A “survey”¹ is useful, we believe, because it traces the retrospective evolution of a science². It also reveals the deficiencies remained after the research of economists―for economic surveys―usually over a rather long³ time accomplished.

During the 40 years since 1973-2013, “Shipping Economics”⁴ reached (in one journal: 984 papers) the 1st position among the other 2 maritime economic branches―“Port Economics” and “Shipbuilding Economics”, with a 60% share (=591 papers) vis-à-vis 40% of Port Economics (=393 papers). Shipbuilding economics―the 3^rd branch―received much less attention⁵, with less than 1/2 a paper per year⁶ (1.5% or 15 papers)! Moreover, shipping economics indicated a reduction in its production of papers by ~8% during the last 7 years or so.

Veenstra [5] argued that shipping is a very useful sector for the application of “advanced” econometric analysis. But only 11% of papers were in this, or near this, area. This means 108 papers out of a total of 984, which have published in Maritime Policy and Management (MP&M thereafter) between 1973 and 2012^8,9.

The amount of research about an industry depends on its cooperation with Academia, we believe. Judging the “openness” of the industry from the number of interviews carried-out as a method of research in producing published papers, interviewing has been applied to only ~3% of papers (~29) over 40 years¹⁰…!

Research greatly helped―in our opinion―not so much by the increased openness of shipowners, but by: 1) the number of “data-providing organizations”¹¹, which emerged; 2) the additional maritime journals published; 3) the increased number of postgraduate maritime centers established in Europe and Asia (China) the last 15 - 20 years; 4) the increased number of international and national maritime conferences; 5) the number of doctoral theses awarded and 6) the increased number of maritime companies listed in stock exchanges, which are transparent and provide quantitative information.

Still, the “market analysis” providing an “understanding of maritime markets” produced only 81 papers over 40 years (~8% of total); and the areas of “dry cargo” and “tankers” had over 1/2 of them (~56%; 45 papers). The above share, of ~8%, we believe, is very low for such an important subject... Certain authors¹², however, deepened and examined the structure of “maritime markets” despite the above general trend.

The IAME 2014 conference provided a list of the future research (conferences had to present what will be published tomorrow); 150 papers were accepted: “Shipping”¹³: 37%, with emphasis on modeling; “Ports”¹⁴: 30%; “Maritime logistics”: 18%; “Environment”:¹⁵ 9%; “Maritime Geography”: 4% and “Maritime Education/Training”: 2% (=100%). “Maritime Geography” appeared there for the first time. “Maritime logistics-ML emerged as a 3rd branch of maritime economics judged by the number of papers published; some [15] argue wrongly that ML is a discipline. ML deals with ports and port areas (hinterlands) with ships and other means of transport [18] .

Modeling for the first time (2014) claimed a serious presence in an IAME conference. There is, however, a deficiency as much as “shipping forecasting” is concerned, which, despite its superior importance in business life, attracted a limited number of papers…

The paper is organized as follows. Part I presents selectively the past research production of maritime economists since 1973 (to 2012). Part II presents briefly the evolution of maritime economic models since 1981 (to 2016). Part III presents separately the maritime economic models using GARCH as this is a milestone. Part IV focuses on the more recent maritime economic models since 2014 (to 2016). Part V focuses on models of forecasting maritime economic markets being of our special interest. Part VI provides a description of the shipping business cycle involving calendar time. Finally, we conclude.

This paper has 7 purposes: 1) to present briefly the (published) research production of maritime economists from 1973 (to 2012), on the economics of “dry cargo” and “tanker” shipping. 2) To show selectively the ability of maritime economists to create seminal models from 1981 (to 2016), with special emphasis in last 11 years (2006-2016). 3) To underline the religious devotion of maritime econometricians to GARCH model. 4) To focus on most recent maritime econometric modeling since 2014 (to 2016). 5) To record the efforts of maritime economists to forecast shipping markets and 6) to give a description of the mechanism of the “shipping economic cycle” taking into account time.

