1

2  
3

7th  Framework  Programme  

SST.2013.6-­‐2.  Towards  a  competitive  and  resource  efficient  port  transport  system  
Collaborative  Project  
Grant  Agreement  no.  605176  

Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

DOCUMENT  ID  
PORTOPIA|D|1.3|DT|2015.06.15  
 
DUE  DATE  OF  DELIVERABLE    
31-­‐05-­‐2015  
ACTUAL  SUBMISSION  DATE  
[Publish  Date]  
 
 
DISSEMINATION  LEVEL  
PU  (Public)  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

DELIVERABLE  1.3  
Port  Traffic  Forecasting  Tool  
 

AUTHORSHIP    

Author(s)   Vonck  Indra,  Prof.  dr.  Theo  Notteboom,  Dr.  Francesco  Parola,    
Dr.  Giovanni  Satta,  Dr.  Luca  Persico  

Beneficiary  Partner   UA-­‐ITMMA  

Issue  Date   2015-­‐11-­‐13  

Revision   M.  Langenus  

Status   Final  draft  version  

Contributors   UNIGE  

Pages   46  

Figures   12  

Tables   12  

Annexes   0  

   

 
SIGNATURES    

Author(s)   Vonck  Indra,  Theo  Notteboom,  Francesco  Parola,  Giovanni  Satta,  Luca  
Persico  
 

Coordinator   Michaël  Dooms  

 

   

Disclaimer

The information contained in this report is subject to change without notice and should not be construed as a commitment by any
members of the PORTOPIA Consortium or the authors. In the event of any software or algorithms being described in this report, the
PORTOPIA Consortium assumes no responsibility for the use or inability to use any of its software or algorithms.

The information is provided without any warranty of any kind and the PORTOPIA Consortium expressly disclaims all implied
warranties, including but not limited to the implied warranties of merchantability and fitness for a particular use.

(c) COPYRIGHT 2013 The PORTOPIA Consortium

 
2  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

This document may be copied and reproduced without written permission from the PORTOPIA Consortium. Acknowledgement of
the authors of the document shall be clearly referenced. All rights reserved.

DELIVERABLE  1.3  
Port  Traffic  Forecasting  Tool  
 

REVISION      

Revision  Date   Reviewer  Name     Reviewer  Signature  

2015-­‐11-­‐13   M.  Langenus   M.  Langenus  

     

     

HISTORY        

Version  Date   Version  Id   Status   Comment  

2013-­‐12-­‐31   V0.0   [DRAFT]   [BRIEF  DESCRIPTION  OF  THE  REVISION]  

2015-­‐11-­‐13   V1.0   Final  Draft   Revision  before  2nd  Annual  Review  (December,  
2015).  

TBD   V1.0   [FINISH]   [BRIEF  DESCRIPTION  OF  THE  REVISION]  

       

 
3  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

DELIVERABLE  1.3  
Port  Traffic  Forecasting  Tool  
Summary  Report  
Deliverable 1.3 (D1.3) is embedded in Task 1.3: Estimation of future traffic flows in the European
port system (in the short, medium and long term). This task focuses on the development and
implementation of a range of methods to increase insight in the expected future (traffic) outlook for
the European port system in the short, medium and long term.
We look at three things:

• A first sub-task explores and applies a range of time-series based forecasting techniques in
order to generate aggregated and top-down forecasts on the maritime traffic evolution of the entire
European port system and significant parts thereof
• A second sub-task aims to develop a forecasting meta-system which groups data and expert
information on medium (1 to 5 years) and long term (> 5 years) developments in port activities in
Europe.
• The third sub-task is focused on the short-term development of the European port system.
It involves the drafting of a survey that will measure short-term traffic expectations (next quarters
up to a year) of relevant stakeholders in the European port business, with a main focus on port
authorities.

 
4  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

DELIVERABLE  1.3  
Port  Traffic  Forecasting  Tool  

1 LIST  OF  CONTENTS  

2   Positioning  of  this  report  in  WP1  ................................................................................................................  6  
3   Forecasting  freight  transport  ......................................................................................................................  7  
3.1   GDP  approach  ..........................................................................................................................................  7  
3.1.1   Desk  and  field  research  approach  ................................................................................................  8  
3.2   General  forecasting  techniques  ...............................................................................................................  8  
3.2.1   Qualitative  forecasting  techniques  ...............................................................................................  8  
3.2.2   Time  series  analysis  ....................................................................................................................  10  
3.2.3   Causal  models  .............................................................................................................................  16  
3.2.4   Overview  .....................................................................................................................................  19  
3.3   Academic  research  in  port  traffic  forecasting  .......................................................................................  20  
3.4   Assessment  of  techniques  in  port  data  forecasting  ..............................................................................  26  
3.5   Combining  port  development  plans  with  forecasting  ...........................................................................  27  
4   Forecasting  examples  in  similar  sectors  ...................................................................................................  32  
4.1   Air  Transport  ..........................................................................................................................................  32  
4.1.1   Boeing  and  Airbus  forecasts  .......................................................................................................  32  
4.1.2   The  International  Air  Transport  Association  (IATA)  Airline  Industry  Forecast  2014-­‐2018  ..........  32  
4.1.3   Lessons  from  the  air  transport  forecasting  .................................................................................  33  
4.2   Rail  transport  .........................................................................................................................................  33  
4.2.1   Railway  Demand  Forecasting  and  Service  Planning  Processes  (Rail  Freight  Service  Review)  ....  33  
4.2.2   Transnet  freight  demand  forecast  ..............................................................................................  35  
4.2.3   Lessons  from  the  rail  transport  industry  ....................................................................................  36  
4.3   Traffic  flow  forecasting  ..........................................................................................................................  36  
4.3.1   Traffic  flow:  Scenario,  Traffic  Forecast  and  Analysis  of  Traffic  on  the  TEN-­‐T,  Taking  into  
Consideration  the  External  Dimension  of  the  Union.  .............................................................................  36  
4.4   Academic  research  ................................................................................................................................  37  
5   Forecasting  in  Portopia  .............................................................................................................................  37  
5.1   Methodology  .........................................................................................................................................  38  
5.1.1   Short  term  ...................................................................................................................................  38  
5.1.2   Medium  term  ..............................................................................................................................  40  
5.1.3   Long  term  ....................................................................................................................................  41  
5.1.4   Implementation  ..........................................................................................................................  42  
5.2   References  .............................................................................................................................................  43  
 
5  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

2 POSITIONING  OF  THIS  REPORT  IN  WP1  

The aim of WP1 within the PORTOPIA project is to further develop the PPRISM indicators on
market trends and structure and to seek meaningful expansions. The specific objectives of the
work package include:

• Improving data availability and comparability of PPRISM indicators

• Collecting and presenting data at a more disaggregated level in terms of goods types and
time periods

• Developing new indicators (ratios and indexes)

• Develop forecasts on short, medium and long term developments in port activities in
Europe using a combination of techniques (modelling, meta-analysis and survey)

• Incorporation in a European Port Observatory (EPO) with link between indicators and
specific policy targets in the EU transport policy.
Deliverable 1.3 (D1.3) is embedded in Task 1.3: Estimation of future traffic flows in the European
port system (in the short, medium and long term). This task focuses on the development and
implementation of a range of methods to increase insight in the expected future (traffic) outlook for
the European port system in the short, medium and long term.
We look at three things:

• A  first  sub-­‐task  explores  and  applies  a  range  of  time-­‐series  based  forecasting  techniques  in  order  to  
generate  aggregated  and  top-­‐down  forecasts  on  the  maritime  traffic  evolution  of  the  entire  
European  port  system  and  significant  parts  thereof  
• A  second  sub-­‐task  aims  to  develop  a  forecasting  meta-­‐system  which  groups  data  and  expert  
information  on  medium  (1  to  5  years)  and  long  term  (>  5  years)  developments  in  port  activities  in  
Europe.  
• The  third  sub-­‐task  is  focused  on  the  short-­‐term  development  of  the  European  port  system.  It  
involves  the  drafting  of  a  survey  that  will  measure  short-­‐term  traffic  expectations  (next  quarters  up  
to  a  year)  of  relevant  stakeholders  in  the  European  port  business,  with  a  main  focus  on  port  
authorities.  

 
6  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

3 FORECASTING  FREIGHT  TRANSPORT  

According to Graham Cox (Director, Maritime Traffic Forecasts Ltd, Hothfield, Kent, UK), there
are two traditional methodologies that have been used to forecast port traffic volumes; one based
on GDP and the other on desk and field research.

3.1 GDP  approach  

The GDP approach is based on the assumption that most economic drivers determine demand for
cargo. In practice this principle is correct but has been oversimplified to apply only the GDP as
causality to traffic and impact on ports.

This method works with multipliers (for e.g. cargo grows 1.5 times for each point of GDP growth
and then forecasted to assess future cargo growth. This method is often combined with other
forms of qualitative forecasting. For example analysts sometimes apply extra multipliers based on
market assessments or econometric modelling. This practice introduces considerable subjectivity
and is hard to audit.

This simplification of forecasting in the maritime industry may have occurred because of data
problems in general and because the GDP approach has been an appropriate simplification for
containerized trade route traffic; given that there is no information about the nature of the cargo
within the boxes. As is shown in deliverable 9.1 data issues are prominent in maritime data isd
therefore quality issues often arise. Unfortunately, analysis based predominantly on GDP trends
has become the norm for all cargo types and right down to the port level, although that some
cargo’s do not have a clear link with GDP evolution (dichotomy between containers and container
content). The large majority of forecasts produced for the liquid and dry bulk trades follow the
same simple GDP methodology. This has happened despite the availability of data on other
economic drivers and more accurate information about the volume and make-up of port trade;
even in some cases for containerized traffic.

It is important to note that there is no fundamental statistical or economics justification for

analysing trade simply in terms of a relationship with GDP. Richer statistical methodologies reveal
underlying relationships with the drivers of this type of traffic that GDP cannot capture. It is
potentially deficient, if for no other reason that it ignores the effect on traffic flows that different
sectors of the economy can exert: such as construction or the retail trade.

An example is that certain sectors move in different paths than others. Agriculture could be failing
in a region but steel industry booming. This development will have a materially different effect on
traffic flows from that indicated by GDP if that sector’s imports or exports constitute a significant
proportion of cargoes. Use of a sole GDP driver will always create a large weighing error with
different kinds of maritime traffic. These different trends in sectors can persist into the medium
and long-term and are hence important for port studies, both in terms of determining the historic
relationships and in particular how they are projected.

A second major issue with this method is that the standard use of fixed relationships: i.e. the
assumption that the multiplier was always the same value and will remain so. An examination of
historical data indicates that such relationships are not static and are unstable for a multitude of
reasons. If there has been structural change during the period that is used to establish the
relationship between GDP, sectors and/or any other drivers that affect port traffic and this change
is not recognized, the multipliers used to project cargo volumes in the forecast period will be
wrong. So in essence this methodology is flawed at detecting turning points in individual cargo
trends.

 
7  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

3.1.1 Desk  and  field  research  approach  

This approach is rather different and comparable to expert opinions and the Delphi method. It
involves interacting with individuals and organizations in the industrial sectors that import and
export to ascertain their views and knowledge of developments. According to Cox this principle is
sound since the supply and demand side are often not distinguishable in statistics. Dangers of this
approach include the fact that interviewees may have a too narrow view of the market, rely
themselves on other forecasts or include external variables which are not relevant.

For short-term forecasting this technique has its benefits. Since it is generaly based on the level of
orders of the main actors in the sector. However useful this approach is for the short-term, it is not
suitable for a port forecast that needs medium- and long-term insight.

A possible alternative to these classic methods could be a multi-factoral methodology. Namely by

conducting an in-depth statistical or econometric analysis, using long runs of data to identify the
key sector drivers for each kind of freight, and how they and other disaggregated data modify the
relationship of cargo type with GDP and hence alter traffic projections. Secondly, a time series
analysis appraising the impact of past shocks to the economy on economic relationships. Thirdly,
the examination of the structure of the market of export and import industries. For example, the
competitiveness of industries in international markets relevant to the traffic of a port needs to be
examined. Changes in government policy and new technology can have significant impacts. Desk
and field research or expert advice in these areas complements statistical analysis to make the
forecast as close to reality as possible.

3.2 General  forecasting  techniques  

In virtually every decision they make, executives today consider some kind of forecast. This is no
different for managers in the maritime or port industry. Sound predictions of demands and trends
are no longer luxury items, but a necessity. In order to cope with seasonality and volatility or large
swings of the economy. Forecasting is a useful tool to help with these issues. This document
provides and overview for the most used methodologies and different approaches and distils the
most suitable techniques for the Portopia project.
There are three basic types of forecasting techniques namely qualitative models, time series
analysis and causal models (both quantitative models). The first uses qualitative data (expert
opinion, for example) and information about special events of the kind already mentioned, and
may or may not take the past into consideration. The second, on the other hand, focuses entirely on
patterns and pattern changes, and thus relies entirely on historical data. The third uses highly
refined and specific information about relationships between system elements, and is powerful
enough to take special events formally into account. As with time series analysis and projection
techniques, the past is important to causal models.
Each of the different techniques has its advantages and disadvantages and suits certain situations.
Often they have alternating levels of success in the difference between short medium and long term
forecasting.

3.2.1 Qualitative  forecasting  techniques  

A qualitative forecasting technique is an Estimating method that relies on expert human judgment
combined with a rating scale, instead of on hard (measurable and verifiable) data.

• used  when  data  are  scarce  or  ambiguous  

• frequently  used  in  new-­‐technology  areas  

 
8  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Advantages

• Flexible  data  needs  allow  for  inclusion  of  non-­‐numerical  data  

• Ambiguous  or  non-­‐complete  data  can  be  processes  
• Allows  experience  and  judgment  of  senior  executives  and  outside  experts.  

