Abstract
We study the impact of transition scenario uncertainty, namely that of future carbon price and electricity demand, on the pace of decarbonization of the electricity industry. To this end, we develop a theory of optimal stopping mean-field games with non-Markovian common noise and partial observation. For mathematical tractability, the theory is formulated in discrete time and with common noise restricted to a finite probability space. We prove the existence of Nash equilibria for this game using the linear programming approach. We then apply the general theory to build a discrete time model for the long-term dynamics of the electricity market subject to common random shocks affecting the carbon price and the electricity demand. We consider two classes of agents: conventional producers and renewable producers. The former choose an optimal moment to exit the market and the latter choose an optimal moment to enter the market by investing into renewable generation. The agents interact through the market price determined by a merit order mechanism with an exogenous stochastic demand. We illustrate our model by an example inspired by the UK electricity market, and show that scenario uncertainty leads to significant changes in the speed of replacement of conventional generators by renewable production.
Similar content being viewed by others
Data availibility
The electricity demand projections used in this paper are available from https://www.gov.uk/government/publications/updated-energy-and-emissions-projections-2019, Annex F. The NGFS scenarios used for illustrative purposes (Fig. 1) are available from https://data.ene.iiasa.ac.at/ngfs/.
Notes
Available at https://data.ene.iiasa.ac.at/ngfs/
Meaning that for all \(t\in I\), \({\mathcal {F}}_t\) is conditionally independent of \({\mathcal {F}}^U_T\) given \({\mathcal {F}}^U_t\).
The installed productive capacity is assumed to be the same for all the conventional producers and fixed over time.
This price refers to peak/off-peak electricity futures price with delivery period \([t,t+\Delta _t]\), observed at t.
As for the conventional producers, the installed productive capacity of the renewable producers is assumed to be the same and fixed over time.
Recall that we are not dealing with day-ahead prices but with futures prices with delivery period \(\Delta _t\). Our merit order mechanism consists simply in matching of demand and supply in the futures market. Electricity futures contracts are traded separately for peak and off-peak periods.
see http://plato.asu.edu/ftp/lpsimp.html for a comparison of different LP solvers.
References
Alasseur, C., Tahar, I.B., Matoussi, A.: An extended mean field game for storage in smart grids. J. Optim. Theory Appl. 184(2), 644–670 (2020)
Alasseur, C., Campi, L., Dumitrescu, R., et al.: MFG model with a long-lived penalty at random jump times: application to demand side management for electricity contracts. arXiv:2101.06031, (2021)
Aliprantis, C., Border, K.: Infinite dimensional analysis: A hitchhiker’s guide. Springer, Berlin Heidelberg (2007)
Aïd, R., Dumitrescu, R., Tankov, P.: The entry and exit game in the electricity markets: A mean-field game approach. Journal of Dynamics and Games 8, 331–358 (2021)
Bagagiolo, F., Bauso, D.: Mean-field games and dynamic demand management in power grids. Dynamic Games and Applications 4(2), 155–176 (2014)
Bauer, N., Baumstark, L., Leimbach, M.: The REMIND-R model: the role of renewables in the low-carbon transformation-first-best vs. second-best worlds. Clim. Change 114(1), 145–168 (2012)
Baxter, J., Chacon, R.: Compactness of stopping times. Probab. Theory Relat. Fields 40(3), 169–181 (1977)
Belak, C., Hoffmann, D., Seifried, F.T.: Continuous-time mean field games with finite state space and common noise. Applied Mathematics & Optimization 84(3), 3173–3216 (2021)
Bertucci, C.: Optimal stopping in mean field games, an obstacle problem approach. Journal de Mathématiques Pures et Appliquées 120, 165–194 (2018)
Bogachev, V.: Measure theory. Springer Science & Business Media, Berlin (2007)
Bolton, P., Despres, M., Da Silva, L. A. P., et al.: The green swan. BIS Books (2020)
Bouveret, G., Dumitrescu, R., Tankov, P.: Mean-field games of optimal stopping: A relaxed solution approach. SIAM J. Control. Optim. 58(4), 1795–1821 (2020)
Buckdahn, R., Goreac, D., Quincampoix, M.: Stochastic optimal control and linear programming approach. Applied Mathematics & Optimization 63(2), 257–276 (2011)
