0 - Cruz & Sarmiento, 2019 Forecast Innacuracy

Transportation
https://doi.org/10.1007/s11116-019-09972-y
Traffic forecast inaccuracy in transportation: a literature

review of roads and railways projects
Carlos Oliveira Cruz1 · Joaquim Miranda Sarmento2
© Springer Science+Business Media, LLC, part of Springer Nature 2019
Abstract
The inaccuracy of traffic forecasts has long stood as a central research theme in the field
of infrastructure and transportation studies. The literature presents several motives for
this phenomenon, ranging from a political bias, insufficient technical preparation, chang-
ing urban patterns, and economic dynamics. Uncertainty due to the inaccuracy of fore-
casts can have a profound impact on the infrastructure development process, right through
from the preliminary studies up until the operation and re-negotiation of contracts (in cases
when projects are developed using a concessions model). This paper provides an extensive
systematic review of forecast inaccuracy in roads and railways projects (analyzing trends,
causes, and results). The research found that: (1) forecasts in rail projects are generally
more optimistic than in road projects; (2) over the last couple of decades the accuracy of
forecasts has not improved significantly, and; (3) there has been a generalized ramp-up
effect in forecasts.
Keywords Traffic forecast · Forecast inaccuracy · Optimism bias · Systematic review ·

Strategic behaviour · Roads · Railways
JEL Classification H54 · J68 · R41 · R42
Introduction
Traffic forecast inaccuracy has been a recurrent topic in the transport research agenda for
several years (Nicolaisen and Driscoll 2014). The planning or construction of any transpor-
tation project or changes to existing infrastructures raises the question of who will use the
infrastructure and/or service, and how the users will benefit. These benefits include time
* Carlos Oliveira Cruz