Disadvantages

• Not  always  accurate  

• If  the  opinion  of  one  person,  whose  view  prevails,  is  incorrect,  the  forecast  is  incorrect  
• Difficult  to  eliminate  the  forecaster’s  personal  bias    

DELPHI METHOD
This is a group technique in which a panel of experts is questioned individually about their
perceptions of future events. The experts do not meet as a group, in order to reduce the possibility
that consensus is reached because of dominant personality factors. Instead, the forecasts and
accompanying arguments are summarized by an outside party and returned to the experts along
with further questions. This process continues until a consensus is reached.
Advantages:

• Quite  effective  for  long-­‐range  forecasting.    

• Eliminates  the  disadvantages  of  group  think.    
• There  is  no  committee  or  debate.    
• The  experts  are  not  influenced  by  peer  pressure    

Disadvantages:

• Low  reliability  
• Lack  of  consensus  from  the  returns.  

MARKET RESEARCH FORECAST

Testing hypotheses based on market trends and general economic knowledge. The best results of
this technique arise when combined with other qualitative forecasts. A first attempt of this
approach was made in deliverable 1.1 ‘market trends’.

PANEL CONSENSUS FORECAST

This technique uses open meetings where all participants provide or exchange ideas. In addition,
instead of only relying on historical data, the consensus method goes one step further. The
historical data is enhanced with the current market trends and events to ensure that the demand
uncertainty is reduced.

VISIONARY FORECAST
Personal insights opinion and facts to predict possible scenarios. This is a non-scientific method
and often prone to errors with a high degree of uncertainty. Delphy method or panel consensus are
preferable alternatives. This technique is the most radical and consists of a prophecy of the future
 
9  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

based on personal insight, judgement, and, when available, historical analogies that can be
extrapolated into possible future forecasts. Indeed, it is characterized by subjective guesswork and
imagination and, in general, the methods used are non-scientific and non-quantitative (Ross, 2011)
HISTORICAL ANALOGY FORECAST
This is a forecasting method that assumes that two different kinds of phenomena share the same
model of behavior. For example, one way to predict the sales of a new product is to choose an
existing product which "looks like" the new product in terms of the expected demand pattern for
sales of the product. Hence, historical analogy forecast constitutes a judgmental forecasting
technique based on identifying a sales history that is analogous to a present situation, such as the
sales history of a similar product, and using that past pattern to predict future sales.

3.2.2 Time  series  analysis  

The second subgroup of analysis is the time series analysis. This technique is used when several
years’ data for a product or product line are available. It measures the rate and changes rate of
change. Time series analysis has the main objective of:

• describing data and obtain simple measures to summarize the main properties of the time
series. The simple analysis of the graph can unveal the existence of trend, seasonality,
outliers and turning points;  
• indentifying the data generating processthat generates the random variable of which a
sequence of observations are available;  
• make forecast estimations of future values. This is based on the principle that the behaviour
of the phenomenon in the past is maintained in the future also.  

To pratically conduct time series analysis two main approaches are known: i) classical approach to
time series analysis and ii) Box-Jenkins approach or modern time series analysis.

CLASSICAL APPROACH TO TIME SERIES ANALYSIS

It is based on the error model: 𝑥! = 𝑓 𝑡 + 𝜀!   t = 1, 2, ..., n

In this case the series 𝑥! is defined as the resultant of an explicit analytical function f(t) plus a
random error term 𝜀!   which must comply with the typical assumptions of the regression model:
𝐸(𝜀! ) = 0; 𝐸 𝜀!! = 𝜎 ! < +∞; 𝐸(𝜀! 𝜀! ) = 0    ∀𝑡 ≠ 𝑠. This model is particularly appropriate for regular
phenomena.

According to the classical approach, every time series 𝑥! can be considered as the result of a
combination of different unobserved factors (stochastic and non stochastic), called the components
of a time series:

• Trend component 𝑇! : can be defined as long-term change in the mean level. Trend component
comprise all cyclic components whose wave length exceeds the length of the observed time series.

• Cyclic component 𝐶! : refers to variations exhibited by the time series at fixed period that cannot
be considered as seasonality. For example, economic data are often affected by business cycle of
about 5 years.

• Seasonality 𝑆! : consists of periodic, repetitive, and generally regular and predictable patterns in
the levels of a time series. Seasonality can repeat on a daily, weekly, monthly or quarterly basis,
these periods of time are structured and occur in a length of time less than a year. Seasonal
fluctuations in a time series can be contrasted with cyclical patterns.
 
10  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

• Irregular fluctuations 𝜀! : after trend, cycle and seasonal variation have been removed something
is left in the residuals that is not predictable from the past history. Making sure that the 𝜀!
component is truly random is a good guarantee that the decomposition of the series into
components is correct.

The classical methods of analysis of time series are concerned with the decomposition of the series
into the four above mentioned components. It is worth noting that the decomposition is, in general,
not unique unless certain assumptions are made.

The most common decomposition models used are:

- additive decomposition model: 𝑥! = 𝑇! + 𝐶! + 𝑆! + 𝜀! (this model assumes that components

have the same unit of measurement of 𝑥! and are independent each other; see Figure 1)
- multiplicative decomposition: 𝑥! = 𝑇! ∗ 𝐶! ∗ 𝑆! ∗ 𝜀! (this model assumes that only the trend
component has the same unit of measurement of 𝑥! , the remaining components are
dimensionless number proportional to the trend component). Applying a log
transformation it is easy to pass from multiplicative to addictive model.
- mixed model: 𝑥! = (𝑇! ∗ 𝐶! ) + 𝑆! + 𝜀! (with additive seasonal component).

Trend
80

60

40

20

0
1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10
Year_1 Year_2 Year_3 Year_4 Year_5 Year_6 Year_7 Year  8 Year_9 Year_10

Cyclical  component Time   Series

6 160
4 150
2 140
0 130
1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10
-­‐2 120
Year_1 Year_2 Year_3 Year_4 Year_5 Year_6 Year_7 Year  8 Year_9 Year_10
-­‐4 110
-­‐6 100
90
Seasonal  component
80
80
70
70
60
60 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10
50 Year_1 Year_2 Year_3 Year_4 Year_5 Year_6 Year_7 Year  8 Year_9 Year_10
40
1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10
Year_1 Year_2 Year_3 Year_4 Year_5 Year_6 Year_7 Year  8 Year_9 Year_10

Irregular  component
15
10
5
0
1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10 1 4 7 10
-­‐5
-­‐10 Year_1 Year_2 Year_3 Year_4 Year_5 Year_6 Year_7 Year  8 Year_9 Year_10

-­‐15

Figure  1  Example  of  additive  decomposition  of  a  time  series  according  to  the  classical  approach  

Source:  own  elaboration.  

The classical approach to time series has a number of advantages:

1. It is based on rather intuitive concepts

2. It is exploitable also when the series length is rahter short

 
11  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

3. It is very useful when the series plot reveals that the variability is strongly dominated by a
component, for example trend or seasonality
It also has a number of disadvantages:
1. The decomposition is not unique

2. The stochastic part is only in the error term

Several methods to estimate trends and seasonal components have been developed.

TREND PROJECTION

The estimation of the trend component (see Figure 2) of an observed time series 𝑥! is based on
regression functions. For example, suppose to have a trend that can be classified as polynomial,
hence we assume that the trend dynamics follows this form: 𝑓 𝑡 = 𝑎! + 𝑎! 𝑡 + 𝑎! 𝑡 ! + ⋯ + 𝑎! 𝑡 ! .
Using least squares estimation methods, it is possible to obtain the estimates of the coefficients.
The polynomial that results can be a good guide to interpret the phenomenon. Much more caution
must be used, instead, when the aim is to make prediction since there is no warrant that the
functional form, supposed to hold over the sample period, does so also outside the time span of the
observed time series.

Figure  2  Estimation  of  linear  trend  in  a  monthly  time  series  by  using  trend  projection  

Source:  own  elaboration.  

SEASONALITY ESTIMATION

The regression method used for the estimation of the trend component can be adopted also for the
estimation of the seasonal component, representable by a periodic function h(t). A first approach to
study h(t) is to assume that h(t) is a sum of dummy variables ℎ 𝑡 = ! 𝑎! 𝐷!" where 𝐷!" is 1 in the j-
th period and 0 otherwise, 𝑎! indicates the level of the phenomenon at the j-th period. A second
possibility is to assume that the seasonal component is given by a sum of harmonic functions (see
Figure 3). Generally a sum of 2, 3 harmonic waves can be a reasonable option to represent a
sufficiently complex seasonal dynamics.
MOVING AVERAGE APPROACH
When the analytic description of the phenomenon is not possible, for example because the dynamic
exhibited by the phenomenon is extremely irregular, the moving averages are a good tool to filter

 
12  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

the time series and indentify the underlying non-linear trend. Moving averages try to capture the
dynamics of the phenomenon without defining an analytical form that holds over the entire time
span as a variation law. Using moving averages it is possible to estimate the trend, remove
seasonality from series and reduce the random component. For example, in order to emphasize the
trend component in a daily time series (see Figure 4) one needs an average over 12 subsequent
values to produce a deseasonalized series not influenceed by seasonal component.
25000
20000
15000
10000

2008 2009 2010 2011 2012 2013 2014

Figure  3  Estimation  of  seasonal  component  in  a  monthly  time  series  by  using  the  harmonic  function  approach  

Source:  own  elaboration.  

Figure  4  Estimation  of  seasonal  component  in  a  monthly  time  series  by  using  the  harmonic  function  approach  

Source:  own  elaboration.  

EXPONENTIAL SMOOTHING APPROACH

Moving averages can be seen as filters. When a trend is estimated and other fluctuations are
smoothed, this is like filtering out the desired component and exclude the non-desired
components. The simple moving average model described above treats the observations on which it
is calculated equally and completely ignores all preceding observations. However, in some context,

 
13  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

a asymmetric filter is preferable in order to assign higher weights to more recent observations (i.e.
the most recent observation should get a little more weight than 2nd most recent, and the 2nd most
recent should get a little more weight than the 3rd most recent, and so on). A popular technique
applied to get smoothed values, is the exponential smoothing (see Figure 5): 𝑥! = ! !!! 𝛼  (1 −
𝛼) ! 𝑥!!! where the weight applied to any observations 𝛼  (1 − 𝛼) ! decrease geometrically.

original time series

a= 0.2
a= 0.6
140
120
100
80

2006 2008 2010 2012 2014

Figure  5  Estimation  of  non-­‐linear  trend  in  a  monthly  time  series  by  using  exponential  smoothing  technique  

BOX AND JENNKINS (1970): THE MODERN APPROACH TO TIME SERIES ANALYSIS
According to the modern approach to time series, the observed time series is conceived as a finite
realization of a stochastic process. The stochastic part coincides with the systematic part of the data
generating process and not simply the error term. This is formally described by the stocahstic
model: 𝑥! = 𝑔 𝜀! , 𝜀!!! , …  
Compared to the classical approach, here the 𝑓 𝑡 component is assumed to be absent or
preliminarily removed and the focus is on 𝑔 𝜀! , 𝜀!!! , … , a correlated component process that
governs the entire data generating mechanism. The modern approach is based on the Box-Jenkins
procedure (Box and Jenkins, 1970). Box - Jenkins Analysis refers to a systematic method of
identifying, fitting, checking and using integrated autoregressive, moving average (ARIMA) time
series models. The method is appropriate for time series of medium to long length (at least 50
observations). ARIMA models are the most general class of models for forecasting a time series
which can be made to be stationary1 by differencing, in conjunction with nonlinear transformations
such as logging or deflating.  
According to this approach, a time series random variable is usually viewed as a combination of
signal and noise and the signal could be a pattern of fast or slow mean reversion or sinusoidal
oscillation or rapid alternation in sign and it could also have a seasonal component. An ARIMA
model therefore is a filter that separate the signal from the noise. As a result, the signal is
extrapolated into the future to obtain forecasts. The ARIMA forecasting equation for a stationary
time series is a linear equation where the predictors consist of lags of the dependent variable
and/or lags of the forecast errors. The forecasted value of the dependent variable (X) is equal to a
constant and/or a weighted sum of one or more recent values of X (autoregressive component)
and/or a weighted sum of one or more recent values of the errors (moving average component).
The acronym ARIMA stands for “Auto-Regressive Integrated Moving Average”. Lags of the

1 A time series random variable is stationary if its statistical characteristics are all constant over time (it has no trend, its
variations around its mean have a constant amplitude and it wiggles in a consistent pattern).
 
14  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

stationarized series in the forecasting equation are called “autoregressive” terms, lags of the
forecast errors, conversely, are called “moving average” terms. A time series which needs to be
differenced to be made stationary is said to be an “integrated” version of a stationary series.  

SARIMA MODELS
In order to taking into account the seasonal component, extant literature developed SARIMA
models. The seasonal part of an ARIMA model has the same structure as the non-seasonal part: it
may have an AR factor, an MA factor and/or an order of differencing. In the seasonal part of the
model, all of these factors operate across multiples of lags (the number of periods in a season). A
seasonal ARIMA model (SARIMA) is classified as an ARIMA(p,d,q)x(P,D,Q) model, where
P=number of seasonal autoregressive (SAR) terms, D=number of seasonal differences, Q=number
of seasonal moving average (SMA) terms
Forecasts from(see Figure 6)  
ARIMA(0,0,2)(0,1,1)[12] with drift
180
160
140
120
100
80

2006 2008 2010 2012 2014 2016

 

Figure  6  Monthly  time  series  forecasting  according  to  SARIMA  model:  ARIMA  (0,0,2)(0,1,1)  [seasonality=12]  

Source:  own  elaboration.  

Therefore, to sum up prior suggestions, time series analysis is one of the most well-known
statistical techniques to make predictions under the assumption that existing patterns will continue
into the future. It helps to identify and explain:

• Any  regularity  or  systematic  variation  in  the  series  of  data  which  is  due  to  seasonality—the  
“seasonals”;  
• Cyclical  patterns  that  repeat  any  two  or  three  years  or  more;  
• Trends  in  the  data;  
• Growth  rates  of  these  trends.  