Carmona, R., Delarue, F.: Probabilistic Theory of Mean Field Games with Applications II. Springer, Switzerland (2018)
Carmona, R., Delarue, F., Lacker, D.: Mean field games with common noise. Ann. Probab. 44(6), 3740–3803 (2016)
Carmona, R., Delarue, F., Lacker, D.: Mean field games of timing and models for bank runs. Appl. Math. Optim. 76(1), 217–260 (2017)
Carmona, R., Dayanıklı, G., Laurière, M.: Mean field models to regulate carbon emissions in electricity production. Dynamic Games and Applications 12(3), 897–928 (2022)
Chenet, H., Ryan-Collins, J., Lerven, F.V.: Climate-related financial policy in a world of radical uncertainty: Towards a precautionary approach. UCL Institute for Innovation and Public Purpose WP 13, (2019)
Cho, M.J., Stockbridge, R.H.: Linear programming formulation for optimal stopping problems. SIAM J. Control. Optim. 40(6), 1965–1982 (2002)
Couillet, R., Perlaza, S.M., Tembine, H., et al.: A mean field game analysis of electric vehicles in the smart grid. In: 2012 Proceedings IEEE INFOCOM Workshops, IEEE, pp 79–84, (2012)
Cox, P.M., Huntingford, C., Williamson, M.S.: Emergent constraint on equilibrium climate sensitivity from global temperature variability. Nature 553(7688), 319–322 (2018)
DeCarolis, J., Daly, H., Dodds, P., et al.: Formalizing best practice for energy system optimization modelling. Appl. Energy 194, 184–198 (2017)
Dianetti, J., Ferrari, G., Fischer, M., et al.: Submodular mean field games: Existence and approximation of solutions. Ann. Appl. Probab. 31(6), 2538–2566 (2021)
Dumitrescu, R., Leutscher, M., Tankov, P.: Control and optimal stopping mean field games: a linear programming approach. Electron. J. Probab. 26, 1–49 (2021)
Dumitrescu, R., Leutscher, M., Tankov, P.: Linear programming fictitious play algorithm for mean field games with optimal stopping and absorption. ESAIM: Mathematical Modelling and Numerical Analysis 57(2), 953–990 (2023)
Edenhofer, O., Lessmann, K., Kemfert, C., et al.: Induced technological change: Exploring its implications for the economics of atmospheric stabilization: Synthesis report from the innovation modeling comparison project. The Energy Journal 27, 57–107 (2006)
Edgar, G.A., Millet, A., Sucheston, L.: On compactness and optimality of stopping times. In: Martingale Theory in Harmonic Analysis and Banach Spaces, pp. 36–61. Springer, Berlin, Heidelberg (1982)
Elie, R., Hubert, E., Mastrolia, T., et al.: Mean-field moral hazard for optimal energy demand response management. Math. Financ. 31(1), 399–473 (2021)
Féron, O., Tankov, P., Tinsi, L.: Price formation and optimal trading in intraday electricity markets. Math. Financ. Econ. 16(2), 205–237 (2022)
Fleming, W., Vermes, D.: Generalized solutions in the optimal control of diffusions. In: Stochastic Differential Systems, pp. 119–127. Stochastic Control Theory and Applications. Springer, New York (1988)
Gomes, D., Patrizi, S.: Obstacle mean-field game problem. Interfaces and Free Boundaries 17(1), 55–68 (2015)
Gomes, D., Saúde, J.: A mean-field game approach to price formation. Dynamic Games and Applications pp 1–25, (2020)
Guéant, O., Lasry, J.M., Lions, P.L.: Mean field games and applications. In: Paris-Princeton lectures on mathematical finance 2010, pp. 205–266. Springer, Berlin Heidelberg (2011)
Guo, X., Hu, A., Zhang, J.: MF-OMO: An optimization formulation of mean-field games. arXiv preprint arXiv:2206.09608, (2022)
He, X., Tan, X., Zou, J.: A mean-field version of Bank-El Karoui’s representation of stochastic processes. arXiv preprint arXiv:2302.03300, (2023)
Hourcade, J., Sassi, O., Crassous, R., et al.: IMACLIM-R: a modelling framework to simulate sustainable development pathways. International Journal of Global Environmental Issues 10(1/2), 5–24 (2010)
Huang, M., Malhamé, R., Caines, P.: Large population stochastic dynamic games: closed-loop McKean-Vlasov systems and the Nash certainty equivalence principle. Commun. Inf. Syst. 6(3), 221–252 (2006)
Keen, S., Lenton, T.M., Garrett, T.J., et al.: Estimates of economic and environmental damages from tipping points cannot be reconciled with the scientific literature. Proc. Natl. Acad. Sci. 119(21), e2117308119 (2022)
Kingman, J.F.C.: Uses of exchangeability. Ann. Probab. 6, 183–197 (1978)
Kizilkale, A.C., Salhab, R., Malhamé, R.P.: An integral control formulation of mean field game based large scale coordination of loads in smart grids. Automatica 100, 312–322 (2019)
Kurtz, T., Stockbridge, R.H.: Existence of Markov controls and characterization of optimal Markov controls. SIAM J. Control. Optim. 36(2), 609–653 (1998)
