LessonThis is a lesson restated in new words. Any figure in it is an example from the source, not this desk's record.
Gambler's ruin asks how likely you are to go broke before you hit a chosen pile of wealth, if you repeat the same bet. A trading system is that question with inputs you cannot see. The win probability, the average winner, and the average loser have to be estimated from closed trades, and those estimates move.
Why it matters
In the fixed game, more starting capital lowers the chance of bankruptcy. A positive expected gain, the lecture's name for an edge, raises the chance of touching a high target before zero. If play never stops, ruin is certain even with that edge, because some losing streak is long enough. Live trading is finite, so the useful number is a hitting probability from today's wealth to a target, built from current estimates.
The idea in plain words
The textbook bet wins one unit with probability p and loses one unit with probability 1 minus p. You do not choose those numbers. The state is Markov: only current wealth matters, and the path that climbed and then fell is irrelevant once you know where you stand. Simulation estimates the chance by playing until the target or zero, then dividing successful paths by all paths. With a fixed distribution, that ratio settles on the hitting probability.
One illustration uses a 60 percent chance of winning one unit, starting at 5 with a target of 10. About 80 percent of the paths reach the target. Starting at 7 lifts the share to about 97 percent. Setting the win chance back to one half drops it by about 20 points, to about 75 percent. Those readings match the direction of the formula.
A trading system is a set of round trips. Its expected gain is the average winner times the win probability, plus the average loser times the loss probability. Positive is an upward model drift. Zero or negative is flat or down. Trades are the data, and a cold streak can be noise or the edge leaving. The blotter does not say which.
How to use it
Freeze the estimates and the problem is ruin with uneven payouts. In the lecture's fixed example, starting at 10 and aiming at 100 produces about a 90 percent chance of missing the target and about a 10 percent chance of hitting it. Raising the start to 50 lifts the success share sharply. In one run with the win probability at 0.55, every plotted path reached the target. That is one simulation, not a property of the next run. A formula for uneven payouts reproduces the 90 and 10 split when the inputs match, and longer samples are expected to settle there while the inputs stay fixed.
Then let the inputs move. One plotted win probability starts near 60 percent, fades, and later recovers. Equity suffers in the fade, including a near even stretch, and then recovers. Another draw lets losses become more common and leaves the curve stuck lower. Each piece is a lever: win probability, loss probability, average win, and average loss. Fixed take-profit and stop levels can hold the payoff sizes nearly constant. Without that cage a rule can cut a trade or let a winner run. The textbook probabilities cannot be changed. A trading rule can change them.
The lecture then reads a win rate and average winner and loser off an empirical system and asks for the chance of going from 5 to 10. Repeated runs sit near 84 percent, then about 82 and 83, clustering around 80 percent. One resample, a win rate of 0.53 with average winner 157 and average loser 149, cuts that chance to about 60 percent. The number you compute sits above or below the chance you would have with the future distribution in hand.
What usually goes wrong
A backtest win rate is not the textbook p. Raw proportions, an exponentially weighted average, a time-series fit, and a filter are all estimates. The lecture's point is that a reading near 10 percent would have to be wildly wrong before the truth sat near 70 to 90 percent, and a very high reading has a similar cushion. Ignoring a system whose own estimates put the target almost out of reach throws away the survival check the trades can support.
Questions people ask
Why does infinite play ruin a winning game?
The model never stops. Any finite pile can be erased by a long enough run of losses, even when the average trade is positive.
Can the blotter show the true edge?
It can estimate the four pieces from closed trades. It cannot show the theoretical values, and a new regime can move all four at once.
What should a 60 percent hitting figure mean?
Treat it as a scenario under today's estimates, then recompute when the win rate or the payoff sizes shift. One resampled book in the lecture moved a similar figure from about 80 percent to about 60 percent with the same formula.