oliveira.cruz@tecnico.ulisboa.pt
Joaquim Miranda Sarmento
jsarmento@iseg.ulisboa.pt
1
CERIS, Instituto Superior Técnico, Universidade de Lisboa, Av. Rovisco Pais, 1049‑001 Lisbon,
Portugal
2
CSG/Advance, ISEG ‑ Lisbon School of Economics and Management, Universidade de Lisboa,
Rua do Miguel Lupi nº20, 1200‑078 Lisbon, Portugal
13
Vol.:(0123456789)
Transportation
travel savings—which can represent up to 80% of the total benefits (Mackie et al. 2001),
lower generalised cost of travel, a lower C O2 footprint, and fewer road accidents, among
other economic, social, and environmental benefits. However, the measurement of these
benefits relies heavily on the forecast of traffic. Since the seminal work of Horowitz and
Emslie (1978), substantial research has focussed on identifying, analysing, and discussing
the main reasons and causes for a successive bias of forecasts towards excessive optimism,
with the subject deserving growing attention over the last years, as observed by an increas-
ing number of studies and papers. This growing attention is a direct result of an increasing
number of transportation projects, particularly in developing economies, but also because
we now have more data and a longer historical pattern for each project. Longer temporal
series allow having a more comprehensive understanding of how forecasted vs real demand
evolves over time.
Traffic forecasts are defined in the literature (Flyvbjerg 2007a; Nicolaisen 2012) as
being any demand, ridership, or traffic estimation, either for the short, medium, or long-
term, for a transportation project, particularly roads and railways. This can either be for
demand (passengers, passengers.km, vehicle, vehicle.km, heavy vehicles, etc.), or rev-
enues (tolls, ticket sales, etc.), and can have varying temporal scales (multiannual, annual,
monthly, daily, peak periods, weekends, business days, holidays, etc.).
Research on traffic forecasts has been in different perspectives and research areas. Some
authors address the problem of traffic forecast inaccuracy within more extensive studies on
the decision-making process of infrastructure investment. Examples of this perspective are
the works of Flyvbjerg (2007b) and Vickerman (2007). Other authors take the perspective
of asymmetric information problems within the relationship between agents in the devel-
opment and management of infrastructure projects, mainly when developed by the private
sector (e.g. Athias and Nuñez 2008) or when studying the overall risk analysis and uncer-
tainty within large infrastructure projects (Næss et al. 2015). Finally, other authors take
the analysis from a political bias perspective (Button et al. 2010). Although many studies
on infrastructure investment start at different points, many of them end up focusing on the
problem of forecast inaccuracy.
For our analysis, it is more relevant to understand what have been the main causes iden-
tified by previous studies that provide some evidence for the comprehension of this phe-
nomenon. But the results have been diverse. The main reasons presented by the literature
for traffic forecast have been: (1) opportunistic behaviour and optimist bias (e.g., Fouracre
et al. 1990); (2) inadequacy in forecast models and data (Welde and Odeck 2011 or Nico-
laisen 2012); (3) overall uncertainty (e.g., Næss et al. 2015); (4) changes in demographics
and land use patterns (e.g. Bain 2009); (5) project causes, such as quality or construction
delays or changes in contract scope, among others (e.g., Parthasarathi and Levinson 2010);
(6) competition and Demand (Bain 2009), and; (7) economic cycle (Mackinder and Evans
1981). A more comprehensive analysis of this motives will be provided later on, but it is
clear that the nature and type of causes are very distinct and cross-disciplinary.
The cross-disciplinary dimension of the subject is reflected by its mention in a wide
variety of journals and scientific areas, such as engineering, transportation, economics,
financing, accounting, law and regulation, and even psychology.
This paper provides a systematic review of past studies on forecast inaccuracy, identify-
ing and analysing the main deviation patterns, the methodologies used, and the main causes
of inaccuracy, among other issues. Following the classification of the literature review
papers developed by Wee and Banister (2016), this research falls within the category of
empirical insights (identifying the existing state of knowledge) and methodologies (provid-
ing an overview of dominant methodologies). The research is based on the construction of
13
Transportation
a database of papers on forecast inaccuracy in the road and rail sectors that were published
in ISI Web of Knowledge and Scopus-indexed journals. Additionally, this paper presents
an analysis of the deviation patterns, methodologies, and causes. Despite this scope, it must
be acknowledged that other sources of information may also be relevant for decision-mak-
ers, such as official reports, Court of Audits/National Audit Office reports and rating agen-
cies (for example Standard and Poor’s, and Moody’s) evaluations on projects. For instance,
a report of Fitch Ratings (2013), based in a set of 5 projects, is more detailed that academic
literature on the main causes for transport inaccuracy.
The research confirms some of the Literature’s previous findings, such as the existence
of non-normality of errors (i.e. the distribution of error does not follow a normal distri-
bution, but it is right skewed) with a bias towards over-optimistic forecasts, but with dis-
tinct patterns among road projects and railway projects. It is important to mention that the
non-normality of error is merely theoretical because even assuming that the error term fol-
lows a normal distribution, the projects with negative error terms would be less likely to
be selected, and developed, thus biasing the sample. We found that rail projects tend to
have higher mean deviations, i.e., differences between real and forecasted traffic, than road
projects, which might be caused by a political bias towards rail projects. Optimism bias is
consistently the main cause for forecast inaccuracy, followed by changing demographics
and land use patterns, and also problems and inaccuracies of forecast models and the data
used to estimate demand. This also helps to explain the fact that, over time, the evolving
methodologies have not been able to significantly and consistently increase the accuracy of
forecasts, even though they have become increasingly complex.
This paper is organised as follows: “Traffic forecasts in the infrastructure lifecycle” sec-
tion presents a theoretical overview of how the several stages of an infrastructure life cycle
are affected by traffic forecast; a systematic review of the literature and the principle results
of the research, with the identification and quantification of the most relevant existing past
studies and respective findings are presented in “Methodology and data” section ; “Sys-
tematic review of the literature” section provides a quantitative analysis of the results of
the Literature; “Data analysis” section presents a case study on the impact of forecast inac-
curacy in estimating the economic benefit of projects; and, finally; “Conclusions and policy
implications” section presents the main conclusions and policy implications.
Traffic forecasts in the infrastructure lifecycle
Bain (2009) developed the first and, to the best of our knowledge, only study to illustrate
how inaccuracy in forecasts affects several of the stages along the life cycle of an infra-
structure. The author argues that traffic forecasts have a fundamental role along the life
cycle of a road, particularly in different areas, such as: cost–benefit analysis (i.e., should
the project be developed or not?); the procurement model (i.e., what should the procure-
ment model be: public–private partnerships—PPP vs. traditional procurement?); setting
the value of the tolls; renegotiations (i.e., what are the outcomes and the compensation to
be paid to the private concessionaire based on the difference between forecasted and real
traffic), and; the “re-bid” (i.e., what is the right base price for the re-bid of a concession
when it finishes?).
How relevant are traffic forecasts in transport infrastructure projects? What impacts can
the erroneous forecasts have on infrastructure projects? Despite several classifications of
the stages of an infrastructure development lifecycle, (see, for instance, Lenferink et al.
13
Transportation
2013), we can divide the lifecycle into a three-stage classification: (1) Decision making, (2)
The Procurement Process, and; (3) Construction and Operation. The remainder of this sec-
tion provides an overview of how traffic forecasts, and, in particular, how inaccuracy, can
impact the different stages in the life cycle of an infrastructure.
Decision making
The decision making stage is the first stage, where the most relevant question for the pro-
ject must the answered: should the project be implemented? To answer this question, sev-
eral additional activities are required. For example, a preliminary planning of the system,
layout, and the capacity/service to be provided, among others. This is included in the plan-
ning and concept development activities, which is supported by technical and environmen-
tal/social and economic studies. This stage involves the development of feasibility studies
for verifying the technical, environmental/social, and economic feasibility. Calculating the
feasibility of a project, with its multidimensional nature, helps to answer the initial ques-
tion and support the go/no-go decision. These studies, particularly those with an economic
and environmental/social analysis, are typically carried out by using a cost–benefit analy-
sis (CBA), which constitutes the most commonly-accepted methodological best practice in
terms of project evaluation (Mackie 2010). In fact, in the EU, there are both national and
European regulation setting nature, type, and methodologies for developing CBA.
The problem of inaccurate forecasts is frequently addressed in the context of cost–ben-
efit analysis (Prest and Turvey 1965; Nash 1993; Damart and Roy 2009; Hyard 2012) or
investment appraisal stages, as these tend to be the most critical input. A social cost–ben-
efit analysis usually considers all direct and indirect costs (e.g., construction costs, opera-
tion costs, environmental costs, etc.), and all expected direct and indirect benefits (e.g.,
direct revenues, if applicable; time savings, C O2 emissions savings; road accidents reduc-
tion; etc.) (Hyard 2012). There are other types of benefits, such as growth in regional com-
petitiveness and job creation, lowering logistical costs, etc.
With regards to transportation projects, the most relevant benefits are revenues (e.g.,
road tolls, or fare revenues in railway systems) and time savings. Regarding tolls, it is
important to mention that these are not a benefit on their own, as they can be considered to
be transfers. Tolls represent a cost for the user, and will be a net benefit for whoever col-
lects the toll, which is usually a concessionaire, a public authority, or a third-party acting
on their behalf. For non-toll roads, the latter is one of the most relevant benefits, although
even for toll roads, time saving can be the greater benefit. Therefore, in a transport project,
traffic inaccuracy will lead to errors in forecasting revenues and also in the estimation of
the financial and economic benefits of a project.
In the decision making process there is an additional layer of analysis which is criti-
cal for the go/no go decision: affordability, i.e., the capacity of the public sector to be
able to provide the necessary long-term funding for the project operation. If the eco-
nomic CBA is positive, and the project should therefore be developed, then it may
require public subsidies, as happens with most rail-based transit services. Public author-
ities (local, regional or federal), need to consider whether the project is affordable, that
is to say, whether it is financeable over time, or does it represent a major stress for
public finances (European Investment Bank 2013). The impact of inaccurate forecasts
in the affordability analysis is critical. An overestimation of forecasts may lead the pub-
lic authorities to expect more future revenues than the project is capable of generating,
13
Transportation
thus jeopardising the project’s affordability. For the case of road projects, see Cruz and
Marques (2012a) for further discussion.
Forecast inaccuracy and procurement
The second stage is the “Procurement process”. This process involves not just the pro-
curement itself (i.e., through a public tender), but also carrying out the preparatory
studies that justify the best procurement option [e.g., traditional procurement vs. pub-
lic–private partnerships (PPPs)].
Even the choice of the PPP model is subjected to variability in traffic forecasts. The
most appropriate procurement model is the Public Sector Comparator (PSC) of Cruz
and Marques (2012b). The PSC calculation is the hypothetical cost of public manage-
ment (Quiggin 2004), which is calculated based on the estimation of capital expendi-
ture (CAPEX), operational expenditure (OPEX), and revenues (Sarmento 2010), and is
risk-adjusted. Based on the “estimated NPV”, the procuring authority verifies whether
the net present value (NPV) of the best private bid is better, or worse, which is referred
to as a ‘value for money analysis’ (Grimsey and Lewis 2005; Morallos and Amekudzi
2008). Different traffic forecasts can lead to a different balance between these variables,
meaning that the ‘value for money’ result will be different for each forecast. As differ-
ent realities emerge, the results can contradict the initial analysis and calculation of the
PSC, which can ultimately affect the decision of the optimal procurement model (tradi-
tional or PPP) for delivering a certain infrastructure project.
Forecast inaccuracy and operation/management
The third stage is that of “Construction and Operation”: this stage follows on from the
procurement process and involves all the activities related with the physical construc-
tion of the infrastructure (or the upgrading of an existing system) and the operation, var-
ying from simpler operation schemes (such as those for non-toll roads), to other much
more complex ones, such as railways operations.
Just as important as in the decision stage, traffic forecasts are extremely relevant dur-
ing the operation stage. Forecast levels of traffic are fundamental for adjusting the val-
ues of tolls and for operation and maintenance plans, and, consequently, for determining
the expected level of financing of the system. This is particularly relevant when PPP
models are used. The concession contract can contain a traffic forecast, which acts as a
guaranteed revenue. Depending on the country, the traffic forecast could be performed
by each of the potential bidders, or it can be provided by the contracting authority as
an input for potential bidders (this is the case in some concessions in Spain, Brazil,
and Portugal) (Oliveira et al. 2016). The potential inability to attain these predeter-
mined volumes can result in the need for a revision of the contract, or even renego-
tiation. Renegotiations of PPP contracts count among the main threats to a project’s
success, and, in studies developed by Guasch et al. (2006), Cruz and Marques (2013a),
Miranda Sarmento and Renneboog (2016a, b), Domingues and Sarmento (2016), Neto
et al. (2018), it has been found that optimistic forecasting ranks among the main reasons
for renegotiating contracts.
13
Transportation
Methodology and data
In this paper, we provide a literature review on traffic forecast inaccuracy in road and rail-
way projects, following the methodology described in Wee and Banister (2016). The analy-
sis of papers in the literature review aims to help readers get up-to-date with the specific
research topic through a well-structured overview of the existing Literature.
We provide a literature review of papers classified as ‘empirical insights’, as we identify
the current “state of art” of this topic. This review is a synthesis of what is already known,
and, in addition, we classify our study by the category of Literature reviews of methodolo-
gies as we provide an overview of the dominant methodologies used.
The first stage of the analysis of the literature involves the construction of a database
consisting of those papers published in the ISI Web of Science/Scopus databases which
include, totally, or partially, the following set of keywords: “traffic forecasts”; “traffic
deviations; “optimism bias”; “rail projects”; “road projects”, and; “traffic analysis”. In this
paper we have used strings such as transport* to both capture transport and transporta-
tion, as well as Boolean operators (such as “and”, “or”, and “not”—again following Wee
and Banister 2016). Additionally, we also use snowballing, as a paper may be cited in the
future by another paper (Jalali and Wohlin 2012). The second filter was to select only those
papers that addressed road and/or rail projects, which was the large majority of the papers
in the database. We ended up with a relatively large number of papers, although not so
large a number that would oblige imposing some kind of “stratification selection”.
The methodology adopted focussed on the following variables: (a) Mean deviation; (b)
Standard deviation; (c) Historical evolution of mean deviation; (d) Historical evolution of
standard deviation; (e) Type of variables used in the analysis, and; (f) Main causes of fore-
cast inaccuracy. For quantitative data, and again following Wee and Banister (2016), we
present the individual result of each study and then use the weighted average results (as
a simple average can be misleading), although we also present the whole range of results.
Systematic review of the literature
Literature on traffic forecast has been mainly focussed on two main areas: (1) an analysis of
the deviations per se (e.g., Horowitz and Emslie 1978)—i.e., a quantification of the main
deviations in a specific project or portfolio of projects, as well as the main causes of such
inaccuracy, or; (2) an analysis of the impacts (consequences) of the inaccuracy of forecasts
(e.g., Nicolaisen and Driscoll 2014; Salling and Leleur 2015). The first category contains
most of the studies published on traffic forecast, although the latter has gained some rel-
evant contributions over time, particularly regarding research related with concessions and
PPP renegotiations, most of which are related with traffic forecast problems. This section
provides an overview of the literature of traffic forecast inaccuracy, focussing first on the
quantitative results, and next on the causes for such inaccuracy.
The measure of inaccuracy in the literature is not standardised. Some authors (e.g., Fly-
vbjerg et al. 2005; Nicolaisen 2012) present the deviation as being the difference in per-
centage between real traffic/demand and that forecasted. For example, if the sign is nega-
tive, this means that the real value is below that forecasted (which was the notation that
we have used in this paper). Other authors (e.g., Baeza and Vassallo 2012; Bain 2009)
present values in distinct forms: lower real traffic is presented as a positive percentage (e.g.,
13
Transportation
a deviation of 30% means that the real traffic was 30% lower), and; deviation is the percent-
age of real traffic from that forecasted (e.g., 30% would mean that the real traffic was only
30% of that forecasted). The other difference is how the deviations are calculated. The most
common methods are: comparisons between annual, monthly or daily traffic/demand, and;
comparisons of accumulated traffic/demand volumes.
The calculation of forecast inaccuracy in year n ( FCn ) used in this research, is the same
as that developed by Flyvbjerg (2007a) and Nicolaisen (2012), which is shown in Formula
(1).
Rn − Fn
FCn = × 100 (1)
Fn
where Rn is the real traffic/ridership in year n, and Fn is the forecast in year n.
“Appendix 1” summarises the literature review, and it contains the studies collected,
the main authors, the number of observations of each study, calculated FCn or average FC
when the study does not provide an annual based calculation, standard deviations, and the
summary of the main causes for inaccuracy presented in the study.
There have been a growing number of publications over the last years, which indicates
that the problem of inaccuracy in forecasts is attracting significant attention from research-
ers. The studies differ in the size of samples (from 1 to 210 projects), but the results are
consistent, as discussed in the next section. Most studies show some evidence of overes-
timated traffic/ridership, although a significant number also demonstrates underestima-
tion. “Appendix 2” provides the summary of the main methodologies and variables that
have been used in forecasts. The cross analysis between “Appendices 1 and 2” will help to
understand the nature of inaccuracy and how it can be improved.
Data analysis
This section aims to provide a quantitative analysis of the actual deviations between fore-
casted and real traffic that has occurred in past studies, based on the above-mentioned
tables. More than just a descriptive analysis, this section identifies potential patterns in
both data and the main conclusions/findings of the research.
Patterns of historical deviations
We conducted this analysis for both road and railway projects, given that in the large major-
ity of studies a clear distinction is made between the two. Flyvbjerg et al. (2005) provide
evidence of some favouritism and strategic bias towards railway projects. This might be
linked with an overall strategy of decreasing dependency on road transport, which favours
rail modes, given their better environmental performance. For example, the Trans-Euro-
pean Transport Network clearly establishes as a priority more rail and port investments,
rather than new roads. It would be relevant to see whether there is evidence of significant
differences between the deviations found in road projects and those of railway projects.
Figure 1 presents the histograms of mean deviations for roads and rail projects, with data
showing a non-normality distribution of error. Flyvbjerg (2008) argues that if inaccuracies
were solely based on technical arguments, then the distribution of error would be normal and
centred on zero. In reality, if the error were random, as would result from non-biased uncer-
tainty, we would still expect some right-skewness because the projects negatively affected
13
Transportation
10
9
8
7
6
4
3
2
0
-60% -40% -20% 0% 20% 40% 60%
10
9
8
7
6
3
2
1
0
-60% -40% -20% 0% 20% 40% 60%
Fig. 1 Histogram of mean deviations for road projects (upper) and rail projects (lower)
by the error, would have a lower probability of selection. The error does affect the project
selection. Flyvbjerg (2008) shows that distributions of error does not follow a normal distribu-
tion, neither does it have a zero average. Our analysis shows some evidence that confirms this
conclusion.
After separating the two types of projects, the next step is to calculate a weighted average
deviation (WAD) found in the studies, considering the size of the sample, according to For-
mula (2):
13
Transportation
∑N � �
Δ × Si
i=1 � i �
WAD = ∑N (2)
S
i=1 i
where Δi is the average deviation of study i, considering N the total number of studies, and,
Si is the sample size of study i [it is important to use the average, as simple average results
could be affected by different sample sizes (Wee and Banister 2016)]. For example, if one
study (sample 100) demonstrated an average error of − 30% and another (sample 5) gave
− 10%, the WAD would be (|− 30%| * 100 + |− 10%| * 5)/(100 + 5) = 29.4%. As mentioned
above, a positive value means that the real traffic is above that forecast, with a negative
value meaning that real traffic is below that forecast.
For road projects, the WAD is 13.3%, whereas for the rail projects it is 29.8%. However,
using the real values (calculating the average with the real positive and negative values for
each one of the deviations, i.e., allowing “compensation” between the traffic above and
below the forecasts), we reach a value of − 9.3% for roads, and − 23.6% for railways. This
means that these results confirm that rail projects in fact have a higher propensity for larger
deviations, as presented by Flyvbjerg et al. (2005). Several reasons can determine this bias.
Rail projects are inherently different from roads, and the demand in rail projects might be
more elastic than road demand, thus being more vulnerable to variations. Another possible
explanation is that rail projects are typically greenfield projects, with no prior historical
data, whereas there are many road projects that are brown field projects, i.e., with previous
usage data that can help to increase the accuracy of forecasts the capacity of the public sec-
tor to be able to provide the necessary long-term funding for the project operation.
However, there is an additional hypothesis whereby rail projects might be favoured by
decision makers, increasing the optimism bias, or strategic bias of rail forecasts. This is an
area that would require more data and analysis to reach definitive answers.
To confirm the differences between rail and road projects deviations, we have developed
a hypothesis test using a t test, summarised in Table 1.
With regards to the road sector, Figs. 2 and 3 present the mean deviation and standard
deviation for all road projects studies, considering the size of the sample and the results
for the standard deviation. Figures 4 and 5 also represent the mean deviations and standard
deviations, but with a temporal perspective.
We can observe higher mean deviation in small samples, and higher standard deviation
in large samples. There is a clear increase in studies carried out in recent years. Further-
more, and most intriguing, is the finding that average deviations and standard deviations
are not decreasing over time. One would argue that with the advancements of science and
mathematical/computational complexity, it would be possible to obtain more accurate and
reliable forecasting models. The access to larger quantities of data, together with the ability
of processing such data, could, in theory increase the ability to forecast. However, it seems
that traditional forecasting models are unable to radically improve forecasting capacity,
which is needed to increase the robustness of the decision-making process. In fact, Odeck
and Welde (2017) argue that in Norway, the accuracy of forecasts has not improved since
transport models became mandatory in 2006. Flyvbjerg (2008) also argues that if the prob-
lem was technical, then one would expect that along time, the ability to solve some of the
technical problems would represent a decrease in inaccuracy. Over the last decades, there
has been a tremendous effort to develop both more sophisticated methodologies (e.g., arti-
ficial intelligence algorithms, etc.) (Xiao et al. 2014), and also the quality of data. However,
for the road sector, a significant improvement in forecast accuracy has yet to be provided.
13
13
Table 1 Hypothesis test (t test)
Null hypothesis t test Decision
Mean deviation p value
(%)
1 The road average deviations is not significantly different from the all 13.3 0.39 Not reject the null hypothesis (road average deviation is significantly
sample average deviation different from the all sample average deviation)
2 The rail average deviations is not significantly different from the all 13.3 0.27 Not reject the null hypothesis (rail average deviation is significantly
sample average deviation different from the all sample average deviation)
3 The road average deviations is not significantly different from the rail 29.8 0.00 Reject the null hypothesis (road average deviation is significantly dif-
sample average deviation ferent from the rail sample deviation)
Transportation
Transportation
0.3
0.2
0.1
0
0 20 40 60 80 100 120 140 160 180 200
-0.1
Mean Deviation
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
Sample size
Fig. 2 Mean deviation per sample size (road projects)
0.7
0.6
0.5
Standard deviation
0.4
0.3
0.2
0.1
0
0 20 40 60 80 100 120 140 160 180 200
Sample size
Fig. 3 Standard deviation per sample size (road projects)
Figures 6, 7, 8 and 9 replicate the results of Figs. 2, 3, 4 and 5, but rather for rail project
studies, with a similar pattern to roads. Our conclusions are similar to those of road pro-
jects, but with the downside of having a lower number of studies.
These analyses of the historical evolution of demand simply demonstrate the value of
the average deviation at the time the study was made. However, they do not provide any
evidence on how average deviation evolves throughout the life of the project. In fact,
few studies actually provide detailed information comparing real and forecasted traffic/
ridership for each year of the project.
Based on the information provided in “Appendix 1”, we have compared how real
traffic evolves, in comparison to that forecasted, during the first 5 years of the project
(Table 2). In general, as a project matures, real traffic grows faster than that forecasted.
Baeza and Vassallo (2012) estimated an overestimation of traffic of 43.64%, but this
13
Transportation
Mean deviaon
0.3
0.2
0.1
0
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
Mean deviation
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
-0.8
Sample size
Fig. 4 Historical evolution of mean deviations (road projects)
0.7
0.6
0.5
Standard deviation
0.4
0.3
0.2
0.1
0
1975 1980 1985 1990 1995 2000 2005 2010 2015 2020
Sample size
Fig. 5 Historical evolution of standard deviations (road projects)
drops to 27.29% in Year 5. The same trend is observed in Odeck and Welde (2017), but
in this case, the study shows an underestimation of traffic of 4%, growing to 24% in Year
5. This data exhibits a trend for demand to grow faster than expected, although, in most
studies, still below the forecast.
Main variables used in the studies
“Appendix 2” presented a summary of the main methodologies and variables used in the
study of forecast inaccuracy. We have developed a classification of these variables, based
on the type of each variable. The classification has the following categories: (1) Project
13
Transportation
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50
Mean Deviation
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
Sample size
Fig. 6 Mean deviation per sample size (rail projects)
0.6
0.5
0.4
Standard deviation
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40 45 50
Sample size
Fig. 7 Standard deviation per sample size (rail projects)
variables: subdivided into technical (technical/physical features of the project), economic