Kurtz, T.G.: Martingale problems for conditional distributions of Markov processes. Electron. J. Probab. 3, 1–29 (1998)
Kushner, H., Dupuis, P.G.: Numerical methods for stochastic control problems in continuous time. Springer, New York (2001)
Lasry, J.M., Lions, P.L.: Jeux à champ moyen. I. Le cas stationnaire. Comptes Rendus Mathématique 343(9), 619–625 (2006)
Lasry, J.M., Lions, P.L.: Jeux à champ moyen. II. Horizon fini et contrôle optimal. Comptes Rendus Mathématique 343(10), 679–684 (2006)
Lasry, J.M., Lions, P.L.: Mean field games. Japan. J. Math. 2(1), 229–260 (2007)
Lions, P.L.: Théorie des jeux à champs moyen et applications. Cours au Collége de France, (2007)
Loulou, R., Labriet, M.: ETSAP-TIAM: the TIMES integrated assessment model part I: Model structure. CMS 5(1), 7–40 (2008)
Meehl, G.A., Senior, C.A., Eyring, V., et al.: Context for interpreting equilibrium climate sensitivity and transient climate response from the CMIP6 Earth system models. Science Advances, 6(26), (2020)
Monasterolo, I., Roventini, A., Foxon, T.J.: Uncertainty of climate policies and implications for economics and finance: An evolutionary economics approach. Ecol. Econ. 163, 177–182 (2019)
Nordhaus, W.: Projections and uncertainties about climate change in an era of minimal climate policies. Am. Econ. J. Econ. Pol. 10(3), 333–60 (2018)
Nutz, M.: A mean field game of optimal stopping. SIAM J. Control. Optim. 56(2), 1206–1221 (2018)
Olshen, R.: A note on exchangeable sequences. Z Wahrscheinlichkeitstheorie verw Gebiete 28, 317–321 (1974)
Parthasarathy, K.R.: Probability measures on metric spaces, vol 352. American Mathematical Soc, (2005)
Peskir, G., Shiryaev, A.: Optimal Stopping and Free-Boundary Problems. Birkhäuser, Basel (2006)
Pham, H., Runggaldier, W.J., Sellami, A.: Approximation by quantization of the filter process and applications to optimal stopping problems under partial observation. In: Monte Carlo Methods Appl, (2005)
Possamaï, D., Talbi, M.: Mean-field games of optimal stopping: master equation and weak equilibria. arXiv preprint arXiv:2307.09278, (2023)
Rieder, U.: Bayesian dynamic programming. Adv. Appl. Probab. 7(2), 330–348 (1975)
Rudin, W.: Functional Analysis, 2nd edn. McGraw-Hill, New York (1991)
Santibáñez, N.H., Jofré, A., Possamaï, D.: Pollution regulation for electricity generators in a transmission network. SIAM Journal on Control and Optimization 61, (2023)
Shokri, M., Kebriaei, H.: Mean field optimal energy management of plug-in hybrid electric vehicles. IEEE Trans. Veh. Technol. 68(1), 113–120 (2018)
Shrivats, A.V., Firoozi, D., Jaimungal, S.: A mean-field game approach to equilibrium pricing in solar renewable energy certificate markets. Math. Finan. 32(3), 779–824 (2022)
Tchuendom, R.F., Malhamé, R., Caines, P.: A quantilized mean field game approach to energy pricing with application to fleets of plug-in electric vehicles. In: 2019 IEEE 58th conference on decision and control (CDC), IEEE, pp 299–304, (2019)
Weidlich, A., Veit, D.: A critical survey of agent-based wholesale electricity market models. Energy Economics 30(4), 1728–1759 (2008)
Weitzman, M.L.: On modeling and interpreting the economics of catastrophic climate change. Rev. Econ. Stat. 91(1), 1–19 (2009)
Acknowledgements
We thank the anonymous reviewer and the guest editor for the insightful comments on an earlier version of this paper, which helped us to improve the presentation of the results.
Funding
Peter Tankov gratefully acknowledges financial support from the ANR (project EcoREES ANR-19-CE05- 0042) from ADEME (Agency for Ecological Transition) in the context of SECRAET project, and from the FIME Research Initiative of the Europlace Institute of Finance.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Conflict of interest
The authors have no competing interests to declare that are relevant to the content of this article.
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendices
Complementary results on the linear programming approach for MFGs of optimal stopping
In this appendix, we provide a theoretical analysis of the linear programming approach in discrete time in an abstract and general case. We show by a duality argument that the relaxed set is equal to the closed convex hull of the set induced by Markovian stopping times. In particular, this result allows us to derive a probabilistic representation of the relaxed set in terms of randomized stopping times (in the sense of [16]).