However, one hidden purpose was to note and induce future research onwards to nonlinear models with a view to improve forecasting. We noted with satisfaction that after our first effort in 2006 [19] , colleagues started to use chaotic models [20] in forecasting and to quote us as well [21] .

This part presents the main points of a previous survey carried-out by Woo et al. (2013) [6] over the last 40 years, restricted to the publications of one journal only (MP&M).

Worth noting is that whatever papers are published is not the unique responsibility of authors. We believe that the subjects of papers published in journals are firstly selected by their editors, and then by their reviewers. For example, the papers about “Marine Economics”¹⁶, when the late Prof. Goss R took over in 1981 as MP&M Editor, stopped. Moreover, journals seek a particular subject in selecting topics for “special” issues. Moreover, research is greatly helped by the degree of how much open (maritime) industry is to joint research with Academia¹⁷, as mentioned.

The progress of sciences in USA, we believe, is due to the joint research between Firms and Academia, especially at the level of doctoral theses. Shipping, unfortunately, is a “secretive” industry and its close cooperation with academia, though much improved, it is still desirable, especially for doctoral students¹⁸.

Modeling is the process of making a representation, (rather simple), of reality, by “observing” it. This definition, inspired by Small [23] , focuses on the fact that the “reality we model, is the reality we observe”.

Models can be derived either: 1) from logic, and from known facts, or from assumptions about reality; e.g. Einstein’s equation E = mc² is such a model; 2) by specifically mimic observations, especially when there is a lack of profound understanding of reality, or 3) from both. Shipping time series models fall in the second category…

The start of modeling of “dry bulk” and “tanker” markets taking into account its topics, methods and data employed, can chronologically be placed by us in 1981. Worth noting is that shipping economics emerged in 1875 as a study of the 1st shipping cartel.

Jing et al. [17] examined the freight rate volatility in dry bulk shipping markets, recognizing the existence of an increased complexity, given the “incredible” boom of freight rates between 2003 and 2005. They examined different sizes and types Cape, Panamax and Handy for their daily rates (BCI, BPI, JEHSI) between 01/03/1999 and 23/12/2005. The “ADF” test showed stationarity, and the JB-test rejected normality due to existing fat tails and spiked peaks in time series. They used GARCH (1, 1) dividing the period into two for obvious reasons: from 01/03/1999 to 31/12/2002 and 01/01/2003 to 23/12/2005; they applied EGARCH [79] . The GARCH found efficient.

Bessler et al. [80] tested the existence of autocorrelation and co-integration in the spot freight rates and the FFA prices. They found that spot freight rates are auto-correlated, while FFAs are not; and both are co-integrated.

Angelidis and Skiadopoulos [81] were the first to apply the “value at risk” model-VaR⁴⁰, (described below in forecasting models), to estimate risk in freight markets. They found that the simple non-parametric methods should be used to measure risk in the freight markets; e.g. the “Rescaled range analysis” is a non-parametric method. Alizadeh and Nomikos [82] analyzed the performance of the “regime-switching volatility models” in the freight market. They found that by incorporating regime changes into volatility models, they improved the precision of VaR estimates.

Greenwood and Hanson [83] argued that ship prices are very volatile, given the observed “mean reversion” in freight earnings. They proposed a behavioral model where firms over-extrapolate exogenous d shocks and partially neglect the endogenous investment responses of their competitors. Their formal estimation of the model confirms that both types of expectation errors are required to account for empirical evidence.

Kalouptsidi [84] examined the impact of the “construction lag” and its extension in times of high investment activity by constructing a dynamic model of entry and exit from shipping markets. She found that moving from time-varying-to-constant-to-“no time” to build this reduces prices, while both increase the level and the volatility of investments.

Adland and Jia [85] tested whether “new building prices” are inelastic to “demand”. They assumed that, for a shipowner, a “newbuilding contract, a resale case, (vessel sold during her construction), and an existing modern vessel” are “near perfect substitutes” (ignoring technological differences). They also assumed that the integration of the newbuilding, 2nd hand ship and freight market is feasible. This, we reckon, is a try to fit a model to theory, instead of fitting a model to reality… Nevertheless the model is interesting.