(essentially costs and revenues), and past forecast (in the case of brownfield projects); (2)
Location: geographical location of the project (e.g., country, region, city); (3) Procurement
process: the level of competition (e.g., number of bidders) and the type of procurement
(e.g., auction or public tender); (4) Macroeconomic variables: GDP and level of the coun-
try’s economic development, and; (5) Legal and regulatory variables: type of legal regime
and governments capacity.
Table 3 summarises the main type of variable used in the previous studies.
13
Transportation
0.3
0.2
0.1
0
Mean deviation
1985 1990 1995 2000 2005 2010 2015

-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
-0.7
Fig. 8 Historical evolution of mean deviations (rail projects)
0.60
0.50
Standard deviation
0.40
0.30
0.20
0.10
0.00
1985 1990 1995 2000 2005 2010 2015
Fig. 9 Historical evolution of standard deviations (rail projects)
Table 2 Ramp up effect
Study Year 1 (%) Year 2 (%) Year 3 Year 4 Year 5
Bain (2009) − 23.00 − 22.00 − 21.00% − 20.11% − 21.00%

Welde and Odeck (2011) − 2.50 − 2.10 + 2.305 n.a. n.a.
Baeza and Vassallo (2012) − 43.64 − 40.03 − 39.69% − 29.17% − 27.39%
Odeck and Welde (2017) 4 10 13% 19% 24%
The trend over time has been the growing complexity of the models. Initial models were
much more based on two types of variables: technical and macroeconomic. More recent
research has evolved towards: incorporating location (i.e., identifying the propensity for
some countries/regions to overestimate forecasts) (Baeza and Vassallo 2012; Button et al.
2010; Oliveira et al. 2016); legal and regulatory differences (i.e., capturing the effect of
more, or less, developed regulators) (Silva 2000), and; the procurement process (which
may influence the strategic behaviour of bidders, which is one of the main reasons for the
overestimation of forecasts (Athias and Nuñez 2008). However, the biggest development
has been on how to incorporate uncertainty in forecasting process (Manzo et al. 2015;
Næss et al. 2015; Salling and Leleur 2015; Nicolaisen and Næss 2015).
13
Transportation
Table 3 Types of variables used in the literature

