Price Discovery in an Agent Based Model Simulation for

Price Discovery in an Agent Based
Model Simulation for Consequential
LCA of Bio-Energy
Sameer Rege, Tomás Navarrete , Antonino Marvuglia, Enrico Benetto
May 20, 2014
CRP Henri Tudor, 6A, avenue des Hauts-Fourneaux, L-4362,Esch/Alzette, Luxembourg.
Structure of the Presentation
 Motivation
 Data Exploration
 Model Structure
 Initial Results
 Discussion
 Conclusion
The Setting
Environment Minister of Luxembourg : Carole Dieschbourg: Luxembourg to
cut GHG emissions by 40% and increase share of renewable energy to
30% by 2030 (Luxemburger Wort March 21, 2014)
Possible thoughts
Farmers : The milk quotas are disappearing, will they give a
higher price for maize to produce biofuel? I could get rid of
cows
Residents: Would we need to pay more for green energy? Is it
going to be imported?
Bureaucrats: 30% share of renewables !!
STATEC data shows 2.64% (4671/176336 TJ) in 2012
We could subsidize biofuels and tax the petrol. How much hit
would we take on tank tourism? Will need a model. How will
people react? I read somewhere we could do an LCA to study
Environmental impacts
Researchers: Which LCA?? Attributional / Consequential?
Consequential will be more appropriate.
Policy Maker: What’s the difference?
Researcher: LCA Perspective
 Attributional LCA
 For a product based on average technology
 Robust, Unambiguous with high level of accuracy
 Stoichiometric relationships between inputs and outputs
 Not suitable for evaluation of policies
 Consequential LCA
 Policy changes impact scale of output of product. Both inside
and outside the life cycle
 Changes (Δ) based on « fragile » economic and financial
relationships rather then more robust physical relationships.
 Greater uncertainty on account of external models
 Includes all indirect effects and works with technology to
produce the marginal unit of output
 Recommendation: You will need to do a CLCA
Researcher: We got two approaches
 Economic Modelling ( <- has same behaviour for All)
 Robust for obtaining prices and computing aggregate changes
 Agent Based Modelling (<- this has behaviour)
 Systems are made of agents, environment and interactions [Ferber, 1999]
 Models are “simulated” and not rooted in optimization behaviour. Agents
have no utility for utility functions!
OpenLCA,
SimaPro,
etc…
Our own simulator
Decision
Environment
deltas
CLCA
So What is your Point??
Policies affect prices
Prices determine Profits or Losses
Leads to changes in behaviour
Implies changes in cropping patterns
Leads to
deltas
CLCA
OK so what??
How should we model the market mechanism to
generate a true picture of price discovery?
Why is this important?
Wrong price discovery => Incorrect prices => Different
Behaviour => different cropping patterns =>incorrect
DELTAS = wrong input for CLCA!
17/05/16
7
Way Forward ….
BUREAUCRAT / POLICY MAKER : OK! So we are
convinced. An Agent Based Model to conduct
CLCA of biofuels
What next?
RESEARCHER:
We briefly explain the Data and the model
Base Data 2009
Luxembourg Farm Size Distribution
Min [
Max )
Min [
Max )
A
0
2
F
30
50
B
2
5
G
50
70
C
5
10
H
70
100
D
10
20
I
100
E
20
30
Agent Based Model - Flowchart
Initialise the ABM with actual data.
Calibrate agent data to match base year values
Model market clearance mechanism. Do What??
Determine the supply of each crop by each farmer and
the selling price
Detemine the demand for each crop by a buying agent
and the buying price
Generate the supply-demand curves
Market Clearance: Barley spring –t0
Profits and prices of cereals
Price
ComparisonActual v/s
ABM
Quantity
Comparison
-Actual v/s
ABM
Discussion
 Only one round of market clearance leads to market
power
 Crop replacement based on profit maximization
criteria leads to higher volatility
 Limited ability to model price forecasts
Conclusion
• Agent Based Models permit granularity that is impossible
to achieve with any other approach (equilibrium models)
• Susceptible to Modeller bias
• Absence of a formal mechanism for closure and is largely
based on statistical simulation
• Potential to study impact of each behavioural rule on the
model outcome thus enabling a potential precise
targetting of policy rules.
• Can carry out a CLCA based on the Δs obtained from
different behavioural rules for the same policy.
Work supported by the Luxembourg
National Research Fund (FNR)
http://www.tudor.lu/musa
Thank you for your attention !
Sameer Rege
R&D Engineer
CRP Henri Tudor
Luxembourg
E-mail: [email protected]