According to business economics [86] , the one of the 3 conditions for goods to be close substitutes is to have the same, or similar, occasion for use [87] . Given also the volatility of the freight market, the timing of entering into the freight market is of great importance (for us) e.g. a Cape earned $200,000/day in June 2008 and less than $5000 in Nov. 2008 (see graph 3 below for 2010). So, the delivery time should be a function of the level of freight rates… (for us).

Authors tried unsuccessfully, we believe to prove that despite subsidies to shipyards and/or to owners, which clearly affect newbuilding prices shipowners are indifferent between 3 cases: 1) ordering a “new building at her contract price 2) buying a resale, or 3) buying an existing vessel”. They also assumed that the 2nd hand ship market for bulk carriers is closely co-integrated with those of the freight rate and new-buildings; moreover, they found no evidence of a “short term asset bubble” in the past boom years… Finally, authors assumed that newbuilding prices are not comparable across time due to the nature of the new contracts (which differ mainly on delivery time and on payments schedule).

They used a single equilibrium equation model, relating today’s “prices” in all 4 ship markets (new-buildings, 2nd hand ships, freight rates and scrapping); it exists a “term structure of newbuilding prices” and a “time-varying delivery lag”, which indicate that newbuilding prices and 2nd hand ship ones are different. The model shows also how a “time varying delivery lag” positively correlated with freight rate will produce inelastic newbuilding prices and the market remaining efficient.

The model uses the classical “net present value” NPV method of a cash flow generated by a modern ship, between now (t) and future (τ): I t , τ = ∑ i = t τ TC t , τ − OPEX / ( 1 + r t , τ ) τ − t (4), where TC is the time-charter today and for τ-t years; r is the (risk-free) rate of interest⁴¹ for the same period and OPEX⁴² is the daily operating cost. Then the scrap value of the vessel is used assuming a linear depreciation method following Adland and Koekebakker [88] with life expectancy constant. They assumed also that investors do not expect the “scrap value” or the “expected useful economic value” to change...

The payments now to shipyard for a new building have the following present value: F t , τ = 1 / ( 1 + r t , τ ) τ − t S τ , 0 (5), where S_τ,0 is the 2nd hand value of the vessel today. This equality may be not true as the amounts agreed to be paid to shipyard may not be equal to vessel’s 2nd hand value on delivery. The 1st owner has to pay the price agreed. The 2nd owner faces a price on delivery, determined by market (supply and demand) then, which may be much higher or much lower than the one paid by 1st owner.

Antiques’ market or market of houses, are good examples of the above. This is why Greeks prefer buying used tonnage at rock bottom prices instead of ordering new ships, which during a boom are very expensive, and run the risk of oversupply, and under-demand, upon delivery...

An existing vessel provides, too, a net present value freight income plus her (uncertain) residual value: S t , 0 = Ι t , τ + 1 / ( 1 + r t , τ ) τ − t S τ , τ (6), where a “linear depreciation formula” is applied S τ , τ = Ζ τ + ( S τ , 0 − Ζ τ ) τ T − τ / Τ τ (7), where S_τ,τ is her 2nd hand price, Z_t is her scrap value, and T_τ is her expected life. In equilibrium, shipowners are assumed to be indifferent between the above 2 alternatives. Equation (7) is then inserted into Equation (6), and by using Equation (5), and assuming “myopic investors”, i.e., Z_τ = Z_t and Τ_τ = Τ_t, one gets: S t , 0 = Ι t , τ + F t , τ ( Τ t − τ ) / Τ t + Z t τ / Τ t 1 / ( 1 + r t , τ ) τ − t (8).

