Anti-Martingale: Inverse Progression and Its Mathematical Model

Inverse progression, known as the Anti-Martingale system, is a position-sizing strategy in which the position increases following a positive outcome and resets to the base value following a negative one. Formally, if S(t) denotes the position size at step t, then S(t+1) = 2·S(t) when X(t) > 0 and S(t+1) = S₀ when X(t) ≤ 0, where X(t) is the iteration result. This recurrence relation generates a stochastic process with a multiplicative structure, fundamentally distinct from additive models. The theoretical foundation of this strategy relates to the concept of positive momentum, whereby serial correlation among successful outcomes is exploited to maximize returns. It is essential to emphasize that under conditions of independent trials, serial correlation is absent, and the strategy confers no statistical advantage.

Exponential growth during a series of positive outcomes is the defining property of inverse progression. After k consecutive successful iterations, the position size reaches S(k) = S₀ · 2ᵏ, and accumulated profit equals P(k) = S₀ · (2ᵏ − 1). The probability of achieving a streak of length k, given a single-trial success probability p, equals pᵏ, which produces a right-skewed profit distribution with a heavy tail. The expected profit from a single streak is computed as E[P] = Σₖ₌₁^∞ pᵏ · S₀ · (2ᵏ − 1) = S₀ · [p·2/(1 − 2p) − p/(1 − p)] when 2p < 1. This formula demonstrates that when p ≥ 0.5 the series diverges, formally implying an undefined expectation and necessitating an upper bound on the maximum streak length.

Crash risk upon streak reversal constitutes the primary vulnerability of inverse progression. Following a streak of k successful iterations, a single negative outcome results in a loss of S₀ · 2ᵏ, which completely negates accumulated profit and generates a net loss of S₀. The Conditional Value at Risk (CVaR) function for this strategy exhibits nonlinear dependence on streak length: CVaR_α(k) = S₀ · 2ᵏ · (1 − p), where α is the confidence level. Critically, the Maximum Drawdown (MDD) in inverse progression is bounded below by S₀ but is unbounded above in terms of single-step loss at peak streak levels. Monte Carlo analysis of drawdown distributions shows that at p = 0.48 and a horizon of 10,000 iterations, the median MDD is approximately 6.2·S₀.

Mathematical comparison with the classical Martingale system reveals a fundamental duality between the two approaches. The classical Martingale increases position size after a negative outcome: S(t+1) = 2·S(t) when X(t) ≤ 0, seeking to recover losses through subsequent iterations. In probability theory, both systems generate martingale processes with respect to different filtrations: the direct system relative to the loss sequence, and the inverse system relative to the profit sequence. The key distinction lies in risk allocation: the classical Martingale concentrates catastrophic risk in the right tail of the loss distribution, whereas the Anti-Martingale distributes moderate risk more uniformly. Formally, the kurtosis of the outcome distribution for the classical Martingale exceeds that of the inverse system by an order of magnitude under identical parameters.

Optimal stopping conditions for inverse progression are derived through the framework of the Optimal Stopping Problem. Let τ denote the stopping time for the position-increase series. The problem is formulated as max E[P(τ)] subject to P(Loss(τ) > L_max) ≤ α, where L_max is the permissible loss threshold and α is the probability of its exceedance. Solving this problem via dynamic programming (Bellman equation) yields the optimal streak length k* = ⌊log₂(L_max/S₀)⌋, which maximizes expected profit subject to the risk constraint. In practice, an adaptive algorithm is recommended that recalculates k* after each update of the parameter p estimate based on Bayesian posterior updating. Production deployment of this algorithm requires real-time stream processing with latency not exceeding 50 ms to ensure correct strategy state updates.