Serious Games as a Tool to Understand 
Complexity in Market Competition: An 
Evolutionary Game Theory Simulation Platform ...
Outline 
 Part 1: Motivations – Making sense in a complex world 
Is there a better tool 
than Excel™ ? 
 Part 2: GTES (G...
Complexity is everywhere in our companies 
 Complexity is everywhere 
 Multiple elements and multiple relations, 
emerge...
From Strategic Planning to Serious Games 
 The solution is not better forecasting  
 With stochastic approaches towards...
History of GTES Development 
 2000: UMTS Bid 
 2004 – 2006 : Distribution Channels Optimization 
 2006 – 2009 : Mobile ...
Systemic Simulation of Smart Grid 
« Regulator » 
• Strategy: reduce CO2 emissions, preserve 
economic throughput (enough ...
Yearly 
Investments 
• Grow / reduce 
nuclear assets 
• Add fossil capacity 
Regulator / 
Environment 
Environment paramet...
Part II 
 Part 1: Motivations – Making sense in a complex world 
 Part 2: GTES (Game-Theoretical Evolutionary Simulation...
Game Theoretical Evolutionary Simulation (GTES) 
GTES is a tool for looking at a complex 
model with too many unknowns 
Pr...
Game-Theoretical Evolutionary Simulation 
GTES 
parametric analysis 
Machine Learning 
Local 
optimization 
Search for Nas...
Evolutionary Algorithms & Machine Learning 
From each actor’s viewpoint, everything being equal, a GTES model defines 
a p...
The Search for Nash Equilibriums 
 When each actor is maximally satisfied, w.r.t. each other actor’s tactic 
, , ( , ) ( ...
Sampling 
 Monte-Carlo 
 The uncertainty regarding the environment parameters 
e is handled through randomization 
 Eac...
Lessons from Practice 
 Defining the satisfaction (w.r.t strategy = set of goals) is critical 
 Additive versus multipli...
Part III 
 Part 1: Motivations – Making sense in a complex world 
 Part 2: GTES (Game-Theoretical Evolutionary Simulatio...
Energy demand Market Share 
Demand generation 
Operator Production 
Max penetration rate 
Price sensitivity 
Dynamic Prici...
S3G : Players’ objectives (optimization functions) 
 Each players tries to optimize three “KPI” (performance indicators) ...
S3G : Simulation & Game Protocol 
S3G Work Plan 
(1) Implement S3G Model 
The model has successively been implemented (100...
Sensitivity to variability 
 Testing the hypothesis that variability favors local operator 
variability 
Demand 
response...
Carbon Tax, Green Power and Storage Cost 
E1 
Reference 
S3: 
CO2 tax 
S3a CO2 tax 
+ solar 
S4 = S3a +cheap 
storage 
S6 ...
Smart Grids : Strategy Matrix 
S3G is a stable model (no war/ chaos) but ESS convergence is approximate 
160 
140 
120 
10...
Oil Price Sensitivity and « De-nuclearization » 
Wholesale 
price 
Operator 
income 
Market 
Share 
Solar 
Investment 
Sto...
S3G Temporary Conclusions 
 Systemic Simulation of Smart Grids 
 A very simple model … 
 … yet which captures a number ...
Part IV 
 Part 1: Motivations – Making sense in a complex world 
 Part 2: GTES (Game-Theoretical Evolutionary Simulation...
Example (1): Distribution Networks 
 Simple model 
 Two-steps phases which 
are distinct from an 
organization viewpoint...
Example (2) : Commercial Costs Optimization 
 Problem: resource allocation between different channels … 
 … while taking...
Example (3) : CGS – Cellular Game Simulation 
 Simulation of competition between n mass-market telephony 
operators 
 Mo...
CGS – Operator Model from CEO’s insight 
 Input 
 ACQ: average acquisition cost per customer 
 FID: loyalty cost 
 “av...
Coupling 
1 
2 3 1 
Orange 
1 
2 3 
Opérateur 
 Acquisition/ price relationship → 
2 3 
SFR 
1 
2 3 
Bouygues 
Nouveaux 
...
CGS – Simulation Architecture 
 5-steps computational model 
Nouveaux 
clients 
Résultats Volumes 
5 4 
Base 
op 2 
Calcu...
Open Mass-Market Naive Competition 
(Example 4) 
Each actor is a company with 
 an ARPU (price) 
 an attractiveness (pre...
Mobile Operators (I) : Playing What-If Scenarios 
4500 
4000 
3500 
3000 
2500 
2000 
1500 
1000 
500 
0 
100% 
80% 
60% 
...
Mobile Operators (II) : Strategy Analysis 
Sensitivity to alpha (aggressive) Sensitivity to alpha (conservative) 
100% 
80...
Part V 
 Part 1: Motivations – Making sense in a complex world 
 Part 2: GTES (Game-Theoretical Evolutionary Simulation)...
Performance Issues 
 Models that have been tested with GTES are computationally 
simple, still running once simulation ra...
Quality of Model Issues 
 Modeling « Torture Bench »  
 Machine learning => zooms on logic faults 
 No mercy for linea...