Equation (8) gives the balance between “prices” of all 4 shipping markets today, where: S_t,0 is the value of the new vessel today, TC is her time charter, Z_t is her scrap value, and F_t,τ is the value of her contract. If τ = 0, her delivery is immediate. Interesting is that this formula can be used to find out the newbuilding (contract) price F, given the price today of a new vessel S, her time charter TC and her scrap value Z. The delivery lag τ-t can also be determined as a function of time. They used also a numerical example using a Cape.

Worth noting is that the Cape market mentioned above in [85] suffered by extremely violent fluctuations (Figure 3).

Interesting is to see what a long boom can do (like the one or two in 2003-2014): 1) it prolongs vessel’s scrapping age by ~145%, from near 21 years in 1993 to 30 years in 2004; and between 25 and 29 years from 2003 to 2014. Vessels therefore remain in service roughly additional 5 - 9 years during a boom. This confirms our experience. This reinforces also our all along argument that scrapping is not a fast equilibrating market mechanism. 2) It prolongs the delivery lag by ~200%; the lag from roughly 1.8 years in 1993, increased to 3.9 years in 2006, 3.8 years in 2009 and fell in 2014 to 1.9 years. So, “shipbuilding demand” cannot be satisfied immediately, but it needs a “construction plus a waiting time”, (for us), which can be extended to almost double during boom years, i.e., from ~2 years to ~4 years (2003-2008). The “delivery lag” shows a faster adaptation than scrapping.

A finding useful for our “theory of depression” is that laid-up tonnage determines the short run fluctuations (ups and downs) of the freight rate from one month to next two say; the delivery lag affects supply after ~2 to ~4 years, depending on the phase of shipping cycle, while scrapping may need 5 to 9 years additional time to exert its full power on supply.

Equation (10) can be solved: F t , τ = [ S t , 0 − I t , τ − Z t ⋅ τ / Τ t 1 / ( 1 + r t , τ ) τ − t ] T t / ( T t − τ ) (11), and it can be used to estimate the “term structure of newbuilding prices”; e.g. the newbuilding price of a Cape started at 75 m$ in 2008 and fell to $64 m in 2009, after a delivery lag of 2.4 years; finally, it fell to $45 m after a delivery lag of 5 years.

Zhang et al. [89] used a Brownian distance correlation method and a Granger causality analysis and applied them to the new ship market considering it relatively distant from 2nd hand ship and freight markets before and after end 2008. Crisis made markets more closely related.

Zheng and Lan [21] , investigated the dynamics of the spot rates for all main sizes of tankers, using a “multifractal de-trended fluctuation analysis” (MF-DFA; due to Kantelhardt J W et al. [90] . Data: Jan. 27^th 1998 to Aug, 5^th 2013; from 2011 to 3884 days; using exponential ln freight rates and their 1st differences data found stationary; the IID hypothesis rejected by the BDS test. Normality is also rejected by the JB test. They concluded that the tanker market is more fractal than the crude oil commodity markets. What does this mean?

Original maritime research dealt with full scale econometric models, like those used by BV; later research, during the last 15 years or so, dealt with direct specifications of freight rate processes. Most studies made the assumption that tanker markets are “efficient” or “weakly efficient” in the sense of Fama in 1991.

Maritime economists almost exclusively used the traditional econometric models of ARIMA, VAR, VECM, and GARCH. They came to the conclusion that EMH does not apply to tanker freight markets, because… they are “untradeable services”… Finally, they [21] justified Goulielmos and Psifia [19] by saying that research reveals that the “tanker freight rates” are nonlinear, non-normal, non-stationary, and have memory effects!

They [21] concluded also that the Hurst exponent of the daily rate returns showed that these markets are fractal but not multifractal⁴³… The terms fractal-multifractal are used wrongly interchangeably by the authors. The paper, therefore, is not clear about fractality and multi-fractality. Fractal means that parts of time series are self-similar with the whole. Fractal is a pattern or object, whose parts echo the whole, only scaled down. But multi-fractality is different, because some parts of the object shrink quickly and others slowly. As argued by Mandelbrot & Hudson [45] “to improve almost any fractal model it is a good idea to replace it by a multifractal one”.

