traffic and revenue projections for toll roads
TRANSCRIPT
TRAFFIC AND REVENUE PROJECTIONS FOR TRANSPORT CONCESSIONS
Luis Willumsen
2
WHY?
Motivation
Real money is at stake Clear focus on specific outcomes Modelling and forecasting are immediately valued Some very bright people are involved Significant modelling challenges Short timescales, quick results Useful lessons for other modelling and forecasting applications
3
CONTEN
TS
Contents
Context Revenue RiskModelling approaches Toll Roads Risk AnalysisWhat makes a good forecaster
4
CONTEXT
Context: the concession process
5
AGEN
TS
Agents
Government, Regulatory Offices Promoters, construction companies Financial Institutions Rating Agencies Monoline insurers and other insurance companies Operators Rolling stock manufacturers Integrators Lawyers, Other Consultants
6
PRO
JECTFINAN
CE
Project Finance
Risk Cost of Finance How to reduce risk
• Better concession contracts• Insurance (several)• Guarantees (especially minimum revenue)• Good Revenue Projections (international level)
Financial close is paramount
Risk Cost of finance
Price of bid
Financial close
7
RISK
PRO
FILES
Risks
General risks Country Currency Political risks
Project specific risks Construction Operating costs Force Majeure Revenue
• Traffic volume, composition and abstraction• Ramp‐up• Growth• Future competition• Toll and fare adjustments (indexing and how)
8
RISK
PRO
FILEOFTO
LLROAD
Typical risk profile of toll road over time
Operating cos ts
Traffic & Revenue
Construction cos ts
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
-2 0 2 4 10
Year
Ris
k (n
omin
al)
Handover
Opening
Pre completion Post completion
Operating cos ts
Traffic & Revenue
Construction cos ts
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
0.80
0.90
1.00
-2 0 2 4 10
Year
Ris
k (n
omin
al)
Handover
Opening
Pre completion Post completion
9
TRAFFICPRO
JECTIONSMAY
BEUNRELIABLE
Past performance of revenue projections is poor
J P Morgan found in 14 Case Studies (USA) 2 underestimates (10 to 30% below actual) 4 moderate overestimates (12 to 25% over actual) 8 ‘blue sky’ overestimates (45 to 75% over actual)
Why? Ignored price and alternatives Ignored willingness to pay tolls Overestimated growth prospects
10
IMPRO
VERELIABILITY
ANDREDU
CERISK
Improve reliability and reduce risk
How?
We need to provide
Confidence in the inputs, extensive and up to date Confidence in the methods Confidence in the results Risk Analysis and treatment Communicate to deliver trust in the Traffic and Revenue
Advisors
11
METHO
DOLO
GIES
Methodologies
• Early Risk Identification and evaluation• Review existing data and collect new, even if just to
validate• Specify models carefully; classic 4/5 stage and AB models
are almost never appropriate• Focus on what really matters!
In‐scope traffic Willingness to pay tolls/fares Growth Competition Contractual obligations Understanding the economic and social context
12
USE
OFM
ODELS
Specification of models
Clients do not trust ‘black boxes’
Excessive complexity high risk
Team more important than software
Excel may be enough
Avoid overlarge zoning systems and networks
Focus on what really matters and demonstrate it
Risk analysis and mitigation
Ability to explain process from input to output
13
Compe
tition
IN SCOPE TRIPS
TRANSPORT MODEL
CAPTURE TRAFFIC AND
REVENUE
WILLINGNESS TO PAY
BENEFITS OFNEW FACILITY
MODELLIN
G
Generic model
Tren
ds
GROWTH
14
WILLIN
GNESS
TOPAY
Willingness to pay
Drivers pay to save time and costs (comfort, security) Subjective Value of Travel Time Depends on Income, Purpose (group), quality of the driving task
Public transport travellers are also willing to pay money for a better journey Travel time, reliability, seat, how crowded, opportunity to
read/work, comfort
Stated and Revealed Preferences Income distribution analysis Segmentation of demand is critical
15
VA
LUESOF
TIME
Value of Time by Country
Chi
le
Arg
entin
a
UK
Swed
enSp
ain
Port
ugal
