, La courbe de la répartition de la Richesse. Reprinted in Oeuvres Completes, 1966, Vol. 1, GENEVE, Librairie Droz. Lausanne, Switzerland
   Google Scholar
• Pareto, V., 1897, Cours d’Economie Politique. Reprinted in Oeuvres Completes, 1966, Vol. 1, GENEVE, Librairie Droz. Lausanne, Switzerland
   Google Scholar
• Krugman, P., 1998, Rashomon in Connecticut. State Magazine, 2nd October
   Google Scholar
• Indicates the amount by which a stock reacts to the market. Beta = 1 means the stock moves in lockstep with the market overall and so it does not magnify market risk
   Google Scholar
• Johansson, A., 1995, Pricing skewness and kurtosis risk on the Swedish stock market. Master thesis, Department of Economics, Lund University of Sweden, September
   Google Scholar
• Fang , H and Lai , T-Y . 1997 . Co-kurtosis and capital asset pricing . The Financial Review , 32 ( 2 ) : 293 – 307 .
   Google Scholar
• Normal distribution or Gaussian distribution is a continuous probability one that describes data clustering around the mean
   Google Scholar
• Skewness is a measure of the asymmetry of the probability distribution of a real-valued random variable. The coefficient of excess skewness is: (=0 for symmetry), where μ 3 is the third moment and μ 2 the second moment. If γ 1 > 0, the long tail is on the right of the distribution (positive skew), as here, like χ 2 distribution. First moment is the mean, second the variance, third skewness and fourth kurtosis
   Google Scholar
• Kurtosis is a measure of the “peakedness” of the probability distribution of a real-valued random variable. The coefficient of excess kurtosis is: γ 2 = μ 4/ − 3 where μ’s are the second and fourth moment. If γ 2 ≤ 0 the distribution is platykurtic with fat in the center and being thinner in the tails than normal and if γ 2 ≥ 0 means leptokurtosis, i.e., thinner in the center and fatter tails, as in many occasions in shipping. Two kinds of excess kurtosis are shown (Appendix)
   Google Scholar
• Standard and Poor’s 500 had a kurtosis of 43.36, 1970–2001; Nasdaq Index 5.78; French CAC-40 4.63, all > 3 [18] (p. 96)
   Google Scholar
• A higher probability than a normally distributed variable with values near the mean
   Google Scholar
• This is a new method. An investor decides first the level of certainty he/she wishes, say 95% probability for losses to be below the risk point [18]
   Google Scholar
• Range = 89 842.5. Minimum = 6655. Maximum 96 497.5. Standard deviation = 16 552
   Google Scholar
• At this point, we would like to note that when analyzing time series, it is necessary for the researcher to subtract the trend, and transform the series into a stationary, non-linear system. A stationary non-linear system is the one that is not subject to temporal dependence based on some outside influence. This process is produced by using “filters” that convert the time-series linearly. The method of first logarithmic difference is used here rather than other filters
   Google Scholar
• Jarque, C. M. and Bera, A. K., 1980, Efficient tests for normality, homoscedasticity and serial independence of regression residuals. Economics Letters, 6(3), 255–259
   Google Scholar
• Brock, W., Dechert, D., Scheinkman, J. and LeBaron, B., 1996, A test for Independence based on the correlation dimension. Econometric Reviews, 15(3), 197–235
   Google Scholar
• Brock , W , Hsieh , D and LeBaron , B . 1991 . Nonlinear Dynamics, Chaos and Instability. Statistical Theory and Economic Evidence , Cambridge : The MIT Press .
   Google Scholar
• As in our sample
   Google Scholar
• Kanzler, L., 1999, Very fast and correctly sized estimation of the BDS statistic. [email protected], http://userd.ox.ac.uk/~econlrk
   Google Scholar
• This is the dimension of a system found by embedding as used in dynamic systems
   Google Scholar
• If the process was anti-persistent ( ), the ratio would be decreasing as n values increase and the V-statistic plot will have a negative slope. Finally, if the system followed random walk (H = 0.5), and thus no correlation between observations existed, the ratio would be steadily increasing at the square root of time and its plot would be a horizontal line (time = number of observations, n)
   Google Scholar
• Hampton, M. J., 1990, published in 1989, ‘Analysis and shipping cycles I and II’, Seatrade Journal, 19–23. Long and Short Shipping Cycles: The Rhythms and Psychology of Shipping Markets, Monograph, 2nd ed. (Cambridge: Cambridge Academy of Transport), March, p. 66
   Google Scholar
• Goulielmos , AM . 2010 . What can we learn from 259 years of shipping cycles? . International Journal of Shipping and Transport Logistics , 2 ( 2 ) : 125 – 150 .
   Web of Science ®Google Scholar