Parrondo’s Paradox Calculator
Two games, each rigged to lose. Play either one forever and your capital drains away; alternate them, and the same two losing games build a fortune. This is Parrondo’s paradox — not a trick of wording but of Markov chains — and the simulation below, seeded so it repeats exactly, lets you watch the impossible happen four strategies at a time.
The Two Games, Precisely
Both games are coin flips for one unit of capital, under the classic parameters published by Harmer and Abbott (1999), with a small bias ε = 0.005 shaving the honest coins. Game A is a single biased coin: win with probability 0.5 − ε = 0.495. Its expectation is −0.01 per play — a slow bleed, undeniable over 10,000 plays. Game B is two coins chosen by your current capital modulo 3: if your capital is a multiple of 3, you must flip the bad coin (win probability 0.1 − ε = 0.095); otherwise you flip the good coin (0.75 − ε = 0.745). Game B looks generous — the good coin dominates naive averages — but it is rigged by occupation: played alone, B herds your capital into the multiple-of-3 states where the bad coin waits (the bad coin’s state is visited about 38% of the time rather than the innocent third), and its true expectation is also negative. Two losing games, each provable on its own terms. The paradox is what the mixtures do, and the panel above is the experiment.
Reading the Panel
The simulation (seeded 1996, the paradox’s birth year in Parrondo’s lectures) runs each strategy for 10,000 plays from zero capital, and this seed’s panel reads: A alone −62, B alone −142, the random mix +104, and the ABB pattern +574. The exact Markov-chain expectations behind them are −100, about −87, about +157 and about +574 per 10,000 plays respectively — the simulation wobbles around its theory by the usual ±100 or so, and the signs never wobble. Note which mixture wins biggest: not the fair coin-flip between games, but ABB — one play of the randomising A followed by two of B — because the pattern’s period of 3 resonates against the trap’s modulus in the most favourable phase, a refinement the next section’s mechanism makes intuitive and the FAQ’s pattern discussion makes precise. Rerun the panel and the same four numbers return: the seed makes the miracle auditable.
The Ratchet: Why Mixing Wins
Game B’s trap is a single doorway: the multiple-of-3 state with its 0.095 coin. Left alone, B’s own dynamics keep returning capital to that doorway (from a multiple of 3, the bad coin usually knocks you to a non-multiple; from non-multiples, the good coin pushes you two steps forward as often as one back, landing on multiples with cruel regularity — the stationary share of trap visits is about 38%, well above the 33% that would make B fair-ish). Game A, played between B’s turns, is a randomising kick: its unbiased-ish steps shuffle the capital’s modulo class, breaking B’s herding and cutting the trap’s occupancy toward the level where B’s good coin (0.745, strongly favourable) dominates the ledger. The mixture wins not because A adds value — A is a losing game — but because A rearranges B’s circumstances, the way a ratchet’s pawl adds nothing but timing and yet the wheel climbs. Physicists recognised the mechanism instantly: it is a flashing Brownian ratchet, the statistical mechanics of molecular motors, wearing a casino’s clothes.
Volatility Pumping: The Paradox at the Trading Desk
Finance met Parrondo wearing a different name: volatility pumping, the observation (formalised in stochastic portfolio theory) that a portfolio rebalanced between volatile assets can grow even when each asset’s own trajectory is flat or declining — the rebalancing rule harvests the alternation the way game A’s kicks harvest B’s good coin. The analogy is real but partial, and this page’s honesty policy requires the boundary to be drawn: rebalancing bonuses arise from genuine diversification and mean-reversion in the assets’ relative prices, not from an accounting illusion, yet neither do they repeal arithmetic — transaction costs and trending markets tax the harvest exactly as ε taxes the coins. What Parrondo contributes to the trading desk is not a strategy but a suspicion: that a system’s behaviour under alternation cannot be inferred from its components in isolation, which is as true of portfolios as of ratchets, and as expensive to learn late in both.