Advantages:

• The  time  series  method  is  a  useful  tool  to  measure  both  financial  and  endogenous  growth  
• The  time  series  method  of  forecasting  is  the  most  reliable  when  the  data  represents  a  broad  time  
period    
• Allows  to  distil  trend,  seasonality  and  clutter  disadvantages  

 
15  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

• Over  correcting  for  trends  

• Lacks  of  precision  
• Time  delays  during  turning  point  assessment  

Disadvantages:

• It  is  not  quite  suitable  for  identifying  turning  points  (see  Figure  7).  

Figure  7  Example  of  unidentified  turning  point  of  a  time  series  analysis  prediction.  

Source:  own  elaboration.  

X11

Is a filter based method based on the Box Jenkins methodology. It provides seasonal adjustment
and is often known as X11 style methods. These are based on the ‘ratio to moving average’
procedure described in 1931 by Fredrick R. Macaulay, of the National Bureau of Economic
Research in the US. The procedure consists of the following steps:

1) Estimate  the  trend  by  a  moving  average  

2) Remove  the  trend  leaving  the  seasonal  and  irregular  components  
3) Estimate  the  seasonal  component  using  moving  averages  to  smooth  out  the  irregulars.  

Seasonality generally cannot be identified until the trend is known, however a good estimate of the
trend cannot be made until the series has been seasonally adjusted. Therefore X11 uses an iterative
approach to estimate the components of a time series. As a default, it assumes a multiplicative
model. The model tests for the presence of identifiable seasonality; moving seasonality; residual
seasonality in addition it automatically removes regression effects before ARIMA modelling and
automatically selects the moving average range.

3.2.3 Causal  models  

These models are the most sophisticated kind of forecasting tool and express mathematically the
relevant causal relationships. They incorporate the results of a time series analysis whilst taking
into account everything known of the dynamics of the flow system.

 
16  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Advantages

• explanatory  power  it  allows  for  a  understanding  of  the  relationships  among  variables  
• The  lack  of  consistent  superior  performance  of  econometric  models  is  disturbing,  particularly  as  the  
extremely  simple  models  perform  nearly  as  well  

Disadvantages

• Variables  with  insignificant  coefficients  have  to  be  rejected  according  to  the  principles  of  
econometrics    
• Change  in  the  relationships  between  variables  increases  with  the  projection  period    
• Substantial  cost  and  time  requirements    

REGRESSION
Regression analysis is widely used for prediction and forecasting, where its use has substantial
overlap with the field of machine learning. This methodology is also used to understand which
among the independent variables are related to the dependent variable, and to explore the forms of
these relationships. In restricted circumstances, regression analysis can be used to infer causal
relationships between the independent and dependent variables. However this can lead to illusions
or false relationships, so caution is advisable.

ECONOMETRIC MODEL
Under this category of models we can find a lot of variations of the basic ordinary least square
(OLS) regression model. Among them panel data regression model emerges as available solution
for forecasting complex phenomena
Panel data (longitudinal time-series data) is a dataset in which the behavior of statical units are
observed across time. Panel data allows to control for variables that cannot be observed or measure
like cultural factors or difference in business practices across companies; or variables that change
over time but not across entities (i.e. national policies, federal regulations, international
agreements, etc.). Therefore panel data models account for individual heterogeneity. With panel
data it’s possible to introduce variables at different levels of analysis (i.e. students, schools,
districts, states) suitable for multilevel or hierarchical modeling. Some drawbacks are data
collection issues (i.e. sampling design, coverage), non-response in the case of micro panels or
cross-country dependency in the case of macro panels (i.e. correlation between countries). There
are two approaches in panel data analysis:

Fixed-effects model. Fixed-effects model explores the relationship between predictor and outcome
variables within an entity (country, person, company, etc.). Each entity has its own individual
characteristics that may or may not influence the predictor variables (for example, being a male or
female could influence the opinion toward certain issue; or the political system of a particular
country could have some effect on trade or GDP; or the business practices of a company may
influence its stock price). In this approach, it is assumed that something within the individual may
impact or bias the predictors or the outcome variable. Fixed-effects model remove the effect of
those time-invariant characteristics so it’s possible to assess the net effect of the predictors on the
outcome variable. Fixed-effects will not work well with data for which within-cluster variation is
minimal or for slow changing variables over time.
 
17  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Random-effects model. Random effects model, unlike the fixed effects model, assumes that the
variation across entities are random and uncorrelated with the predictor or independent variables
included in the model. An advantage of random effects is that it’s possible to include time invariant
variables (i.e. country of origin) that in fixed effects models are absorbed by the intercept. Random
effects assume that the entity’s error term is not correlated with the predictors which allows for
time-invariant variables to play a role as explanatory variables. In random-effects is crucial to
specify all the individual characteristics that may influence the predictor variables. However,
sometimes some of those variables may not be available therefore leading to omitted variable bias
in the model.

INTENTION TO BUY MODEL

An intention to buy model is a combination of surveys performed on the possible clients. And trend
mapping of projected sales.

IO MODEL
This model is a quantitative economic technique that represents the interdependencies between
different branches of a national economy or different regional economies. Because the input–
output model is fundamentally linear in nature, it lends itself to rapid computation as well as
flexibility in computing the effects of changes in demand. It is also used to identify economically
related industry clusters and also so-called "key" or "target" industries (industries that are most
likely to enhance the internal coherence of a specified economy).

Figure  8  Example  of  IO  model

ECONOMETRIC IO MODEL

This technique is a combination of a IO model and an econometric model.

DIFFUSION INDEX
Diffusion indexes measure the proportion of the components that contribute positively to the
index. The first step in computing the diffusion indexes is to calculate if a component increased,
decreased, or had no change. Components that rise more than 0.05 percent are given a value of 1,
components that change less than 0.05 percent are given a value of 0.5, and components that fall
more than 0.05 percent are given a value of 0. Next, sum the values of the components. Third,
divide by the number of components. Finally, multiply by 100.
 
18  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

LEADING INDICATOR
Leading indicators are events or measures that are a prelude to, and can thus help predict, another
event or measure. If you have ever experienced a thunderstorm you know that a bolt of lightning
always precedes a clap of thunder. Thunder is the sound made by lightning, but due to the relative
speeds of light and sound, lightning is a leading indicator of thunder. This relationship is a good
analogy for how leading indicators are useful in business. Many underlying economic conditions
move through value chains and geographies at different speeds. Understanding and measuring the
fastest moving conditions will give you better insight into those measures that are important to
you, but do not move at the same rates. As an example, large computer hardware manufacturers
such as IBM and HP are keenly interested in monthly measures of corporate IT spending, as these
serve as indicators for how much hardware they will sell in future periods.
LIFE CYCLE ANALYSIS

Is a technique to assess the environmental aspects and potential impacts associated with a product,
process, or service, by: Compiling an inventory of relevant energy and material inputs and
environmental releases. Evaluating the potential environmental impacts associated with identified
inputs and releases.

Figure  9  Life  cycle  model  

3.2.4 Overview  

Table 1 below provides an overview of the forecast methods available. Accuracy depicts the
usefulness of the tools on short (0-1y) medium (1-5y) and long term forecasts (+5y). Identification
of turning points expresses the ability of the methodology to indicate trend reversals. Data required
offers an overview of the necessary data.

 
19  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Delphy Market  research Panel  Conscensus Visionary  forecast Historical  analogy

Accuracy
Short  Term Fair  to  good Excellent Poor  to  fair Poor Poor
Qualitative  

Medium  term Fair  to  good Good Poor  to  fair Poor Fair  to  good
Long  Term Fair  to  good Fair  to  good Poor Poor Fair  to  good
Identification  of  turning  points Fair  to  good Fair  to  good Poor  to  fair Poor Poor  to  fair
Data  required Questionnaires Market  data Expert  pannel Scenario  set Raw  data
Possible  within  Portopia  forecast Yes Yes Yes Yes Yes

Moving  average Exponential  Smoothing Box-­‐Jenkins X11 Trend  projections

Accuracy
Short  Term Poor  to  good Fair  to  very  good Excellent Excellent Very  good
Medium  term Poor Poor  to  good Poor  to  good Good Good
Time  series

Long  Term Very  poor Very  poor Poor  to  good Good Good
Identification  of  turning  points Poor Poor Very  poor Very  poor Good
Data  required 2  year  raw  data 2  year  raw  data 2  year  raw  data 3  year  raw  data 5  year  raw  data
Possible  within  Portopia  forecast Yes Yes Yes Yes Yes

Regression Econometric  model Intention  to  buy IO  model Economic  IO  model
Accuracy
Short  Term Good  to  very  good Good  to  very  good Poor  to  good NA NA
Medium  term Good  to  very  good Very  good  to  e xcellent Poor  to  good Good  to  very  good Good  to  very  good
Long  Term Good  to  very  good Good Very  poor Good  to  very  good Excellent
Identification  of  turning  points Very  good Excellent Good Fair Good  to  very  good
Data  required Quarterly  data Quarterly  data raw  data raw  data raw  data
Causal  methods

Possible  within  Portopia  forecast Yes Yes No No No

Diffusion  index Leading  indicator Life  cycle  analysis

Accuracy
Short  Term Poor  to  good Poor  to  good Very  poor
Medium  term Poor  to  good Poor  to  good Poor  to  good
Long  Term Very  poor Very  poor Poor  to  good
Identification  of  turning  points Good Good Poor  to  good
Data  required raw  data raw  data raw  data
Possible  within  Portopia  forecast Yes Yes No

Table  1  overview  of  the  forecast  methods  available  

3.3 Academic  research  in  port  traffic  forecasting  

Over the last two decades, both academics and practitioners awarded a growing attention to the
study of forecasting techniques in the port domain.
In particular, significant research efforts have been dedicated to forecast both port cargo
throughput and container throughput (Lam et al., 2014). Indeed, cargo throughput has been
proved to play a valuable role for each port. First, it constitutes the most relevant production index
for evaluating port growth and success; second, it affects future port planning and development
(Zhang et al., 2013). Analogously, extant literature recognizes that reliable models capable to
forecast container throughput are necessary in order to support decisions on planning, upgrading
and upsizing of container terminal facilities. The capacity of a container terminal, in fact, cannot
easily be adjusted or increased in the short term by leveraging on inventory, outsourcing or
overtime working (Peng and Chu, 2009). Capacity management and long-term investments,
indeed, require accurate forecasts. The forecasting of handling volumes may provide terminal
operators and port authorities (PAs) with useful data, for taking decisions and implementing new
strategies more effectively (Talluri and Ryzin, 2004).
Several scholars investigated the accuracy of several forecasting techniques, as previously defined,
in predicting port cargo throughput or container throughput. Both qualitative and quantitative
methods have been considered. As concern qualitative methods, for example, Rashed et al. (2015)
debated on scenario-based forecasting as a viable technique for predicting container throughput,
focusing on the Hamburg - Le Havre range (Port of Antwerp). As uncertainty related to explanatory
 
20  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

variables is a serious concern in long-term forecasting, the Authors suggest scenario-based analysis
as a viable tool for investigating the impact of various sources of the economic and transport trends
on the container throughput’s future trajectories.
Referring to quantitative methods based on time series analysis, Jiang and Lei (2009) compared
nonlinear gray model with gray model to forecast cargo throughput while Xu (2011) proposed an
autoregressive forecast model to predict cargo throughput. Peng and Chu (2009) investigated
which model is capable of generating the most accurate prediction of container throughput. In
particular, they tested six univariate models (classical decomposition model, trigonometric
regression model, regression model with seasonal dummy variables, grey model, hybrid grey model
and SARIMA model) on three major Taiwanese ports (i.e. Keelung, Taichung and Kaohsiung), in
the 2003-2006 timeframe. In their study, the classical decomposition model emerges as the best
forecasting model since it has the highest values of accuracy. Peng and Chu’s study (2009), also
suggest that more sophisticated and complex statistical methods do not necessarily perform better
that simpler ones.

A number of prior contributions focus on causal models. Lam et al. (2004), challenging port cargo
throughput forecasting, propose and develop a neural network analysis in order to forecast 37 types
of freight movements in the Port of Hong Kong. The explanatory factors included in the model are
trade value of imports/exports/re-exports at 1990 prices, population, electricity demand, and Hong
Kong Gross Domestic Product (GDP). The outputs of the analysis demonstrate that neural network
models are more accurate than regression analysis when their results are compared with the
observed data, but the reliability of the proposed neural network forecasting model decreases in the
long-term. Chou et al. (2008), by focusing on import containers throughput, propose a modified
regression model for forecasting the volumes of Taiwan’s import containers. In their study, they
demonstrate that within quantitative causal model, the total forecast error for modified regression
model is significantly lower than that for traditional regression models, which do not consider and
modify the errors produced by the non-stationary contribution coefficient.
Some recent academic contributions tempted to compare or combine various forecasting methods
and techniques. Chen and Chen (2010), for example, focus on time series analysis and create an
optimal predictive model of volumes of container throughput at ports by using genetic
programming (GP), decomposition approach (X-11) and seasonal auto regression integrated
moving average (SARIMA). In their analysis performed on Taiwanese ports, within the Jan. 1978 -
Dec. 2006 period, they discover that both in short- and long-term forecasting the sequence of
container throughputs at sampled ports has a valuable trend and seasonal cycle nature. X-11,
SARIMA and GP forecasts provide all accurate predictions, but the GP is proved to be the optimal
method for predicting container throughput. Moreover, Gosagang et al. (2011), explore and
compare the neural network method and linear regression technique for predicting the container
throughput at Bangkok port. By testing these two forecasting techniques in the 1999-2010
timeframe, the Authors demonstrate that the accuracy of linear regression technique is less than
neural network approach, using multilayer perceptron. Xiao et al. (2012) propose the application of
a hybrid methodology, which includes qualitative and quantitative analysis to the container
throughput forecast. In particular, they develop a feed forward neural network based on the
improved particle swarm optimization with adaptive genetic operator for forecasting and then
adjust the forecast results of this model with the knowledge from port experts. Empirical results on
the Tianjin Port suggest that forecasts developed with their hybrid model are significantly more
accurate than some other methods. Analogously, de Langen et al. (2012) argue that extant
forecasting approaches adopted in port domain rely uniquely on trend forecasts and trend based
models. These models, are not capable to capture disruptions of historical patterns, especially for
certain types of commodities. As a result, they propose a forecasting approach that uses a model, in
combination with expert judgements and commodity specificy research. This approach is
applicated to forecasts for 2030 volumes of all major commodities handled in the Hamburg - Le
 
21  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Havre range, in four differen scenarios. By incorporating expert-based analysis the proposed
methodology provide valuable modifications to the assumption and results of the traditional
transport model used in the study. Moreover, qualitative analysis (expert judgements) allows to
identify throughput drivers for a number of commodify and likely disruptions of prior growth
trajectories are identified. In addition, the paper demonstrated how relevant commodity specific
trends are in forecasting overall port cargo throughput. Instead, Zhang et al. (2013) develop a
combined model composed of grey-forecast and logistic-growth curve model in order to improve
the accuracy of forecast model of cargo throughput for ports. They test this model on Chinese ports
in the 2002-2011 timeframe, and demonstrate that combined models can obtain relatively higher
forecast accuracy when it is difficult to gather sufficient data through field / desk research. In
addition, the Authors state that forecast obtained with the combined model they propose are more
accurate than any in individual ones. Finally, Xie et al. (2013) propose three hybrid approaches
based on least squared support vector regression (LSSVR) model in order to forecast container
throughput in two main Far East ports. Their outcomes demonstrate that the proposed hybrid
approaches can achieve better forecasting performance than individual approaches.

