Abstract
This paper studies a dynamic collective choice model in the presence of an advertiser, where a large number of consumers are choosing between two alternatives. Their choices are influenced by the group’s aggregate choice and an advertising effect. The latter is produced by an advertiser making investments to convince as many consumers as possible to choose a specific alternative. In schools, for example, teenagers’ decisions to smoke are considerably affected by their peers’ decisions, as well as the ministry of health campaigns against smoking. We model the problem as a Stackelberg dynamic game, where the advertiser makes its investment decision first, and then the consumers choose one of the alternatives. On the methodological side, we use the theory of mean field games to solve the game for a continuum of consumers. This allows us to describe the consumers’ individual and aggregate behaviors, and the advertiser’s optimal investment strategies. When the consumers have sufficiently diverse a priori opinions toward the alternatives, we show that a unique Nash equilibrium exists between them, which predicts the distribution of choices over the alternatives, and the advertiser can always make optimal investments. For a certain uniform distribution of a priori opinions, we give an explicit form of the advertiser’s optimal investment strategy and of the consumers’ optimal choices.
Similar content being viewed by others
References
Anderson B, Moore J (2007) Optimal control: linear quadratic methods. Courier Dover Publications, New York
Baar T, Geert JO (1998) Dynamic noncooperative game theory. Society for Industrial and Applied Mathematics, Philadelphia
Bensoussan A, Chau M, Yam P (2015) Mean field stackelberg games: aggregation of delayed instructions. SIAM J Control Optim 53(4):2237–2266
Bensoussan A, Chau M, Yam P (2016) Mean field games with a dominating player. Appl Math Optim 74(1):91–128
Bhat C, Guo J (2004) A mixed spatially correlated logit model: formulation and application to residential choice modeling. Transp Res 38:147–168
Brock W, Durlauf S (2001) Discrete choice with social interactions. Rev Econ Stud 68(2):235–260
Clarke F (2013) Functional analysis, calculus of variations and optimal control. Springer Science & Business Media, Berlin
Dockner EJ, Jørgensen S (1988) Optimal advertising policies for diffusion models of new product innovation in monopolistic situations. Manag Sci 34:119–130
Erickson G (1995) Differential game models of advertising competition. Eur J Oper Res 83(3):431–438
Freiling G (2002) A survey of nonsymmetric Riccati equations. Linear Algebra Appl 351:243–270
Huang M (2010) Large-population LQG games involving a major player: the Nash certainty equivalence principle. SIAM J Control Optim 48(5):3318–3353
Huang M, Caines PE, Malhamé R P (2003) Individual and mass behaviour in large population stochastic wireless power control problems: centralized and Nash equilibrium solutions. In: Proceedings of the 42nd IEEE conference on decision and control, Maui, Hawaii, pp 98–103
Huang M, Caines PE, Malhamé RP (2007) Large-population cost-coupled LQG problems with nonuniform agents: individual-mass behavior and decentralized epsilon-Nash equilibria. IEEE Trans Autom Control 52(9):1560–1571
Huang M, Malhamé RP, Caines PE (2006) Large population stochastic dynamic games: closed-loop McKean–Vlasov systems and the Nash certainty equivalence principle. Commun Inf Syst 6(3):221–252
Koppelman F, Sathi V (2005) Incorporating variance and covariance heterogeneity in the generalized nested logit model: an application to modeling long distance travel choice behavior. Transp Res 39:825–853
Lasry JM, Lions PL (2006) Jeux à champ moyen. I-le cas stationnaire. C R Math 343(9):619–625
Lasry JM, Lions PL (2006) Jeux à champ moyen. II–horizon fini et contrôle optimal. C R Math 343(10):679–684
Lasry JM, Lions PL (2007) Mean field games. Jpn J Math 2:229–260
Little J (1979) Aggregate advertising models: the state of the art. Oper Res 27(4):629–667
Little J, Lodish L (1969) A media planning calculus. Oper Res 17:1–35
McFadden D (1974) Conditional logit analysis of qualitative choice behavior. In: Zarembka P (ed) Frontiers in econometrics. Academic Press, New York, pp 205–142
Nakajima R (2007) Measuring peer effects on youth smoking behaviour. Rev Econ Stud 74(3):897–935
Rudin W (1987) Real and complex analysis, 3rd edn. International series in pure and applied mathematics. McGraw-Hill Inc, New York
Rudin W (1991) Functional analysis. International series in pure and applied mathematics. McGraw-Hill Inc, New York
Salhab R, Malhamé R P, Le Ny J (2016) A dynamic game model of collective choice: stochastic dynamics and closed loop solutions. arXiv preprint arXiv:1604.08136
Salhab R, Malhamé RP, Le Ny J (2017) A dynamic game model of collective choice in multi-agent systems. IEEE Trans Autom Control. https://doi.org/10.1109/TAC.2017.2723956
Von Stackelberg H (2010) Market structure and equilibrium. Springer Science & Business Media, Berlin
Salhab R, Malhamé R P, Le Ny J (2016) A dynamic collective choice model with an advertiser. In: Proceedings of the 55th IEEE conference on decision and control, Las Vegas, NV, pp 6098–6104
Author information
Authors and Affiliations
Corresponding author
Additional information
This work was supported by NSERC under Grants 6820-2011 and 435905-13.