Net
herla
nds
Italy
Irela
nd
Ger
man
y
Fran
ce
Finl
and
Den
mar
k
Aus
tria
Bel
gium
Can
ada
Nor
way
USA
-
10,000
20,000
30,000
40,000
50,000
60,000
Country
GD
P p
er C
apita
$
0
2
4
6
8
10
12
14
16
18
20
Val
ue o
f Tim
e $/
hr
"GDP per capita US$" VOT Business VOT General
16
WILLIN
GNESS
TOPAY, TRU
CKS
Trucks and freight
Who chooses route/mode? Clients – shipper – haulier ‐ drivers Value of Time for Freight Depends on the value of the commodity and type of
contract (JIT) Vehicle Operating Costs Savings (gradients, stops) Stated and Revealed Preferences Company maturity
Large and small operators Large and small trucks
17
Sydney Orbital System: an example
39 km long15 junctionsToll 25 cents/km, all vehiclesToll capped at $5 (20 km)Two lanes each wayTolled using free‐flow ETC30 years concession Completes orbital route around SydneyServes main area of expansion of Sydney
18
Modelling capped tolls
55 5 5 5
5555 55 55 55 55 5 5 5V1
2020 20 20 2020
V2
V=V1+V2V
CUBE can do this and more general pricing schemes
19
WSO Heavy Veh VTTS 2001
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0-5 5.01-10 10.01-20 20.01-30 30.10-40 40.01-50 50.01-60 60.01-70 70.01-80 80.01-90 90.01-100
100.01-110
110.01-120
120.01-130
130.01-140
140.01-150
VTTS ($/hour)
Freq
uenc
y
Heavy vehicle Free Flow Heavy vehicle Slowed Down Total time
WSO Heavy Veh VTTS 2001
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0-5 5.01-10 10.01-20 20.01-30 30.10-40 40.01-50 50.01-60 60.01-70 70.01-80 80.01-90 90.01-100
100.01-110
110.01-120
120.01-130
130.01-140
140.01-150
VTTS ($/hour)
Freq
uenc
y
Heavy vehicle Free Flow Heavy vehicle Slowed Down Total time
VALU
EOFTIM
EFO
RFREIG
HT
Willingness‐to‐Pay, heavy commercial vehicles
20
MARKET
SEGMEN
TATION
Values of time in model
Category Level VTTS AUS$/hr % of users in category
Low 2.46 30%
Medium 15.90 48%
High 57.95 8%
Employer pays 50.50 14%
Commuters
Average 20.32 100%
Low 2.45 36%
Medium 13.26 36%
High 47.41 12%
Employer pays 50.50 16%
Non-Commuters
Average 19.40 100%
Low 3.79 50%
Medium-High 23.42 50%
Light Commercial
Average 13.55 100%
Heavy Commercial Low 13.96 52%
Medium-high 49.54 48%
Average 31.06 100%
21
RAM
P‐UP
Ramp up forecasting
The transitional period from opening to stable (modelled) flows, or ramp up, is very important as early revenues are critical to the financial success of a concession (and banks look very closely to early revenues)
Sadly, the ramp up period is almost impossible to model using conventional tools It depends on how quickly can people become familiar with the
advantages (and costs) of the new facility Therefore commuter traffic has a shorter ramp up than inter‐city Significant advantages result in shorter ramp‐ups Complex pricing leads to longer ramp up periods Information and marketing can shorten the ramp up
The best guidance is other similar cases and judgement
22
REVEN
UES
Revenue model
Traffic is never directly converted into revenue; there will be: Ramp up Seasonal variations and the need for annualisation Exempted vehicles Non‐payers, evasion Leakage Revenue sharing
In the case of public transport concessions Sharing of revenue with integrated fare systems is complex There are more discounted fares, season tickets
Usually a spreadsheet models take care of these issues
23
LRT, BRT ANDM
ETRO
Public transport concessions
Increasing participation of private sector Infrastructure usually needs public sector contribution But revenue risk is often allocated to operations (fraud,
fare collection and evasion, marketing, integration) Mode transfer from cars is an issue (policy and revenue) Assumptions about what the competition will do are key Mode Specific Constant, the inherent advantage of the
new mode, is a problem Benchmarking is useful
24
KEY
ISSUES
Risk analysis
Confidence that the model represents the present and future well Build different scenarios to cope with a changing future De‐construction of model outputs to show dependencies Stochastic risk analysis to provide envelope of uncertainty