Relations with System Dynamics 
 System Dynamics 
 Models based on interaction 
networks between state 
variables 
(e.g....
Serious Games : Key Take-aways 
What-if Scenarios, 
Players’ strategies 
Results 
Game 
Analysis 
Models’ torture bench  ...
Failed SNS Experiment 
 I applied GTES to study the “Google+ versus 
Facebook” fight two years ago. 
 Value of SNS exper...
RTMS : Repeated Tender Market Share 
 Problem: find a model to reproduce the behavior of buyers / 
sellers in a closed re...
Future : Connected Health Trust Game 
 The problem : data privacy issues with connected devices 
 Would you share your h...
Conclusion 
« The difficulty lies, not in the 
new ideas, but in escaping 
from the old ones.” 
J. M. Keynes 
 “Serious G...
Prochain SlideShare
Chargement dans…5
×

GTES UTC 2014

804 vues

Publié le

Serious Games as a Tool to Understand Complexity in Market Competition: An Evolutionary Game Theory Simulation Platform

Presentation to Labex MS2T, UTC Compiegne

Publié dans : Sciences
0 commentaire
1 j’aime
Statistiques
Remarques
  • Soyez le premier à commenter

Aucun téléchargement
Vues
Nombre de vues
804
Sur SlideShare
0
Issues des intégrations
0
Intégrations
16
Actions
Partages
0
Téléchargements
12
Commentaires
0
J’aime
1
Intégrations 0
Aucune incorporation

Aucune remarque pour cette diapositive

GTES UTC 2014

  1. 1. Serious Games as a Tool to Understand Complexity in Market Competition: An Evolutionary Game Theory Simulation Platform November 28th, 2014 – v0.3 UTC – Labex MS2T Yves Caseau National Academy of Technologies – AXA Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 1/42
  2. 2. Outline  Part 1: Motivations – Making sense in a complex world Is there a better tool than Excel™ ?  Part 2: GTES (Game-Theoretical Evolutionary Simulation)  Part 3: Smart Grid Systemic Simulation Example  Part 4: Mass Market Telephony Simulation Examples Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 2/42
  3. 3. Complexity is everywhere in our companies  Complexity is everywhere  Multiple elements and multiple relations, emergent behavior (ecosystems)  Feedback loops and delays  Uncertainty  Planning / Forecasting is still a major corporate activity  Budget, marketing, business plans, …  Excel™ is still the preferred tool  Complicated spreadsheets … at best, a few scenarios and sensitivity analysis  Today’s business practices are suited to a complicated world, not a complex world  Taking competitors & markets into account (adaptation)  Taking uncertainty into account  Enough of linear extrapolations ! Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 3/42 Part I : Motivations
  4. 4. From Strategic Planning to Serious Games  The solution is not better forecasting   With stochastic approaches towards uncertainty …  With multi-variable optimization …  That’s what experience and complexity theory say  We need to develop skills to better prepare for whatever the future is bringing (situation potential)  Cf. the Art of military war games  Practice of multiple simulated situations develop reactive skills (reflexes) and systemic understanding  Lessons from multiple strategic thinkers (Julien, Taleb, …)  Serious Games  Play against “smart” opponents  Experience feedback loops  Each scenario (game) is plausible, even if not likely Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 4/42 Part I : Motivations
  5. 5. History of GTES Development  2000: UMTS Bid  2004 – 2006 : Distribution Channels Optimization  2006 – 2009 : Mobile Operator Competition Model  2009 – 2010 : Extension to Free  2010 – 2012 : Smart Grid Model  Reconcile three geographic visions of Smart Grids  Reconcile two corporate visions of Smart Grid Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 5/42 Part I : Motivations The « Utility » view The “Utility view” defines a smart grid as adapting the power network to: • local sources (as opposed to a one-way distribution network), • intermittent production sources (though storage and favoring flexible production units) • using price incentives to “shave” demand peaks. The « Google » view The “Google view” defines a smart grid as: • change from a tree structure to a network structure (centralized to de-centralized), • the use of market forces to create a dynamic and more efficient equilibrium between supply and demand, • the use of IT to provide information to all actors, including end consumers. The « Japanese » view The “Japanese view” is human-centered instead of being techno-centered. The goal is to change human behavior to adapt to new challenges (lack of resources, global warming, …). Smart grids are the backbone of a multi-scale architecture (smart home, neighborhood, city, region, country) where each level has its own resources and autonomy.
