Martingale and anti-martingale are position-sizing rules, not entry systems. Martingale increases the next position after a loss; anti-martingale increases it after a win. Neither can turn an untested signal into a trustworthy edge. They change the size of the next bet, and therefore the path through risk.
The distinction is set out in Robert Pardo's Wiley trading-systems text and Richard Weissman's price-risk management chapter. Both treat the progression as a sizing decision separate from the logic that decides when to enter and exit.
The rule is about the next size
In the trading sense, a martingale is a loss-progression rule. After a losing trade, the next position is increased, often by doubling it, and a win resets the sequence. The recovery story assumes that the next win can cover the accumulated losses and leave the base gain intact.
An anti-martingale, also called a reverse martingale, flips the state. The next position grows after a win and is reduced or reset after a loss. The pure version is usually described with the same step-up idea, but the direction of escalation changes. These definitions and their variations are described in Pardo's sizing examples and Weissman's discussion of volumetric risk.
The clean comparison is easiest to see with a base risk unit and a multiplier greater than one. The following is the pure progression, not a claim that every strategy uses it:
martingale: loss -> larger size -> larger size -> ... ; win -> base size
anti-martingale: win -> larger size -> larger size -> ... ; loss -> base sizeThat is why the two rules can produce very different equity curves from the same entries and exits. A martingale makes the largest bet during a losing run. An anti-martingale makes the largest bet during a winning run, which can be helpful if the run persists and painful when it ends.
What the recovery story leaves out
The classic martingale only has its neat recovery property under restrictive assumptions: the payoff per unit stays stable, a win arrives before the available capital or a wager limit is reached, and costs do not consume the recovery. Trading adds variable payoff, spread, slippage, commission, financing and gaps. The arithmetic is no longer a casino thought experiment.
This is not just a warning copied from a trading forum. A paper in The Mathematical Gazette analyses finite-step martingale systems and identifies exponential growth as the reason finite budgets and table limits matter (Lazowski's paper). An earlier analysis in the Journal of Economic Behavior & Organization studies the same escalation and notes that a finite cap on wagers does not necessarily stop people from continuing after consecutive losses (Rational escalation of costs). Those papers examine gambling, not a cTrader strategy. They support the mechanism and the human difficulty of stopping, not a forecast for market returns.
Anti-martingale removes the specific problem of increasing after losses. It does not make exposure harmless. A winning streak can build a large position just before the strategy enters a choppy or adverse regime. The rule is then exposed to the end of the streak, not to the start of it.
There is also a naming trap. A fixed-fractional rule sizes from current equity, so the cash amount tends to rise after gains and fall after losses. That makes it anti-martingale-like in one sense, but it is not the same as a step progression keyed only to the last trade. Pardo's sizing examples separate equity-based sizing from both progressions (Wiley text); Weissman's risk-management chapter makes the same distinction between a unit's risk and the account-level position size (Weissman chapter). The implementation must be written down; labels are not specifications.
Evidence supports caution, not a winner
The literature does not establish a universal winner between martingale and anti-martingale. The sources above are useful for definitions and failure modes, but they do not show that an anti-martingale overlay creates a market edge.
The more formal sizing literature makes the same point from another direction. Kelly's framework chooses allocations to maximise expected logarithmic growth under a repeated-gamble model, while later work adds an explicit drawdown constraint rather than treating growth as the only objective (Hsieh and Barmish on Kelly's limitations, Busseti, Ryu and Boyd's risk-constrained Kelly paper). That is a different problem from increasing size because the last trade won.
The distinction matters in a backtest. If the entries, exits and trade outcomes stay fixed, a sizing overlay can change terminal wealth, drawdown, volatility, exposure and loss clustering. It cannot tell you whether the entry logic has predictive value. The result is also path-dependent: the same collection of wins and losses can produce a different equity path when their order changes.
How you'd actually test it
The fair test is a controlled comparison. Keep the signal, exits and trade universe fixed. Change only the sizing rule.
- Define the base risk unit, the starting size, the step or multiplier, the reset rule, the maximum size and the maximum progression length.
- Compare a constant-size or fixed-risk benchmark with a martingale and an anti-martingale version. Keep the risk unit comparable; a larger final return bought with a larger risk budget is not a clean victory.
- Apply the new size only after the previous trade is complete. The next size must use information that existed at that decision time.
- Charge spread, slippage, commission and financing. The costs that make a backtest honest are especially relevant when the rule changes turnover and exposure.
- Hold back an untouched out-of-sample period. Do not tune the multiplier, cap or progression length on the same trades used to judge the idea. The parameter-sensitivity guide explains why that distinction matters.
- Report more than terminal return: expectancy per base risk unit, max drawdown, largest position, exposure, turnover, time under water, loss-limit breaches and the contribution of costs.
- Stress the order. Shuffle trade outcomes as a diagnostic, and use block resampling when preserving runs or serial dependence matters. The Monte Carlo drawdown explainer covers why one historical sequence is not a worst-case guarantee.
For a prop-style account, the relevant question is not whether a progression can produce a smooth-looking curve in one sequence. It is whether the sizing path stays inside the account's actual loss constraints when the sequence is less cooperative. A strategy that reaches a higher terminal value by breaching the constraint has failed the test being asked of it.
The realbacktesting standard is reproducibility: rebuild the same sizing path in cTrader from fixed rules, completed observations and real costs, then compare the result with the benchmark (how realbacktesting verifies a cTrader backtest). If the size history cannot be reconstructed, the headline curve is not evidence yet.
Frequently asked
Does anti-martingale create an edge?
No. It changes exposure after previous outcomes. Any edge must come from the entry and exit process and remain visible when the sizing rule is tested out-of-sample.
Is martingale always unprofitable?
No single label settles the result of every finite test. It does, however, create an escalating-loss exposure that must be evaluated against finite capital, costs and loss constraints; a short sample can hide that risk.
Is a fixed-fractional rule the same as anti-martingale?
No. A fixed-fractional rule sizes from equity and trade risk. An anti-martingale progression sizes from the sequence state. They may move in the same direction after a win, but they are different specifications.
Takeaway
Martingale adds size when the recent sequence is going badly; anti-martingale adds size when it has been going well. Neither earns trust until its sizing path survives real costs, unseen data and the account's actual loss limits.