The cycles found [21] were all short term, from 2.1 months to 6.6 months (on a 30 days calendar month). The Hurst exponent found between 0.4 and 1. It is not the same thing to have H < 0.50 and H > 0.50. Large tankers have relatively stronger multi-fractality and higher volatility, being commercially inflexible and having thin market conditions [21] .

Most tanker markets have long term memory. Multi-fractality comes from both memory effects and fat tails, where the first is stronger. Finally, they found [21] that the crude oil markets are less multi-fractal than tanker markets; due mainly to fat tails; they exhibit longer non-periodic cycles.

Tsioumas and Papadimitriou [91] investigated the “excess returns” issue, between “time charter” and its corresponding “voyage” (using a TCE) in dry bulks (Cape and Panamax) for 2 t/c trips of BPI and BCI. They used weekly data: Jan. 2003 to Jan. 2014, to find-out a long term co-integration. TCE = time charter equivalent.

Their paper falls into “technical analysis” pioneered in 1980s by the “Norwegian School”, i.e., by Norman [92] , Adland [93] , Adland and Koekebakker [56] , Alizadeh and Nomikos [67] [71] and Adl and and Strandenes [68] . Co-integration [3] [4] found in the long term. They then used the MMTM (“Modified Momentum Trading Model”) due to Wilder in 1978 to formulate a chartering strategy, which is to get into voyage charters every week. This was compared with the benchmark strategy of always chartering in vessels on a t/c trip. The existence of excess returns confirmed. t/c = time charter.

They also used the “Relative Strength Index”-RSI at t + 1 using a 4 week moving average. The “voyage” is profitable when RSI moves above its MA (4), where MA(4) is a moving average of 4 periods. Authors could especially with Capes use monthly moving averages, which according to Figure 3 indicate the greatest fluctuations since June 2010.

Unfortunate for them is that their finding that all rates have right tales (+ skew) and kurtosis greater than 3 (leptokurtosis) and significant departures from normality… destroy the accuracy of their findings. Time series found non-stationary, if not in log-1st differences. Using MTMM one could improve excess returns from 15% to 31% for the Cape over Carter/Rotterdam; and from 6% to 10% for Panamax over Maracaibo/Rotterdam. The study confirmed that voyages are more volatile and that Cape had higher margins than Panamax [94] .

This area is one of the most important, but as Stopford [24] said: “maritime forecasting has a poor reputation”. This opinion is shared also by the President of the shipping companies “Dynacom” and “Dynagas” Mr. George Procopiou G. [95] . Maritime economists have indeed paid hard efforts to forecast freight markets and especially dry bulk freight rate markets, but in vain.

As we saw above (Table 4), Kavussanos and Nomikos [61] argued that the VECM generates the most accurate forecasts of spot prices, but it fails in forecasting future prices… It would be interesting to know why.

Batchelor et al. [96] , compared ARIMA, VAR and VECM in predicting spot and forward freight rates, where ARIMA outperformed in FFAs; VAR and VECM outperformed, slightly, ARIMA in spots.

Randers and Goluke [59] argued that in shipping, 2 cycles interplay, one 4 years and one 20 years in accordance with the “theory of waves” and on the capability to recognize patterns and construct indicator charts in search of clues for a move up or down; their forecasting for a turning point up in Aframax rates in 2010, proved, however, wrong.

Chen et al. [97] applied ARIMA and VAR to predict freight rates of several dry bulk routes. VAR outperformed out of the sample ARIMA. Leonova and Nikolov [98] used a model based on “wavelet” and “neural network” in order to predict dry bulk freight rates. Bulut et al. [99] developed a “VAR fuzzy integrated logical” forecasting model for time charter rates. Duru et al. [100] used a “fuzzy-DELPHI adjustment method” to improve the accuracy of statistical forecasts in the dry bulk shipping index.