Markovian and randomized stopping times in the canonical space
We consider a discrete time setting with horizon \(T\in \mathbb N^*\) and we let \(I=\{0, \ldots , T\}\) be the set of time indices and \(I^*=\{1, \ldots , T\}\). We are given a nonempty compact metric space (E, d) and consider two canonical spaces: \(\Omega _X=E^I\) and \({\overline{\Omega }} = E^I\times I\), where \(E^I\) is the set of functions from I to E endowed with the pointwise convergence topology (which coincides with the uniform convergence topology here). Denote by \({\mathcal {B}}(\Omega _X)\) and \({\mathcal {B}}(\overline{\Omega })\) the respective Borel \(\sigma \)-algebras. Let X be the identity map on \(\Omega _X\) and let \({\overline{X}}\) and \(\overline{\tau }\) be the projections on \({\overline{\Omega }}\), i.e. \(\overline{X}({\overline{\omega }})=x\) and \({\overline{\tau }}({\overline{\omega }})=\theta \) for each \({\overline{\omega }}=(x, \theta )\in {\overline{\Omega }}\). On \((\Omega _X, {\mathcal {B}}(\Omega _X))\) we define the filtration \({\mathcal {F}}^X_t:=\sigma (X_s:s\le t)\), for \(t\in I\), and on \(({\overline{\Omega }}, {\mathcal {B}}({\overline{\Omega }}))\) we define the filtrations \({\mathcal {F}}^{{\overline{X}}}_t:=\sigma ({\overline{X}}_s:s\le t)\) and \({\mathcal {F}}^{{\overline{\tau }}}_t:=\sigma (\{{\overline{\tau }}=0\}, \ldots , \{{\overline{\tau }}=t\})\), for \(t\in I\). Note that \({\mathbb {F}}^{{\overline{\tau }}}\) is the smallest filtration for which \({\overline{\tau }}\) is a stopping time. Define \(\overline{{\mathcal {F}}}_t:={\mathcal {F}}^{{\overline{X}}}_t\vee {\mathcal {F}}^{{\overline{\tau }}}_t\), for \(t\in I\). For any probability \(\overline{\mathbb {P}}\) on \(({\overline{\Omega }}, {\mathcal {B}}({\overline{\Omega }}))\), we denote by \(\mathbb F^{{\overline{X}}, \overline{\mathbb {P}}}=({\mathcal {F}}_t^{{\overline{X}}, \overline{\mathbb {P}}})_{t\in I}\) and \(\overline{\mathbb F}^{\overline{\mathbb {P}}}=(\overline{{\mathcal {F}}}^{\overline{\mathbb {P}}}_t)_{t\in I}\) the filtrations
We are given transition kernels \((\pi _t)_{t\in I^*}\) on E and an initial law \(m_0^*\) on E. We assume that the transition kernels are continuous as maps from E to \({\mathcal {P}}(E)\). Let \(\nu \in {\mathcal {P}}(E^I)\) be the unique law of the Markov chain with transition kernels \((\pi _t)_{t\in I^*}\) and initial law \(m_0^*\). We denote by \(\mathbb F^{X, \nu }=({\mathcal {F}}_t^{X, \nu })_{t\in I}\) the filtration
Definition 12
Denote by \({\mathcal {T}}\) the set of measurable functions \(\tau :\Omega _X\rightarrow I\) such that \(\tau \) is an \(\mathbb F^{X, \nu }\)-stopping time.
Definition 13
Let \({\mathcal {A}}_0\) be the set of probabilities \(\overline{\mathbb {P}}\) on \(({\overline{\Omega }}, {\mathcal {B}}({\overline{\Omega }}))\) such that
-
1.
Under \(\overline{\mathbb {P}}\), \({\overline{X}}\) is a Markov chain with initial distribution \(m_0^*\) and transition kernels \((\pi _t)_{t\in I^*}\).
-
2.
\({\overline{\tau }}\) is an \(\mathbb F^{{\overline{X}}, \overline{\mathbb {P}}}\)-stopping time.
Lemma 10
(Characterization of \({\mathcal {A}}_0\) by Markovian stopping times) It holds that
Proof
Let \(\overline{\mathbb {P}}\in {\mathcal {A}}_0\). By the first condition, the first marginal of \(\overline{\mathbb {P}}\) coincides with \(\nu \). Since \({\overline{\tau }}\) is \({\mathcal {F}}_T^{{\overline{X}}, \overline{\mathbb {P}}}\)-measurable, there exists some measurable function \(\tau :\Omega _X\rightarrow I\) such that \(\overline{\mathbb {P}}({\overline{\tau }}= \tau ({\overline{X}}))=1\). In particular we can disintegrate \(\overline{\mathbb {P}}\) as \(\overline{\mathbb {P}}(dx, d\theta )=\nu (dx)\delta _{\tau (x)}(d\theta )\). Using that \(\overline{\tau }\) is an \(\mathbb F^{{\overline{X}}, \overline{\mathbb {P}}}\)-stopping time and that \({\overline{\tau }}= \tau ({\overline{X}})\) \(\overline{\mathbb {P}}\)-a.s., we get that \(\tau ({\overline{X}})\) is an \(\mathbb F^{{\overline{X}}, \overline{\mathbb {P}}}\)-stopping time. Moreover, for any \(t\in I\) and \(B\in {\mathcal {F}}_t^{\overline{X}, \overline{\mathbb {P}}}\), it is easy to check that \((X, \tau (X))^{-1}(B)\in {\mathcal {F}}_t^{X, \nu }\). In particular, taking \(B=\{\tau (\overline{X})\le t\}\), we get \(\{\tau (X)\le t\}\in {\mathcal {F}}_t^{X, \nu }\), implying that \(\tau \in {\mathcal {T}}\).
