Transcript Slide 1

Transportes, Inovação e Sistemas, S.A.
Using @RISK for Traffic Forecast Analysis
Case Study: Marão Tunnel Concession
Palisade User Conference
London, 22nd April 2008
Inês Teles Afonso
Using
@RISK
for Traffic35,
Forecast
Analysis, London
Av. da
Republica,
6º 1050-186
Lisboa | |22.04.2008
Portugal
TIS.PT – Transportes Inovação e Sistemas, S.A.
| www.tis.pt
Slide 1 | 29
Table of Contents
Case Study Presentation
Context – What is a traffic study?
What are the advantages of using @RISK?
Traffic Modelling Model (VISUM) with @RISK
Methodology @RISK
Results Analysis
The data presented in this presentation was modified so that we
could ensure the privacy of our client
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 2 | 29
Case Study Presentation
National Politics
TUNNEL MARÃO CONCESSION
Strategy
Identical opportunities
of Development
Similar mobility
conditions
Traffic Study
Main Objective
Traffic Forecast in
concession sections
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 3 | 29
Case Study Presentation
IP4
Junction 4
EN312
EN210
Vila
Real
Junction 5
EN304
EN15
EN15
Quintã
Amarante
IP4
Campeã
Junction 1 Junction 3
IP4
A4/IP4
IP4
Vila Real
MARÃO TUNNEL
Carvalhais
CONCESSION
A4/IP4
A24/IP3
A4/IP4
EN2
EN15
EN210
Junction 2
EN101
EN313-2
EN304
Penaguião
EN101-5
EN101
A24/IP3
Main characteristics:
Connection between cities of Amarante and Vila Real
length → 30 km
Cross-Section → 2x2
Free Flow Speed → 100 km/h
Tolled motorway → open scheme with toll charge in road section → 3 - 4
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 4 | 29
Case Study Presentation
Traffic Model was built on the VISUM
platform
The obtained results allow to predict the
traffic demand on the studied sections
over the period of analysis
4
1
2
5
3
Note: These values doesn't correspond to the real project values
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 5 | 29
Context – What is a traffic study?
Transport Demand
Characteristics
Transport Supply
Characteristics
INPUT
Traffic Study /
Traffic Model
Traffic Forecasts for the study infra-structure
Finance Analysis
(...income estimates)
Project Analysis
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
OUTPUT
Environmental
Impact Assessment
Slide 6 | 29
Context – What is a traffic study?
It is a very important issue to know the expected
evolution for the traffic (OUTPUT) and what are
INPUT
the associated risks.
Usually we reflect the uncertainty of the model
Traffic Study /
Traffic Modal
results in three scenarios such as “central”,
“optimistic” and “pessimistic” allowing for a
very limited deterministic analysis.
Is it possible to present a clearer and
stricter outcome?
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
OUTPUT
ANALYSIS
Slide 7 | 29
What are the advantages of using @RISK?
YES!
Using @RISK, the OUTPUT (traffic
forecast)
probability
is
represented
distribution
by
variable
A
variable
B
variable
C
a
which
improves the quality of decision-
making.
Traffic Model
Relations..
HOW?
With
simulation
(Monte
Carlo
OUTPUT
simulation). In each iteration @RISK
tries all valid combinations of the
values of INPUT to simulate all
possible outcomes (OUTPUT).
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
ANALYSIS
Slide 8 | 29
Traffic Modelling Model (VISUM) with @RISK
PROBLEM
Due the complexity of traffic model, it
takes some time to get its outcome
(traffic forecast). Therefore it is not
feasible to do a traffic assignment
(VISUM) for each @RISK iteration.
4 minutes for a traffic assignment
SOLUTION
10.000 @RISK iterations
Draw adjustment curves representing
iterations duration
833 hours!
the relationship between OUTPUT
(traffic forecast) and INPUT variables -
35 days!!!
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Elasticity Curves
Slide 9 | 29
Traffic Modelling Model (VISUM) with @RISK
2. Traffic
Assignments
3. Elasticity curves
adjustment
OUTPUT - AADT.km
1. Change INPUT
values
Variable
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 10 | 29
Methodology @RISK Application
1. OUTPUT Variable definition
2. INPUT Variables
variable A
variable B
variable C
variable D
INPUT variables definition
Definition of the probability
distribution for each one
Traffic Model
Relations..
Analysis of correlation
between them
3. Interaction between INPUT and
OUTPUT
OUTPUT variables
4. @RISK Simulation
5. Results Analysis
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 11 | 29
OUTPUT Variable Definition
OUTPUT Variable
Traffic Forecast (AADT.km) for Tunnel Section (2011, 2020, 2030,
2040 and Accumulated Revenue)
IP4
Junction 4
EN312
EN210
Vila
Real
Junction 5
EN304
EN15
EN15
Quintã
Amarante
IP4
Campeã
Junction 1 Junction 3
IP4
A24/IP3
A4/IP4
IP4
Vila Real
A4/IP4
Carvalhais
A4/IP4
EN2
EN15
EN210
Junction 2
EN101
EN313-2
EN304
Penaguião
EN101-5
EN101
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
A24/IP3
Slide 12 | 29
INPUT Variables Definition
INPUT Variables
Transport Demand
GDP Annual Variation Rate (after 2009)
Traffic growth
factors
Toll value f(VAT)
Value of Time (VoT) Variation Rate
Fuel Cost Annual Variation Rate
GC= l (length).Co (Operational Cost) + t (travel time).VOT + l.T(unit toll)
Generalized Cost
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 13 | 29
Probability Distribution for INPUT Variables
variable A
variable B
variable C
variable D
Around each input variable, there
was a very deep discussion to
decide
which
probability
distribution should the variable
assume.