Type of variable Examples
Project
Technical Length
Stations per mile
No. of bypass schemes
Online improvements
Junction improvements
Greenfield/brownfield
Economic CAPEX
Toll revenues
Past forecasts AADT
No. of vehicles in peak hour
No. of vehicles per day
Location Location of the project
Procurement process No. of bidders
Type of process (e.g. auction)
Macroeconomic High income country
GDP
Legal and regulatory Civil law
Government learning capacity (e.g. no. of conces-
sions the public administration has awarded in
the past)
This table, based on the previous tables, presents the type of variables considered in the literature to be
related with traffic inaccuracies
Main causes for transport inaccuracy
When looking at Table 4, we can observe that the main causes for traffic inaccuracy are
(ordered by frequency): (1) opportunistic behaviour and optimist bias (e.g., Fouracre et al.
1990; Button et al. 2010); (2) inadequacy in forecast models and data (Welde and Odeck
2011 or Nicolaisen 2012); (3) overall uncertainty (e.g., Odeck 2013; Næss et al. 2015);
(4) changes in demographics and land use patterns (e.g., Horowitz and Emslie 1978; Bain
2009); (5) project causes, such as quality or construction delays, among others (e.g., Par-
thasarathi and Levinson 2010); (6) competition and Demand (Bain 2009), and; (7) eco-
nomic cycle (Mackinder and Evans 1981). Table 4 presents these main causes, by author.
Few studies have objectively identified one main reason for transport inaccuracy,
although the large majority of them identify a set of two to four main reasons for the inac-
curacy. Flyvbjerg et al. (2002, 2004, 2005) considered the following classification for
causes of inaccuracy in forecasts: technical causes; psychological causes, and; political-
economic causes.
As we can see, opportunistic behaviour and optimism bias is the most frequent cause for
traffic inaccuracy, as identified by past literature. This can originate from the public side
through political bias, or from the private side, from opportunistic behaviour.
Political bias emerges as being the main cause for forecast inaccuracy. This fact can
be understood by the fact that most transportation projects are highly visible, frequently
13
13
Table 4 Main causes for forecast inaccuracy by author
Causes 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Total
Opportunistic behaviour and optimism bias × × × × × × × × × 9

Changing demographics and land use patterns × × × × 4
Forecast models and data × × × × × × 6
Overall uncertainty × × × × × 5
Project related (quality, construction delays, etc.) × × × 3
Competition and demand × × 2
Economic cycle × × 2
This table, based on “Appendix 1”, presents the causes for forecast inaccuracy, by author
1—Horowitz and Emslie (1978); 2—Mackinder and Evans (1981); 3—Pickrell (1989); 4—Fouracre et al. (1990); 5—Flyvberg et al. (2006); 6—FTA (2008); 7—Bain (2009);
8—Button et al. (2010); 9—Parthasarathi and Levinson (2010); 10—Welde (2011); 11—Welde and Odeck (2011); 12—Baeza and Vassallo (2012); 13—Nicolaisen (2012);
14—Odeck (2013); 15—Sanko et al. (2013); 16—Black (2014); 17—Button and Chen (2014); 18—Næss et al. (2015); 19—Nicolaisen and Næss (2015); 20—Salling and
Leleur (2015)
Transportation
Transportation
requiring a strong political commitment and are used in political campaigns. The political
decision to advance with such projects not always complies with the technical decision-
making process. When detailed forecasts are carried out to support the CBA, the politi-
cal commitment to the project is frequently so high that it would be very difficult to stop
the project. This is also related with the existing infrastructure-planning model. In sev-
eral countries (e.g., UK, USA, or France), politicians generally present large infrastructure
development plans, containing several distinct projects. This means that a commitment is
made to a portfolio of projects, which may (and probably will) include less attractive pro-
jects. It is more difficult for governments to step back from projects that already have some
level of commitment and public support, even though the CBA might not provide robust
positive results.
These plans are defined more by political arguments, rather than technical ones, how-
ever they create an expectation among the public that the projects will go ahead. Forecasts
to support the CBA are published further along he process, and if the results are negative,
then it would be difficult, at this stage, to take a backwards step to question the sustainabil-
ity of the project.
Simultaneously, with the increase in the use of PPP models, opportunistic behaviour by
concessionaires has become more frequent, such as increasing forecasts, with the hope of
maximising public subsidies to compensate lower traffic in the future. When concession-
aires are responsible for forecasts, due to the lack of a traffic risk-mitigation mechanism
in the contract, which is common, the incentive is for the concessionaire to overestimate
traffic revenues in order to win the contract. This has been one of the relevant causes for the
poor financial performance of PPPs.
Another important cause is related with changing demographics and land use patterns.
Most roads and rails have a lifespan of over 50 years, which means that during this period
it is very difficult to forecast urban development and the evolution of demographics.
The lack of sophistication of models and data appears only as the third most frequent
cause. One could argue that improving the ability to forecast is more related with external
factors, rather than the project or the decision tools (forecasting models) themselves.
Overall, “inherent” uncertainty in previewing the future is also relevant for explain-
ing inaccuracy in projections, particularly in long-term projections. In the literature, pro-
ject variables are also found to be a cause of inaccuracy in traffic projections. This can be
related with imprecise project concept design planning, risk management and implementa-
tion, and also poorly-organised bidding processes, as well as with quality issues or con-
struction delays. Other motives include competition and lower demand, and also the effect
of the economic cycle. Forecasting demand and traffic implies making assumptions about
macroeconomic variables (such as GDP growth, or inflation) which are extremely difficult
to assess in the medium and long-term.
It is important to mention that although many papers presented some possible explana-
tions (or causes) for traffic forecasts inaccuracy, as shown in the Appendices and discussed
earlier, the reality is that few papers focus on a quantitative understanding and justification
of such reasons. Most times, the rationale is merely a speculative motive. The contribution
of most papers has been more on providing evidence of such a phenomenon.
Over the last three decades, the papers have focused their contribution on understanding
the magnitude and prevalence of inaccuracy in forecasts. Those papers that have tried to
understand the reasons for such inaccuracy fully have not concluded by one single motive.
There are quantitative variables that are easy to demonstrate that have been poorly fore-
casted (e.g. GDP growth, population, land use patterns, etc.). In these cases, when compar-
ing the assumptions and the reality, is easy to conclude that, in many cases, the baseline
13
Transportation
assumptions were, in general, too optimistic. But even in the cases where quantitative
reasons are found, the authors typically invoke other, more qualitative reasons, typically
around governance and optimism bias, etc. It is almost as if the authors acknowledge that
there is something else behind the inaccuracy.
But among those that acknowledge that there are (also) governance and political reasons
behind traffic forecasts, different lines of thought emerge. On the one hand, authors such as
Flyvbjerg et al. 2002; Flyvbjerg 2008) argue that the main reasons are psychological and
misrepresentation while other, such as Salling and Leleur (2015), present overall uncer-
tainty and a constantly changing reality as the main reason.
Bottom line, there are diverse reasons, of various nature, and it is unlikely to expect to
be able to pinpoint an exact explanation for each error in forecasts. But it is clear that there
are “hidden” reasons that need to be better understood.
Conclusions and policy implications
Irrespective of the technological and methodological improvements, the problem of fore-

cast inaccuracy will remain as one of the most relevant factors for the appraisal and man-
agement of transportation projects. This problem is exacerbated by the increasing usage
of private sector participation (through concessions, or PPPs), where initial forecasts can
have a large impact on the future success of a project and its cost. Aware of the typical
excessive optimism of governments’ forecasts, the private sector typically enforces risk
mitigation strategies, such as “band systems” and financial and economic re-equilibrium
models to compensate for potential revenue losses (see more in Cruz and Marques 2013b).
The “band system” is a risk-mitigation strategy, based on the definition of upper and lower
limits for demand variation. If the demand varies within certain boundaries, then this is
the risk incurred by the private concessionaire. If the demand is lower than the lower limit,
then the concessionaire is entitled to compensation, and if it is above the upper limit, then
it may have to share the earnings with the grantor.
The systematic review presented in this paper confirms some of the existing conclusions
of the Literature, such as the non-normality of error, and the inability to increase accuracy
over time. In addition, it also brings some new elements to light, such as an average devia-
tion less than that of previous studies, although previous studies have significantly lower
samples. When dividing projects by type (roads vs. railways), there is evidence of a higher
optimism in railway projects. Large public programmes to increase investment in railways
(e.g., in Europe) thus favour rail investments over roads, which ultimately could lead to a
high-level strategic public bias.
However, at the same time, there is evidence of a “ramp-up” effect in traffic growth. As
the infrastructure matures, it seems that it generates an annual growth higher than expected,
even in the few cases were the initial forecast was underestimated. It would be relevant to
have more studies with longer data sets, in order to better understand the performance of
traffic forecasts in the very long term. This a relevant gap in the Literature. Most studies
focus on the first years of operation, and, due to the “ramp-up” effect of demand, the inac-
curacy tends to decrease over time. What is the performance for the first 10, 20 or 30 years
of operation? This is question that remains under discussed.
13
Transportation
Governance, with involuntary (optimism) or voluntary (strategic) bias, is one of the