Tables 2 and 3 report some of the most recent studies on forecasting technique in the port domain,
focusing on port cargo throughput and container throughput. For each contribution key research
dimensions are highlighted, including: authors, year of publication, main topic, aims of the paper,
geographical coverage, method, techniques/models performed, sample timeframe, temporal
perspective, time unit, accuracy measures and main conclusions.

Overall, the review performed on extant port literature facing the issue of forecasting methods and
techniques provide some useful insights and recommendations applicable within the PORTOPIA
project.

- Both qualitative and quantitative forecasting methods appear viable within the port context.
- Port cargo throughput and container throughput emerge as the predominantly investigated
phenomena in extant port literature, due to their relevance for port planning and
development.
- Both short-term and long-term perspectives are adopted in forecasting the above-
mentioned phenomena. In this perspective, qualitative methods, e.g. Delphi method,
market research, scenario-based forecasts, etc., as well as some quantitative techniques,
which paved on time series analysis, e.g. classical decomposition model (X-11), SARIMA,
etc., are the solutions preferred by academics for short-term predictions in the port domain.
Also neural network approach is demonstrated to be an accurate technique but requires
more data inputs. Conversely, time series analysis and causal models (traditional regression
models, modified regression models, etc.) are commonly used in forecasting both port cargo
throughput and container throughput in the long term.
- The results of prior studies on this topic unveil a certain degree of uncertainty in
determining the best solution to adopt. In this vein, diverse accuracy measures used to
assess the goodness of the statistical models proposed and developed can lead to different
conclusions. Therefore, in the PORTOPIA Project, the adoption and the comparison of
various accuracy measure is suggested. Root mean squared error (RMSE), mean absolute
error (MAE) and mean absolute percent error (MAPE) are the preferred accuracy measures
used in similar studies.
- The selection of a specific forecasting technique does not depend only on the chosen
temporal perspective (short/medium/long term) but also on the object of the forecast (i.e.
the variable investigated in the forecast). So different forecasting methods and ad-hoc
forecasting models should be implemented for each type of variable. Of course, this
approach is not preferable within the PORTOPIA dashboard, which includes many relevant
 
22  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

indicators. Hence, we could use (more) time-consuming and expensive techniques for
developing some models capable to forecast the most significant variables in the port
domain. Port cargo throughput and container throughput seem to be the most valuable
variables to assess.
- Even when focusing on a single variable, forecasting methods should consider several
aspects related to the investigated variable. If container throughput is estimated, for
example, diverse types of traffic, e.g. seaborne traffic and inland waterways traffic (Lam et
al., 2004), or import/export traffic flows (Coto-Millán et al., 2005), should be treated in a
different manner by the predicting model. In this perspective, a forecasting model which
fits well for a gateway port should not be optimal for a transshipment hub. As a results,
various alternative specifications are expected to be introduced in order to take this concern
into account. For example, de Langen et al. (2012) propose to consider three types of
container volumes: direct deepsea, shortsea, and transshipment flows.

 
23  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Geographical   Sampled   Temporal  

Author Year Topic Aims  of  t he  paper Method Technichs  and  m odels Time  unit Accuracy  m easures  used Main  conclusions
coverage timeframe perspective
The  a nalysis  demonstrates  that  neural  network  (NN)  models  
The  paper  proposes  a nd   Mean  a bsolute  error  
Lam  ,  Asce,   are  more  a ccurate  than  r egression  a nalysis  when  their  
develops  neural  network   Quantitative   (MAE);  R2  (squared  
Ng,   Port  c argo   Short  term;   Yearly   results  a re  c ompared  with  the  observed  data.  Further  
2004 analysis  for  forecasting  37   Hong  Kong   (Causality   Neural  Network  a pproach 1983  -­‐  2003 multiple  c orrelation  or  the  
Seabrooke,   throughput Long  term. data robustness  c hecks,  moreover,  s hows  that  the  r eliability  of  the  
types  of  freight  movements  i n   analysis) coefficient  of  
Hui   proposed  NN  forecasting  models  decreases  with  the  
Hong  Kong. determination).
increasing  time  horizon.
The  paper  proposes  a   The  outcomes  demonstrate  that  the  total  forecast  errors  for  
Import   modified  r egression  model   Quantitative   the  proposed  modified  r egression  model  i s  l ower  than  that  
Chou,  Chu   Yearly  
2008 containers   for  forecasting  the  volumes   Taiwan (Causality   Modified  r egression  model 1989  -­‐  2001 Long  term Total  error. for  traditional  r egression  models,  which  do  not  c onsider  a nd  
and  Liang data
throughput of  Taiwan's  i mport   analysis) modify  the  errors  produced  by  the  non-­‐stationary  
containers. contribution  c oefficient.
Six  univariate  models:  
The  purpose  of  the  s tudy  i s  to   classical  decomposition  
Taiwan  (3   Root  mean  s quared  error   The  empirical  i nvestigations  demonstrate  that  the  c lassical  
search  a  model  that  i s   model  (X-­‐11);  trigonometric  
major  ports:   Quantitative   (RMSE);  Mean  a bsolute   decomposition  model  i s  the  best  forecasting  model  s ince  i t  
Container   capable  of  generating  the   regression  model;  r egression   Jan.  2003  -­‐   Monthly  
Peng  a nd  Chu 2009 Keelung,   (Time  s eries   Short  term. error  (MAE);  Mean   has  the  l owest  values  of  a ll  the  three  used  performance  
throughput most  a ccurate  prediction  of   model  with  s easonal  dummy   Dec.  2006 data
Taichung  a nd   analysis) absolute  percent  error   measures.  The  s easonal  dummy  r egression,  c onversely,  i s  the  
container  throughput  useful   variables;  grey  model;  hybrid  
Kaohsiung) (MAPE). second  best  model  for  forecast  a ccuracy.
for  port  a uthorities. grey  model  a nd  SARIMA  
model.
The  c ontribution  a ims  a t  
creating  a n  optimal  
predictive  model  of  volumes  
Genetic  programming  (GP);   The  outcomes  s uggest  that  the  s equence  of  c ontainer  
of  c ontainer  throughput  a t  
Quantitative   decomposition  a pproach  (X-­‐ throughputs  a t  s ampled  ports  has  a  valuale  trend  a nd  
Chen  a nd   Container   ports  by  using  genetic   Jan.  1978  -­‐   Short  Term;   Monthly   Mean  a bsolute  percent  
2010 Taiwan (Time  s eries   11);  s easonal  a uto   seasonal  c ycle  nature.  X-­‐11,  SARMIA  a nd  GP  forecasts  provide  
Chen throughput programming  (GP),   Dec.  2006 Long  term data error  (MAPE).
analysis) regression  i ntegrated  moving   all  a ccurate  predictions,  but  the  GP  i s  proved  to  be  the  
decomposition  a pproach  (X-­‐
average  (SARIMA) optimal  method  for  predicting  c ontainer  throughput.
11)  a nd  s easonal  a uto  
regression  i ntegrated  moving  
average  (SARIMA).
The  manuscript  a ims  to   The  r esults  c ontribute  to  the  knowledge  of  what  factors  a re  
Gosasang,   explore  a nd  c ompare  the   important  a nd  a ffect  to  the  c ontainer  volumes  which  
Quantitative   Neural  Network  a pproach   Root  mean  s quared  error  
Chandraprak Container   neural  networks  method  a nd   Thailand   Jan.  1999  -­‐   Monthly   forecasting  technique  i s  the  most  a ppropriate  for  predicting  
2011 (Causality   and  l inear  r egression   Long  term (RMSE);  Mean  a bsolute  
aikul  a nd   throughput linear  r egression  technique   (Bangkok  Port) Dec.  2010 data container  throughput.  The  outcomes,  i n  fact,  demonstrate  that  
analysis) technique error  (MAE).
Kiattisin for  predicting  the  c ontainer   the  a ccuracy  of  l inear  r egression  technique  i s  l ess  than  
throughput  a t  Bangkok  Port. neural  network  a pproach  (using  multilayer  perceptron).

Table  2  Forecasting  methods  and  techniques  in  the  port  domain:  recent  academics  contributions  on  cargo  and  container  throughput  

Source: own elaboration.

 
24  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Geographical   Sampled   Temporal  

Author Year Topic Aims  of  t he  paper Method Technichs  and  m odels Time  unit Accuracy  m easures  used Main  conclusions
coverage timeframe perspective
The  paper  proposes  a  
forecasting  a pproach  that  
uses  a  model,  i n  
combination  with  expert  
The  Authors  a rgue  that  the  forecasting  a pproaches  
judgements  a nd  c ommodity   Regression  model  c ombined  
de  Lange,   Qualitative   predominantly  a dopted  i n  l iterature  r ely  uniquely  on  trend  
Port  c argo   specificy  r esearch.  This   Hamburg  -­‐  Le   with  expert  j udgements,   Yearly  
Meijeren,   2012 and   1998  -­‐  2008 Long  term Not  a pplicable. forecasts  a nd  trend  based  models.  These  models,  a re  not  
throughput approach  i s  a pplicated  to   Havre  r ange market  r esearches  a nd   data
Tavasszy Quantitative capable  to  c apture  disruptions  of  historical  patterns,  
forecasts  for  2030  volumes   scenarios
especially  for  c ertain  types  of  c ommodities.  
of  a ll  major  c ommodities  
handled  i n  the  Hamburg  -­‐  Le  
Havre  r ange,  i n  four  differen  
scenarios.
The  paper  a ims  to  develop  a  
combined  model  c omposed   The  outcomes  unveil  that  the  c ombined  model  c an  obtain  
of  grey-­‐forecast  a nd  l ogistic-­‐ Quantitative   Organic  c ombination  of  grey   relatively  higher  forecast  a ccuracy  when  i t  i s  not  easy    to  find  
Zhang,  Huang,   Port  c argo   Yearly  
2013 growth-­‐curve  model  to   China (Time  s eries   forecast  a nd  l ogistic  c urve   2002  -­‐  2011 Long  term Mean  s quare  error  (MSE). in-­‐depth  i nformation.  I n  a ddition  the  forecast  obtained  
Zhao throughput data
improve  the  a ccuracy  of   analysis) model throughout  the  c ombined  model  a re  more  a ccurate  than  a ny  
forecast  model  of  c argo   in  the  i ndividual  ones.
throughput  for  ports.
Three  hybrid  a pproaches  
based  on  l east  s quares  
Quantitative   The  r esults  s uggest  that  the  proposed  hybrid  a pproaches  c an  
Xie,  W ang,   Container   support  vector  r egression   Shangai,   Hybrid  a pproaches  based   Jan.  2001  -­‐   Monthly  
2013 (Time  s eries   Short  term RMSE,  MAE  a nd  MAPE achieve  better  forecasting  performance  than  i ndividual  
Zhao  a nd  Lai throughput (LSSVR)  model  for  c ontainer   Shenzhen LSSVR  model   Aug.  2012 data
analysis) approaches.  
throughput  forecasting  a t  
ports  a re  proposed.  
Uncertainty  r elated  to  explanatory  variables  i s  a  s erious  
concern  i n  l ong-­‐term  forecasting.  Hence,  s cenario  a nalysis  i s  
The  a im  of  the  manuscript  i s   a  viable  tool  for  i nvestigating  the  i mpact  of  various  s ources  
to  provide  a  decision   of  the  economic  a nd  transport  trends  on  the  c ontainer  
Rashed,  
support  i nstrument  for   Hamburg  -­‐  Le   throughput  future  trajectories.  The  l iterature  r eview  
Meersman,  
Container   decision  makers  a nd   Havre  r ange   Yearly   performed  i n  the  s tudy  unveils  that  further  r esearch  a re  
Sys,  Van  de   2015 Qualitative Scenario-­‐based  forecast 1986  -­‐  2014 Long  term Not  a pplicable.
throughput stakeholders  by  estimating   (Port  of   data needed  i n  this  a rea  a nd  a re  beneficial  both  to  the  i ndustry  a t  
Voorde,  
long-­‐term  forecasts  of  the   Antwerp) the  port  l evel  a nd  to  the  policy  makers  a s  c oncerns  
Vanelslander
container  throughput  a t  the   investments  a nd  planning  decisions.  Data  demonstrates  that  
port  l evel. the  behaviour  of  the  r elationship  between  the  c ontainer  
throughput  a nd  the  economic  a ctivity  c hanged  a fter  the  
financial  c risis  i n  2008.

Table  3  Forecasting  methods  and  techniques  in  the  port  domain:  recent  academics  contributions  on  cargo  and  container  throughput  (continued)  

Source: own elaboration.