This discussion was based mainly
on expert judgement.
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 14 | 29
Probability Distribution for INPUT Variables
GDP Annual Variation Rate (after 2009)
The source for GDP before 2009 was the Bank of Portugal
After 2009 it is considered a stochastic variable
To avoid to have negative
values of traffic, which is
a non sense, the
distribution was
truncated.
Normal distribution
Mean= 2,3%
Standard Deviation =
0,5%
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 15 | 29
Probability Distribution for INPUT Variables
Fuel Cost Variation Rate
It was considered that the likeliness of fuel prices reaching very high
levels in the long or medium term is higher than that of regressing to
lower levels
The modelled variable
consists of the Fuel Cost
Variation until 2020.
Weibull distribution
Percentile 5% = 0,8
Percentile 50% = 1
Percentile 95% = 1,5
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 16 | 29
Probability Distribution for INPUT Variables
Value of Time (VoT) Variation Rate
VoT is one of the most decisive parameters for the route choice model;
Research on VoT growth over time indicates annual growth
ranging from 30% to 100% of
annual GDP growth rate
In the deterministic
approach it was used 70%
Triangular Distribution
minimum = 0,3
Most likely = 0,7
Maximum = 1
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 17 | 29
Probability Distribution for INPUT Variables
Toll value
Toll = €0,07.(1+VAT).(paid length)
The toll value is changed when VAT changes. VAT is the INPUT variable;
In 2007 Portuguese VAT was 21%;
It is not likely that VAT can
increase much more;
The probability of simulating a
lower VAT than the most likely is
higher than getting a higher
most likely value
Weibull distribution
Percentile 5% = 0,18
Percentile 50% = 0,21
Percentile 95% = 0,23
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 18 | 29
Correlation Analysis Between INPUT Variables
The correlation matrix was constructed considering the following variable
relations:
Negative correlation between GDP and VAT, and GDP and Fuel Costs
Positive correlation between GDP and VoT
Positive correlation between VAT and Fuel Costs
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 19 | 29
Interaction between INPUT and OUTPUT variables
variable C
Traffic Model
Relations..
variable D
AADT.km
variable B
AADT.km
variable A
Fuel Cost rate
AADT.km
OUTPUT
AADT.km
GDP rate
Toll f (VAT)
Value of Time
GDP Annual Variation Rate and VoT Annual Variation rate have positive
elasticity with the traffic forecast, which means that when they increase, the
traffic demand on the Tunnel also increases
Fuel Cost and Toll Annual Variation Rate have negative elasticity with the
traffic forecast. Their growth implies a traffic demand decrease on the Tunnel
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 20 | 29
The data presented in this presentation was modified so that we
could ensure the privacy of our client
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 21 | 29
Results Analysis – Output Distributions Graphs
1
2
Traffic Model
Traffic Model
3
4
Traffic Model
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Traffic Model
Slide 22 | 29
Results Analysis – Output Distributions Graphs
The red line
represents the
deterministic output
(traffic forecast.km)
of the Traffic Model.
The deterministic outcome is always on the right side of the mean value of
the distribution.