main causes of inaccuracy, which is certainly one of the most difficult to solve. Flyvb-
jerg (2008) claims that psychological and political-economic effects are the main cause.
Political-economic motives are linked to misrepresentation or competition for scarce
funds, which strategically leads to excess optimism, in order to guarantee that the project is
selected over its direct competitors (Wachs 1989, 1990).
Reality has shown that these two reasons may coexist, however the argument of authors
is that when political pressure is high, strategic misrepresentation is more likely to occur,
and when this pressure is low, the involuntary optimism bias is predominant. This discus-
sion neglects a fundamental aspect, which is the inability to determine exactly where a
naïve optimism stops, and a strategic bias begins.
The independent assessment of projects is crucial, however this suggestion often
emerges as a panacea for most technical/governance problems of large-scale project devel-
opment. Most forecasts are based on project-specific models, built on local/regional socio-
economic characteristics, urban patterns, and demographics. Forecasts are then evaluated
case by case, on a stand-alone basis. It would be helpful to identify potentially over-opti-
mistic scenarios and to benchmark projects with similar ones. For example, for a given
geographical area (country, or continent), to compare the existing levels of traffic per km of
road, or the number of trips per population, in the area of influence of the project. Ideally,
an international database of road or rail projects would provide a crucial tool for bench-
marking forecasts.
This benchmark would help to decrease any opportunistic or strategic behaviour by both
the public and private sector. However, traditional research on traffic forecasts has been
more directed towards the mathematical development of computational forecast models,
using more sophisticated methodologies, such as artificial intelligence. There are advance-
ments in other sectors, where artificial intelligence is applied to long-term forecasting,
which can provide useful tools for the transportation sector. See for instance, Yaseen et al.
(2016), for an application in the water sector, or Sonmez et al. (2017), for energy consump-
tion in the transportation sector. Based on the evidence of this research, and past similar
studies, future research regarding inaccuracy and forecast improvement should focus on the
macro assessment of forecasts, benchmarks, and project governance, in order to be able to
significantly decrease the likelihood of inaccuracy.
Ackowledgements Joaquim Miranda Sarmento gratefully acknowledges the financial support received from
FCT-Fundação para a Ciência e Tecnologia (Portugal), and also the national funding obtained through a
research Grant (UID/SOC/04521/2019).
Appendix 1
See Table 5.
13
Table 5 Literature review on traffic forecasts and the main reasons for inaccuracy
Author Sector Location Sample dimension Traffic forecast inaccuracy
Mean SD Target year Summary of the main
13
causes for inaccuracy
Road projects
Horowitz and Emslie Road US 78 highway segments 1968: − 24.00% N.A. Opening year Changing demograph-
(1978) ics
1972: − 21.00% Optimism bias
Mackinder and Evans Road UK 44 road projects − 7.00% N.A. Opening year Changing demograph-
(1981) ics
Optimism bias
Economic recession
Flyvbjerg et al. (2005) Road 14 European coun- 183 road projects Road: − 9.50% Road: 44.30 Varies Overall uncertainty
tries Road projects: changes
are land use patterns.
Bain (2009) Road USA 104 projects Year 1: − 23.00% Year 1: 26.00% Opening year Economic recession;
Year 2: − 22.00% Year 2: 23.00% the forecast assumed
Year 3: − 21.00% Year 3: 22.00% a stable long term
Year 4: − 20.00% Year 4: 24.00% growth, but the real-
Year 5: − 21.00% Year 5: 25.00% ity turned out to be
different
Different land use
patterns
Wrong assumptions in
willingness-to-pay
for tolls
Existence of (more
attractive) competing
roads
Overestimation of travel
time savings
Transportation
Table 5 (continued)
Transportation
Parthasarathi and Road US 108 projects − 6.41% 40.65% Opening year Lack of sophistication
Levinson (2010) of existing forecast
models, unable to
cope with uncertainty
Delays in construction
Welde (2011) Road Norway 30 toll projects Average 23.30% Opening year N.A.
Fixed link: 5.7%
Highway: − 1.5%
Toll: − 11.1%
Year 1: − 6.5%
Year 3: − 2.1%
Year 4: 2.4%
Welde and Odeck Road Norway 25 toll projects and Toll projects: Toll projects: Years 1–5 Misspecification of
(2011) 25 toll free road Year 1: − 2.50% Year 1: 22.0% models
projects Year 2: − 2.10% Year 2: 20.00%
Year 3: +2.30% Year 3: 23.20%
Toll free Roads: Toll free Roads:
19.00% 20.50%
Baeza and Vassallo Road/motorway Spain 15 toll concessions Year 1: − 43.64% Year 1: − 14.46% Years 1–5 Strategic behaviour
(2012) Year 2: − 40.03% Year 2: − 13.87% from government and
Year 3: − 39.69% Year 3: − 15.33% concessionaires; rene-
Year 4: − 29.17% Year 4: − 17.91% gotiations extended
Year 5: − 27.39% Year 5: − 22.22% the duration of the
concessions (in some
cases duplicating the
initial duration)
13
13
Nicolaisen (2012) Road Denmark, Norway, 31 rail projects − 11.12% 35.07% Opening year Poor quality of data
Sweden and UK Optimism bias
Deliberate manipula-
tion
Odeck (2013) Road Norway Regional and national 1.31% N.A. N.A. Lack of sophistication
data (20 regions) of existing forecast
models, unable to
Black (2014) Road Australia 8 tolled roads − 48.00% N.A. Opening year Political influence
to inflate traffic
forecasts, in order that
the project could be
completed;
Strategic behaviour
by bidders (more
optimistic, better
proposal)
Button and Chen Road US 26 highway invest- Public-sector: N.A. Opening year Strategic influence of
(2014) ment − 67.7% the forecast to benefit
PPPs: − 37.2% the terms of the PPP
contract
Nicolaisen and Næss Road Denmark and UK Do-nothinga: 35 Do-nothing: Do-nothing: Opening year Lack of sophistication
(2015) projects − 6.98% 12.3 of existing forecast
Do-somethingb: Do-something: Do-something: models, unable to
146 projects 11.12% 35.07 cope with uncertainty
Salling and Leleur Road Great Britain, Den- 122 road projects; − 22.30% 44.00 Overall uncertainty
(2015) mark, Sweden and 15 fixed link projects
Norway
Transportation
Transportation
Odeck and Welde Road Norway 68 toll road Year 1: 4% 23% 1975–2014 Toll traffic forecasts
(2017) Year 2: 10% in Norway perform
Year 3: 13% fairly well, compared
Year 4: 19% to the forecasts of
Year 5: 24% other countries
Rail projects
Pickrell (1989) Railway US 10 projects 65.30% 17.20% Varies Lack of sophistication
of existing forecast
models, unable to
Delays in construction
Lower level of quality
than initially planned
Fouracre et al. (1990) Rail Developing countries 13 metro projects − 44.44% 26.63% General optimism of
planners (optimism
bias)
Flyvbjerg et al. (2005) Rail 14 European coun- 27 rail projects − 51.40% 28,1% Varies Overall uncertainty
tries Rail projects: strategic
bias towards favour-
ing rail projects rather
than road projects
Road projects: changes
is land use patterns
Button et al. (2010) Transit US 47 transit projects 27.87% 45.42% Varies Unclear (probably
political strategic
behaviour)
13
13
Nicolaisen (2012) Rail Denmark, Norway, 31 rail projects 18.48% 32.84% Opening year Poor quality of data
Sweden and UK Optimism bias
Deliberate manipula-
tion
Salling and Leleur Rail Great Britain, Den- 39 rail projects − 22.30% 44.00% Overall uncertainty
(2015) mark, Sweden and
Norway
This table presents the main literature review on traffic forecasts and the main reasons for traffic forecast inaccuracy. The papers are organised by chronological order. N.A.
means not applicable. The table contains the following information: sector (road and/or railway); sample dimension (number of projects in the study); mean deviation (relative
difference between real traffic and forecasted traffic); standard deviation (standard deviation of the sample); target year (year, or period, in which the comparison between real
and forecasted traffic was performed), and; the main reasons for inaccuracy (motives found by the authors to explain the observed differences)
NA not available
a
Do-nothing projects are projects that have been evaluated, but were not fully implemented (e.g., widening of a road)
b
Do-something projects are projects that have been implemented (e.g., new road)
Transportation
Appendix 2
See Table 6.
Transportation
Table 6 Literature review on methodologies and variables used on traffic forecasts and real traffic deviations
Author Region/country Period Sector Main variables used in the analysis Summary
Horowitz and USA 1968 and 1972 Road Measurement of average daily traffic This study intends to compare forecasts with measurements
Emslie (1978) (ADT)
ADT 1968 forecasts Most forecasts were made using the sequential transportation
forecasting technique
ADT 1972 forecasts Regression analysis was used to investigate forecasting bias
Mackinder and UK 1962–1971 Road N.A. Compares the forecasts obtained by some transport studies with
Evans (1981) the actual values and accesses the accuracy of such forecasts
Presents some reasons for forecast inaccuracy
Studies how more recent forecasts carried out using more recent
and sophisticated model perform when compared with older
forecasts, using traditional models
Pickrell (1989) USA 1971–1987 Railway Ridership (average weekday pas- Accesses the accuracy of ridership and costs forecasts by measur-
sengers) ing the percentage gap between forecasted values and actual
values
Capital costs Addresses the reasons behind the inaccuracy of forecasts
Operating expenses Proposes some methods to improve forecast accuracy
Fouracre et al. Developing N.A. Rail N.A. Studies the performance of Metro projects from 21 developing
(1990) countries cities
Accesses how forecasts deviate from actual values and presents
the reasons for such deviations
Presents some major reflections regarding the financial viability
of Metro projects, focusing especially on their planning and
financial management
13
13
Davis and Nihan N.A. N.A. N.A. N.A. Uses the k-nearest neighbour approach to solve some of the prob-
(1991) lems generated when performing short-term traffic forecasting
through non-parametric regression models
Van Arem et al. N.A. N.A. Road N.A. Presents some relevant issues regarding traffic forecasting and
(1997) dynamic traffic management (DTM)
Lam and Tam N.A. N.A. N.A. N.A. Performs Monte Carlo simulation in order to evaluate risk, which
(1998) allows incorporating uncertainty when developing traffic fore-
cast models
Boyce (1999) N.A. N.A. N.A. N.A. Addresses the uncertainty behind traffic flow forecasting models
and proposes a model to estimate that uncertainty
Chrobok et al. Duisburg 1998, 1999 Road Vehicles/minute Creates a framework for traffic on-line simulations
(2000) Develops an algorithm for traffic forecasts
Silva (2000) Developing 1990–1999 Road Toll roads investment (in US$) Discusses the main trends regarding toll roads investment in
countries developing countries
Wahle et al. N.A. N.A. N.A. N.A. Analysis of the impact of dynamic information in traffic patterns
(2000a) and proposes a model for drivers’ behaviour
Wahle et al. N.A. N.A. N.A. N.A. Analysis of the impact of dynamic information in traffic patterns
(2000b)
Prakash et al. UK 1949–1997 Road Total activity of traffic on the road in The Granger non-causality methoda is used to evaluate whether
(2001) vehicle km (log) traffic increases after expenditure on new and existing roads
Total expenditure on the road (log
and deflated to yield real road
expenditure)
Chrobok et al. North Rhine- N.A. Road Traffic flow This study creates traffic forecasts using a micro simulation model
(2001) Westphalia based on traffic flows
Wahle and Duisburg and 1998 and 1999 Road Simulation on traffic Uses simulations and historical data to provide real-time informa-
Schreckenberg North tion on traffic
(2001) Rhine-West- Historical data on traffic
phalia
Transportation
Clark (2003) London N.A. Road N.A. The author uses a pattern matching technique to develop a traffic
Transportation
forecast model
Vlahogianni et al. N.A. N.A. N.A. N.A. Addresses short-term traffic forecasting by reviewing previous
(2004) forecasting models
Proposes a framework for developing a forecasting algorithm,
considering three elements: determination of scope, conceptual
specification of the output, and modelling
Forecasting models are affected by the design (e.g. selection of
data and data resolution) and modelling (methodology, type of
inputs and output, and quality of data) of the process
Chrobok et al. Duisburg 1998–2000 Road Daily traffic demand Uses a constant model a linear model to estimate traffic forecasts
(2004) The two models are compared using heuristics prediction.
Bain and Plan- N.A. N.A. Road N.A. Compares the accuracy of different studies on toll road forecasts
tagie (2004) Compares the accuracy of traffic forecasts for toll roads and toll-
free roads
Flyvbjerg et al. 14 European 1969–1998 Road and Traffic forecasts Compares the forecasts obtained for road and rail projects with
(2005) countries rail the actual traffic
Actual traffic Debates the reasons for the big difference between traffic forecasts
and actual traffic and advances some suggestions in order to
reduce such difference
13
13
Vassallo (2006) Chile 1998–2002 Road Traffic volume Addresses mechanisms for traffic risk mitigation, namely, Mini-
mum Income Guarantee, Least Present Value of the Revenues,
and Revenue Distribution Mechanisms
Discusses the application of these methods during recession