Dormancy, Bet-Hedging and the Bacterial Ratchet
Biology may be the paradox’s most literal employer: bacteria that switch stochastically between a growing state and a dormant, antibiotic-proof persister state are running Parrondo’s strategy in flesh — each state alone is a losing bet across the environments the colony will meet (growth dies by antibiotic, dormancy dies by starvation-by-irrelevance), while the alternation survives both. Evolution tuned the switching rates the way the pattern analysis tunes ABB: not by slogan but by the stationary arithmetic of states and traps. The parallel has been drawn formally in the bet-hedging literature, and it completes the paradox’s tour of the sciences — a casino puzzle that describes molecular motors, bacterial survival and the phase discipline of mixtures, all because one Markov chain refused to average the way intuition demanded.
Related Tools
More from the Statistics Calculators hub: Bertrand’s Paradox Calculator — probability bending the other way — and the Monty Hall Problem Calculator, the hub’s third great intuition-repair kit.
Frequently Asked Questions
Is this a real effect or a simulation artefact?
Real, and proven analytically as well as simulated: the Markov-chain stationary distributions of B alone versus B perturbed by A are computable in closed form, and the mixture’s positive expectation is a theorem, not a sampling fluke. The seeded simulation exists so the theorem can be watched — and so sceptics can rerun it identically. The effect has been replicated in electronic and mechanical ratchets in physics laboratories, which is either reassuring or alarming depending on how one feels about casinos teaching statistical mechanics to do work.
Can I use Parrondo’s paradox to beat real games?
No — and the reason is instructive: real casino games are memoryless in the way that matters (each bet’s odds do not depend on your capital modulo anything), so there is no trap state for a second game to rescue you from. Parrondo’s mechanism needs state-dependent odds, which gambling does not offer but nature and markets sometimes approximate: the paradox’s genuine applications are in physics (ratchets, molecular motors), biology (life-history strategies that alternate between losing environments), and as a parable in finance about volatility and rebalancing — where the honest version of the claim is subtler and the dishonest version is a sales pitch. This page is the parable, computed.
Why does game B lose if its good coin wins 74.5%?
Because the good coin does not get a fair share of the flips. Played alone, B’s capital dynamics settle so that roughly 38% of plays happen at multiples of 3, where the 0.095 coin rules; weight the two coins by their occupancy (0.38 × 0.095 + 0.62 × 0.745 ≈ 0.498 as an effective win rate before the ε arithmetic finishes the job) and the generosity evaporates — the bad coin’s rare visits cost more than the good coin’s frequent wins pay. Occupancy, not the sticker probabilities, is a state-dependent game’s true odds: the lesson generalises to every queue, market and ecosystem where where-you-are decides what-you-face.
What does ε (epsilon) do, and why 0.005?
ε is the small bias that makes the components strictly losing: with ε = 0, game A is exactly fair and B is close to fair, and the mixture’s winning, while still present, dances on a knife’s edge that confuses the demonstration. Harmer and Abbott’s canonical choice ε = 0.005 makes A provably losing (−0.01 per play in expectation), pushes B under water as well, and leaves the mixture’s gain intact and dramatic — a paradox with its controls set so the effect cannot be explained away as fair-game noise. The parameters are the literature’s, not this page’s invention, and the calculator labels them on the panel for exactly that reason.
Does the switching pattern matter, or will any mixture win?
It matters enormously, and the landscape is this page’s favourite plot twist: at these parameters the random mix wins (+0.0157 per play in expectation), AAB wins modestly (+0.0092), ABB wins handsomely (+0.0574 — the panel’s champion) — and the strict ABAB metronome still loses (−0.0067 per play), because its period-2 rhythm locks into a fixed, unfavourable phase against B’s period-3 trap instead of scrambling it. Mixing is not magic; it is phase management. The patterns that rescue B are those that keep knocking the capital’s modulo class out of the rhythm B’s dynamics try to impose — which is why the paradox rewards analysis over slogans, and why its calculator prints mechanisms instead of mottos.
Who was Parrondo, and where is this used?
Juan Parrondo, a Spanish physicist, devised the games in 1996 to illustrate flashing ratchets — systems (formalised by Feynman among others) in which asymmetric potentials plus random switching rectify noise into directed motion. The paradox leapt fields: it models molecular motors hauling cargo in cells, informs discussions of dormancy-switching in bacteria (two losing environments, one surviving alternation), and serves economics as a cautionary emblem about aggregation. Its calculator here is deliberately the canonical version — the same coins, the same ε, the same trap — because a paradox this portable deserves to travel with its papers in order.