Walking Through an
Uncertain Forest
Imagine you are standing at the edge of a vast, ancient forest. You have no map. The only rule is that every few seconds, you must take one step — forward, left, or right. Where you end up is not fate; it is the accumulated result of every uncertain choice made along the way. This is a stochastic process.
A Single Random Variable: Your First Step
Before you step into the forest, you pause. You look at the ground ahead. There is a muddy path to the left (probability 40%), a stony trail to the right (30%), and a mossy clearing straight ahead (30%). You close your eyes, breathe, and step.
That one decision — that one moment of chance resolved into a single outcome — is a random variable. It has no history. It has no future. It simply is: one roll of the universe's dice, producing a value from a set of possibilities, each with its own weight.
A random variable is a snapshot. A single measurement. A point, not a path.
Independence: A Forest With No Memory at All
Now imagine the strangest possible forest. After every step, the trees rearrange themselves. The terrain you just crossed disappears. Where you can go next has absolutely nothing to do with where you have been — the mud and the stones and the moss are scattered fresh each time you look up.
This is independence. Your second step is not influenced by your first. Your tenth step knows nothing of your ninth. Each moment is a brand new random variable drawn from scratch, uncorrelated with all the others.
Independence is the clean theoretical baseline — the forest at its most forgetful. Most real-world paths are not like this, but it is the foundation on which everything else is built.
The Markov Property: A One-Step Memory
Now the forest becomes more realistic. The terrain persists — the muddy path you just walked has soaked your boots, and that matters. But here is the twist: only your current position determines where you can go next. Not the full history of every step you have ever taken. Just where you are right now.
Did you arrive here from the north or the south? From a sunny clearing or a dark thicket? It does not matter. The forest only sees your present footprint.
The Markov property states that the future is independent of the past, given the present. Your current position is a sufficient summary of everything relevant about your history. The forest is not amnesiac — it sees where you stand — but it is incurious about how you arrived.
This elegant constraint makes the mathematics tractable, and it describes a surprising number of real processes: stock prices, board-game positions, weather states, the spread of a rumour.
Martingales: A Path With No Inherent Slope
You have been walking for hours. You stop and ask yourself: on average, will the next clearing I reach be higher, lower, or at the same elevation as where I stand now? If the answer is always the same elevation — regardless of where you are or how you got there — then the path you are walking is a martingale.
A martingale is a fair game. There is no systematic hill. No gravity pulling you down, no escalator lifting you up. Your best forecast for tomorrow's position is simply today's position. All the uncertainty is noise, not trend.
"The martingale path does not wander randomly — it wanders symmetrically. No uphill bias, no downhill slide. Just honest, undirected uncertainty."
Martingales are central to finance: a fair market price should be a martingale — the expected future price equals the current price, because any visible drift would be traded away the moment it appeared.
Stationarity: The Forest That Never Changes
You have now walked for days. And something strange and beautiful occurs to you: the forest looks exactly the same as it did on day one. Not that you have gone in a circle — you have not. But the statistical texture of the forest is identical. The distribution of clearings and thickets, the typical distances between trees, the average light level — all of it is the same.
This is stationarity. The rules of the forest do not change over time. The probability laws governing your walk today are the same laws that governed it yesterday and will govern it tomorrow.
Stationarity is enormously convenient for inference. Data gathered early in your walk tells you just as much about the forest's laws as data gathered later. The whole journey is, statistically speaking, one long repeated experiment.
Many real processes violate stationarity — climate is shifting, markets evolve, languages drift. Identifying when a process has stopped being stationary is one of the central challenges of modern data science.
From Probability to Stochastic Processes: Geometry Becomes Physics
Classical probability is like geometry. It deals in shapes — static, timeless, perfectly defined. A probability distribution is a shape: here are all the possible outcomes, here is how the weight is distributed across them. You can measure it, rotate it, describe its symmetry. It is beautiful and it is still.
Stochastic processes are like physics. Physics takes geometry's shapes and asks: what happens when they move through time? A particle does not just exist at a point — it has a trajectory, a velocity, a history, a future that depends on forces acting on it right now. The shapes are still there, but they are animated. They evolve.
The random variable is a geometric object: a distribution, a shape, a snapshot. The stochastic process is that same object set in motion — a whole sequence of random variables indexed by time, each one influencing or independent of the last, the entire sequence obeying rules that govern how randomness propagates forward.
"The forest was always there. Stochastic processes are what happens when you finally start walking through it."
Stochastic Processes &
Time Series Analysis
The distinction between a stochastic process and a time series is fundamental, and it maps cleanly onto the distinction between a scientific theory and an experiment.
The Theoretical Model: The Stochastic Process
A stochastic process is a mathematical framework that models a sequence of random variables indexed by time, capturing the probabilistic structure of how a system evolves. It is the underlying data-generating mechanism — the invisible machine that produces outcomes.
Examples of such models include ARMA processes (capturing autocorrelation in stationary data), GARCH models (capturing volatility clustering in financial returns), and Brownian motion (the continuous-time limit of a symmetric random walk). Each of these is a precise specification of how randomness accumulates and propagates across time.
The stochastic process is the grammar — the complete set of rules that governs what sentences are possible, and with what probability.
The Empirical Reality: The Time Series
A time series, on the other hand, is a concrete sequence of observations — typically a single realisation or sample path — obtained from a stochastic process over time. It is what we actually observe: a sequence of dated measurements of temperature, price, unemployment rate, or any other quantity evolving through time.
Crucially, we never observe the stochastic process itself — only one of its possible paths. The process is the ensemble of all paths that could have occurred; the time series is the single path that did occur.
The time series is a sentence — one utterance actually spoken, drawn from the infinite space of grammatically valid possibilities defined by the process.
The Relationship: Theory Grounding Empirical Practice
The connection between the two is the engine of statistical modelling and prediction. In practice:
Thus, stochastic processes form the theoretical foundation, while time series analysis is the empirical application. The analyst's task is to work backwards from a single observed path — the time series — to infer the properties of the invisible process that generated it.
This inferential challenge is deep precisely because the process is only ever observed once. We cannot re-run history to generate a second sample path from the same process. Instead, we exploit the structure of stationarity and ergodicity — walking long enough through one path eventually reveals the full statistical texture of the forest — to make inference from a single trajectory possible.
"Stochastic processes form the theoretical foundation, while time series analysis is the empirical application. The analyst works backwards from a single observed path to infer the invisible process that generated it."
A Summary Comparison
| Dimension | Stochastic Process | Time Series |
|---|---|---|
| Nature | Theoretical model | Empirical observation |
| What it is | Full probability law over all possible paths | One realised path from that law |
| Observable? | Never directly observed; always inferred | Directly measured and recorded |
| Role | Describes the data-generating mechanism | Used to estimate and validate the model |
| Examples | Brownian motion, ARMA, GARCH, Markov chains | Daily stock prices, temperature records, GDP |
| Analogy | The grammar (rules of the forest) | A sentence (one walk through the forest) |
The forest was always there. Stochastic processes describe the forest. Time series analysis is the art of mapping it from a single walk.