Now let \(\tau \in {\mathcal {T}}\) and define \(\overline{\mathbb {P}}(dx, d\theta ):=\nu (dx)\delta _{\tau (x)}(d\theta )\). Since \(\overline{\mathbb {P}}\circ \overline{X}^{-1}=\nu \), the first condition is verified. Define the set \(C:=\{\overline{\tau }=\tau ({\overline{X}})\}\in \overline{{\mathcal {F}}}_T\) which has probability 1 under \(\overline{\mathbb {P}}\). We have
where \(C_1\in {\mathcal {F}}_t^X\) and \(C_2\in {\mathcal {N}}_\nu ({\mathcal {F}}_T^X)\) are such that \(\{x\in \Omega _X:\tau (x)\le t\}=C_1\cup C_2\). We can deduce that \(C_1\times I\in {\mathcal {F}}_t^{\overline{X}}\) and \(C_2\times I\in {\mathcal {N}}_{\overline{\mathbb {P}}}(\overline{{\mathcal {F}}}_T)\), so that \(\{\tau ({\overline{X}})\le t\}\in {\mathcal {F}}_t^{\overline{X}, \overline{\mathbb {P}}}\). Finally,
\(\square \)
Let \({\mathcal {L}}\) be the generator associated to \((\pi _t)_{t\in I^*}\). For any \(\varphi \in C(I\times E)\), define the process \(M(\varphi )\) by \(M_0(\varphi )=\varphi (0, {\overline{X}}_0)\) and
By taking the conditional expectation, one can show the following lemma.
Lemma 11
(Equivalence between the Markov property and the martingale problem) Let \(\overline{\mathbb G}\) be a filtration including \({\mathbb {F}}^{\overline{X}}\) and \(\overline{\mathbb {P}}\) be a probability on \((\overline{\Omega }, {\mathcal {B}}(\overline{\Omega }))\) such that \(\overline{\mathbb {P}}\circ \overline{X}_0^{-1}=m_0^*\). Then, the following are equivalent:
-
1.
Under \(\overline{\mathbb {P}}\), \({\overline{X}}\) is a \(\overline{\mathbb G}\)-Markov chain with transition kernels \((\pi _t)_{t\in I^*}\).
-
2.
Under \(\overline{\mathbb {P}}\), for any \(\varphi \in C(I\times E)\), the process \(M(\varphi )\) is a \(\overline{\mathbb G}\)-martingale.
Definition 14
Let \({\mathcal {A}}_1\) be the set of probabilities \(\overline{\mathbb {P}}\) on \((\overline{\Omega }, {\mathcal {B}}({\overline{\Omega }}))\) such that under \(\overline{\mathbb {P}}\), \({\overline{X}}\) is an \(\overline{\mathbb F}\)-Markov chain with initial distribution \(m_0^*\) and transition kernels \((\pi _t)_{t\in I^*}\). By Lemma 11, \({\mathcal {A}}_1\) is exactly the set of probability measures \(\overline{\mathbb {P}}\) on \(({\overline{\Omega }}, {\mathcal {B}}({\overline{\Omega }}))\) such that \(\overline{\mathbb {P}}\circ \overline{X}_0^{-1}=m_0^*\) and under \(\overline{\mathbb {P}}\), for any \(\varphi \in C(I\times E)\), the process \(M(\varphi )\) is an \(\overline{\mathbb F}\)-martingale.
Lemma 12
We have the inclusion \({\mathcal {A}}_0\subset {\mathcal {A}}_1\).
Proof
Let \(\overline{\mathbb {P}}\in {\mathcal {A}}_0\). Since \({\overline{\tau }}\) is an \({\mathbb {F}}^{{\overline{X}}, \overline{\mathbb {P}}}\)-stopping time, we have that \(\overline{{\mathbb {F}}}^{\overline{\mathbb {P}}}=\mathbb F^{{\overline{X}}, \overline{\mathbb {P}}}\). For any \(\varphi \in C(E)\), we obtain
\(\square \)
Lemma 13
(Immersion property) If \(\overline{\mathbb {P}}\in {\mathcal {A}}_1\), then \(\mathbb F^{{\overline{X}}}\) is immersed in \(\overline{\mathbb F}\) under \(\overline{\mathbb {P}}\).
Proof
We need to show that for all \(t\in I\), under \(\overline{\mathbb {P}}\), \(\overline{{\mathcal {F}}}_t\) is conditionally independent of \({\mathcal {F}}^{{\overline{X}}}_T\) given \({\mathcal {F}}^{{\overline{X}}}_t\). Let \(\varphi _t:{\overline{\Omega }}\rightarrow \mathbb R\) be bounded and \(\overline{{\mathcal {F}}}_t\)-measurable, \(\psi _t:E^I\rightarrow \mathbb R\) be bounded and \({\mathcal {F}}_t^{{\overline{X}}}\)-measurable, \(\psi _{t+}:E^I\rightarrow \mathbb R\) be bounded and \(\sigma ({\overline{X}}_{t+1}, \ldots , {\overline{X}}_T)\)-measurable and \(\phi _{t}:E^I\rightarrow \mathbb R\) be bounded and \({\mathcal {F}}_t^{{\overline{X}}}\)-measurable. We have
This shows that
This is sufficient to prove the claim. \(\square \)
Applying a discrete time version of Proposition 1.10 in Chapter 1 in [14], we get the following corollary.