periods
Nunez (2007) N.A. N.A. N.A. N.A. Studies the reasons for forecast inaccuracy, more specifically—
forecast overestimation
Athias and N.A. N.A. Road Forecasted traffic/actual traffic Considers the bids offered in auctions for toll road concessions
Nuñez (2008) Number of bidders Each bidder makes an offer based on their own traffic appraisal
Civil law (dummy = 1 if yes) Demand forecasts are subject to several uncertainties, each bidder
will attempt to understand the amount and relevance of the
information possessed by other bidders based on their offer
High income country (dummy = 1 The winning bidder is the one with the most overestimated traffic
if yes) forecast
Government Learning—“Number of The behaviour of bidders in auctions may reflect the accuracy of
concessions the public authority has their traffic forecast estimations
awarded before the present project”
Length (in Km)
Transportation
Federal Transit USA 2002–2007 Railway Capital Costs (inflation-adjusted) Compares the actual with the forecasted performance of ridership
Transportation
Administration and capital costs for projects which were funded by the New
(2008) Starts programme
Ridership Addresses the methods used to gather information that was used
by project decision-makers
Bain (2009) USA 2002–2005 Road Toll revenues Summarises the results of some studies on traffic forecasting;
Traffic volume An analysis of the accuracy of forecasts and the reasons for the
deviations of the forecast from actual traffic
Button et al. USA 1972–2005 N.A. Ridership Evaluates whether accuracy in forecasts has in fact improved, as
(2010) Capital costs some authors postulate
System characteristics (length, in
miles; vehicles; stations per mile)
Type (new system or extension)
Technology (Heavy rail transit, light
rail transit, bus rapid transit, other)
Time (year the system planning was
completed)
Pickrell Effect [whether system plan-
ning was completed before or after
Pickrell’s (1992)]
13
13
Parthasarathi Minnesota 1961–1964 Road Actual Traffic OLS regressions are used to estimate the factors affecting forecast
and Levinson Traffic forecasts inaccuracy
(2010) Compares forecasted traffic with actual traffic
OLS models:
Evaluates the accuracy of obtained forecasts and identifies the fac-
Forecast accuracy tors of inaccuracy
Number of years between the year in
which the report was prepared and
the forecast year;
Forecast vehicle kilometres travelled
(VKT) by project;
Roadway type;
Roadway functional classification;
Roadway segment direction;
Decade of report preparation;
Project status (existing/new facility) at
the time of report preparation.
East; middle; middle north; middle
south; north; northeast; northwest;
south; southeast; southwest; west
Reports prepared between 1971 and
1980;
Reports prepared between 1981 and
1990;
Reports prepared after 1990
Transportation
Welde (2011) Norway 1990–2009 Road Traffic volume forecasts Compares the forecasted values of traffic and operating costs with
Transportation
Operating costs forecasts actual values

Actual traffic volume
Actual operating costs
Welde and Odeck Norway 1990–2007 Road Forecasted and actual traffic levels Studies the accuracy of traffic forecasts comparing tolled roads
(2011) Forecasted and actual costs with toll free roads
Forecasted and actual revenues
Net present value
Peng et al. (2011) N.A. N.A. Road Traffic flow Develops a lattice model to forecast traffic flow, which incorpo-
Traffic stability rates the drivers’ forecasts effects (DFE)
Highways UK 2002–2009 Road Bypass schemes Presents the Meta-Analysis, developed by Post Opening Project
Agency (2011) Evaluation (POPE)
Online improvements Evaluates the major aspects regarding schemes, focussing more
precisely on the accuracy of the appraisals
Junction improvements Addresses the impacts of appraisal inaccuracy and the reasons for
such inaccuracies
Baeza and Vas- Spain 1996–2007 Road/ Annual average daily traffic Studies the accuracy of forecasts for toll motorway concessions
sallo (2012) motor- Traffic forecasts in Spain
way
Nicolaisen Denmark, Nor- N.A. Road and Traffic forecasts and actual values Reviews concepts regarding traffic forecasts, as well as the most
(2012) way, Sweden rail relevant literature review on the topic
and UK Costs forecasts and actual values Analyses the accuracy of traffic forecasts
Uses a set of databases, documents, questionnaires and interviews
to access how traffic forecasts can have an impact on political
decisions
Studies how uncertainty regarding projects is managed
Sanko et al. Japan (Nagoya) 1991 Railway Passengers per day (forecasted and Estimates the accuracy of the forecasts for the number of passen-
(2013) actual) gers of the Tokadai Line
13
13
Odeck (2013) Norway 1996–2008 Road Forecasted traffic growth rates Studies the accuracy of traffic forecasts by calculation the percent-
(national and regional) age error between the forecasted values and observed values
Actual traffic growth rates (national Comparess the accuracy of forecasts before and after these fore-
and regional) casts were revised
GDP growth-rate forecasts Debates how inaccurate traffic forecasts lead to wrong policy-
making
Black (2014) Australia N.A. Road Number of vehicles per day Reviews several studies that compare tactual traffic with traffic
forecasts
Actual/forecast ratio An analysis of the accuracy of traffic forecasts and the reasons for
such optimistic forecasts
An analysis of the reasons why the government moved forward
with these PPP projects
Button and Chen USA 1986–2004 Road Actual/forecast traffic flow Compares the accuracy of private sector traffic demand forecasts
(2014) with public sector traffic demand forecasts
Nicolaisen and N.A. N.A. Road and N.A. Presents some literature review of articles that have focussed on
Driscoll (2014) rail traffic forecast accuracy
Manzo et al. Denmark N.A. Road Trip generation Uses a four-stage model (combining trip generation, trip distribu-
(2015) tion, mode choice, and trip assignment), taking into account
uncertainty in forecasting
Trip assignment A scenario analysis is also performed
Trip distribution Considers how uncertainty regarding model inputs affects fore-
casting
Mode choice
Næss et al. N.A. N.A. N.A. N.A. Based on the results of a survey, interviews, and other studies.
(2015) Tries to understand the reasons behind the frequent inaccuracy
of traffic forecasts
Transportation
Nicolaisen and Denmark and 1970–2010 Road Annual average daily traffic Studies the accuracy of demand forecasts of do-nothing alterna-
Transportation
Næss (2015) England tives (compared to implementing new projects)