This means that traffic study may have assumed optimistic values
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 23 | 29
Results Analysis – Output Distributions Graphs
The
Percentile 5%
uncertainty
Percentile 50%
increases with time
Percentile 95%
This
evolution
of
is
the
an
model
intuitive
perception
AADT.km
But the stochastic model allows to
2010
see that the uncertainty is bigger for
the lower demand values
2020
2030
Percentile 5%
Percentile 95%
-5%
-24%
-28%
-28%
4%
20%
25%
25%
2040
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 24 | 29
Results Analysis – Tornado Graphs
Regression Sensitivity for AADT.km TUNNEL /
2011/F16
1
GDP Annual Evoluation Tax .../C4
,766 GDP Annual Evoluation Tax .../C4
Fuel Cost Annual Evoluatio.../C7
-,281
,242
VOT (relation GDP & VOT) /.../C13
-0,75
-0,5
-0,25
0
0,25
1 -1
0,75
0,5
3
,593
-0,5
-0,25
-0,5
0,25
VAT (Toll charge) / SIMULA.../C10
-0,25
0
0,75
0,5
0,75
1
Std b Coefficients
4
,575
-,559
Fuel Cost Annual Evoluatio.../C7
VOT (relation GDP & VOT) /.../C13
0,5
0,25
Regression Sensitivity for AADT.km TUNNEL /
2040/AI16
VAT (Toll charge) / SIMULA.../C10
0
,242
GDP Annual Evoluation Tax .../C4
,326
-,03
-0,75
-0,75
Fuel Cost Annual Evoluatio.../C7
VOT (relation GDP & VOT) /.../C13
-1
Fuel Cost Annual Evoluatio.../C7
-,07
Std b Coefficients
-,58
,647
VOT (relation GDP & VOT) /.../C13
Regression Sensitivity for AADT.km TUNNEL /
2030/Y16
GDP Annual Evoluation Tax .../C4
2
-,566
VAT (Toll charge) / SIMULA.../C10
-,227
-1
Regression Sensitivity for AADT.km TUNNEL /
2020/O16
,344
-,125
1 -1
-0,75
Std b Coefficients
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
-0,5
-0,25
VAT (Toll charge) / SIMULA.../C10
0
0,25
0,5
0,75
Std b Coefficients
Slide 25 | 29
1
Results Analysis – Tornado Graphs
Regression Sensitivity for AADT.km TUNNEL /
2011/F16
1
GDP Annual Evoluation Tax .../C4
,766 GDP Annual Evoluation Tax .../C4
-,281
These results shows
,242
-,227
-1
uncertainty of the
-0,75
-0,5
0
0,25
VOT (relation GDP & VOT) /.../C13
0,5
0,75
3
,593
-,58
outcome
1 -1
-1
-0,75
-0,5
-0,25
-0,5
0,25
0
4
0,75
0,5
0,75
-,559
Fuel Cost Annual Evoluatio.../C7
,344
-,125
1 -1
-0,75
-0,5
Std b Coefficients
-0,25
VAT (Toll charge) / SIMULA.../C10
0
0,25
0,5
0,75
Std b Coefficients
What factors cause higher uncertainty on the traffic forecast?
•
GDP is the INPUT with more influence. Fuel Costs are the second most
influential and become more relevant until 2020 where the variable value
remains constant
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
1
,575
VOT (relation GDP & VOT) /.../C13
0,5
0,25
Regression Sensitivity for AADT.km TUNNEL /
2040/AI16
VAT (Toll charge) / SIMULA.../C10
0
-0,25
GDP Annual Evoluation Tax .../C4
,326
-,03
-0,75
VAT (Toll charge) / SIMULA.../C10
Std b Coefficients
Fuel Cost Annual Evoluatio.../C7
VOT (relation GDP & VOT) /.../C13
,242
-,07
Regression Sensitivity for AADT.km TUNNEL /
2030/Y16
GDP Annual Evoluation Tax .../C4
,647
Fuel Cost Annual Evoluatio.../C7
VAT (Toll charge) / SIMULA.../C10
-0,25
2
-,566
Std b Coefficients
INPUT variables on
uncertainty of output
Fuel Cost Annual Evoluatio.../C7
VOT (relation GDP & VOT) /.../C13
the importance of the
Regression Sensitivity for AADT.km TUNNEL /
2020/O16
Slide 26 | 29
1
Results Analysis - Revenues
What is the possibility of having the revenues 15% less than the deterministic
forecast?
•
The model allows to estimate that that outcome can occur with a probability
of 13%
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 27 | 29
Conclusions
With Deterministic model:
•
The outcome sensitivity analysis is given by deterministic results by changing the
values of the input variables;
•
It is not possible to measure the probability of those results.
With stochastic approach (@RISK)
The outcome sensitivity analysis is based on a probabilistic distribution;
It improves the deterministic analysis answering to the following questions:
what are the expected variation for the traffic forecast results?
what are the factors that cause higher uncertainty on the traffic forecast?
What are the risks of having less revenue than the deterministic forecast?
For all of these, the decision (expert and client) can obtain a more transparent and
accurate approach of the outcome presented by traffic model using @RISK analysis
software.
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 28 | 29
Conclusions
The undertaken analysis allow to identify the main RISKS associated with the
Concession Traffic Forecast.
Usually, on the deterministic model we assume the most likely values for the
input variables.
The @RISK results, in this case, allows to observe that the deterministic
outcome could have been too optimistic
The information supplied by @RISK analyses allows to add information to the
traffic forecast results, improving the interpretations of the results
In future analysis we remain with two main challenges:
to accurately replicate the relevant relations of the traffic model in
Excel (VISUM with @RISK)
to improve the methodology for the setting of the probability
distributions
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 29 | 29
Results Analysis – Tornado Graphs
Thank You
Using @RISK for Traffic Forecast Analysis, London | 22.04.2008
TIS.PT – Transportes Inovação e Sistemas, S.A.
Slide 30 | 29
Transportes, Inovação e Sistemas, S.A.
Using @RISK for Traffic Forecast Analysis
Case Study: Marão Tunnel Concession
Palisade User Conference
London, 22nd April 2008
Inês Teles Afonso
Using
@RISK
for Traffic35,
Forecast
Analysis, London
Av. da
Republica,
6º 1050-186
Lisboa | |22.04.2008
Portugal
TIS.PT – Transportes Inovação e Sistemas, S.A.
| www.tis.pt
Slide 31 | 29