Corollary 14
We have the equality \({\mathcal {A}}_1=\{\nu (dx)\kappa (x, d\theta ):\kappa \in {\mathcal {K}}\}\), where \({\mathcal {K}}\) is the set of transition kernels from \(\Omega _X\) to I such that for all \(t\in I\) and \(B\in \sigma (\{0\}, \ldots , \{t\})\), the mapping \(x\mapsto \kappa (x, B)\) is \({\mathcal {F}}_t^{X, \nu }\)-measurable.
In particular, the set \({\mathcal {A}}_1\) can be interpreted as the set of randomized stopping times since each kernel \(\kappa \in {\mathcal {K}}\) gives a probability to stop given the observation of the underlying process (in an adapted way through the filtration of the process). The set \({\mathcal {A}}_0\) corresponds to the set of Markovian stopping times in the sense that up to null sets, the stopping rule is determined by the information given by the underlying Markov process and the kernel is given by a Dirac measure (cf Lemma 10).
Proposition 15
The set \({\mathcal {A}}_1\) is compact and convex.
Proof
The relative compactness of \({\mathcal {A}}_1\) follows since \({\mathcal {P}}({\overline{\Omega }})\) is compact. Let us show that \({\mathcal {A}}_1\) is closed. Let \((\overline{\mathbb {P}}_n)_{n\ge 1}\subset {\mathcal {A}}_1\) converging weakly to some \(\overline{\mathbb {P}}\). Using the continuity of the map \((x, \theta )\mapsto x_0\) and passing to the limit, we get \(\overline{\mathbb {P}}\circ {\overline{X}}_0^{-1}=m_0^*\). Let \(\varphi \in C(E)\) and \(\psi \in C({\overline{\Omega }})\) an \(\overline{{\mathcal {F}}}_{t-1}\)-measurable function, then
This shows that \({\mathcal {A}}_1\) is closed and henceforth compact. The convexity follows by the linearity of the conditions with respect to the probability measure. \(\square \)
For any \(\overline{\mathbb {P}}\in {\mathcal {A}}_1\), we define the occupation measures
Definition 15
Define the sets
Proposition 16
The set \({\mathcal {R}}_1\) is compact and convex.
Proof
For the compactness it suffices to show that \({\mathcal {A}}_1\ni \overline{\mathbb {P}}\mapsto (\mu ^{\overline{\mathbb {P}}}, m^{\overline{\mathbb {P}}})\) is continuous, since \({\mathcal {A}}_1\) is compact. Let \((\overline{\mathbb {P}}_n)_{n\ge 1}\subset {\mathcal {A}}_1\) converging weakly to \(\overline{\mathbb {P}}\in {\mathcal {A}}_1\). For any \(\varphi \in C(E)\) and \(t\in I{\setminus }\{T\}\),
where we used the continuity of the function \({\overline{\Omega }}\ni (x, \theta )\mapsto \varphi (x_t)\mathbb {1}_{t<\theta }\). This shows that for each \(t\in I{\setminus }\{T\}\), \((m_t^{\overline{\mathbb {P}}_n})_{n\ge 1}\) converges to \(m_t^{\overline{\mathbb {P}}}\). By the same argument, for each \(t\in I\), \((\mu _t^{\overline{\mathbb {P}}_n})_{n\ge 1}\) converges to \(\mu _t^{\overline{\mathbb {P}}}\). We conclude that \({\mathcal {R}}_1\) is compact. By the linearity of the same mapping \({\mathcal {A}}_1\ni \overline{\mathbb {P}}\mapsto (\mu ^{\overline{\mathbb {P}}}, m^{\overline{\mathbb {P}}})\) and the convexity of \({\mathcal {A}}_1\), we get that \({\mathcal {R}}_1\) is convex. \(\square \)
We define the relaxed set in this framework as follows.
Definition 16
Let \({\mathcal {R}}\) be the set of pairs \((\mu , m)\in {\mathcal {P}}^{sub}(E)^I\times {\mathcal {P}}^{sub}(E)^{I{\setminus }\{T\}}\) such that for all \(\varphi \in C(I\times E)\),
Proposition 17
We have the inclusions \({\mathcal {R}}_0\subset {\mathcal {R}}_1\subset {\mathcal {R}}\). In particular \(\overline{\text {conv}}({\mathcal {R}}_0)\subset {\mathcal {R}}_1\subset {\mathcal {R}}\).
Proof
The first inclusion is a consequence of the inclusion \({\mathcal {A}}_0\subset {\mathcal {A}}_1\) proved in Lemma 12. For the second inclusion, let \(\overline{\mathbb {P}}\in {\mathcal {A}}_1\) and consider the associated measures \((\mu ^{\overline{\mathbb {P}}}, m^{\overline{\mathbb {P}}})\). Using that for any \(\varphi \in C(I\times E)\), \(M(\varphi )\) is an \(\overline{\mathbb F}\)-martingale under \(\overline{\mathbb {P}}\) and \({\overline{\tau }}\) is an \(\overline{\mathbb F}\)-stopping time, we get
which implies by taking the expectation and replacing the expression of \(M_{{\overline{\tau }}}(\varphi )\),
By the definition of the measures we obtain
The last inclusions follow by Proposition 16. \(\square \)
Probabilistic representation
We want to show that the sets \({\mathcal {R}}\) and \({\mathcal {R}}_1\) are equal. We will follow similar arguments to [30] in order to show that the closed convex hull of \({\mathcal {R}}_0\) is equal to \({\mathcal {R}}\), which is sufficient to obtain the desired equality according to Proposition 17.