Annual average weekly traffic Accuracy is measured by the difference between the forecast and
the observed traffic, as a percentage of the forecast
Salling and Great Britain, 2009–2013 Road and Actual and estimated demands Presents the UNITE-DSS Decision Support Model. This is a com-
Leleur (2015) Denmark, rail plex model developed to improve decision-making by incor-
Sweden and porating not only a deterministic component with cost–benefit
Norway analysis and optimist bias, but also a stochastic component
Actual and estimated costs Considers a socio-economic analysis, risk and uncertainty, and
also database information
Xu et al. (2015) N.A. N.A. N.A. N.A. Studies the possibility of selection bias in project appraisals
Oliveira et al. Portugal and N.A. N.A. N.A. Develops a framework that allows one to understand what drives
(2016) Brazil investment towards risky projects (what leads investors to
finance projects with higher risk)
Applies the developed framework to the Portuguese case and
compares it with the Brazilian case
This table presents the main literature review on methodologies and variables used on traffic forecasts. Papers are organised by chronological order
Period means the temporal interval of each study
NA not applicable
a
The granger non-causality method is a statistical analysis to determine the relationship between two time series. It was developed in Granger (1969)
13
Transportation
References
Athias, L., Nuñez, A.: Winner’s curse in toll road concessions. Econ. Lett. 101(3), 172–174 (2008)
Baeza, M.A., Vassallo, J.M.: Traffic uncertainty in toll motorway concessions in Spain. Proc. Inst. Civ. Eng.
165(TR2), 97–105 (2012)
Bain, R.: Error and optimism bias in toll road traffic forecasts. Transportation 36(5), 469–482 (2009)
Bain, R., Plantagie, J.: Traffic forecasting risk: study update 2004. In: Proceedings of the European Trans-
port Conference, Strasbourg, France, 4–6 Oct 2004
Black, J.: Traffic risk in the Australian toll road sector. Public Infrastruct. Bull. 1(9), 3 (2014)
Boyce, A.M.: Risk analysis for privately funded transport schemes. In: Transportation Planning Methods.
Proceedings of Seminar F, European Transport Conference, Cambridge, UK, 27–29 Sept 1999
Button, K., Chen, Z.: Demand forecasting errors and the ownership of infrastructure. Appl. Econ. Lett.
21(7), 494–496 (2014)
Button, K.J., Hardy, M.H., Doh, S., Yuan, J., Zhou, X.: The accuracy of transit system ridership forecasts
and capital cost estimates. Int. J. Transp. Econ. 37(2), 155–168 (2010)
Chrobok, R., Kaumann, O., Wahle, J., Schreckenberg, M.: Three categories of traffic data: historical, cur-
rent, and predictive. In: Control in Transportation Systems. Proceedings of the 9th IFAC Symposium,
Braunschweig, Germany, 13–15 June 2000
Chrobok, R., Wahle, J., Schreckenberg, M.: Traffic forecast using simulations of large scale networks. In:
Intelligent Transportation Systems, pp. 434–439 (2001)
Chrobok, R., Kaumann, O., Wahle, J., Schreckenberg, M.: Different methods of traffic forecast based on real
data. Eur. J. Oper. Res. 155(3), 558–568 (2004)
Clark, S.: Traffic prediction using multivariate nonparametric regression. J. Transp. Eng. 129(2), 161–168
(2003)
Cruz, C.O., Marques, R.C.: Risk-sharing in highway concessions: contractual diversity in Portugal. J. Prof.
Issues Eng. Educ. Pract. 139(2), 99–108 (2012a)
Cruz, C.O., Marques, R.C.: Using probabilistic methods to estimate the public sector comparator. Comput.
Aided Civ. Infrastruct. Eng. 27(10), 782–800 (2012b)
Cruz, C.O., Marques, R.C.: Endogenous determinants for renegotiating concessions: evidence from local
infrastructure. Local Gov. Stud. 39(3), 352–374 (2013a)
Cruz, C.O., Marques, R.C.: Using the economic and financial reequilibrium model to decrease infrastruc-
ture contract incompleteness. J. Infrastruct. Syst. 19(1), 58–66 (2013b)
Damart, S., Roy, B.: The uses of cost-benefit analysis in public transportation decision-making in France.
Transp. Policy 16(4), 200–212 (2009)
Davis, G.A., Nihan, N.L.: Nonparametric regression and short-term freeway traffic forecasting. J. Transp.
Eng. 117(2), 178–188 (1991)
Domingues, S., Sarmento, J.: Critical renegotiation triggers of European transport concessions. Transp. Pol-
icy 48, 82–91 (2016)
European Investment Bank: The Economic Appraisal of Investment Projects at the EIB. European Invest-
ment Bank, Luxembourg (2013)
Federal Transit Administration: The predicted and actual impacts of new starts projects—2007. US Depart-
ment of Transportation (2008)
Fitch Ratings: Bliss, Heartburn, and Toll Road Forecasts. Public Finance Special Report, New York (2013)
Flyvbjerg, B.: Five misunderstandings about case-study research. In: Seale, C., et al. (eds.) Qualitative
Research Practice: Concise Paperback Edition, pp. 390–404. Sage, London and Thousand Oaks, CA
(2007a)
Flyvbjerg, B.: Policy and planning for large-infrastructure projects: problems, causes, cures. Environ. Plan.
B Plan. Des. 34(4), 578–597 (2007b)
Flyvbjerg, B.: Curbing optimism bias and strategic misrepresentation in planning. Reference class forecast-
ing in practice. Eur. Plan. Stud. 16(1), 3–21 (2008)
Flyvbjerg, B., Holm, M.S., Buhl, S.: Underestimating costs in public works projects: error or lie?. J. Am.
Plan. Assoc. 68(3), 279–295 (2002)
Flyvbjerg, B., Holm, M.K.S., Buhl, S.L.: What causes cost overrun in transport infrastructure projects?
Transp. Rev. 24(1), 3–18 (2004)
Flyvbjerg, B., Holm, M.K.S., Buhl, S.L.: How (in)accurate are demand forecasts in public works projects?
The case of transportation. J. Am. Plan. Assoc. 71, 131–146 (2005)
Fouracre, P.R., Allport, R.J., Thomson, J.M.: The Performance and Impact of Rail Mass Transit in Develop-
ing Countries. Transport and Road Research Laboratory, Wokingham (1990)
Granger, C.W.J.: Investigating causal relations by econometric models and cross-spectral methods. Econo-
metrica 37(3), 424–438 (1969)
13
Transportation
Grimsey, D., Lewis, M.K.: Are public private partnerships value for money? Evaluating alternative
approaches and comparing academic and practitioner views. Acc Forum 29(4), 345–378 (2005)
Guasch, J.L., Laffont, J.-J., Straub, S.: Renegotiation of concession contracts: a theoretical approach. Rev.
Ind. Organ. 29(1–2), 55–73 (2006)
Highways Agency: Post opening project evaluation of major schemes (2002 to 2009): meta-analysis main
report. UK Highways Agency (2011)
Horowitz, J., Emslie, R.: Comparison of measured and forecast traffic volumes on urban interstate highways.
Transp. Res. 12(1), 29–32 (1978)
Hyard, A.: Cost-benefit analysis according to Sen: an application in the evaluation of transport infrastructure
in France. Trans. Res. Part A Policy Pract. 46(4), 707–771 (2012)
Jalali, S., Wohlin, C.: Systematic literature studies: database searches vs. backward snowballing. In: Pro-
ceedings of the ACM-IEEE International Symposium on Empirical Software Engineering and Meas-
urement. ACM (2012)
Lam, W.H.K., Tam, M.L.: Risk analysis of traffic and revenue forecasts for road investment projects. J.
Infrastruct. Syst. 4(1), 19–27 (1998)
Lenferink, S., Tillema, T., Arts, J.: Towards sustainable infrastructure development through integrated
contracts: experiences with inclusiveness in Dutch infrastructure projects. Int. J. Proj. Manag. 31(4),
615–627 (2013)
Mackie, P.: Cost-benefit analysis in transport. A UK perspective. Discussion Papers from International
Transport Forum. Mexico (2010)
Mackie, P.J., Jara-Diaz, S., Fowkes, A.S.: The value of travel time savings in evaluation. Transp. Res.
Part E Logist. Transp. Rev. 37(2–3), 91–106 (2001)
Mackinder, I.H., Evans, S.E.: The Predictive Accuracy of British Transport Studies in Urban Areas.
Transport and Road Research Laboratory, Crowthorne (1981)
Manzo, S., Nielsen, O.A., Prato, C.G.: How uncertainty in input and parameters influences transport
model: output a four-stage model case-study. Transp. Policy 38, 64–72 (2015)
Miranda Sarmento, J., Renneboog, L.: Renegotiating public-private partnerships. European Corporate
Governance Institute (ECGI)-Finance Working Paper (416) (2016a)
Miranda Sarmento, J., Renneboog, L.: Anatomy of public-private partnerships: their creation, financing
and renegotiations. Int. J. Manag. Proj. Bus. 9(1), 94–122 (2016b)
Morallos, D., Amekudzi, A.: The state of the practice of value for money analysis in comparing public
private partnership to traditional procurements. Public Works Manag. Policy 13(2), 114–125 (2008)
Naess, P., et al.: Forecasting inaccuracies: a result of unexpected events, optimism bias, technical problems,
or strategic misrepresentation?. J Transp. Land Use 8(3), 39–55 (2015)
Næss, P., Nicolaisen, M.S., Andersen, A., Strand, A.: Forecasting inaccuracies: a result of unexpected
events, optimism bias, technical problems, or strategic misrepresentation? J. Transp. Land Use
8(3), 39–55 (2015)
Nash, P.: Cost-benefit analysis of transport projects. In: Williams, A., Giardina, E. (eds.) Efficiency in
the Public Sector: The Theory and Practice of Cost-Benefit Analysis. Edward Elgar, Aldershot
(1993)
Neto, D., Cruz, C.O., Sarmento, J.M.: Understanding the patterns of PPP renegotiations for infrastruc-
ture projects in Latin America: the case of Brazil. Competition and Regulation in Network Indus-
tries (2018)
Nicolaisen, M.S.: Forecasts: fact or fiction? Uncertainty and inaccuracy in transport project evaluation.
Ph.D. thesis, Department of Development and Planning, Aalborg University (2012)
Nicolaisen, M.S., Driscoll, P.A.: Ex-post evaluations of demand forecast accuracy: a literature review.
Transp. Rev. 34(4), 540–557 (2014)
Nicolaisen, M.S., Næss, P.: Roads to nowhere: the accuracy of travel demand forecasts for do-nothing
alternatives. Transp. Policy 37, 57–63 (2015)
Nunez, A.: Sources of errors and biases in traffic forecasts for toll road concessions. Doctoral thesis,
Faculté des Sciences Economiques, Université Lumière Lyon 2 (2007)
Odeck, J.: How accurate are national road traffic growth-rate forecasts? The case of Norway. Transp.
Policy 27, 102–111 (2013)
Odeck, J., Welde, M.: The accuracy of toll road traffic forecasts: an econometric evaluation. Transp. Res.
Part A Policy Pract. 101, 73–85 (2017)
Oliveira, M., Ribeiro, J., Macário, R.: Are we planning investments to fail? Consequences of traffic forecast
effects on PPP contracts: Portuguese and Brazilian cases. Res. Transp. Econ. 59, 167–174 (2016)
Parthasarathi, P., Levinson, D.: Post-construction evaluation of traffic forecast accuracy. Transp. Policy
17(6), 428–443 (2010)
13
Transportation
Peng, G.H., Cai, X.H., Liu, C.Q., Cao, B.F.: A new lattice model of traffic flow with the consideration of
the driver’s forecast effects. Phys. Lett. A 375(22), 2153–2157 (2011)
Pickrell, D.H.: Urban Rail Transit Projects: Forecast versus Actual Ridership and Costs. Transportation
Systems Center, US Department of Transportation, Cambridge (1989)
Pickrell, D.H.: A desire named streetcar fantasy and fact in rail transit planning. J. Am. Plan. Assoc. 58(2),
158–176 (1992)
Prakash, A.B., Oliver, E.H.D.A., Balcombe, K.: Does building new roads really create extra traffic?
Some new evidence. Appl. Econ. 33(12), 1579–1585 (2001)
Prest, A.R., Turvey, R.: Cost-benefit analysis: a survey. Econ. J. 75(300), 683–735 (1965)
Quiggin, J.: Risk, PPPs and the public sector comparator. Aust. Account. Rev. 14(33), 51–61 (2004)
Salling, K.B., Leleur, S.: Accounting for the inaccuracies in demand forecasts and construction cost esti-
mations in transport project evaluation. Transp. Policy 38, 8–18 (2015)
Sanko, N., Morikawa, T., Nagamatsu, Y.: Post-project evaluation of travel demand forecasts: implica-
tions from the case of a Japanese railway. Transp. Policy 27, 209–218 (2013)
Sarmento, J.M.: Do public-private partnerships create value for money for the public sector? The Portu-
guese experience. OECD J. Budg. 10(1), 93 (2010)
Silva, G.F.: Toll roads: recent trends in private participation. World Bank Other Operational Studies (2000)
Sonmez, M., Akgungor, A.P., Bektas, S.: Estimating transportation energy demand in Turkey using the
artificial bee colony algorithm. Energy 122(1), 301–310 (2017)
Van Arem, B., Kirby, H.R., Van Der Vlist, M.J.M., Whittaker, J.C.: Recent advances and applications in the
field of short-term traffic forecasting. Int. J. Forecast. 13(1), 1–12 (1997)
Vassallo, J.M.: Traffic risk mitigation in highway concession projects: the experience of Chile. J. Transp.
Econ. Policy 40(3), 359–381 (2006)
Vickerman, R.: Cost—benefit analysis and large-scale infrastructure projects: state of the art and challenges.
Environ. Plan. B Urban Anal. City Sci. 34(4), 598–610 (2007)
Vlahogianni, E.I., Golias, J.C., Karlaftis, M.G.: Short-term traffic forecasting: overview of objectives and
methods. Transp. Rev. 24(5), 533–557 (2004)
Wachs, M.: When planners lie with numbers. J. Am. Plan. Assoc. 55(4), 476–479 (1989)
Wachs, M.: Ethics and advocacy in forecasting for public policy. Bus. Prof. Ethics J. 9(1–2), 141–157 (1990)
Wahle, J., Schreckenberg, M.: A multi-agent system for on-line simulations based on real-world traffic data.
In: Proceedings of the 34th Annual Hawaii International Conference on System Sciences (2001)
Wahle, J., Bazzan, A.L.C., Klüglc, F., Schreckenberg, M.: Anticipatory traffic forecast using multi-agent
techniques. In: Helbing D., Herrmann H.J., Schreckenberg M., Wolf D.E. (eds.) Traffic and Granular
Flow’99, pp. 87–92. Springer, Berlin (2000a)
Wahle, J., Bazzan, A.L.C., Klüglc, F., Schreckenberg, M.: Decision dynamics in a traffic scenario. Phys. A
278(3), 669–681 (2000b)
Wee, B.V., Banister, D.: How to write a literature review paper? Transp. Rev. 36(2), 278–288 (2016)
Welde, M.: Demand and operating cost forecasting accuracy for toll road projects. Transp. Policy 18(5),
765–771 (2011)
Welde, M., Odeck, J.: Do planners get it right? The accuracy of travel demand forecasting in Norway. Eur. J.
Transp. Infrastruct. Res. 11(1), 80–95 (2011)
Xiao, Y., Liu, J.L., Hu, Y., Wang, Y., Lai, K.K., Wang, S.: A neuro-fuzzy combination model based on
singular spectrum analysis for air transport demand forecasting. J. Air Transp. Manag. 39, 1–11 (2014)
Xu, X., Chen, A., Wong, S.C., Cheng, L.: Selection bias in build-operate-transfer transportation project
appraisals. Transp. Res. Part A Policy Pract. 75, 245–251 (2015)
Yaseen, Z.M., Kizi, O., Demir, V.: Enhancing long-term streamflow forecasting and predicting using perio-
dicity data component: application of artificial intelligence. Water Resour. Manag. 30(12), 4125–4151
(2016)
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.
Carlos Oliveira Cruz Assistant Professor at the Civil Engineering Department at the Technical University
of Lisbon (Portugal) where he obtained his PhD in Civil Engineering (Public Private Partnerships). Senior
Researcher at CERIS - Civil Engineering Research and Innovation for Society.
Joaquim Miranda Sarmento Assistant Professor of Finance at ISEG – Lisbon School of Economics and
Management and member of Advance/CSG, ISEG, Universidade de Lisboa. Ph.D. in Finance from Tilburg
University.
13