 
25  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

3.4 Assessment  of  techniques  in  port  data  forecasting  

Not all discussed methodologies are applicable in the Portopia forecast. Some issues may be
present due to lack of available data, on other cases some methodologies might not allow the
necessary precision or achievement of goals.
In order to provide the best possible results we will only take into account the methods which score
above “fair” (all green values in the table).

For short term forecasting this would limit the selection (in order of quality) to:

• Market  research  
• Box  Jenkins  
• X11  
• Trend  projections  
• Exponential  smoothing  
• Econometric  model    
• Regression  

At first sight it would seem that market research, a qualitative methodology would be optimally
combined with a Box Jenkins methodology. However Box Jenkins and X11 techniques are quite
complex and hard to implement in a system. Trend projections and exponential smoothing on the
other hand, both time series techniques are a good fit and both good to very good in accuracy.
Implementation should be less costly and when combined with a qualitative aspect be perfectly
able to forecast 1-3 quarters. The best match for short term forecasting would be a combination of a
qualitative techniques with a time series analysis. Market research is already implemented in the
project and could be complemented with external validation like the Delphy Method. For time
series analysis trend projections or exponential smoothing are advised due to their low complexity
and fast possible implementation. A slight advantage is put on trend projections since they allow to
identify turning points better than exponential smoothing.
For medium term forecasting:

• Econometric  model  
• Regression  
• Market  research  
• X11  
• Trend  projections  

Medium term forecasting is best performed using causal models according to the table.
Regressions and econometric models are best in accuracy and identification of turning points. This
is a valuable option for this type of forecasting, however these methods require a lot of data and
indirect indicators. Market research is a valuable option and already performed within the project.
For X11 implementation the same issues exists as with short term forecasting. Trend projections
are on the other hand an easy to implement guide for medium term forecasting and as an added
bonus allow for a good identification of turning points. As best match a more in depth market
research is proposed with trend projections allowing for a fast implementation and good accuracy.
For long term forecasting:

• Econometric  model  
• Regression  
 
26  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

• Trend  projections  

Long term forecasting is often the most complex and sensible in selection of techniques. As with
medium term forecasting econometric models and regression analysis are the best match. If data is
available these two are an option but require a lot of input and work. Trend projections are also in
this case a good match making it a good to very good option for all types of forecasting. The
Commission already has a long term forecasting model based on econometric principles, it would
seem advisable that this model is maintained and combined with qualitative forecasting techniques
or trend projections from the Portopia program in order to validate the results.

3.5 Combining  port  development  plans  with  forecasting  

Cargo demand forecasting is a crucial activity for planning business development and building
logistics infrastructures. For this reason, seaports are required to carefully monitor market changes
and estimate industry trends before undertaking huge investments in new facilities that notably
commit resources in the long-term. Hence, the realization of port infrastructures needs a great
amount of financial public and private resources and an endowment of technical and organizational
capabilities. Port planning consists in a complex analytical work that should be able to match cargo
flow projection and future demand estimation with the setting of a suitable supply of infra- and
supra-structures.
In this regard, Port Authority (PAs) is the public management body entitled to take executive
decisions on port planning with the aim to enlarge port spaces and achieve a more efficient
exploitation of the overall area. In so doing, the PA is expected to attract the interest of private
partners for co-financing, developing and running new logistics facilities. The undertaking of sound
forecasting activities by PA should unveil future market opportunities in various maritime business
segments, facilitating a smooth allocation of technical and financial resources in the new ventures.
The analysis of a number of European Core Ports shows a quite dis-homogeneous behavior of PAs
in relation to planning activities (i) and cargo forecasting (ii) (Tables 4 and 5). Overall, 28 PAs
located in 16 EU countries are scrutinized, taking into account website contents, and focusing on
statistics and historical traffic data, forecasting exercises and key archival documents, e.g., Master
Plans, Annual Reports, Green Reports, etc. The degree of sophistication and thematic scope of the
available planning documents are heavily affected by the diverse managerial approach of the PA in
relation to strategic decisions and development plans. Indeed, in a number of cases, port planning
activity is not practically formalized into an overarching document capable to summarize the main
contents of the planning activity. Especially in the past, in fact, PAs were used to develop port areas
and infrastructures by performing “step by step” expansion programs. These short/mid-term actions
resulted into temporally fragmented development projects, drove to forget an unitary and logically-
interconnected vision in port growth. In this perspective, PAs frequently undertook investments in
specific business areas (and not in other), attracted by some favorable demand forecasts (that, in the
end, may materialized in practice or even not). This mode of acting was also partially justified by
those imitative choices of PAs that simply followed the mainstream behavior of the closest
competitors. This definitely represents a myopic way of conceiving port planning, thus easily
leading to inefficient and sub-optimal solutions. In recent years, however, a growing number of PAs
was able to realize overarching long-term documents and planning efforts, ensuring economies of
scale and scope as well as a more conscious analytical perspective in the modelling of port

 
27  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

expansion programs. In Europe, ports like Rotterdam, Hamburg, Antwerp made pioneering attempts
in conjugating forecasting and planning activities, keeping the pace of the quick and profound
modifications of the competitive landscape (e.g., containerization, diffusion of ICT, deployment of
mega-vessels, growing market concentration, etc.).
As mentioned, the analysis of planning documentation and website contents regarding sample PAs
enables to identify relevant differences in their strategic and forecasting behavior. In particular, the
main addressed issues are the following.
1) A growing number of PAs develops formalized and well-established planning documents
whereas other (medium-sized and minor) ports are acting on an informal and fragmented basis, by
approving a (individual) project after the other. In the latter case, port development is not built
around a common backbone but it is more the result of spot opportunities or political/market
pressures.
2) Where available, port planning documents may have various forms and typologies depending on
their role within the overall planning exercise. In particular, such documents may differ in terms of
time horizons (mid-term, long-term) and nature (stand-alone vs. updating documents, strategic vs.
tactical documents). Quite often the same PA can realize a two-tier planning activity based on the
realization of documents addressing diverse temporal ranges (short/mid vs. long-term). Therefore,
planning documents have a differentiated structure and a diverse portfolio of contents. Such
documents are also characterized by a different degree of flexibility: some documents strictly define
future investments and port development objectives, whereas others only sketch broader strategic
views and project ideas/proposals. In practice, this intrinsic heterogeneity translates into a diverse
naming:
• Business Plan
• Strategic Plan
• Master Plan
• 3-year Operating Plan
• Strategic Development Plan
• Progress Report.

3) In its planning documents, the PA may have a diverse attitude to the explicit disclosure of the
contents and outcomes of the forecasting exercise. The release of detailed information regarding
forecasting techniques and input data may provide insights on the accuracy and reliability of long-
term visioning and strategic objectives. In other words, the disclosure of data and method regarding
purpose-made forecasting might reinforce or weaken the arguments in support of investment
decisions both in economic and social terms. Generally speaking, PAs are quite reluctant to provide
information on this, preferring to keep some information asymmetries between themselves and the
community of stakeholders. Apart from a few exceptions, forecasts seem almost “taken for granted”
as planning documents are fully concentrated to describe the contents of expansion programs and
investment timing and content.
4) Next to this, the PA attitude to publicly disclose on their company website traffic data, expansion
programs, forecasting information, and planning documents is also a theme that should deserve
more attention by practitioners and policy makers. The scrutiny of PA websites unveils a diverse
willingness of PAs to disseminate information about planning and forecasting studies. Some PAs,
like Rotterdam and Hamburg, extensively disclose throughput statistics and traffic data time-series,
 
28  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

as well as show their estimations for (long-term) economic and demand growth. On the contrary,
other PAs report a minimum amount of historical data and do not provide (on the website) any
information regarding planning documents and forecasting exercises.
5) In the PAs showing an attitude to the systematic realization of demand forecasts, it is relevant to
understand the “developer” of the forecasting elaborations and outcomes (e.g., ad-hoc task force,
internal bureau, business consultants, etc.). Typically, within their own organization, major PAs
dedicate a specific function to port planning. In this regard, a team of experts might be dedicated to
forecasting activities for supporting investment decisions. In other port contexts, instead, where the
traffic size is rather small or the experience of managers in forecasting is weak, PA normally
resorts to external advisors for the definition of some predicting variables and the set-up of an
appropriate forecasting methodology.

 
29  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Country Core  Ports Statistics  and  historical  data Annual  Reports  (on-­‐line) Website  contents  and  other  (on-­‐line)  documents

"2020  Ma s ter  Pl a n"  i s  a pproved  a nd  i ncl udes  s ome  foreca s ti ng  e xerci s e  (the  ful l  vers i on  i s  
BE ANTWERP La rge  a mount  o f  d a ta  a nd  s tati s ti cs  a re  a va i l a bl e From  2006  o nwa rds
not  a va i l a bl e  o n-­‐l i ne)  (LF)
A  d etai l ed  S ecti on  o n  tra ffi c  s tati s ti cs  p er  s egment  i s   "Stra tegi c  Port  I nfra s tructure  Project  2013-­‐2022"  i s  u nder  d evel opment.  F ul l  d ocumentati on  i s  
BE ZEEBRUGGE From  2008  o nwa rds
provi ded  (from  2004  o nwa rds ) not  p rovi ded  o n-­‐l i ne.  S ome  foreca s ti ng  i s  i ncl uded  i n  thi s  l ong-­‐term  p roject   (LF)
"Stra tegi c  Pl a n  2010-­‐2020"  w a s  m a de  p ubl i c  i n  2010  a nd  i t  i s  a va i l a bl e  o n-­‐l i ne.  S i mpl i fi ed  
BE GHENT Some  d a ta  a nd  s tati s ti cs From  2009  o nwa rds
tra ffi c  foreca s t  e s ti ma ti ons  a re  s hown  i n  the  p l a n  (LF)
BU BURGAS Onl y  a  few  o pera ti ona l  i nforma ti on  a va i l a bl e Not  a va i l a bl e No  i nforma ti on  a bout  foreca s ti ng
In  the  w ebs i te  there  i s  S ecti on  ca l l ed  "Pl a ns "  tal ki ng  a bout  e nvi ronmental  p rotecti on  
Some  o pera ti ona l  s tati s ti cs  o n  p ort  o pera ti ons  a nd   progra mmes  i n  the  m i d-­‐term  (2010-­‐2013);  i n  a ddi ti on,  there  i s  a  S ecti on  ca l l ed  "Bus i nes s  
CRO RIJEKA Not  a va i l a bl e
throughput  (from  2009  o nwa rds ) potenti a l "  d ea l i ng  w i th  the  d evel opment  a nd  i mprovement  o f  i nfra s tructures  i n  contai ner  
a nd  p a s s enger  s egments .  S ome  rough  l ong-­‐term  foreca s ts  a re  i ncl uded   (LF)
"Port  D evel opment  Pl a n  to  2025"  (on-­‐l i ne).  S ome  foreca s t  e s ti ma ti ons  a re  i ncl uded  i n  l ong-­‐
From  2008  o nwa rds .  S ome  d a ta  a nd  
Long-­‐term  h i s tori ca l  d a taba s e  o n  p ort  throughput  p er   term  p l a nni ng  d ocuments  (LF).  A  ra ther  s ophi s ti ca ted  a na l ys i s  i s  p erformed  i n  rel a ti on  to  
DE HAMBURG foreca s ti ng  i nforma ti on  a re  i ncl uded  
tra ffi c  s egment.  La rge  e mpi ri ca l  b a s e. foreca s ti ng.  I n  the  l ong-­‐term  p l a nni ng  d ocumentati on  s ome  m ethodol ogi ca l  n otes  a nd  
(LF)
foreca s ti ng  o utcocomes  a re  d i s cl os ed  (per  b us i nes s  s egment)  (LF)
Detai l ed  d a ta  a nd  s tati s ti cs  p er  commodi ty  (from  
DE BREMEN  PORTS From  2011  o nwa rds No  rel eva nt  i nforma ti on  a bout  foreca s ti ng
2005  o nwa rds ).
Somme  b a s i c  s tati s ti cs  a re  a va i l a bl e  o n  the  H TML  
DK COPENHAGEN  MALMÖ pa ges  o f  the  w ebs i te.  D eeper  contents  a re  i ncl uded   From  2001  o nwa rds No  rel eva nt  i nforma ti on  a bout  foreca s ti ng
i n  the  Annua l  R eport.
"III  S tra tegi c  Pl a n    2015-­‐2020"  i s  a pproved.  The  contents  o f  the  Ma s ter  Pl a n  h a ve  b een  
Detai l ed  d a ta  a nd  s tati s ti cs  p er  commodi ty  (from  
ESP BARCELONA From  2006  o nwa rds di s cl os ed  b y  the  Port  Authori ty  Pres i dent  o n  F ebrua ry  2015.  The  ful l  (el ectroni c)  vers i on  i s  
1996  o nwa rds ).
not  a va i l a bl e  yet.  H owever,  s ome  l ong-­‐term  foreca s t  e s ti ma ti ons  h a ve  b een  rel ea s ed   (LF)
"Stra tegi c  Pl a n  for  2020"  i s  a pproved  a nd  currentl y  u nder  d evel opment.  The  ful l  vers i on  i s  
Detai l ed  tra ffi c  s tati s ti cs  a nd    i nforma ti on  (2013  
ESP VALENCIA From  2005  o nwa rds not  a va i l a bl e  o n-­‐l i ne.  S ome  fra gmented  i nforma ti on  d i s cl os es  fi gures  o n  future  tra ffi c  
onwa rds )  a re  d i s cl os ed
es ti ma ti ons  (LF)
ESP ALGECIRAS Some  tra ffi c  s tati s ti cs  from  1999  o nwa rds From  2013  o nwa rds No  rel eva nt  i nforma ti on  a bout  foreca s ti ng
ESP BILBAO Some  h i s tori ca l  d a ta  a nd  i nforma ti on  a re  a va i l a bl e From  2003  o nwa rds No  rel eva nt  i nforma ti on  a bout  foreca s ti ng
Some  d a ta  a nd  i nforma ti on  a re  a va i l a bl e  i n  a  
s tati s ti ca l  S ecti on,  b ut  l i mi ted  i n  ti me  (2013   A  s ecti on  o n  "Performa nce  res ul t  a na l ys i s "  i s  p rovi ded  (2007  o nwa rds ),  b ut  n o  rel eva nt  
EST TALLINN From  2000  o nwa rds
onwa rds ).  Mos t  d a ta  a re  i ncl uded  i n  the  Annua l   i nforma ti on  a bout  foreca s ti ng  i s  d i s cl os ed
Report.