  6. 6. Systemic Simulation of Smart Grid « Regulator » • Strategy: reduce CO2 emissions, preserve economic throughput (enough energy at « acceptable price »), keep a balanced budget • Tactical play: incentives to invest in green energy, CO2 Tax, storage requirement for intermittent sources of energy Open Questions • What part does local storage play? • What CO2 price would change significantly the cost/benefits analysis? • What is the systemic benefit of local management? • What could be the large-scale effect of dynamic pricing on self-optimization of customer demand? • Does Smart Grids provide better resilience ? • Is the relationship between supplier and operators a “coopetition” or a competition ? « Supplier » • Strategy: maintain EBIDTA, reduce exposure to demand peaks, maintain market share • Tactical play: Variable pricing (higher price when demand & production costs are high), power plant investments « Operator » • Strategy: grow turnover, grow EBITDA, increase market share • Tactical play: Storage utilization policy, Dynamic pricing, when to invest on additional capacity (green, storage, fossil) « City » • Strategy: maintain low energy average price, avoid peak prices, preserve comfort (limit “shaving”) • Tactical play: choose local operator or “classical” supplier, invest into energy savings (megawatts) CO2 Tax (Regional fossil/nuclear) Supplier Regulator Green CO2 Tax Incentives/ constraints energy Wholesale « classical » distribution of energy CCitiytCyity City price (Local fossil/green) Operator Energy @ dynamic price Variable demand Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 6/42 Part I : Motivations
  7. 7. Yearly Investments • Grow / reduce nuclear assets • Add fossil capacity Regulator / Environment Environment parameters • Demand growth • Oil Price Trend • Nuclear Growth / Reduction trend • CO2 tax Technology Parameters • Cost of green tech (yearly trend • Cost of storage (yearly trend) Systemic Parameters • Demand variability • NegaWatt generation (alpha) • Peak shaving (beta) • Market share sensitivity (gamma) Yearly • Adjust market Operator City Supplier Operator’s customers Supplier’s Customers MM price supply Open Market The supplier buys electricity on the open Market when demand exceeds capacity, at a very high price demand Buffer Reserve Storage is divided into Two separate units with Different logics supply o.buy o.sell o.inReserve Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 7/42 shares • Invest into « negawatt » energy saving equipments Yearly Investments • Add « green » capacity • Add fossil capacity • Add storage capacity demand o.inBuffer o.fossilePower o.sellReserve o.greenPower o.outBuffer o.outReserve Wholesale price MM price Simulation over 15 Years Systemic Simulation of Smart Grid (Model)
  8. 8. Part II  Part 1: Motivations – Making sense in a complex world  Part 2: GTES (Game-Theoretical Evolutionary Simulation)  Part 3: Smart Grid Systemic Simulation Example  Part 4: Mass Market Telephony Simulation Example Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 8/42
  9. 9. Game Theoretical Evolutionary Simulation (GTES) GTES is a tool for looking at a complex model with too many unknowns Problem (fuzzy, complex, …) Abstrac-tions Model: Set of equations with too many unknown parameters ! Split parameters « Players » Environment (DIP) Strategy (DDP) Tactical (DV) “tactical” may be derived from “strategy” Parameters which are meaningful (e.g., oil price future) Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 9/42 (local optimization) Parameters which describe the player’s goals eParameters sParameters Scenario-defined Obscure & unknown randomize Game Theoretical Approach Player’s degrees of freedom Wolter Fabrycky: DDP/DIP/DV Part II : GTES (Game Theoretical Evolutionary Simulation)
  10. 10. Game-Theoretical Evolutionary Simulation GTES parametric analysis Machine Learning Local optimization Search for Nash Equilibriums Tactical Strategic analysis of the player’s goals Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 10/42 Part II : GTES (Game Theoretical Evolutionary Simulation)  Two ways to look at GTES  Solving a complex undefined optimization problem  Game theory in a complex environment  An approach inspired by Robert Axelrod pioneering work on Agent-based models of cooperation & competition  E.g.; experimental/evolutionary validation of TIT-for-TAT strategy in a repeated prisoner dilemma game Sampling Monte-Carlo Global Parameterized Optimization Problem Parameters Strategy External Non-cooperative Repeated Game Classification Taking the uncertainty of the Actors model into account Context
  11. 11. Evolutionary Algorithms & Machine Learning From each actor’s viewpoint, everything being equal, a GTES model defines a parametric optimization problem: f x e e  E p max , , Multi-actor maximization « game/problem » (fp  Rn) i for each actor represents the « strategy» x  X  The objective function fp  A set of state variables and associated target values (e.g., EBITDA, share, …)  Linear combination + concave valuation of difference  The set x of free variables represent the « tactic »  Finding the best solution (called BR: Best Response) requires to solve an optimization problem  Machine learning means that finding the best tactic is automated  An approximate model requires a heuristic solution  Hill-climbing / meta-heuristics (SA or genetic algorithms) Bounded rationality   Local moves according to a neighborhood structure + dichotomy search  Choosing the proper neighborhood structure is a key modeling choice Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 11/42 Part II : GTES (Game Theoretical Evolutionary Simulation)