0 - Cruz & Sarmiento, 2019 Forecast Innacuracy

Uploaded by

Copyright:

Available Formats

0 - Cruz & Sarmiento, 2019 Forecast Innacuracy

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

0 - Cruz & Sarmiento, 2019 Forecast Innacuracy

Uploaded by

Copyright:

Available Formats

Transportation

Traffic forecast inaccuracy in transportation: a literature

© Springer Science+Business Media, LLC, part of Springer Nature 2019

Keywords Traffic forecast · Forecast inaccuracy · Optimism bias · Systematic review ·

JEL Classification H54 · J68 · R41 · R42

* Carlos Oliveira Cruz

Traffic forecasts in the infrastructure lifecycle

Forecast inaccuracy and procurement

Forecast inaccuracy and operation/management

Systematic review of the literature

Patterns of historical deviations

Fig. 2 Mean deviation per sample size (road projects)

Fig. 3 Standard deviation per sample size (road projects)

Fig. 4 Historical evolution of mean deviations (road projects)

Fig. 5 Historical evolution of standard deviations (road projects)

Main variables used in the studies

Fig. 6 Mean deviation per sample size (rail projects)

Fig. 7 Standard deviation per sample size (rail projects)

variables: subdivided into technical (technical/physical features of the project), economic

1985 1990 1995 2000 2005 2010 2015

Fig. 8 Historical evolution of mean deviations (rail projects)

Fig. 9 Historical evolution of standard deviations (rail projects)

Bain (2009) − 23.00 − 22.00 − 21.00% − 20.11% − 21.00%

Table 3 Types of variables used in the literature

Main causes for transport inaccuracy

Opportunistic behaviour and optimism bias × × × × × × × × × 9

Conclusions and policy implications

Irrespective of the technological and methodological improvements, the problem of fore-

Governance, with involuntary (optimism) or voluntary (strategic) bias, is one of the

causes for inaccuracy

causes for inaccuracy

Discusses the application of these methods during recession

Operating costs forecasts actual values

Næss (2015) England tives (compared to implementing new projects)

You might also like

Fig. 2 Mean deviation per sample size (road projects)

Fig. 3 Standard deviation per sample size (road projects)

Fig. 4 Historical evolution of mean deviations (road projects)

Fig. 5 Historical evolution of standard deviations (road projects)

Fig. 6 Mean deviation per sample size (rail projects)

Fig. 7 Standard deviation per sample size (rail projects)

Fig. 8 Historical evolution of mean deviations (rail projects)

Fig. 9 Historical evolution of standard deviations (rail projects)

Table 3 Types of variables used in the literature