Notes :
SF Short-­‐term  F oreca s t  (0-­‐1  yea r)
MF Mi d-­‐term  F oreca s t  (1-­‐5  yea rs )
LF Long-­‐term  F oreca s t  (beyond  5  yea rs )

Table  4  The  disclosure  of  forecasting  contents  in  Port  Authority’s  website,  planning  documents  and  annual  report.  

 
30  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Country Core  Ports Statistics  and  historical  data Annual  Reports  (on-­‐line) Website  contents  and  other  (on-­‐line)  documents
Ba s i c  tra ffi c  s ta ti s ti cs  a nd  Key  Perfoma nce  I ndi ca tors  
FR MARSEILLE-­‐FOS From  2012  o nwa rds No  rel eva nt  i nforma ti on  a bout  foreca s ti ng
(KPIs )  a re  a va i l a bl e  ( from  2011  o nwa rds )
The  i ntegra ted  s tra tegi c  p l a n  "HAROPA  2030"  a i ms  to  d evel op  a  l ong-­‐term  d evel opmen  o f  
Some  tra ffi c  s ta ti s ti cs  a re  a va i l a bl e  ( from  2010  
FR LE  H AVRE From  2011  o nwa rds the  fol l owi ng  p orts :  Le  H a vre,  R ouen  a nd  Pa ri s .  The  d ocumenta ti on  i s  a va i l a bl e  o n-­‐l i ne  a nd  
onwa rds )  p er  b us i nes s  s egment
conta i ns  l ong-­‐term  foreca s ti ng  a nd  trend  a na l ys i s  from  va ri ous  p ers pecti ves   (LF)
Some  h i s tori ca l  d a ta  a nd  i nforma ti on  a re  a va i l a bl e   From  2006  o nwa rds .  S ome  s hort-­‐term   "Bus i nes s  Pl a n  2014-­‐2018"  h a s  b een  a pproved  ( a va i l a bl e  o n-­‐l i ne).  S ome  m i d-­‐term  foreca s ts  
GR PIRAEUS
(from  2008  o nwa rds ) foreca s t  i s  i ncl uded  (SF) a re  p rovi ded  p er  tra ffi c  s egment  (MF)
"The  Ma s terpl a n  2012-­‐2040"  h a s  b een  a pproved  ( a va i l a bl e  o n-­‐l i ne).  I t  conta i ns  a n-­‐hoc  
Some  tra ffi c  s ta ti s ti cs  a re  a va i l a bl e  ( from  2007  
IRL DUBLIN From  2006  o nwa rds Secti on  "Foreca s ts "  w hi ch  i ncl udes  a  ra ther  s ophi s ti ca ted  l ong-­‐term  a na l ys i s  rega rdi ng  
onwa rds )  p er  b us i nes s  s egment
economi c  p rojecti ons  a nd  e xpected  tra de  vol umes  (LF)

IT GIOIA  TAURO Some  b a s i c  s ta ti s ti cs  a nd  tra ffi c  d a ta  a re  a va i l a bl e From  2013  o nwa rds  ( ba s i c) "3-­‐yea r  Opera ti ng  Pl a n"  i s  a pproved  b ut  n o  rel eva nt  i nforma ti on  a bout  foreca s ti ng  i s  a dded

In  the  w ebs i te  there  i s  a n  a d-­‐hoc  S ecti on  o n  "Pl a ns  a nd  foreca s ts "  i ncl udi ng  s ome  b a s i c  
tra ffi c  e xpecta ti ons  p er  ca rgo  s egment  ( mi d-­‐term)  (MF);  "3-­‐yea r  Opera ti ng  Pl a n"  ( 2015-­‐2017)  
Some  h i s tori ca l  d a ta  a nd  i nforma ti on  a re  a va i l a bl e  
IT GENOVA From  2009  o nwa rds i s  a pproved  b ut  n o  rel eva nt  i nforma ti on  a bout  foreca s ti ng  i s  i ncl uded;  the  n ew  "Ma s ter  
(from  2008  o nwa rds )
pl a n"  h a s  b een  i nforma l l y  ci rcul a ted  a mong  s ta kehol ders :  i t  conta i ns  l ong-­‐term  foreca s ts  
(LF)
Not  a va i l a bl e  o n  the  PA  Authori ty  
webs i te.  An  a bs tra ct  o f  the  Annua l  
"3-­‐yea r  Opera ti ng  Pl a n"  h a s  b een  i s s ued  b ut  n o  rel eva nt  i nforma ti on  a bout  foreca s ti ng  i s  
Some  s ta ti s ti cs  a re  a va i l a bl e  i n  d i fferent  s ecti ons  o f   Report  i n  i ncl uded  i n  the  Annua l  
IT LA  S PEZIA a ddd;  i n  a ddi ti on,  the  "Ma s ter  Pl a n"  a pproved  i n  2006  ( a va i l a bl e  o n-­‐l i ne)  d oes  n ot  i ncl ude  
the  w ebs i te  ( more  s ys tema ti ca l l y  from  2011  o nwa rds ) Report  o n  I ta l i a n  Ports  rel ea s ed  b y  
a ny  s ubs ta ti a l  i nforma ti on  a bout  tra ffi c  foreca s t
the  Mi ni s try  o f  Tra ns port  a nd  
Infra s tructures
Deta i l ed  d a ta  a nd  s ta ti s ti cs  p er  commodi ty  ( from   Ma s er  p l a n  h a s  n ot  b een  a pproved  d es pi te  the  fi rs t  d ra ft  w a s  i s s ued  b y  the  Port  Commi ttee  
IT NAPOLI From  2013  o nwa rds  ( ba s i c)
2001  o nwa rds ). i n  2000;  there  i s  n o  o ffi ci a l  i nforma ti on  a bout  foreca s ti ng  o n  the  PA  w ebs i te.
Some  h i s tori ca l  d a ta  a nd  i nforma ti on  a re  a va i l a bl e   "3-­‐yea r  Opera ti ng  Pl a n"  i s  a pproved  ( 2013-­‐2015)  b ut  n o  rel eva nt  i nforma ti on  a bout  
IT VENEZIA From  2007  o nwa rds
(from  2006  o nwa rds ) foreca s ti ng  i s  a dded
La rge  a mount  o f  d a ta  a nd  s ta ti s ti cs  a re  a va i l a bl e  i n  a   "Bus i nes s  Pl a n  2011-­‐2015"  i ncl udes  m i d-­‐term  foreca s ti ng   (MF);  i n  the  "Port  Vi s i on  2030"  p l a n  
NL ROTTERDAM s ophi s ti ca ted  S ecti on  o f  the  w ebs i te.  La rge  e mpi ri ca l   From  2000  o nwa rds s ome  l ong-­‐term  tra ffi c  foreca s ti ng  i s  p rovi ded.  Es ti ma ted  fi gures  a re  regul a rl y  u pda tes  i n  
ba s e. the  "Progres s  R eports "  (LF)
"Port  Vi s i on  2008-­‐2020"  p l a n  i s  u nder  d evel opment  a nd  i ncl udes  l ong-­‐term  foreca s ti ng  
La rge  d a ta ba s e  o n  s ta ti s ti cs  a nd  h i s tori ca l  tra ffi c  
NL AMSTERDAM From  2008  o nwa rds es ti ma tes ;  a  "Port  Vi s i on  2030"  d ocument  h a s  b een  a l s o  d ra fted:  i t  u da tes  tra ffi c  foreca s ts  
da ta  ( from  e a rl y-­‐1990s  o nwa rds )
(LF)
Some  h i s tori ca l  d a ta  a nd  tra ffi c  i nforma ti on  a re   "Stra tegy  2027"  p l a n  h a s  b een  i s s ued  i n  2013.  I t  conta i ns  l ong-­‐term  e conomi c  a nd  tra ffi c  
PL GDANSK Not  a va i l a bl e
a va i l a bl e  ( from  2004  o nwa rds ) foreca s ts  (LF)
"Stra tegi c  D evel opment  Pl a n  ( 2025)"  h a s  b een  i s s ued  b y  PA;  the  ful l  d ocument  i s  n ot  
Tra ffi c  s ta ti s ti cs  a re  a va i l a bl e  b ut  contents  a re  
PT LISBON From  2006  o nwa rds a va i l a bl e  o n-­‐l i ne.  Onl y  a  s hort  b rochure  i s  p rovi ded  a nd  thi s  d oes  n ot  conta i n  a ny  reference  
i rregul a r  a nd  fra gmented  ( from  1997  o nwa rds )
to  l ong-­‐term  tra ffi c  foreca s ti ng
Some  h i s tori ca l  d a ta  a nd  tra fi c  i nforma ti on  a re   "Ma s ter  Pl a n  2014-­‐2020"  h a s  b een  i s s ues  b ut  d ocumenta ti on  i s  n ot  a va i l a bl e  o nl i ne;  the  PA  
RO CONSTANTA From  2001  o nwa rds
a va i l a bl e  p er  commodi ty  ( from  2006  o nwa rds ) devel oped  s ome  l ong-­‐term  foreca s ti ng  b ut  i nforma ti on  a re  fra gmented  (LF)
"Ma s ter  Pl a n  2035"  w a s  a dopted  i n  2014  b y  the  PA;  ful l  d ocumenta ti on  i s  n ot  a va i l a bl e  o n-­‐
La rge  d a ta ba s e  o n  s ta ti s ti cs  a nd  h i s tori ca l  tra ffi c  
SW GOTHENBURG Not  a va i l a bl e l i ne,  b ut  s ome  n ews  o n  thi s  u nvei l  the  p res ence  o f  foreca s ti ng  a cti vi ty  i ns i de  the  p l a nni ng  
da ta  ( from  e a rl y-­‐1970s  o nwa rds )
document   (LF)

Notes :
SF Short-­‐term  F oreca s t  ( 0-­‐1  yea r)
MF Mi d-­‐term  F oreca s t  ( 1-­‐5  yea rs )
LF Long-­‐term  F oreca s t  ( beyond  5  yea rs )

Table   5   The   disclosure   of   forecasting   contents   in   Port   Authority’s   website,   planning   documents   and   annual   report   (continued).

 
31  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

4 FORECASTING  EXAMPLES  IN  SIMILAR  SECTORS  

4.1 Air  Transport  

4.1.1 Boeing  and  Airbus  forecasts  

The most frequently consulted literature in air cargo business on forecasting air cargo
are the market forecasts by Airbus and Boeing. These largest airplane manufacturers
issue yearly reports on the state of air cargo and the expected future demand. The
forecasts in both manufacturers’ reports are created using basic econometric
techniques, such as trend analysis. The expected demand is discussed for continental
trade flows, such as Asia to North America or Europe to Africa without a distinction in
transported commodities. An issue with these two forecasts is that there is often a bias
present for ‘positive growth results’ due to the fact that the authors are airplane
manufacturers and therefore benefit from growing cargo flows.

The companies release demand forecasts with compound annual growth rate (CAGR)
for aggregated trade flows, such as Europe to North-America. The forecast approach is
very similar for both companies and includes:

• Econometric  modeling,  a  tool  in  determining  the  overall  importance  of  economic  
factors  
• Explicitly  named  techniques  are  not  available  in  the  report,  but  additional  
correspondence  with  an  analyst  from  Boeing  clarified  the  econometric  modeling  to  be  
trend  analysis      
• Explicitly  named  economic  factors  are  GDP,  transportation  costs,  exchange  rates,  real  
income,  investments,  export/imports,  industrial  production  and  relative  prices  
• Judgmental  modifications  to  account  for  changes  in  growth  factors.  Examples  include  
air  service  agreements,  trade  quotas,  restrictions  on  night  operations,  increase  in  
capacity,  route  re-­‐structuring  

4.1.2 The   International   Air   Transport   Association   (IATA)   Airline   Industry  

Forecast  2014-­‐2018    

The IATA forecast also works with an annual growth rate (CAGR) for the calculation of
its 5 year based forecasts. (total market growth prediction is currently of 4.1% over the
next five years). Emerging economies, particularly in the Middle East and Africa, will
be the fastest-growing markets. The Airline Industry Forecast Is (according to the
authors) the most comprehensive tool for analyzing the future freight traffic demand,
as well as the evolution of global air travel market.
The report provides detailed five-year traffic forecasts for over 3,000 individual
country-pairs, plus aggregate results at regional and global levels. The forecast is
derived from the results of a survey done to the industry’s major airlines, civil aviation
and airport authorities. The aggregation rules for forecasting include: over 1,000
unduplicated international country pairs, including aggregated values for 6 world
regions, 17 world sub regions, and over 900 country to sub region forecasts.

Following indicators are provided by the IATA platform:

• Traffic  volume,  number  of  tonnes  of  freight    

• Compound  Annual  Growth  Rate  (CAGR)    
 
32  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

• Traffic  growth  in  volumes  

A summary report by IATA Economics provides analysis of the macroeconomic events

shaping the industry and their direct implications for airlines’ projections. It also
identifies key challenges and opportunities across top routes and regions. The users get
access to annual reports and gain the fundamental tools and data that are required for
basic forecasting. This report allows them to understand the market shape and identify
the most important new emerging markets as well as where to maximize business
investment and build effective strategy plans with intelligence directly from the
industry's leading experts.

