The concept of a martingale originates in the world of gambling, but the idea extends far beyond the roulette or poker table. Imagine a simple game: you start with €0, and in each round you toss a fair coin. If you win, you gain €1; if you lose, you lose €1. Each round is completely independent of the previous one; what happened before does not influence the outcome of the next toss.

Suppose we denote your wealth at a given time t by  X_t​. The question is then: based on everything you know up to now, what is the best prediction of your wealth tomorrow? In a fair game, the answer is obvious: it is simply your current wealth. The game contains no hidden tricks or patterns that can systematically help you win. Mathematically, we write this as:

Here lies the essence of a martingale: a process in which, given all available information, you have no predictable advantage. Your wealth may fluctuate and make large jumps, but those changes are purely due to chance. The past does not help you to systematically predict the future.

To define this idea mathematically, we need three building blocks. First, a probability space, the universe in which all possible outcomes and their probabilities are contained. Next, a stochastic process  (X_t)_t≥0​, which describes the evolution over time of a random quantity, such as wealth in the gambling game or a stock price. Finally, we need a filtration F_t, a collection representing all the information available at time t. You can think of it as an archive that grows as time passes: everything you know is preserved, and nothing disappears.

With these building blocks, we can formally define a martingale. A process (X_t)( is called a martingale with respect to F_t if it satisfies three conditions: it is integrable, it is adapted to the filtration (meaning it is based on the available information), and, most importantly, the expected value of tomorrow given everything we know now is equal to today:

This means nothing more than that, no matter how complicated the past may be, no function of earlier outcomes can systematically predict the future better than the current value itself. It is a precise mathematical translation of the idea of a fair game.

A well-known example is the random walk. Suppose your wealth changes from day to day by small, random shocks that have mean zero and are independent of previous outcomes. Then the expected value tomorrow turns out to be exactly equal to today, precisely as in a martingale. Yet this is only one example: a martingale can be much more complex. The variation may fluctuate, the dependence between moments may be intricate, and it may appear as though there is structure or pattern in it. The only thing that matters is that the average change, given the past, is zero.

In short, the martingale is both simple and profound. It is the mathematical model of a fair game, but in a universal format that also applies outside the gambling world, for example in financial markets and economic theory. It makes clear that, sometimes surprisingly, knowledge of the past provides no systematic advantage in predicting the future.