Uncertainty about future
inputs
Errors and imperfections
in model
Uncertainty about
forecasts
GDP, Population,Competitors,PricesPolicies
SegmentationBehavioural responsesSVTTrips/toursZones & Networks
TrafficRevenuestreams
25
KEY
ISSUES
Model and future uncertainties
Update the model and forecasts to recent data and changes Check model produces sensible and defensible results Agree basis for scenarios with bidder and financial advisors Produce at least three scenarios: stressed (downside), expected (base) and Optimistic (upside) forecastsMore scenarios may be needed if uncertainty about competitors responses or other factor is critical Undertake sensitivity tests Willingness to pay Revenue mix – joint tickets Evasion Different customer groupsGrowth in population and GDP Reorganisation of public transport, parking policy
26
OUTPU
TMAKE
UP
Deconstruct the model in its components
Modelling Demand captured from other roads or PT Demand growth What is capturable from other modes Re‐distribution Pure generated traffic
Demand segmentation By type of vehicles Urban and long distance travel Week day and week end ETC and conventional tolls Season ticket and single fare users Discounts
27
PUBLIC
TRANSPO
RTDECO
NSTRU
CTION
Conceptual LRT forecasts
28
STOCHASTIC
RISKAN
ALYSIS
Stochastic risk analysis
Beloved by Financial Institutions Generally based around Base Case Take 2‐3 key variables : GDP, SVT, etc ..and look into their historical variability (standard deviation ) The model is used to track variability in revenue resulting from variability in key inputs, usually via a simplification in Excel FutureRevenue = RevenueFactor * Base Case Revenue A value for RevenueFactor of 1 indicates Base Case Also presented as the level of revenue that is likely to be exceeded 90 or 95% of the time (P90 and P95)
29
TYPICALOUTPU
T
Example of output
0.85
0.9
0.95
1
1.05
1.1
1.15
2003 2005 2007 2009 2011 2013 2015
Year
Rev
enue
Fac
tor
GDP variation σ: 0.5%
SVT variation σ : 2.5%*mean VST
30
CASE
STUDY
A Mexico toll –road transaction
The main driver of future demand was GDP growthIt affects underlying (in‐scope via car ownership) demand and growth in willingness to payObservations on past variations of GDP growth suggested:
Normal distribution of variations (once outliers were removed) Standard Deviations around 1.5
Undertook the P90 analysis using 5000 Monte Carlo iterations with Normal Distribution and SD =1.5P90 is the level of demand that will be exceeded in practice with a probability of 90% (9 out of 10 times)
31
FORECASTSC
ASE2P90 and P95 for toll road
'Four Roads- Case 2 WRe P90 sd 1.5
2
7
12
17
22
2008 2010 2012 2014 2016 2018 2020 2022 2024 2026 2028 2030 2032 2034 2036
Mx$
Bill
ions
32
PRACTICALASPECTS
Some recommendations
Bear in mind the need to reach financial close and still have a project that generates a profit to investors
Transparency and traceability are paramount You will be audited by financial institutions sooner or later If problem is very simple, use Excel; otherwise, use a well
established software package: CUBE is good Use at least 6 user classes for cars and 4 for trucks with
different willingness to pay Collect up to date data, at least counts and OD surveys Confirm willingness to pay Always model tolls as tolls and not as time penalties Beware of perverse incentives
33
CLO
SURE: THE
TRANSPO
RTMODELLER
PROFESSIO
NAL
What makes a good modeller forecasting professional?
Curiosity: broad range of interests to enable you to understand the underlying reasons for demand beyond any model or theory
Good listening skills Communication skills Analytical ability Questioning mind, in particular of your own results Realism: all modelling projects take longer than you initially anticipate; plan accordingly
The final test: would you put your retirement fund on this project?
Luis Willumsen Consultancy
THANK YOU