The Math of Luck, Risk, and Daily Choices
Summary
This article explores the mathematics of stochastic processes, which describe systems evolving randomly over time. It explains concepts like random walks, emphasizing that drift from the starting point grows with the square root of time and random walks tend to return to their origin. It introduces the concept of Markov chains and stationary distributions as eigenvectors for long-term behavior. Finally, it defines martingales as fair games where strategy cannot alter the expected outcome, illustrating that winning requires an inherent edge in the game itself, not just clever play.
Key Insights
Position in a random walk typically deviates from zero by the square root of time.
Contrary to intuition, the typical distance from the starting point in a random walk grows with the square root of the number of steps. For 100 steps, it's about 10 units; for 10,000 steps, about 100 units.
A random walk's position grows without bound but much slower than time, and it inevitably returns to the starting line.
Two key consequences of a random walk are that its position grows without bound (though slowly) and that it is guaranteed to return to its starting point infinitely often. It's impossible to get permanently stuck in a losing region.
Winning chances in a fair game are determined by the ratio of what you risk to what you chase.
This result (1/3 chance of winning) is independent of the coin's fairness. It highlights that the ratio of the stop-loss amount to the profit target dictates success probability, not the edge in the game itself.
Long-term behavior of a Markov chain is its stationary distribution, equivalent to the eigenvector with eigenvalue 1.
Assuming the system settles, the long-run state distribution is a probability distribution that remains unchanged after one step. This corresponds to the eigenvector associated with eigenvalue 1 of the transition matrix, guaranteed by the Perron-Frobenius theorem for positive matrices.
The optional stopping theorem states that strategy cannot beat a fair game.
For a martingale with a valid stopping rule (dependent only on past information and bounded in time), the expected balance upon stopping is equal to the starting balance. This means no strategy can guarantee a gain or loss.
Winning requires an inherent edge in the game, not clever playing or stopping rules.
The optional stopping theorem implies that no amount of strategy or cleverness in when to quit can create an advantage in a fair game. To achieve a positive expectation, one must have an edge built into the game's rules themselves.
Luck is governed by rules (distributions), not the absence of rules.
Luck is not random chaos but a set of underlying rules governing distributions. Understanding these structures, not just individual outcomes, is key to navigating uncertainty and identifying where advantages truly lie.
Randomness has structure: spread grows as sqrt(time), with inevitable returns.
Even unpredictable outcomes follow structures. The deviation from the start grows as the square root of time, and random walks inevitably return to their origin infinitely often.
Fair games are unbeatable by strategy; winning requires an intrinsic edge.
No strategy or stopping rule can alter the expected outcome of a fair game (martingale). A positive expectation can only be achieved if the game itself offers an advantage.
Sections
Introduction to the Game and Stochastic Processes
A coin flip game with stopping conditions illustrates uncertainty over time.
The article presents a game involving flipping a fair coin: win $1 for heads, lose $1 for tails. The game stops when one reaches $100 profit or $50 loss. The question of winning chances highlights the counterintuitive nature of risk over time.
Stochastic processes model step-by-step evolution under uncertainty.
The article defines stochastic processes as collections of random variables indexed by time. These model phenomena like stock prices, weather, or bank balances where the next state is random but not arbitrary. The evolution can be continuous or discrete.
Two perspectives on stochastic processes: moment-by-moment vs. entire path.
Stochastic processes can be viewed as a value evolving randomly at each moment, or as a probability distribution across all possible complete future paths. While the first is easier to visualize, the latter is more mathematically useful.
Key questions for stochastic processes: past influence, long-term behavior, and extreme events.
When analyzing a stochastic process, three main questions arise: how much does the past predict the future, what is the long-term stable state (if any), and how likely are extreme outcomes.
The Random Walk
A simple random walk involves independent steps of +1 or -1 with equal probability.
The simplest stochastic process is a random walk, starting at zero, where each step is either +1 or -1 with a 50% probability, independent of previous steps. Plotting the running total creates a jagged line.
Position in a random walk typically deviates from zero by the square root of time.
Contrary to intuition, the typical distance from the starting point in a random walk grows with the square root of the number of steps. For 100 steps, it's about 10 units; for 10,000 steps, about 100 units.
A random walk's position grows without bound but much slower than time, and it inevitably returns to the starting line.
