Atlas models of equity markets (Q2496492)

Size is one of the most important descriptive characteristics of assets: one can understand a great deal about an equity market by observing, and making sense of, the continual ebb and flow of small-, medium- and large-capitalization stocks in its midst. Thus it is important to have models which describe (if not explain) this flow, and which exhibit stability properties for the resulting distributions of capital that are in agreement with actual observation. The simplest such model is the Atlas model for equity markets. This is a constant-coefficient model for the values (capitalizations) of stocks represented by their relative rank and driven by independent Brownian motions. It assigns the same, constant volatility to all stocks; zero growth rate to all stocks but the smallest; and positive growth rate to the smallest stock. Because it is responsible for all the growth (or support) in the market, this smallest stock is then called the Atlas stock. Somewhat more precisely: with \(g > 0\), \(\sigma > 0\) given constants, with independent Brownian motions \(W_1(\cdot),\dots, W_n(\cdot)\), and with \(X_i(t)\) representing the capitalization at time \(t\) of the stock with index (name) \(i\), the Atlas model postulates the dynamics \[ d(\log X_i(t))=\gamma_i(t)\,dt+\sigma_i(t)\,dW_i(t),\quad i=1,\dots,n,\tag{1} \] where the growth rates and volatilities are specified by \[ \gamma_i(t) = ng\cdot\mathbf 1_{\{X_i(t)=X_{p_t(n)}(t)\}},\quad \sigma_i(t)=\sigma.\tag{2} \] Here the ``reverse order-statistics'' notation is used for the capitalizations of stocks ranked in descending order from largest to smallest, \[ \max_{1\leq i\leq n} X_i=:X_{(1)}(t)\geq X_{(2)}(t)\geq \dots\geq X_{(n-1)}(t)\geq X_{(n)}(t):= \min_{1\leq i\leq n} X_i(t).\tag{3} \] It is also considered the random permutation \((p_t(1), ..., p_t(n))\) of \((1, ..., n)\), for which \( X_{p_t(k)}(t)\!= X_{(k)}(t)\), \(p_t(k)<p_t(k +1)\) if \(X_{(k)}(t) = X_{(k+1)}(t)\) hold with \(k = 1, \dots , n\). Roughly speaking, this means that \(p_t(k)\) is the name (index) of the stock with the \(k\)th-largest relative capitalization at time \(t\), and that ties are resolved by resorting to the lowest index. More generally, suppose there are given real numbers \(\gamma\), \(g_1,\dots,g_n\), \(\sigma_1 > 0, \dots , \sigma_n > 0\) such that \(g_1<0\), \(g_1+g_2<0,\dots,g_1 + \cdots+ g_{n-1}<0\), \(g_1 + \cdots+ g_n=0\). Corresponding to these parameters, the general model considered in this paper postulates the dynamics of (1) for the stock capitalizations \(X_1(t),\dots,X_n(t)\), but now with growth rates and volatilities given by \[ \gamma_i(t) = \gamma +\sum_{k=1}^ng_k\mathbf 1_{ \{X_i(t)=X_{p_t(k)}(t) \} }, \quad\sigma_i(t) =\sum_{k=1}^n\sigma_k\mathbf 1_{ \{X_i(t)=X_{p_t(k)}(t)\}}\tag{4} \] in place of (2). In other words, this more general model specifies \(\gamma + g_k\) as the growth rate, and \(\sigma_k\) as the volatility, for the stock with rank \(k\) at any given time. The model of (1)--(4) is called the \textit{first-order model}. Clearly \[ \gamma=g>0,\quad g_k =-g\quad\text{for}\quad k =1,\dots,n- 1\quad \text{and}\quad g_n = (n-1)g, \tag{5} \] in the case of the Atlas model of (1)--(2). A model of the type (1)--(4) with parameters that satisfy (5), though with possibly different volatilities, is called \textit{generalized Atlas model}. All these models have strictly nondegenerate volatility structures and bounded drift coefficients, so they admit a (unique) equivalent martingale measure on any given time-horizon; thus there are no relative arbitrage opportunities for such models. The first question that arises for the first-order model of (1)--(4) is the rigorous formulation of, and the study of existence/uniqueness of solution to, the resulting system of stochastic differential equations. The authors start by constructing a diffusion process corresponding to the stochastic equation of (1)--(4), and then study the behavior of the resulting ``ranked capitalization'' (reverse-order-statistics) processes of (3). Ergodic properties of these processes are studied. The authors introduce portfolios in the context of the model (1)--(2) and study the growth rates of a few relatively easy-to-implement investment rules. Some detailed comparisons of long-term-growth performance are carried out. The authors conclude with considerations of diversity. Some elementary computations show that models of this sort capture very well the intuitive notion that ``no stock can be allowed to dominate the entire market with anything but extremely low probability'' -- despite the fact that such models fail to be diverse in a strict, almost sure sense.