The dynamic evolution of preferences (Q2373374)

The paper characterizes dynamic evolution of preferences in a class of strategic interactions providing simple conditions characterizing the limiting distribution of preferences and apply their results to study the evolutionary emergence of overconfidence and interdependent preferences. Consider a class of strategic interactions in which two players \(i\) and \(j\) with identical strategy sets as a subset of the reals engage in a symmetric game whose payoffs are given by the same underlying payoff function \(\Pi :\mathbb{R}^2\to \mathbb{R},\) i.e., \(\Pi ^i( x^i,x^j) =\Pi ( x^i,x^j) \) and \(\Pi ^j( x^i,x^j) =\Pi ( x^j,x^i) \) where \(x^i,x^j\) is a typical strategy pair and \(\Pi ^i( x^i,x^j) ,\Pi ^j( x^i,x^j) \) denote the respective payoffs. \ Let \(\tau ^i,\tau ^j\) denote player types drawn from a compact sub-interval \(T\) of the reals and a function \(B:R^{3}\to R\) provide the (symmetric) dispositions of the two players, \(B^i( x^i,x^j,\tau ^i) =B( x^i,x^j,\tau ^i) \) and \(B^j( x^i,x^j,\tau ^j) =B( x^j,x^i,\tau ^j) ,\) which differentiate their perceived payoffs \(U^i()\) and \(U^j()\) from their realized or true payoffs \(\Pi ^i\) and \(\Pi ^j\), i.e., \(U^i( x^i,x^j,\tau ^i) =\Pi ^i( x^i,x^j) +B^i( x^i,x^j,\tau ^i) \) and \(U^j( x^i,x^j,\tau ^j) =\Pi ^j( x^i,x^j) +B^j( x^i,x^j,\tau ^j) .\) Let \(\Gamma \) denote the game in which the players \(i\) and \(j\) choose strategies \(x^i,x^j\) to maximize their perceived payoffs \(U^i(.,\tau ^i)\),\(U^j(.,\tau ^j)\) and obtain true payoffs \(\Pi ^i\) and \(\Pi ^j. \) Assumption A. \(\Gamma \) has a unique pure strategy Nash equilibrium \(\widehat{x}^i( \tau ^i,\tau ^j) , \widehat{x}^j( \tau ^i,\tau ^j) \) for each pair of player types \(( \tau ^i,\tau ^j) \in T\times T.\) Under this assumption, the true payoffs in the equilibrium are given by the following fitness functions: \(f^i( \tau ^i,\tau ^j) =\Pi ^i( \widehat{x}^i( \tau ^i,\tau ^j) ,\widehat{x} ^j( \tau ^i,\tau ^j) ) \) and \(f^j( \tau ^i,\tau ^j) =\Pi ^j( \widehat{x}^i( \tau ^i,\tau ^j) ,\widehat{x}^j( \tau ^i,\tau ^j) ) .\) To study the evolution of disposition the authors adopt an indirect evolutionary approach by considering a ``types game'' in which each player chooses a type \(\tau ^i,\tau ^j\) in \(T\) and receives payoffs according to the fitness functions. Assumption B. The types game is dominance solvable. They consider a continuous time model in which at each point of time \(t\geq 0,\) the population of individual types is characterized by the distribution \(G_{t}\in \Delta ( T) \), where \( \Delta ( T) .\) They assume that the initial distribution \(G_{0}\) has full support over \(T\). At each \(t\), individuals are randomly matched in pairs to play the game \(\Gamma \). \ Dropping the superscripts \(i\) and \(j\) due to the symmetry of the problem, the average fitness levels of individuals of type \(\tau \) at time \(t\) is given by \(\int f( \tau ,\tau ^{\prime }) dG_{t}( \tau ^{\prime }) .\) \ They assume that the distribution of types evolves according to the differential equation: \[ \frac{d}{dt}G_{t}( S) =\int_{S}g( \tau ,G_{t}) dG_{t}( \tau ) ,S\subset T\text{ Borel measurable} \] where \(g:T\times \Delta ( T) \to R\) is a continuous growth-rate function satisfying: \[ g( \tau ,G_{t}) >g( \widetilde{\tau },G_{t}) \Longleftrightarrow \int f( \tau ,\tau ^{\prime }) dG_{t}( \tau ^{\prime }) >\int f( \widetilde{\tau },\tau ^{\prime }) dG_{t}( \tau ^{\prime }) \] and the normalization, to ensure that \(G_{t}\) remains a distribution function, \(\int_{S}g( \tau ,G_{t}) dG_{t}( \tau ) =0\) for each \(t.\) \ The evolutionary dynamical system reflects the idea that the proportion of more successful types increases at the expense of the less successful types perhaps because they have more descendants or because they are imitated more often. The formulation is also compatible with the assumption that successful types translate into stronger influence rather than numerical proliferation. Under certain regularity conditions the dynamical system has a well-defined solution. The main theorem shows that any serially dominated strategy must eventually become extinct under any payoff-monotonic population dynamics and, in particular, in a dominance solvable game any such dynamics must converge in distribution to a point mass at the unique action surviving the iterated elimination of strictly dominated strategies. By this theorem, the support of any limiting distribution must be a subset of the set \(U\) of serially undominated strategies. \ In games where the set \(U\) can be characterized or computed easily, the theorem gives useful predictions regarding the dynamic evolutionary outcomes. \ Two such special cases, game which are dominance solvable and the case of strategic complementarities, are dealt with in two further theorems. \ The authors go on to consider two applications of their main result, interpreting the dispositions as perception biases representing over and under-confidence, optimism or pessimism and considering interdependent preferences in which players can give different weights to their opponents' payoffs in their utility functions allowing for envy or altruism.