Growth Is Not Survival Two Readings of One Function, and the Opposite Effects of Concentration

Abstract Economic and policy discussion often treats growth as a proxy for survival: as long as the aggregate is rising, the system is taken to be healthy. This paper argues that growth and survival are two readings of one function and can move in opposite directions. Let the one-period growth factor be b and write ψ(s) = ln E[bˢ]. Time-average growth is the slope of ψ at zero; ensemble growth, the quantity read by aggregate indicators, is the value of ψ at one; survival reads the other zero of ψ. Under an absorbing wall this zero gives the exponent of the ruin probability; in a system with additive reinjection it is the tail index of the cross-sectional distribution. So the same number reads as extinction where there is no recovery term and as concentration where there is one. In the lognormal case ensemble growth is the "sum" of time growth and the ergodicity gap, while survival reads their "ratio": turning up the ergodicity gap raises growth, lowers survival and deepens concentration at the same time. Four groups of results follow. (1) "Survival first, then growth" is a chance-constrained problem and can be represented by a real-valued function; the trade-off exists genuinely within the survival set and has no object outside it; the shadow price of the boundary ε diverges as ε → 0 when measured in probability and tends to zero when measured in log-odds. (2) Under Kelly positioning, three objectives have three optima: ensemble growth rises without bound with leverage, time growth peaks at full Kelly, and survival peaks at zero position; every position above Kelly is strictly dominated by its symmetric point; with reinjection the tail index is 1 − 2/c, with its infimum at the ruin point rather than at the most aggressive position. (3) Along a line of equal growth, the survival exponent and the tail index vanish on the same ruin line while the growth indicator stays constant: for the same aggregate growth, the closer to the ruin line, the harder survival and the deeper concentration. For a reversal of the current trend, distinguishability also vanishes on this line; this statement depends on the convention for the change, and for a change of fixed size the direction on the concentrated side is opposite. (4) Under regime change, the worst change is not a reversal but the one that scales the edge by c − 1: the larger the position, the milder the change that kills. The detection-period criterion robust to every change is exactly B(2/c − 1) ≥ h, whose left side is the exponent of the stationary ruin probability; combined with the chance constraint, the largest allowed position is 2B/[B + max(h, ln 1/ε)], so the value choice ε and the detector parameter h lie on the same axis. Under the worst change, the cumulative-sum statistic tuned to it is exactly (2 − c)/c times the agent's drawdown, so the survival probability has the closed form min{1, (e^Λ − 1)/(e^h − 1)}: for Λ ≥ h survival is pathwise, not merely in expectation; an agent whose reserve is calibrated to the symmetric flip survives the worst change with probability only 0.73, 0.28 and 0.054 (c = 0.5, 1, 1.5). At the level of a population, the probability that "at least one unit survives" is not determined by the correlation coefficient and is bounded above by the common-mode probability, and redistribution proportional to wealth acts on the multiplicative channel and changes the tail index. The paper closes with a table of prohibitions (prohibitions that exclude observable events kept apart from definitional rules), seven falsifiable predictions and an execution ledger: the number of executions on external data is zero.