  12. 12. The Search for Nash Equilibriums  When each actor is maximally satisfied, w.r.t. each other actor’s tactic , , ( , ) ( , ) i i i i i i i x T f x t f t t       The simplest way to find a NE is to iterate the computation of the « Best Response » function  An iterative loop that may be nested with the local optimization loop  A heuristic version may be derived according to a neighborhood structure V  There does not necessarily exist a « pure » Nash Equilibrium  The loop may not converge ( “destructive war” or “chaos”)  The convergence rate increases with a « maxmin » approach  The valuation function is extended to take one level of feedback ( , ) min( ( , ( ), ) * i i i f t t f t BR j t   i i V i j j  i    Hence producing the concept of « Forward looking Nash Equilibrium » RAIRO - Operations Research Vol. 43 No. 4 (October- December 2009)  Nash Equilibrium (NE) , , ( , ) ( , ) * * i i i i i i x T f x t f t t     Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 12/42 Part II : GTES (Game Theoretical Evolutionary Simulation)
  13. 13. Sampling  Monte-Carlo  The uncertainty regarding the environment parameters e is handled through randomization  Each parameter from E is drawn between an min/max  Example: a  [1.0, 3.0]  Scenarios  Are used to implement « what-if » analysis (though e)  Boundaries for Monte-Carlo sampling  Experiences  Sample Size x Scenario x Strategies  For each sample, we search for a NE through a fixed number of iterations  Result is a triplet  Classification (% of stable, war, chaos)  Typical values of key “business” status variable (mean + confidence intervals)  Stability metric (rate of convergence, 9000 8000 7000 6000 5000 4000 3000 2000 1000 0 1 18 35 52 69 86 103 120 137 154 171 188 205 222 239 256 273 290 7000 6000 5000 4000 3000 2000 1000 0 1 35 69 103 137 171 205 239 273 307 341 375 409 443 477 511 545 579 7000 6000 5000 4000 3000 2000 1000 standard deviation ratios) 0 1 37 73 109 145 181 217 253 289 325 361 397 433 469 505 541 577 Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 13/42 Part II : GTES (Game Theoretical Evolutionary Simulation) stable chaos ?
  14. 14. Lessons from Practice  Defining the satisfaction (w.r.t strategy = set of goals) is critical  Additive versus multiplicative formulas  Use multiple strategy objects to avoid the fine tuning of sensitivity  One way to resolve the relative weight issue  Local optimization = Neighborhood + meta-heuristics  Simple local-climbing seems enough …  … but the need for “multiple simultaneous changes” (e.g., 3-opt) is an indication of the game’s interest  Experiences with meta-heuristics (Tabu, genetic, random walks) are interesting but do not change the nature of the result  Sensitivity to initial values for “tactics” is a quality indicator of local search strategy  Meta-principle : increase the “opt power” until stability is reached  Measuring “Nash convergence” is tricky (unless infinite time)  Easy : define a N-uple distance over tactics  Harder : evaluate if a small distance is acceptable  Linear regression on major Business KPIs Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 14/42 Part II : GTES (Game Theoretical Evolutionary Simulation)
  15. 15. Part III  Part 1: Motivations – Making sense in a complex world  Part 2: GTES (Game-Theoretical Evolutionary Simulation)  Part 3: Smart Grid Systemic Simulation Example  Part 4: Mass Market Telephony Simulation Example  Part 5: Conclusion Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 15/42
  16. 16. Energy demand Market Share Demand generation Operator Production Max penetration rate Price sensitivity Dynamic Pricing MW Random noise pattern Time (hourly/daily) « NegaWatt » compensation (yearly) Generates investments Peak « shaving » (hourly) Peak price → partial cutoff Price ($) Wholesale base price Production cost × D Production (GW) D Nuclear capacity City Energy Demand (MW) Extra demand Extra capacity buffer + resell • use buffer if full • Use reserve if (buy) price is high • Fill reserve is (buy) price is low • Sell from reserve if (sell) price is Extra demand buffer reserve Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 16/42 By City % savings Electricity sell price a1 a2 b1 b2 % cutoff price As price rise, Cities invest in energy saving Self-motivated or operator-controlled g1 g2 Operator market share Price ratio (supplier/operator) Supplier (wholesale/ customer) Operator • local → production price + margin1 • supplier → wholesale price + margin2 + customer costs Use local « green » power • Green power is intermittent • The operator controls & monitor all green production from the city Use local storage (buffer / reserve) Use local « fossil » power Wholesale purchase (Supplier) high • Fossil production is variable • Fossil production generates CO2 taxes • Unmet demand is bought wholesale - +/- Reserve « threshold » prices → tactical parameters (policy) S3G : A Collection of Simple Models Difference between constrained /unconstrained demand Part III : Smart Grids Systemic Simulation