Appendix A
Appendix A
1.1 Quantities Related to the mA Nash Equilibrium
For all \(x\in L_2([0,T],\mathbb {R}^n)\),
where \(\phi \) is the unique solution of \(\frac{d}{dt}\phi (t,s)=\left( \frac{1}{r}\varGamma (t) B^{(2)}-A'\right) \phi (t,s)\), \(\phi (s,s)=I_n\), and
Proof of Theorem 2
The cost functional \(\bar{J}_0\) is positive and coercive with respect to \(v\in L_2([0,T])\), i.e., \(\lim \limits _{\Vert v\Vert _2 \rightarrow \infty }\bar{J}_0(v)/\Vert v\Vert _2=\infty \). If we show that \(\bar{J}_0\) is continuous in the reflexive Banach space \(L_2([0,T])\) with respect to v, then by Tonelli’s existence theorem [7, Theorem 5.51], \(\bar{J}_0\) has a finite minimum. Thus, we need only show that \(\bar{J}_0\) is continuous. The state y is continuous with respect to v. The fixed points \(\lambda (y)\) of F are continuous with respect to v. In fact, consider v and \(v'\) in \(L_2([0,T])\) and denote by \(y, y'\) the corresponding MA trajectories and by \(\lambda \) and \(\lambda '\) the corresponding fixed points. We have
Therefore, \( \Big (1-\sup \limits _{s \in [0,1]}\Big |\frac{dF}{d\lambda } (s,y)\Big |\Big ) |\lambda - \lambda '|\le |F(\lambda ',y) -F(\lambda ',y')|. \) Under Assumption 3, \(\sup \limits _{s \in [0,1]}\Big |\frac{dF}{d\lambda } (s,y)\Big |<1\). Under Assumption 2, \(\bar{F}\) is continuous. Moreover, \(\varDelta \) is continuous with respect to the \(L_2\) norm \(\Vert \Vert _2\). Hence, F is continuous with respect to y, and \(|F(\lambda ',y)-F(\lambda ',y')|\) converges to zero as \(\Vert y-y'\Vert _2\) converges to zero. Therefore, the fixed points \(\lambda \) of F are continuous. In view of (12) and the continuity of the fixed points \(\lambda \), \(\bar{x}(T)\) is continuous. Therefore, \(\bar{J_0}\) is continuous.\(\square \)
Proof of Theorem 3
We derive the condition on \(v^*\) (17) by studying the first variation of the cost functional in (13) with respect to a perturbation \(v=v^*+\eta \delta v\), where \(\eta \in \mathbb {R}\), and \(\delta v \in L_2([0,T],\mathbb {R}^{m_1})\). To this end, we need to derive at first an explicit form of the constraint on \(\bar{x}_v\). We have that \(\bar{x}_v=\bar{x}^\lambda \) defined in (12), where \(\lambda \) is the unique fixed point of \(\lambda \mapsto F(\lambda ,y)\). By taking the derivative of \(\bar{x}^\lambda \) with respect to time, we obtain that,
where
We compute now the Gâteaux derivatives [7] of y and \(\bar{x}\) at \(v^*\) in the direction \(\delta v\):
where,
and \(\mathcal {L}_1\) (resp. \(\mathcal {L}_2\)) is a continuous linear operator from the Hilbert space \(L_2([0,T],\mathbb {R}^{n_1})\) (resp. \(L_2([0,T],\mathbb {R}^{n})\)) to \(L_2([0,T],\mathbb {R}^{n})\) such that for all \(z_1 \in L_2([0,T],\mathbb {R}^{n_1})\) and \(z_2 \in L_2([0,T],\mathbb {R}^{n})\),
Using Fubini–Tonelli’s theorem [23], one can show that the adjoint operators of \(\mathcal {L}_1\) and \(\mathcal {L}_2\) are, respectively, \(\mathcal {L}_1^*\) and \(\mathcal {L}_2^*\) defined in (16). We recall from [24] that the adjoint operator of a linear continuous operator \(\mathcal {G}\) defined from the Hilbert space \((H_1,\langle ,\rangle _1)\) into the Hilbert space \((H_2,\langle ,\rangle _2)\) is the linear continuous operator \(\mathcal {G}^*\) defined from the Hilbert space \((H_2,\langle ,\rangle _2)\) into the Hilbert space \((H_1,\langle ,\rangle _1)\) and satisfying for all \(x\in H_1\) and \(y \in H_2\) \( \langle \mathcal {G}(x), y \rangle _2 = \langle x, \mathcal {G}^*(y) \rangle _1. \) Here, we use the explicit form of the operator \(\varDelta \) (24). The Gâteaux derivative of \(\bar{J}_0\) is
We have
By integrating (27) from 0 to T we get \(0=\Big \langle B_0'P,\delta v \Big \rangle -\Big \langle \mathcal {L}_1^* (Q)(t),\delta y \Big \rangle \). Similarly, we have