Two key consequences of a random walk are that its position grows without bound (though slowly) and that it is guaranteed to return to its starting point infinitely often. It's impossible to get permanently stuck in a losing region.
The Math Behind the Coin Game
Probability of winning a bounded random walk depends on the ratio of risk to reward.
In a game with barriers, the probability of hitting the upper barrier (gain B) before the lower barrier (loss A) is A / (A + B). For the game with +$100/-$50 barriers, this is 50 / (50 + 100) = 1/3.
Winning chances in a fair game are determined by the ratio of what you risk to what you chase.
This result (1/3 chance of winning) is independent of the coin's fairness. It highlights that the ratio of the stop-loss amount to the profit target dictates success probability, not the edge in the game itself.
Success probability in bounded random walks can be calculated using recursive relations and boundary conditions.
The probability function can be determined by knowing the probability at the boundaries (1 at the top, 0 at the bottom) and the recursive relation that the probability at any point is the average of the probabilities one step up and one step down.
Properties Enabling Analysis
Independent increments and stationarity simplify the analysis of stochastic processes.
Stochastic processes are easier to analyze when they have independent increments (past intervals don't affect future ones) and stationarity (interval distributions are constant over time).
Markov Chains and State Transitions
A Markov chain's future depends only on its present state, not its entire history.
A Markov chain is a process where the entire past's influence on the future is fully captured by the current value. The random walk is an example of a Markov process.
Transition matrices describe movement probabilities between states in finite Markov chains.
For finite state systems, a transition matrix (where rows sum to one) defines the probability of moving from any state to any other state in one step. Matrix powers predict multi-step probabilities.
Machine reliability can be modeled with a Markov chain using a transition matrix.
A machine's working/broken states can be modeled with a Markov chain. With a transition matrix defining probabilities of staying working, breaking, or being repaired, long-term behavior can be analyzed.
Long-Run Behavior: Stationary Distributions
Long-term behavior of a Markov chain is its stationary distribution, equivalent to the eigenvector with eigenvalue 1.
Assuming the system settles, the long-run state distribution is a probability distribution that remains unchanged after one step. This corresponds to the eigenvector associated with eigenvalue 1 of the transition matrix, guaranteed by the Perron-Frobenius theorem for positive matrices.
Eigenvalues less than 1 govern the fading of initial conditions in Markov chains.
The smaller eigenvalues (in magnitude) of the transition matrix dictate how quickly the system converges to its stationary distribution. Their magnitude less than 1 ensures initial conditions eventually become irrelevant.
Martingales and Fair Games
A martingale is a process where the expected value of the next step equals the current value.
A martingale formally defines a fair game: at any point, the expected value of the process in the next step is precisely the current value. This holds regardless of volatility.
The optional stopping theorem states that strategy cannot beat a fair game.
For a martingale with a valid stopping rule (dependent only on past information and bounded in time), the expected balance upon stopping is equal to the starting balance. This means no strategy can guarantee a gain or loss.
Winning requires an inherent edge in the game, not clever playing or stopping rules.
The optional stopping theorem implies that no amount of strategy or cleverness in when to quit can create an advantage in a fair game. To achieve a positive expectation, one must have an edge built into the game's rules themselves.
The coin game's probabilities can be derived from the optional stopping theorem.
Applying the optional stopping theorem to the coin game (expected final balance of zero) allows calculation of the winning probability: 100p - 50(1-p) = 0, yielding p = 1/3.
Key Takeaways
Randomness has structure: spread grows as sqrt(time), with inevitable returns.
Even unpredictable outcomes follow structures. The deviation from the start grows as the square root of time, and random walks inevitably return to their origin infinitely often.
Long-term behavior of state-limited systems is a unique, stable eigenvector equilibrium.
Systems with memory restricted to their current state (Markov chains) converge to a single, stable equilibrium distribution, regardless of starting conditions, determined by the transition matrix's eigenvector.
Fair games are unbeatable by strategy; winning requires an intrinsic edge.
No strategy or stopping rule can alter the expected outcome of a fair game (martingale). A positive expectation can only be achieved if the game itself offers an advantage.
Luck is governed by rules (distributions), not the absence of rules.
Luck is not random chaos but a set of underlying rules governing distributions. Understanding these structures, not just individual outcomes, is key to navigating uncertainty and identifying where advantages truly lie.
Ask a Question
*Uses 1 Wisdom coin from your coin balance