Equality of the values. We are given two functions \(f\in C(I{\setminus } \{T\}\times E)\) and \(g\in C(I\times E)\). We place ourselves on the filtered probability space \((\Omega _X, {\mathcal {B}}(\Omega _X), \mathbb F^X, \nu )\). For each \((t, x)\in I\times E\), define the probability measure on \(\nu _{t, x}\in (\Omega _X, {\mathcal {B}}(\Omega _X))\) by
Denote by \({\mathcal {T}}_{t, x}\) the set of \(\{t, \ldots , T\}\)-valued \(\mathbb F^{X, \nu _{t, x}}\)-stopping times. Define the value function
The value function verifies the dynamic programming principle (Theorem 1.9 in Chapter 1 in [55]):
By backward induction, since g, f and the transition kernels are continuous, we obtain that \(v\in C(I\times E)\). Define the quantities
By Proposition 17, we have \(V^S\le V^{LP}\). The Snell envelope \(Y=(Y_t)_{t\in I}\) associated to (f, g) is \(Y_t=v(t, X_t)+\sum _{s=0}^{t-1}f(s, X_s)\). We say that \(\tau ^\star \in {\mathcal {T}}\) is an (f, g)-optimal stopping time if
The random variable \(\tau _{\text {min}}:=\inf \{t\in \{0, \ldots , T\}: v(t, X_t) =g(t, X_t)\}\) is an (f, g)-optimal stopping time and \((Y_{t\wedge \tau _{\text {min}}})_{t\in I}\) is an \(\mathbb F^X\)-martingale. In particular,
Now, for all \((\mu , m)\in {\mathcal {R}}\), using v as a test function in the constraint, we get
where the inequality is a consequence of the dynamic programming principle. Taking the supremum over \((\mu , m)\in {\mathcal {R}}\), we deduce that \(V^S(f, g)\ge V^{LP}(f, g)\), which implies that \(V^S(f, g)= V^{LP}(f, g)\).
Theorem 18
The set \({\mathcal {R}}\) is equal to the closed convex hull of \({\mathcal {R}}_0\). As a consequence, \({\mathcal {R}}={\mathcal {R}}_1\), meaning that we can represent any \((\mu , m)\in {\mathcal {R}}\) with some randomized stopping time.
Proof
Assume that there exists some \((\mu ^0, m^0)\in {\mathcal {R}}{\setminus } \overline{\text {conv}}({\mathcal {R}}_0)\). Recall that for any compact metric space K, the topological dual of \({\mathcal {M}}_s(K)\) endowed with the weak topology \(\sigma ({\mathcal {M}}_s(K), C(K))\) is C(K). In particular, by Theorem 3.4 (b) p. 59 in [59], there exists some \((f^0, g^0)\in C(I{\setminus }\{T\}\times E)\times C(I\times E)\) and \(c\in \mathbb R\) such that for all \((\mu , m)\in \overline{\text {conv}}({\mathcal {R}}_0)\),
Taking the supremum over \((\mu , m)\in {\mathcal {R}}_0\), we get \(V^{S}(f^0, g^0)<V^{LP}(f^0, g^0)\), which is a contradiction. \(\square \)
Other technical results
1.1 Exchangeable random variables and De Finetti’s theorem
We recall a version of De Finetti’s theorem adapted to our setting. Let (E, d) and \((F, \rho )\) be two complete and separable metric spaces. Consider a complete probability space \((\Omega , {\mathcal {F}}, \mathbb P)\), a sequence of E-valued random variables \((X_n)_{n\ge 1}\) and an F-valued random variable Y.
Definition 17
We say that \((X_n)_{n\ge 1}\) is i.i.d. given Y if the following two conditions are satisfied.
-
1.
For all \(n\ge 1\) and all \((B_k)_{k\le n}\in {\mathcal {B}}(E)^n\),
$$\begin{aligned} \mathbb P\left( \cap _{k=1}^n\{X_k\in B_k\}|Y\right) =\prod _{k=1}^n\mathbb P(X_k\in B_k|Y)\quad a.s. \end{aligned}$$ -
2.
For all \(n, k\ge 1\) and all \(B\in {\mathcal {B}}(E)\), \(\mathbb P(X_n\in B|Y)=\mathbb P(X_k\in B|Y)\) a.s.