Chang et al. [101] , applied the “Value-at-risk” (VaR⁴⁴) model to evaluate the risk of dry bulk freight rates (BDI, BPI, BCI, BHSI, BSI), showing an asymmetric long-memory volatility process, including expected shortfalls. They used the FIGARCH (fractionally integrated), the Hyperbolic GARCH and FIAPARCH with 3 distributions. The precise VaR estimates are obtained from an asymmetric long-memory volatility system with skewed Student-t distribution. The asymmetric FIAPARCH model outperformed all other models in the out of the sample forecasting! The innovative element of this is the inclusion of a long-memory structure. As we said in the footnote, VaR has indeed ignored long memory/dependence in the time series originally.

Santos et al. [102] , attempted to forecast 1 month to 6 months ahead, the 1-year and 3-year “period charter rates” of the VLCC tankers, using “Artificial Neural Networks-ANNs” model, out-of-the sample. They covered: 1994 (Oct.) to 2010 (Sept.), or 192 observations. They found that the literature about both ANNs and period charter rates, vis-à-vis spots, is scarce. They used monthly time series, and their innovative element is the use of “Radial basis function”-RBF due to Haykin [103] . They compared it with ARIMA. They found that ANNs plus RBF outperform ARIMA and NN-MLP (multi-layer perceptron). Dougherty [104] found only 2 models concerning shipping sector out of 52 (0.04%) using ANNs.

The above attempt [103] is the 6th in a series between 1997 and 2016 using “Artificial Neural Networks” among Li & Parsons [105] ; Lyridis et al. [106] ; Goulielmos et al. [107] ; Mehrara et al. [108] ; Leonova and Nikolov [98] ; and Zheng et al. [21] . At last maritime economists used nonlinear regressions. Li and Parsons [105] , were the first to try to forecast 1 - 24 months ahead the tanker freight rates, by using ARIMA and ANNs. The ANNs outperformed, especially in longer forecasting horizons.

Looking at the results of Santos et al. [102] : the 1 month forecasts using NN-RBF (2004-2010), were quite close; at times, a $5000 maximum deviation occurred; similarly the 2 months had ~$30,000 deviations in 2008-2009 and in 2005 (3 year time charter) and 3 months forecasts with again ~$25,000 deviations (3 year time charter) in 2008-2009 and in 2005. Their 6 months forecasts were inaccurate in most years. ARIMA and the MLP model underperformed. It seems that the above model shows shortcomings in years of radical changes up, like 2005 and 2008-2009 (peak rates); also when forecasting period is longer than one month…

Zeng et al. [109] , forecast BDI, using the “empirical mode decomposition”-EMD (due to Premanode & Toumazou in 2013, and others before) and the “artificial neural networks-ANNs”. They found that the “EMD-ANN” outperforms both ANN and VAR. The index decomposed into a number of “independent intrinsic mode functions”-IMF, consisting of 3 components: short term fluctuations, extreme events impact and long term trends. Authors recognized that in the dry bulk shipping market a great deal of non-linearity exists [110] ; Goulielmos [111] . As the authors argued, it is difficult for traditional econometric and statistical methods to capture the non-linear characteristics hidden in the dry bulk freight rates time series! ANN often, however, suffers from local minima.

Shipping cycle is the “outcome of the relationship of demand to supply”, and vice versa. If only demand is cyclical and/or only supply is cyclical, then the freight rate will be cyclical. If we want to have no shipping cycles, then demand variations should be smoothly and timely addressed by a very elastic supply. But demand cannot be, and is not, adaptive to supply. The reverse is true.

Maritime economists are confused because, as other economists did, they kept time out of their analysis (except Shimojo as mentioned by McConville [112] . Let us use a graphical presentation in an effort to bring time into the picture (Figure 4).

As shown, as demand increased from D₁ to D₂, supply increased by 0T₂-0T₁, due to ships that they were laid up and brought-in (T₁T₂). This may take from 3 to 6 months for ships on average to be reactivated. The freight rate increased from F₁ to F₂. As demand remained high, shipowners ordered ships, which when delivered increased supply by 0T₃-0T₂ (T₂T₃); this caused freight rate to fall to F₃. This took 2 years on average. If demand does not increase to D₃, then the new-buildings delivered will cause the fall in the freight rate to F₃ instead of going up to F₄. If demand falls to D₁ then a surplus of Supply will be created equal to T₁T₃.