  17. 17. S3G : Players’ objectives (optimization functions)  Each players tries to optimize three “KPI” (performance indicators)  Using a linear combination  Measuring the difference between current and target value (defines a strategy)  Regulator  To maintain total output (economy = electricity consumed + negaWatt)  To reduce CO2  To keep a balanced budget (subsidies < taxes)  City  To keep average electricity bill as low as possible  To keep the current level of demand-response shaving  To reduce the « feared worst price » = peak price + 5 x anual growth rate.  Supplier (global)  To keep income at current level  To protect market share (80% when simulation starts)  To keep the number of hours when foreign supply is needed to a minimum  Operator (local)  To grow market share (from 20%)  To grow turn-over  To increase income (sales – expenses) Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 17/42 Part III : Smart Grids Systemic Simulation
  18. 18. S3G : Simulation & Game Protocol S3G Work Plan (1) Implement S3G Model The model has successively been implemented (1000 lines of CLAIRE code) “Rules of Play” (2) What-if analysis and validation S3G has been checked though a number of what-if scenarios (both as a debugging method and a first output) (3) Machine Learning : « Tactic optimization » A crude version of “hill climbing / local search” optimization is operational. More complex methods are required because of pricing structure (4) Search for Nash equilibrium This is how we address the question of competition vs. cooperation. (5) Randomize unknown environment parameters Full-blown GTES simulation includes a Monte-Carlo sampling of unknown systemic parameters to asses the robustness of phase (4) : classification. Randomization is extended to demand generation to study the impact of variability (6) Search for robust strategies and robust equilibriums The search for best tactics is extended to take robustness into account. The analysis of the competition landscape is revised accordingly (7) Scenario Analysis The last phase is to decompose the parametric space into relevant scenarios to address the questions/issues from slide # 1. Environment parameters • Demand growth • Oil Price Trend • Nuclear Growth • CO2 tax Reflects one’s vision of World economy & policies Technology Parameters • Cost of green tech • Cost of storage Reflects one’s confidence in technology progress Systemic Parameters • Demand variability • NegaWatt generation • Peak shaving • Market share sensitivity S3G tool Regulator Supplier Operator City Reflects one’s understanding of energy ecosystem Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 18/42 Part III : Smart Grids Systemic Simulation
  19. 19. Sensitivity to variability  Testing the hypothesis that variability favors local operator variability Demand response shaving Reaction to price increase NegaWatt Investment Local vs centralized fossile production  The results show only non-significant improvements for SG operator (small compared to the overall “local resell business” equilibrium !) Wholesale price Operator EBITDA Market Share negaWatt Fossile investment Fear : projected 5- yr price DR shaving E1: regular 85,13€ 789M€ 22% 7,08TWh 2MW 14,04% E2: more variation 87,50€ 711M€ 22% 7,15TWh 4MW 14,87% E3: local variation 84,39 703M€ 22,56% 6,34 12MW 14,7%  Variability “pushes” the system in the “right direction” (favorable to smart grids), but is a “small scale” change and most of it seems absorbed in the complex loop interactions Very sensitive to environment parameters Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 19/42 Part III : Smart Grids Systemic Simulation
  20. 20. Carbon Tax, Green Power and Storage Cost E1 Reference S3: CO2 tax S3a CO2 tax + solar S4 = S3a +cheap storage S6 cheap Solar (100€/MWh)  Carbon Tax Wholesale price Operator income Market Share Solar Investment Storage Investment NegaWatt DR 84,12€ 862M€ 21,8% 0MW 0MW 7,3TWh 14,3% 35,8Mt 93,7€ 1106M€ 19,8% 0MW 0MW 9,37TWh 14,5% 24.7Mt 92,3€ 1006M€ 18,8% 1670MW 6,8MW 8,63TWh 14,8% 25.4Mt 87.8€ 992M€ 21% 416MW 50MW 7,7TWh 15,25% 29.3Mt 83,57€ 786M€ 21,7% 2461MW 0MW 6,9TWh 14,2% 35.7Mt  Does not help: reinforces the advantage of nuclear energy shaving  Carbon tax to Solar subsidy: positive (PV industry ) but marginal  Green Power : still too expensive …  (all economic parameters drawn from Web search – orders of magnitude)  Storage Cost  Local optimization finds the optimal buffer/reserve ratio & when to buy/sell  Efficiency = average difference (buy/sell) price -> depend on price structure !  Yields a price threshold at [50% to 100%] of wholesale price   Resilience (e.g., Japan) or co-usage (electric car) is not factored in Total CO2 Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 20/42 Part III : Smart Grids Systemic Simulation
  21. 21. Smart Grids : Strategy Matrix S3G is a stable model (no war/ chaos) but ESS convergence is approximate 160 140 120 100 80 60 40 20 0 1 3 5 7 9 11 13 15 17 19 EBITDA (M€) price (€/MWh) Nash distance (%) 160 140 120 100 80 60 40 20 0 1 3 5 7 9 11 13 15 17 19 The strategies of suppliers and operators may be aligned or conflicting Operator: Soft Strategy Supplier : Soft strategy Supplier: 11635 M€ @ 79.7€ Operator: 1282 M€ : 19.9% MS Supplier: Hard strategy Supplier: 7119 M€ @ 70.3€ Operator: 1253 M€ : 20.58% MS The effect is regulation is very important  EBITDA (M€) price (€/MWh) Nash distance (%) Operator: Hard Strategy Focus on Marketshare Supplier: 11775 M€ @ 80.7€ Operator: 667 M€ : 21.6% MS Supplier: 7225 M€ @ 68.7€ Operator: 737 M€ : 19.9% MS • wholesale boundaries • fixed/variable price structure Surprise ? Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 21/42 Part III : Smart Grids Systemic Simulation
