Capital distribution and portfolio performance in the mean-field Atlas model (Q2351635): Difference between revisions

This paper deals with the asymptotic behavior of a mean-field version of the rank-based model. For a market containing a fixed number \(n\geq1\) of stocks, with respective capitalizations \(X_{n}^1(t),\dots,X_{n}^{n}(t)>0\) at time \(t\), the log-capitalizations \(Y_{n}^{i}(t)=\log X_{n}^{i}(t)\) are assumed to satisfy the relation \(dY_{n}^{i}(t)=\gamma_{n}^{i}(t)dt+\sigma_{n}^{i}(t)dB^{i}(t)\), \(\forall i\in\{1,\dots,n\}\), where \(B^{i}(t), i\geq1\), are independent Brownian motions, \(\gamma_{n}^{i}(t)=\sum_{j=1}^{n}{\mathbf 1}_{\{Y_{n}^{i}(t)=Y_{n}^{(j)}(t)\}}\gamma_{n}^{j}\), \(\sigma_{n}^{i}(t)=\sum_{j=1}^{n}{\mathbf 1}_{\{Y_{n}^{i}(t)=Y_{n}^{(j)}(t)\}}\sigma_{n}^{j}\), where \(Y_{n}^{(1)}(t)\leq\cdots\leq Y_{n}^{(n)}(t)\) is the increasing reordering of \(Y_{n}^{1}(t),\cdots,Y_{n}^{n}(t)\), \(\gamma_{n}^{j}=\gamma(j/n)\), \(\sigma_{n}^{j}=\sigma(j/n)\), \(\gamma,\sigma:\;[0,1]\to\mathbb R\) are continuous functions, and the initial log-capitalizations \(Y_{n}^{1}(0),\dots,Y_{n}^{n}(0)\) are i.i.d. according to a given probability distribution \(m\) on \(\mathbb R\). The following propagation of chaos phenomenon is established: as \(n\to\infty\), the log-capitalization processes asymptotically behave like independent copies of a stochastic process \(Y(t)\), such that, for all \(t\geq0\), \(E[Y(t)]=E[Y(0)]+gt\), where \(g=\int_{0}^{1}\gamma(u)du\). The long time behavior of the fluctuation \(\tilde Y(t)=Y(t)-gt\) is described. The authors prove that the law of \(\tilde Y(t)\) converges toward an explicit equilibrium distribution. The weighted capital measure is defined by \[ \Pi_{n}^{p}(t)=\sum_{j=1}^{n}{(X_{n}^{(j)}(t))^{p}\over (X_{n}^{1}(t))^{p}+\cdots+(X_{n}^{n}(t))^{p}}\delta_{j/n}. \] It is proved that \(\lim_{n\to+\infty}\Pi_{n}^{p}(t)=\Pi^{p}(t)\), where the asymptotic weighted capital measure \(\Pi^{p}(t)\) is a deterministic probability distribution on \([0,1]\), with an explicit expression in terms of the law of \(Y(t)\). The long-time behavior of \(\Pi^{p}(t)\) is studied. Then the authors recover the Pareto-like shape of the capital distribution curves and provide a new description of the phase transition phenomenon. They introduce a family of portfolio rules, called \(p\)-diversity weighted portfolios. The performance of such portfolios is described in terms of the weighted capital measures. It is obtained a law of large numbers for the grows rates of these portfolios. The long time behavior of these asymptotic grows rates is studied.