Therefore, \(\delta \bar{J}_0= \Big \langle B_0'P,\delta v \Big \rangle + r_0 \Big \langle v^*, \delta v \Big \rangle .\) By optimality, \(\delta \bar{J}_0=0\) for all \(\delta v \in L_2([0,T])\). Hence, \(v^*=-\frac{1}{r_0}B_0'P\).\(\square \)
Proof of Lemma 2
The idea of the proof is to replace the term
in the expression of \(\mathcal {L}_2^*(Q)\) by an assumed known constant \(K_1\). Equation (15) is then a linear differential equation parameterized by \(K_1\) whose solution is a linear operator of \(K_1\). By replacing this solution in the term (28), and by requiring that \(K_1\) is equal to (28), one can show that the unique solution of (15) is \( Q(t)= \varPhi (T,t)'\Big (\alpha \xi ^*\int _t^T \varPhi (\sigma ,T)'H(\sigma )\mathrm {d \sigma }Y +M_0(\bar{x}_{v^*}(T)-p_2)\Big ) \), where Y is the unique solution of the following linear algebraic equation:
\(\square \)
Proof of Theorem 4
Let v and \(v'\) in \(L_2([0,T])\), and denote by \(y, y'\) the corresponding MA trajectories, and by \(\lambda \) and \(\lambda '\) the corresponding fixed points. We have,
The rest of the proof is similar to the proof of Theorem 2.\(\square \)
Proof of Lemma 3
The uniqueness follows from Assumption 5. Let \(v\in L_2([0,T])\). The path \(\bar{x}_v\) defined in (12) is uniformly bounded with c (with respect to the \(L_2\) norm). Therefore, the optimal cost \(\bar{J}_0(v^*,\bar{x}_{v^*})\le \bar{J}_0(v,\bar{x}_{v})\) of the MA optimal control problem defined in (13) is uniformly bounded with c. Hence, the optimal control law \(v^*\) and the optimal state \(y_{v^*}\) are uniformly bounded with c. Consequently, the term \( \varDelta \left( K(p_2)y_{v^*}+ \alpha \bar{x}^\lambda \right) \), where \(\bar{x}^\lambda \) is defined in (12), is uniformly bounded with c by a positive constant \(L_1\). This means that \(-L_1\le \varDelta \left( K(p_2)y_{v^*}+ \alpha \bar{x}^\lambda \right) \le L_1\). Hence, \(\bar{F}(-L_1) \le F(\lambda ,y) \le \bar{F}(L_1)\). If we choose \(-L_1>a-c/2\) and \(L_1<a+c/2\), that is, \(c>\max (2(a+L_1),2(-a+L_1)):=c_0\), then the map F takes its values in (0, 1). Therefore, F has a unique fixed point \(\lambda \in (0,1)\).\(\square \)
Proof of Theorem 5
By Theorem 4 we know that there exists \(v^*\) an optimal investment strategy. Moreover, we know that \(v^*\) and \(\bar{x}_{v^*}\) should be equal to \(v^*=-\frac{1}{r_0}B_0'P^\lambda \) and \(\bar{x}_{v^*}=\bar{x}^\lambda \), where \(P^\lambda \) and \(\bar{x}^\lambda \) are defined in (21)–(22) for a fixed point \(\lambda \) of \(F_u\). It remains to show that \(F_u\) has a unique fixed point \(\lambda _*\). Let \(\lambda \) and \(\lambda '\) be two distinct fixed points of \(F_u\). Then, \(\lambda \) and \(\lambda '\) are, respectively, the fixed points of \(s\mapsto F(s,y^\lambda )\) and \(s\mapsto F(s,y^{\lambda '})\), where F is defined above (11) and \(y^\lambda \) in (21). Following Lemma 3, \(\lambda \) and \(\lambda '\) belong to (0, 1). But, \(\beta 'x^0 - \delta \) has a uniform distribution, which implies that \(F_u\) has a shape similar to that of the cumulative distribution function of a uniform distribution. Thus, all the real numbers in the interval [0, 1] are fixed points of \(F_u\). In particular, \(\lambda =0\) is a fixed point of \(s \mapsto F(s,y^{\lambda =0})\). This leads to a contradiction and shows that \(F_u\) has a unique fixed point. \(\square \)
Rights and permissions
About this article
Cite this article
Salhab, R., Malhamé, R.P. & Le Ny, J. A Dynamic Collective Choice Model with an Advertiser. Dyn Games Appl 8, 490–506 (2018). https://doi.org/10.1007/s13235-018-0254-x
Published:
Issue Date:
DOI: https://doi.org/10.1007/s13235-018-0254-x