4.1.3 Lessons  from  the  air  transport  forecasting  

The air business is an interesting sector since it is closely related to the maritime
industry. Notable links are:

• High  demand  for  freight  and  passenger  transport    

• Forecasting  mostly  done  for  fleet  development      
• Fleets  are  also  high  in  capital  investment  
• Often   commissioned   by   liner   companies   in   order   to   assess   current   order   book  
evolution  

Off course it is not possible to compare all aspects of the airline industry with the
maritime sector. Main differences include:
• Cargo  is  often  time-­‐sensitive  and  high  value:  perishables  
• Cargo  is  often  process  critical;  medicine  or  machine  parts  
• Cargo  is  of    very  high  value;  high-­‐tech  or  precious  metals  

Keeping these similarities and differences in mind the following lessons can be derived
from the existing forecasts.

• There   are   many   freight   rates:   for   different   cargoes,   different   vessels,   and   different  
contract   types.1   But   these   different   freight   rates   are   often   surprisingly   strongly  
correlated  over  time,  when  seen  in  a  multi-­‐year  perspective.  This  is  mainly  because  of  
the   strong   degree   of   substitutability   of   cargoes   among   vessels   across   routes:   most  
cargoes   can   be   transported   on   a   number   of   different   ships,   so   conditions   spread  
quickly   from   trade   to   trade.   Secondly,   investors   and   financial   institutions   are  
undifferentiated  as  to  specific  trades  and  thus  provide  additional  substitutability.  
• Aggregation  should  be  available  on  a  multitude  of  levels  with  a  maximum  of  flexibility  
• Objectivity  should  be  a  major  parameter  avoiding  the  skewedness  created  by  industry  
forecasters  
• The   IT   system   should   allow   for   a   maximum   of   user   freedom   and   interpretation   but   the  
added   value   of   an   annual   comprehensive   report   is   substantial   as   is   proven   by   the   IATA  
example  

4.2 Rail  transport  

4.2.1 Railway   Demand   Forecasting   and   Service   Planning   Processes   (Rail  

Freight  Service  Review)  

 
33  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

These reports are based on extensive interviews with CN and CP subject matter experts
directly involved in and responsible for the development of the railways’ demand
forecasting and service planning processes. QGI has also relied on the experience and
background of the team members that participated in the interviews and prepared this
report. There are two types of forecasting included, demand forecasting for medium-
long and short term planning. Many railway planning processes including financial,
asset, capital investment and service planning are predicated on the railways’ estimate
of the freight volumes (demand) to be handled within a given time period. For this
information to be meaningful and support the railways’ planning processes demand
forecasts must not only provide an estimate of the total volumes but also identify the
commodities, timing, and the origins and destinations of the traffic. Medium to long
term planning activities provide the railways with a highly aggregated view of demand
for time series forecasting.

Figure  10  Transnet  level  of  detail  

Forecast information can be input at any level of detail and the planning system will
automatically aggregate up or disaggregate down. Disaggregation of forecast data to the
lowest level of detail is done on the basis of historical traffic movements. Detailed
traffic forecasts, including those at the individual customer level, can subsequently be
adjusted manually to introduce more current customer or market information.
While shippers have expressed the view that they believe railway demand forecasts do
not always reflect the true demand they communicate to the railways the railways argue
that they must weigh the input of shippers against the expected performance of broader
markets and consider potential volatility in demand. QGI’s interviews with railway
representatives revealed that many believe that some of their customers have difficulty
providing accurate medium to longer term demand forecasts. Railway staff identified
many reasons for this including:
• Inefficiencies in customers’ logistics management processes;
• unrealistic or inaccurate market forecasts and market share projections;
 
34  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

• an unwillingness of customers to reveal their own marketing projections.

At an aggregate level railway forecasts of demand are accurate within approximately
10% over the course of a year

4.2.2 Transnet  freight  demand  forecast  

The transnet forecasting methodology has been included for its completeness and high
level of complexity. The model encompasses multiple layers and is used for all strategic
matters ranging from allocation of resources to the creation of a long term capacity
planning framework.
Planning layer Source Forecast period Objective
Operational Demand Contracts and customer Daily/weekly/monthly/ To allocate resources
needs, resource annually according to contracted
planning and demand.
scheduling.
MDS Demand Market intelligence, 7 – 10 years To plan and fund
(Corporate Plan) commercial agreements, business objectives
30-year Demand Book, against confirmed and
capacity planning anticipated customer
demand.
LTPF Demand Modelled 30 years To provide a planning
macroeconomic growth, framework for capacity
industry sector growth expansions.
projections, historic
data, 30-year Demand
Book.

Table  6  The  transnet  forecasting  timeframe    

It is a combination of 4 models namely:

• Freight  Demand  Model  (FDM):  which  investigates  the  sources  of  supply  and  demand  in  
the  economy,  disaggregated  to  352  districts  and  72  commodities.  This  model  
essentially  translates  economic  activity  in  the  form  of  currency  (Rand)  into  production  
and  consumption  of  goods  in  the  form  of  tons.  The  forecasts  generated  by  this  model  
are    based  on  macroeconomic  growth  scenarios  
• Liquid  Fuels  Demand  Model  (LFDM):  The  LFDM  uses  supply-­‐side  capacity  intelligence  
and  matches  the  fuel  production  forecasts  with  demand  and  then  supplies  the  
projected  shortfalls  with  fuel  imports.  
• Transportation  Model  (TTM)  The  TTM  uses  gravity  flow  methodology,  a  well-­‐
established  technique  to  model  the  flow  of  goods,  people,  and  so  forth.  The  route  that  
freight  will  follow  is  determined  by  the  least  total  resistance  of  the  connecting  links  
from  origin  to  destination.  
• Market  Share  Model  (MSM):  is  created  in  order    to  calculate  the  rail  addressable  
market  and  to  determine  rail  targets  over  the  longer  term.  By  comparing  planned  
seven-­‐year  volumes  with  what  is  available  in  the  market,  it  enables  the  planners  to  
check  the  realism  of  short  term  targets  and  make  informed  longer-­‐term  projections  for  
each  commodity  and  on  each  route.  

 
35  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Figure  11  The  transnet  forecasting  methodology  

The input drivers for freight forecasting are a combination of factors which ultimately
are converted to sector and commodity-specific compound annual growth rates.
Macroeconomic inputs include, import/ export forecasts, industry trends, Government
expenditure and global trends both regionally and worldwide.

4.2.3 Lessons  from  the  rail  transport  industry  

• Always  aggregate  based  on  commodities  and  geographical  area  

• Keep  a  clear  split  between  long,  medium  and  short  term  
• Look  at  proxies  and  drivers  and  make  sure  that  the  right  drivers  are  linked  to  the  right  
underlying  commodities  

4.3 Traffic  flow  forecasting  

4.3.1 Traffic   flow:   Scenario,   Traffic   Forecast   and   Analysis   of   Traffic   on   the  
TEN-­‐T,  Taking  into  Consideration  the  External  Dimension  of  the  Union.  

This forecast was included since it has a direct link with the Portopia work due to its
focus on the European transport sector and the TENT network. Since it was published
in 2009 we will focus on the methodology and not the results.
The study uses the TRANS-TOOLS model and has provided a coherent and reliable
forecast for the future traffic flows in EU and the neighboring countries for 2020 and
2030 with particular focus on flows between old and new Member States, between new
Member States and between EU and the neighboring countries. The analysis has
covered all transport modes, both freight and passenger traffic on links and through
nodes with a focus on the TEN-T network.

The study uses proxies as drivers in order to estimate the future traffic flows and
creates 2 scenarios for the situation in 2020 and 2030 a Baseline scenario and a
scenario termed “Sustainable Economic Development”. Generally speaking, there are
two types of drivers respectively increasing or decreasing transport demand. Examples
of the first type could be economic growth or population growth. An example of the
second type could be increasing energy prices.

 
36  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

The baseline is a “Business as Usual” scenario including already agreed infrastructure

and policy measures and in line with the current trends. The Sustainable Economic
Development scenario describes a faster economic and demographic development,
higher fuel costs and provide for an intensive development of the road and rail
networks.
The Baseline and the Sustainable Economic Development scenarios have been
illustrated by short descriptions of the possible futures and by parameters
corresponding to the input for the TRANS-TOOLS model. On this basis forecasts for all
modes for 2020 and 2030 in the Baseline and Sustainable Economic Development
scenarios for both freight and passenger transport have been produced. Subsequently,
these forecasts have then been applied for identifying the most important
infrastructure axes for cohesion, development of the internal market and the relations
to neighboring countries.

4.4 Academic  research      

The airline research More academic literature uses, amongst other techniques: The
Weighted Majority Algorithm (WMA) lets a set of macroeconomic variables make
individual forecasts which assigns normalized weights to the predictors based on their
in-sample forecasting accuracy, variations on the Box-Jenkins method described in
previous sections (Faraway and Chatfield, 1997) ,A self-organized, five-layer neuro-
fuzzy model is developed to model the dynamics and forecast air passengers (Chen,
2012), A system dynamics simulation (Suryani et al., 2012) and one of the more recent
approaches uses a support vector machine to test a different number of input nodes to
optimize the predicted cargo own from Beijing to Shanghai (Heng, 2013).

Academic attempts will not be discussed in detail in this deliverable since they often
require too much time or resource investment and are too complex to implement.

5 FORECASTING  IN  PORTOPIA  

To handle the increasing variety and complexity of managerial forecasting problems,
many forecasting techniques have been developed in recent years. Each has its special
use, and care must be taken to select the correct technique for a particular application.
The selection of a method depends on many factors—the context of the forecast, the
relevance and availability of historical data, the degree of accuracy desirable, the time
period to be forecast, the cost/ benefit (or value) of the forecast to the company, and the
time available for making the analysis.
The factors below have been taken into account with the selection of forecasting
techniques:
What is the cost associated with developing the forecasting model, compared with
potential gains resulting from its use?

• How  complicated  are  the  relationships  that  are  being  forecasted?  

• Is  it  for  short-­‐run  or  long-­‐run  purposes?  
• How  much  accuracy  is  desired?  
• Is  there  a  minimum  tolerance  level  of  errors?  
• How   much   data   points   are   available?   Techniques   vary   in   the   amount   of   data   they  
require.  
 
37  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

5.1 Methodology  

The goal of this deliverable is to determine the forecasting technique to be used in the
system which is applicable to all developed indicators. Each indicator is investigated
based on its compatibility with the proposed methodologies. The forecasting will be
split up in short- medium and long term periods for maximum relevance and should be
implementable into the IT system supporting the Portopia platform.

5.1.1 Short  term  

As can be seen from the previous sections the short term forecasting timelines are often
between a few days and a couple of months in length. The less complex models often
give quite accurate matches limiting allocation of resources.

Figure  9  

Figure  12  variability  on  timeframes  

Time frame

In order to create a relevant forecasting methodology a difference has to be made

between forecasting timeframes. As a rule of thumb the closer the forecasting frontier
the more reliable the result. Therefore different methodologies have to be applied to
differe nt timeframes.

Within the system the short term forecasting will focus on a period of 1 to 4 quarters.
The quarter (3 months) is used as a basis since it is the smallest possible timeframe
within the Portopia data reporting system for all proposed indicators. This means that
the system should allow for a quarterly update of the indicator forecast.

Applicable methodologies

For the short term forecasting system it is important that all proposed methodologies
are totally automated and can run with minimum input from system administrators. In
line with section [nr] we will only use the selected methodologies which are applicable
to the project and compare them to the different possible indicators

 
38  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Market  trends  and  structure  indicators Forecast  possible Smallest  timeframe Added  value  of  forecast Timeline
Call  size Yes Annually High M
GDP  growth  vs  tonne  per  type  of  cargo Yes Annually High M L
Intra-­‐European  traffic  i ncidence  i n  Eu  ports Yes Quarterly High S M L
Maritime  traffic Yes Quarterly High S M L
Market  share Yes Quarterly High S M L
Modal  split Yes Annually High L
Traffic  Growth Yes Quarterly High S M L
Transhipment  i ncidence  i n  EU  ports Yes Quarterly High S M L

Table  7  Market  trends  and  structure  indicatos  forecast  overview  

The market trends and structure indicators are best suited for forecast applications.
Due to their high data availability and comparability forecast can be made on multiple
levels and multiple forecast periods.
Socio  economic  indicators Forecast  possible Smallest  timeframe Added  value  of  forecast Timeline
Direct  employment Yes Annually High M L
Direct  value  added Yes Annually High M L

Table  8  Socio  economic  indicatos  forecast  overview    

Socio economic indicators are suited for medium and long term forecasts. The
complexity involved renders them difficult to incorporate into the IT systems.
Environmental  and  safety  indicators Forecast  possible Smallest  timeframe Added  value  of  forecast Timeline
Air  quality Yes Annually Medium M L
Carbon  footprint Yes Annually Medium M L
Days  l ost Yes Annually Medium M L
Definition  of  objectives  and  targets  for  e nvironmental  i mprovement No
Differentiated  fees  for  clean  shipping No M L
Energy  consumption Yes Annually Medium
Environmental  management  i ndex No
Environmental  policy  makes  reference  to  e spo  guideline  documents No
Environmental  responsibilities  of  key  personnel  documented No
Environmental  training  program  for  port  e mployees No
Existence  of  an  e nvironmental  management  system No
Existence  of  an  e nvironmental  monitoring  program No
Existence  of  an  e nvironmental  policy No
Fatal  work  accidents Yes Annually Medium M L
Inventory  of  relevant  e nvironmental  l egislation   No
Inventory  of  significant  e nvironmental  aspect  for  the  port  area No
Investment  i n  protection No
LIquefied  natural  gas  bunkering No
Marine  e cosystems No
Nautical  accidents Yes Annually Medium M L
Noise Yes Annually Medium M L
On-­‐shore  power  supply No
Port  security  i ncidents Yes Annually Medium M L
Publication  of  a  publicly  available  e nvironmental  reports No
Sediment  quality No
Soil  quality No
Terrestrial  habitats No
Top  ten  e nvironmental  priorities No
Waste No
Water  consumption Yes Annually Medium M L
Water  quality No
Work  related  accidents Yes Annually Medium M L
Logistics  chain  and  operational  performance  i ndicators No
Average  port  dues  per  ton No
Average  THCs No
Landside  congestion Yes Annually Medium M L
Maritime  connectivity No
Maritime  fluidity  i ndicator No
Roro  connectivity No