  22. 22. Oil Price Sensitivity and « De-nuclearization » Wholesale price Operator income Market Share Solar Investment Storage Investment NegaWatt DR E1 Reference 84,12€ 862M€ 21,8% 0MW 0MW 7,3TWh 14,3% 35,8Mt E4: oil price increase 91,2€ 647M€ 21,4% 0MW 5MW 8,3TWh 15,5% 28Mt S2: Government « de-nuclearizes » 88€ 785M€ 21,7% 0M 4MW 8TWh 14,6% 45Mt H1: 3 cities (vs 10) 82.05€ 866M€ 21.7% 0MW 0MW 6.9TWh 12.67% 36,1Mt H2: 20 cities 83.6€ 568M€ 23.6% 0MW 0MW 7TWh 13,56% 37Mt  Oil Price increase does not favor smart grids operators  De-nuclearization is a more favorable scenario … Shaving  When storage cost is lowered (1% market share gain at 100$/W)  Studying scale-sensitivity would require more time/computers/faster machines   One game with approximate results (10 samples) : 1 day of CPU  Based on other GTES application, typical sample size should be 1000  Total CO2 Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 22/42 Part III : Smart Grids Systemic Simulation
  23. 23. S3G Temporary Conclusions  Systemic Simulation of Smart Grids  A very simple model …  … yet which captures a number of interaction loops between players  Demonstrates an interesting level of complexity …  … shown by the “relative difficulty” to get stable Nash Equilibrium  Serious Gaming as a learning tool (not a forecasting tool !)  Takes the various stakeholders viewpoint into account   Build systemic knowledge (understand the environment as a system with feedback loops and delays)  GTES is an interesting approach for serious gaming  A Few Lessons Learned  Importance of regulation  Competition regulation (dynamic & wholesale pricing)  Technology / CO2 incentives  Smart behavior starts with storage  When it becomes affordable at local scale (vs. STEP) Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 23/42 Part III : Smart Grids Systemic Simulation
  24. 24. Part IV  Part 1: Motivations – Making sense in a complex world  Part 2: GTES (Game-Theoretical Evolutionary Simulation)  Part 3: Smart Grid Systemic Simulation Example  Part 4: Mass Market Telephony Simulation Examples  Conclusions Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 24/42
  25. 25. Example (1): Distribution Networks  Simple model  Two-steps phases which are distinct from an organization viewpoint  Année 1  Coupling with other operators through distributors  Serious Games at the excom level … successful impact R5: agences Parc SRO Année 1 100 clients(forfaits) « statistiques » R2: WEBT Parc BT Année 1 SOCC Ne fait rien R4: DCT Churne (interne ou externe) R1 R2 R3 R6 Parc SRO Année 2 R3: GSAT renouvelle Parc BT Année 2 R1 : RCBT R6: WEB @SRO Année 2 Calcul PdM BT/ SRO Calcul Renouvellement Par réseau Agrégation Résultat : - Bilan par réseau - Parc total par opérateur Calcul Ventes Par réseau Base 100.0 « ajustée » Répartition Calcul PdM BT/ SRO Courbe « en S » d’appétence Courbe « en S » de renouvellement Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 25/42 Part IV : Telephony Simulation Examples
  26. 26. Example (2) : Commercial Costs Optimization  Problem: resource allocation between different channels …  … while taking competition between distribution channels into account (hard to evaluate)  Goal (met) : start discussion between channels   method: what-if scenarios  First round : calibration  Second round : global simulation Adjustment Optimization delta OPEX 2006 data sales, fixed/ variable costs delta CAPEX result Euros C2 @ p2 Dist : 100% C3 @ p3 Dist : 100% Competition Matrix Par canal C1 @ p1 Dist : 60% C2 @ p2 Dist : 100% Sensibility Price -> Sales Flux si p1 < p2 C[1,2] = 50% C[2,1] = 50% Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 26/42 Part IV : Telephony Simulation Examples
  27. 27. Example (3) : CGS – Cellular Game Simulation  Simulation of competition between n mass-market telephony operators  Mobile telephony, internet access provider, most operator business  Follow-up of UMTS 2000 simulation  While taking distribution channels into account  Make use of only published data (Annual reports) … which required successive simplification iterations  Time unit = year (3/5 year => 3/5 iterations)  Model that has been played with 3 and 4 players …  First to evaluate 3 years plans (3YP) robustness  Then, to simulate the arrival of Free on the French market (2009-2010)  Independently, we run a “war game” organized by McKinsey, with a more sophisticated model but no automated reaction Serious game … are serious : cannot comment !  Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 27/42 Part IV : Telephony Simulation Examples
  28. 28. CGS – Operator Model from CEO’s insight  Input  ACQ: average acquisition cost per customer  FID: loyalty cost  “average service basket” price (PPM: typical average package price)  Internal Variables  Customer Base  # acquisitions, # renewals  Consumption (faction of expected average)  ARPU (summation of PPM x usage)  Simple Financial Model  Operations expenses (including annual trend)  Inbound / Outbound Turnover, interco (TA)  Ebitda = CA – DO – FID – ACQ - Interco PU ACQ PU FID 2 Opérateur prix 3 MVNO inclus renouvellement 1 acquisition churn Aggregation: •MVNO •Voice / data • MM / Business • Pre-/post-paid Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 28/42 Part IV : Telephony Simulation Examples