If \((X_n)_{n\ge 1}\) is i.i.d. given Y, then one can show that the sequence \((X_n, Y)_{n\ge 1}\) is exchangeable. In particular, using the same argument as in [39], we obtain that for all measurable functions \(\psi :E\times F\mapsto \mathbb R\) such that \(\mathbb E[|\psi (X_1, Y)|]<\infty \),
where \({\mathcal {G}}\) is the exchangeable \(\sigma \)-algebra. Using Theorem 3 in [53], one can replace \({\mathcal {G}}\) in (17) by \(\sigma (Y)\) in order to derive the following result.
Theorem 19
If \((X_n)_{n\ge 1}\) is i.i.d. given Y, then, for all measurable functions \(\psi :E\times F\mapsto \mathbb R\) such that \({\mathbb {E}}[|\psi (X_1, Y)|]<\infty \), we have
1.2 Dynkin’s formula for randomized stopping times
Proposition 20
Let \(\varphi \in C(I\times E)\) and \(\kappa \) be a randomized stopping time. Then,
Proof
Let \(\overline{\Omega }:=\Omega \times I\), \(\overline{{\mathcal {F}}}:={\mathcal {F}}\otimes 2^I\), \(\overline{{\mathbb {P}}}(d\omega , ds):=\mathbb P(d\omega )\kappa (\omega , ds)\). Consider \({\overline{X}}_t(\omega , s):=X_t(\omega )\) and \(\theta (\omega , s):= s\). Define the filtration \(\mathbb F^\theta :=({\mathcal {F}}^\theta _t)_t\) with \({\mathcal {F}}_t^\theta :=\sigma (\{\theta =0\}, \ldots , \{\theta =t\})\). Now, construct the filtration \(\overline{{\mathbb {F}}}:=(\overline{{\mathcal {F}}}_t)_t\), with \(\overline{{\mathcal {F}}}_t:={\mathcal {F}}_t\otimes {\mathcal {F}}^\theta _t\). By construction \(\theta \) is an \(\overline{\mathbb F}\)-stopping time. We are going to show that \(\overline{X}:=(\overline{X}_t)_t\) is an \(\overline{{\mathbb {F}}}\)-Markov chain. Let \(\psi \in C(E)\). We show first that
Let \(B\in {\mathcal {F}}_{t-1}\) and \(C\in {\mathcal {F}}_{t-1}^\theta \). We have
where in the second equality we used the measurability property of \(\kappa \). We finally have
This shows that \(\overline{X}\) is an \(\overline{\mathbb F}\)-Markov chain. Now, by Dynkin’s formula,
By applying Fubini’s theorem we obtain the desired result. \(\square \)
1.3 A continuity property for the price function
Lemma 21
Let (E, d) be a metric space. Consider the function \(P:E\rightarrow \mathbb R_+\) defined by
where \(r:E\rightarrow \mathbb R_+\) and \(s:E\times {\mathbb {R}}_+\rightarrow \mathbb R_+\) are continuous, for each \(x\in E\), \(p\mapsto s(x, p)\) is increasing with \(s(x, 0)=0\) and there exists \({\tilde{p}}\ge 0\) such that \(\inf _{x\in E}s(x, {\tilde{p}})\ge \sup _{x\in E}r(x)\). Then there exists for each \(x\in E\) a unique \(p^\star (x)\in [0, {\tilde{p}}]\) such that
Moreover, \(x\mapsto p^\star (x)\) is continuous and as a consequence P is also continuous.
Proof
For each \(x\in E\), since \(r(x)\le s(x, {\tilde{p}})\), we have that \({\tilde{p}}\in \{p\ge 0:r(x)\le s(x, p)\}\). Using that \(s(x, 0)=0\le r(x)\) and \(p\mapsto s(x, p)\) is continuous and increasing we must have that there exists a unique \(p^\star (x)\in [0, \tilde{p}]\) such that \(r(x)=s(x, p^\star (x))\) and for all \(p<p^\star (x)\), \(r(x)>s(x, p^\star (x))\). This allows us to write \(P(x)=p^\star (x)\wedge p_{\text {max}}\). Let us show that \(p^\star (x)\) is continuous. Let \((x_n)_{n\ge 1}\subset E\) converging to \(x\in E\). Since \([0, {\tilde{p}}]\) is compact, there exists at least one limit point of the sequence \((p^\star (x_n))_{n\ge 1}\). Let \(p_0\) be a limit point of the above sequence and assume without loss of generality that the whole sequence converges to that point. Using the continuity of r and s and passing to the limit in the equality \(r(x_n)=s(x_n, p^\star (x_n))\), we obtain \(r(x)=s(x, p_0)\), and henceforth \(p_0=p^\star (x)\). \(\square \)
Rights and permissions
Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted manuscript version of this article is solely governed by the terms of such publishing agreement and applicable law.
About this article
Cite this article
Dumitrescu, R., Leutscher, M. & Tankov, P. Energy transition under scenario uncertainty: a mean-field game of stopping with common noise. Math Finan Econ (2024). https://doi.org/10.1007/s11579-023-00352-w
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11579-023-00352-w
Keywords
- Electricity market
- Energy transition
- Scenario uncertainty
- Mean-field games
- Common noise
- Optimal stopping
- Partial information