Table  9  Environmental  indicatos  forecast  overview  

 
39  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Environmental and safety indicators are harder to forecast since a lot of these
indicators are binary or qualitative. The indicators that are possible to forecast can only
be forecasted in medium to long term forecasts.
Governance  indicators Forecast  possible Smallest  timeframe Added  value  of  forecast Timeline
Autonomous  management No
Extent  of  performance  management No
Integration  of  the  port  cluster No
Intermodal  container  connectivity No
Market  openness No
Port  authority  e mployee  productivity Yes Annually Low M L
Port  authority  i nvestments Yes Annually Low M L
Reporting  Corporate  social  responsibility No
Safety  and  security  i ndicator No

Table  10  Governance  indicatos  forecast  overview  

Like the socio economic indicators, governance indicators are also difficult to forecast.
Only investments and productivity can be forecast but the added value of forecasting
these indicators is rather limited.
User  perception  indicators Forecast  possible Smallest  timeframe Complexity  of  forecast Timeline
Overall  user  perception  quality  i ndex No

Table  11  User  perception  indicatos  forecast  overview  

Short term forecast product development

Time series forecast system

The time series forecast will be applies to all indicators with sufficient data points and
is based on a moving average or exponential smoothing mother due to their good
matches with the forecasting timeframe and low resource requirement. They can be
incorporated into the dashboards fairly easily and offer a quick overview of the
expectations of the coming months.
Once the forecast methodology is further developed other options could be included
into the system allowing the user to switch between different forecast techniques.
Qualitative forecast/ confidence index

In addition to the basic times series forecast based on the indicators developed during
the project we propose the introduction of the confidence index. A forecast technique
based on the qualitative market research/Delphi method methodology. Inclusion of this
parameter has multiple advantages:

• It  can  be  used  as  a  check  for  the  time  series  forecast  
• A  new  indicator  is  created  within  the  Portopia  project  
• If  can  be  introduced  within  the  system  in  combination  with  the  quarterly  update  of  
data  
• Combines  the  strengths  of  qualitative  and  time  series  forecasting  

5.1.2 Medium  term  

Timeline

 
40  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

The medium term forecasting timelines runs from 1 to 5 years, due to the low update
requirement the automation level of this type of forecast is not as high as with the short
term forecast tool. This will allow for a comparison to other sectors.
Methodologies
Possible methodologies for allowing medium term forecasting include market research,
historical analogy, X11, Trend projections, regression and econometric IO models. The
proposed approach for medium term forecasting would be to use proxy forecasting
applied in other sectors. This means linking the indicators to certain sectors which
drive the underlying proxy.
Medium term forecast product
Annual forecast update
This product would combine the medium and long term forecasting sections. This
annual update and extraction from the system will allow an overview of the forecasting
data and work performed.

5.1.3 Long  term  

Timeline
Long term forecasting will deal with a timespan of over 5 years and with a maximum of
20 years comparable to the masterplan forecasting done by ports. This renders the
forecast quite complex and with a high degree of uncertainty.
Applicable methodologies

Trend projections is the only time series technique applicable for long term forecasts
with acceptable standards. Due to the high degree of complexity the more resource
intensive methodologies are better suited for this type of analysis. A full econometric
model, regression analysis or input output model would be best suited for this
timeframe. The issue is that these type of analyses are not possible within this project
since they require too much resources, be it data storage or man hours.
Long term forecasting product

Update of DG move model

The PwC/NEA (2013) and TRANStools and Primes/Tremove models were used for the
traffic forecast analysis conducted for the preparation of the White Paper on EU
Transport Policy in 2011. The forecasts are based on a scenario that assumes a status
quo of existing policies and already planned policy reforms, but at the same time
assumes that these policy reforms do not create a level playing field for all 319 TEN-T
ports (of which 83 ports are part of the core network). Following the low growth
scenario the cargo throughput in the ports of EU27 would grow from 3.6 billion tons in
2011 to 5.8 billion tons in 2030. The EC argues that this growth would cause capacity
problems and an unbalanced use of the port system and associated network.

 
41  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Region   Container Dry  Bulk Liquid  Bulk Other  Cargo RoRo Total
UK/Ireland   125.74 155.43 297.49 137.46 35.26 751.39
Nordic   50.53 187.66 240.30 122.01 81.87 682.37
South  Baltic 19.91 158.09 88.92 17.68 39.39 323.98
Hamburg-­‐France   595.58 434.53 571.20 186.83 138.26 1,926.40
Iberia   217.28 176.38 213.45 38.34 50.98 696.44
Italy/Malta   179.00 112.67 261.87 80.05 64.24 697.83
Balkan/Aegean   120.80 156.28 122.21 50.50 128.72 578.51
Black  Sea 8.22 69.73 28.90 1.53 37.81 146.19
Total   1,317.06 1,450.77 1,824.34 634.40 576.53 5,803.11

Table  12  EU  2030  port  traffic  by  region  of  loading/unloading  according  to  PWC  (2013)  

Source:  EC  2013

The forecasts were obtained by running a combination of several models namely a GDP
growth model PRIMES (which calculated the long term average GDP growth rates in
the EU of 1.4%) and the Trans-tool model. As mentioned in the original document these
projections must be taken with caution because of the multiple underlying
assumptions. Particular attention must be given to new developments, for example the
introduction of new or raising trade barriers or further world trade liberalization. The
scenario used as a baseline assumes that the current state of affairs will prevail; it does
not consider a sensitive analysis about possible trade agreements. However, the
forecasting methodology used by PwC/NEA does not take into account competitive
cargo shifts between ports. The baseline forecast only allows mapping the geographical
distribution of trade to change, but it does not contain any assumptions about
competition between ports.
The proposition for the long term forecast is to interpret the results obtained from the
DG move model and update them into the annual forecast document provided by the
system.

5.1.4 Implementation  

The next phase of the project will include implementation of the IT system. During the
first phase of implementation the forecast system will be created for all compatible
indicators which are already online.
During the second phase a more complex forecast system will be integrated into the
existing system allowing the users more freedom in analysis. Also the annual forecast
update will be developed and made available on the website.

 
42  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

5.2 References  

Blauwens, G., & Van Steelandt G. (1992). Prognose van het Containerverkeer in de
Haven van Antwerpen, Antwerp.
BTRE – Bureau of Transport and Regional Economics (2002). Australia’s Seaborne
Containerised Freight. Forecasts to 2010-11, Working Paper #50, Department of
Transport and Regional Services, Commonwealth of Australia, Canberra, August.
Chen S-H., Chen J-N. (2010). Forecasting container throughputs at ports using genetic
programming. Expert Systems with Applications, 37: 2054-2058.
Chou, C. C., Chu, C. W., Liang, G. S. (2008). A modified regression model for
forecasting the volumes of Taiwan’s import containers. Mathematical and Computer
Modelling, 47(9), 797-807.
Coto-Millán, P., Baños-Pino, J., Castro, J. V. (2005). Determinants of the demand for
maritime imports and exports. Transportation Research Part E: Logistics and
Transportation Review, 41(4), 357-372.
Dagenais, M. G., & Martin, F. (1985), “Forecasting Containerized Traffic for the Port
of Montreal (1981-1995)”, Cahier de recherche #8504, Département de Sciences
économiques, Université de Montréal.
De Langen, P.W., van Meijeren, J., Tavasszy, L.A. (2012). Combining Models and
Commodity Chain Research for Making Long-Ter Projections of Port Throughput: an
Application to the Hamburg-Le Havre Ragnge. European Journal of Transport and
Infrastructure Research, 12(3): 310-331.
Drewry Shipping Consultants (2006). The Drewry Annual Container Market Review
and Forecast 2006/07.
ESCAP – Economic and Social Commission for Asia and the Pacific (2005). Regional
Shipping and Port Development Strategies (Container Traffic Forecast), United Nations,
New York.
Flemish Port Commission (1996), Annual Report.
Gosasang, V., Chandraprakaikul, W., Kiattisin, S. (2011). A comparison of traditional
and neural networks forecasting techniques for container throughput at Bangkok port.
The Asian Journal of Shipping and Logistics, 27(3), 463-482.
Hamal, K., James, B., & Cregan, M. (2006). Port Level Forecasts of Container and Ship
Movements in Australia: 2004-05 to 2024-25, paper presented at the 29th Australian
Transport Research Forum Conference, Gold Coast, 27-29 September.
Hui, E. C. M., Seabrooke, W., & Wong G. K. C. (2004). Forecasting Cargo Throughput
for the Port of Hong Kong: Error Correction Model Approach, in Journal of Urban
Planning and Development, Vol. 130, No. 4.
Iannone, F., (2007). Modelli di previsione a lungo termine della domanda di
movimentazione di contenitori marittimi nel sistema portuale campano. Società Italiana
degli Economisti dei Trasporti - IX Riunione Scientifica. Napoli.

 
43  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Jiang, F., Lei, K. (2009). Grey Prediction of Port Cargo throughput Based on GM(1,1,a)
Model. Logistics Technology, 9, 68-70.
Jugović, A. Hess, S., & Poletan Jugović T. (2010). Traffic Demand Forecasting for Port
Services Promet – Traffic & Transportation, Vol. 23, 2011, No. 1, 59-69
Lam, W.H.K., Asce, M., Ng, P.L.P., Seabrooke, W., Hui, E.C.M. (2004). Forecasts and
reliability Analysis of Port Cargo Throughput in Hong Kong. Journal of urban planning
and development, 130(3): 133-144.
Langen, P. W. de (2003). Forecasting Container Throughput: a Method and
Implications for Port Planning. Journal of International Logistics and Trade, Vol. 1,
No.1.
MDS Transmodal (2006). UK Port Demand Forecasts to 2030. Final Report.
MDS Transmodal (2007). Supply and Demand in the Container Trades: Global and
Regional Trends and Forecasts 2006.
Meersman, H., Moglia, F., & Van de Voorde, E. (2002). Forecasting Potential
Throughput, in Huybrechts M., Meersman H., Van de Voorde E. Van Hooydonk E.,
Verbeke A., Winkelmans W. (eds.), Port Competitiveness. An Economic and Legal
Analysis of the Factors Determining the Competitiveness of Seaports, Editions De
Boeck Ltd., Antwerp.
Ocean Shipping Consultants (1997). The European Container Market Prospects to 2008.
Ocean Shipping Consultants (2004). Marketing of Container Terminals.
Ocean Shipping Consultants (2006). The European & Mediterranean Containerport
Markets to 2015.
Peng, W.-Y., Chu, C.-W. (2009). A comparison of univariate methods for forecasting
container throughput volumes. Mathematical and Computer Modelling, 50: 1045-1057.
Rashed, Y., Meersman, H., Sys, C., Van de Voorde, E., Vanelslander, T. (2015).
Scenario-based forecasting of container throughput: A conceptual framework. IAME
Conference 2015, Kuala Lumpur, Malaysia.
RMPA - Rotterdam Municipal Port Authority (1998). Annual Report.
Rodríguez García, T., Gonzales Cancelas, N., Soler-Flores F. (2013). Forecasting
models in ports transport systems Are ANNs Applications the solution?. Electronic
International Interdisciplinary Conference September, 2. - 6. 2013, 509-514.
Ross, D.F. (2011). Distribution Planning and Control: Managing in the Era of Supply
Chain Management, Edition 2, Springer Science & Business Media.
SEA - Study Centre for the Expansion of Antwerp (till 1985). Prognose van de Trafiek
van de Haven van Antwerpen, Antwerpen, several editions.
Smyth, T. D. (1998). Forecasting Trade for Port Projects in the Developing World. Data
sources and interpretation. Seaport Newsletter, No. 2, Seaport Consultants Canada Inc.

 
44  
 Deliverable  1.3  
Port  Traffic  Forecasting  Tool  

Sun, N. C., Bunamo, M. C. (1973). Competition for Handling U.S. Foreign Trade
Cargoes: The Port of New York’s Experience, in Economic Geography, Vol. 49, No. 2.
Talluri, K.T., Ryzin G.V. (2004). The theory and practice of revenue management.
Kluwer Academic Publishers.
Ueda H., Kado H., Miyake K., Nagano H. (2005). An Analysis of Marine Container
Transportation in the Asian Region, in Proceedings of the Eastern Asia Society for
Transportation Studies, Vol. 5.
Van Straelen, R. (1993). Econometrische Ramingen op Lange termijn van het
Containerverkeer in de Haven van Antwerpen, in Blauwens G., De Brabander G., Van
de Voorde E. (eds.), De dynamiek van een haven: liber amicorum aangeboden bij het
emeritaat van prof. Fernand Ridder Suykens, Kapellen, Pelckmans.
Verbeke, A., Declercq, E., Teurelincx, D. (1996), The Future Growth of Container
Traffic in the Port of Antwerp, in Winkelmans W., Notteboom T. (eds.), The Issue of
Prognoses in Maritime Transport and Seaport Production in Belgium and Poland: a
Conceptual and Empirical Study, Garant, Leuven.
Xiao, Y., Xiao, J., Wang, S. (2012). A hybrid forecasting model for non-stationary time
series: an application to container throughput prediction. International Journal of
Knowledge and Systems Science (IJKSS), 3(2), 67-82.
Xie, G., Wang, S., Zhao, Y., Lai, K. K. (2013). Hybrid approaches based on LSSVR
model for container throughput forecasting: a comparative study. Applied Soft
Computing, 13(5), 2232-2241.
Xu, Z.G. (2011). Based on the prediction of self-regression model of the Port of
Shanghai port cargo throughput. Journal of Wuhan Technical College of
Communications, 13(1), 71-74.
Zhang, C., Huang, L., Zhao, Z. (2013). Research on combination forecast of port cargo
throughput based on time series and causality analysis. Journal of Industrial
Engineering and Management, 6(1): 124-134.

 
45