  29. 29. Coupling 1 2 3 1 Orange 1 2 3 Opérateur  Acquisition/ price relationship → 2 3 SFR 1 2 3 Bouygues Nouveaux  S-curve to model sensitivity  + competition model Cf. INRIA  Change / price relationship [churn / renewal] → similar model with 2010 Talk different parameters Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 29/42 Part IV : Telephony Simulation Examples
  30. 30. CGS – Simulation Architecture  5-steps computational model Nouveaux clients Résultats Volumes 5 4 Base op 2 Calcul Churn, Renouvellement, migration Canaux PdM Par canal Calcul Ventes Par canal 3YP – tactique : fid, acq, pricing Base op 1 f,f' : Courbes en S + compétition (opérateur y 1 Renouvellement (global – non ventilé par canal) 2 3  Each iteration produces yearly results (model’s variables) for each operator Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 30/42 Part IV : Telephony Simulation Examples
  31. 31. Open Mass-Market Naive Competition (Example 4) Each actor is a company with  an ARPU (price)  an attractiveness (premium)  a customer base  fixed costs + variables costs  migration fluidity  a structural churn Crude estimates ! Market share follows a (price + premium)a distribution (new) price (old) price migration migration … Churn follows a constant × priceb distribution CAVEAT  - closed market - retail channels ignored - no segmentation ⇒ 50 lines of code, easy to reproduce migration (new) price (old) price Customer base 1 Unknown: - a - b Customer base N Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 31/42 Part IV : Telephony Simulation Examples
  32. 32. Mobile Operators (I) : Playing What-If Scenarios 4500 4000 3500 3000 2500 2000 1500 1000 500 0 100% 80% 60% 40% 20% 0% B O S F M 100% 80% 60% 40% 20% 100% 80% 60% 40% 20% 4500 4000 3500 3000 2500 2000 1500 1000 500 0 100% 80% 60% 40% 20% Stable WAR Chaos Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 32/42 Part IV : Telephony Simulation Examples -500 2011 2012 2013 2014 2015 B O S F M 3 mobile operators 4 mobile operators 4 mobile operators, loose strategies 4 mobile operators, tight strategies Stable WAR Chaos dev sat% result 5000 4500 4000 3500 3000 2500 2000 1500 1000 500 4500 4000 3500 3000 2500 2000 1500 1000 500 0 -500 2011 2012 2013 2014 2015 B O S F M Stable WAR Chaos 0% B O S F M dev sat% result -500 2011 2012 2013 2014 2015 B O S F M Stable WAR Chaos 0% B O S F M dev sat% result 0 2011 2012 2013 2014 2015 B O S M 0% B O S M dev sat% result
  33. 33. Mobile Operators (II) : Strategy Analysis Sensitivity to alpha (aggressive) Sensitivity to alpha (conservative) 100% 80% 60% 40% 20% 100% 80% 60% 40% 20% 4500 4000 3500 3000 2500 2000 1500 1000 500 0 If the fourth operator builds a network ? If the third operator flattens its costs ? 100% 80% 60% 40% 20% 4500 4000 3500 3000 2500 2000 1500 1000 500 0 100% 80% 60% 40% 20% Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 33/42 4500 4000 3500 3000 2500 2000 1500 1000 500 4500 4000 3500 3000 2500 2000 1500 1000 500 0 -500 2011 2012 2013 2014 2015 B O S F M Stable WAR Chaos 0% B O S F M dev sat% result -500 2011 2012 2013 2014 2015 B O S F M Stable WAR Chaos 0% B O S F M dev sat% result 0 -500 2011 2012 2013 2014 2015 B O S F M Stable WAR Chaos 0% B O S F M dev sat% result Stable WAR Chaos 0% B O S F M dev sat% result -500 2011 2012 2013 2014 2015 B O S F M Part IV : Telephony Simulation Examples
  34. 34. Part V  Part 1: Motivations – Making sense in a complex world  Part 2: GTES (Game-Theoretical Evolutionary Simulation)  Part 3: Smart Grid Systemic Simulation Example  Part 4: Mass Market Telephony Simulation Example  Future Directions & Conclusions Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 34/42
  35. 35. Performance Issues  Models that have been tested with GTES are computationally simple, still running once simulation ranges from tens of milliseconds (most of them) to one second.  Tactics’ optimization requires from 100 to 1000 simulation cycles  The search for Nash equilibrium requires many hundreds of optimization cycles (interlacing between the two loops helps by a factor of 10)  Monte-Carlo Sampling requires a few hundreds to a few thousands of Nash equilibriums searches Consequently,  Computation time quickly becomes a problem (from a day to a year)  Parallel computation is straightforward Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 35/42 Part V : Future Directions & Conclusion
  36. 36. Quality of Model Issues  Modeling « Torture Bench »   Machine learning => zooms on logic faults  No mercy for linear approximation that are “locally right”  « Model tuning » takes time !  A good practice is to limit oneself to status variable for which an value history is available  « stability » requirement  Limit / boundary behaviors  Ex: S-curve versus linear formulas  Concavity/convexity –  Neighborhood structure and exploration strategy  Tradeoff between efficiency and relevance (should mimic actors)  Scale –sensitive : need to work on the real size problem (even if abstracted)  Monte-Carlo sampling must be introduced early on to avoid over-engineering Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 36/42 Part V : Future Directions & Conclusion