This is a shipping cycle. This cycle took 27 - 30 months to be accomplished, but this is quite symptomatic and it is dangerous to consider this as fixed and as valid for all cases. The “construction time⁴⁵” can take from six months to up and near 4 years. The frequent and short cycles are due to the lay-up of ships depending on the tonnage accumulated there from last depression. New-buildings and the time these need to appear―given demand―is the decisive factor, as deliveries may be as much as 100m dwt for only one year (e.g. in 2012 for dry cargoes) and scrapping is unable to reduce it, but only by 33%. In shipping all are matters of time. So, time must return to maritime economy analysis to derive finally more sound conclusions.

Maritime economists since especially 2000 focused more to port economics. Moreover, a new branch formed titled “Maritime Logistics”. Cruising and Shipbuilding have been neglected, though during the last 5 years or so we saw a certain number of papers dealing with them.

To renovate maritime research, and modeling, industry’s journals have to direct research through planned “special issues”. Moreover, non-listed shipping companies must boost joint research and create systematic links with Academia.

There is a serious progress made 1) in the number of maritime doctoral theses produced; 2) in the number of postgraduate degree courses including the e-learning ones 3) in the number of maritime conferences; 4) in the increasing number of published maritime papers and 5) in the number of maritime journals, which published the last 20 years or so.

Quality has to surpass quantity though. This is the golden age of maritime research and maritime education, we believe. Journals’ editors shape the various research directions. Maritime research produced (in only one journal), since 1973, almost 1000 papers by using 3 methods: “empirical” 1/3; “scientific” 1/3 and “conceptual” (34%).

Econometric modeling and statistical analysis which drew our closer attention produced a slim number of 108 studies or ~3 papers per year. Moreover, market analysis, which we consider most important, had only 81 studies or 2 papers per year. Studies for shippers, transport chains. Maritime economists followed general econometrics but with long delays.

Modeling had a slim presence despite its rigor and power of mathematics but more serious is the lack of studies on maritime forecasting. We believe that if Academia wants to help the industry, it must improve forecasting.

Maritime economists are troubled, but at the same time helped, by general econometrics after long delays during this period. This period, stationarity emerged as a must and co-integration appeared as the proper solution not only econometrically, but also from the side of economic theory. We have been supplied with VAR (& VARMA) over the previous defected simultaneous equation models (SEMs)…

Between 1981 and 2004, 19 models have been worked out by maritime economists using: co-integration; VAR; ARCH; GARCH; ARIMA; ARIMA-X; EGARCH and ECM-error correction model. There is devotion to linear models. A pleasant exception was the “artificial neural networks” models with 5 studies. Modeling was extended to maritime “futures” (FFAs), as well, with 5 studies (2003-2004).

After 2006, the “Norwegian School” which was very active produced 5 - 6 models using the “technical” and “fundamental” analyses. Moreover, Goulielmos and Psifia introduced for a change the nonlinear chaotic model of “Rescaled Range Analysis” in maritime forecasting in 2006 and 2007. Eight models that have appeared from 2008 till 2016 continued to use GARCH, EGARCH, VaR, technical analysis and co-integration.

In the area of forecasting, 8 models emerged since 2003, using VECM, ARIMA, VAR, Wavelet/ANN, VAR/fuzzy, fuzzy-DELPHI, VaR, AFIAPARCH and ANN. We are glad that maritime economists have discovered nonlinearity. They used “fractional integration” together with “empirical mode decomposition” and “ANN”.

The author declares no conflicts of interest regarding the publication of this paper.

Goulielmos, A.M. (2018) Maritime Economic Modeling: A Survey, 2006-2016. Modern Economy, 9, 2064-2091. https://doi.org/10.4236/me.2018.912129