  37. 37. Relations with System Dynamics  System Dynamics  Models based on interaction networks between state variables (e.g., CGS example)  Proximity  Very close to the work of J. Forrester or J. Sterman  These networks are a good first step towards GTES modesl  Differences PU ACQ PU FID prix 3 2 1 Prix PU FID Prix TA Usage Acquisition Interco EBITDA PU ACQ FID Base Renouvellement TCO Offre CA Entrant ARPU CA Sortant ACQ  What’s inside the model (detail = interaction formulas) is critical and has a deep impact on results …  … especially when the system is coupled with a stochastic input flow  Polarities (+/- ) between state variables are not enough, not even the values of local derivatives (elasticity => linear model). Dépenses Churn Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 37/42 Part V : Future Directions & Conclusion
  38. 38. Serious Games : Key Take-aways What-if Scenarios, Players’ strategies Results Game Analysis Models’ torture bench  The model is wrong … or too complex (e.g. Social Networks) The model is wrong … but may be fixed (e.g. CGS) The model is right … our thinking was wrong Systemic education  Continuous improvement The more embarrassing, the more useful  eg: Market Share , Technology introduction … The model is right … and shows a feedback loop that we missed Many successful instances over the years  Sales Channel, Customer Lifecycle, LTE bid The expert disagrees and the fun starts  Simple Model + Key Business Variables Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 38/42 Part V : Future Directions & Conclusion
  39. 39. Failed SNS Experiment  I applied GTES to study the “Google+ versus Facebook” fight two years ago.  Value of SNS experience = quality of social content X quality of Edge ranking  The model is interesting, it shows the recursive value equation (strong reinforcement)  Quality of content = f(size of network x time spent on network)  Time spent on SNS = f(Quality of content / effort)  The model was easy to implement (using Duncan Watts principles for social network growth)  Showcases the difference between adoption and usage !  However, the results are immensely sensitive to frequencies (of visit) and delays  Bottom-line : GTES is not universal   Start with a few scenarios, then heavy sampling with Monte-Carlo …  If unstable, you need more real-life measures to narrow the incertitude about systemic parameters Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 39/42 Part V : Future Directions & Conclusion
  40. 40. RTMS : Repeated Tender Market Share  Problem: find a model to reproduce the behavior of buyers / sellers in a closed repeated tender market  For instance, IT division buying software development man days  See if there is a systemic justification for observed practices  First Model  Bid price is a combination (either fixed or randomized) of  Balanced price (economic optimization)  Dynamic price (based on previous bidding history) Two sets of coefficient according the previous  Selection is based on price (cheapest wins) with a bias towards diversity (cf. the two-sourcing rules)  Very similar to repeated Prisoner's dilemma game (Axelrod) but more complex (than TIT-for-TAT)  Preliminary results  Interesting : Nash equilibriums are found … not always (and required a lot of tuning).  “Forward Nash Equilibrium” – interesting but expensive  Meaningful simulations : collusions, bluffing, … Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 40/42 Part V : Future Directions & Conclusion
  41. 41. Future : Connected Health Trust Game  The problem : data privacy issues with connected devices  Would you share your health data with your insurance company to get a better price ?  Would you react socially - as a group – to selective price increases ?  Insurance issues :  Anti-selection (if another insurer gets the “lower risk” group)  Asymmetry of information  The model:  Segments of population with different health risk behaviors, which are revealed (or not) by connected devices  Group of insurance companies with different policies (fixed or variable prices according to behavior)  Macro parameters  Precision of determination - link between behavior & risk  Stability of determination – evolution in time  Importance of social behavior  No results yet … stay tuned (open for collaboration) Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 41/42 Part V : Future Directions & Conclusion
  42. 42. Conclusion « The difficulty lies, not in the new ideas, but in escaping from the old ones.” J. M. Keynes  “Serious Games” approach will become mainstream in the future  Forecasting does not work any longer  Need to develop reflexes and skills  Build systemic knowledge (understand the environment as a system with feedback loops and delays)  GTES is an interesting platform for serious gaming  Combination of “classical techniques”  Evolutionary game theory … will become popular with faster computers  A workbench for model tuning – CAVEAT – not all models adapt to GTES  Not a panacea, but proven utility over the last 10 years  Still, a lot of work is required  Parallelization (good MapReduce candidate)  Making “Forward Nash” (look-ahead) practical  Leveraging evolutionary meta-heuristics (e.g. genetic algorithms) Yves Caseau – 2014 – Serious Games as a Tool to Understand Complexity in Market Competition 42/42 Part V : Future Directions & Conclusion

×