Step 03 · Stochastic Series

Random Sequences over Time

A single random quantity is a snapshot. Repeat the same experiment at regular moments and you obtain a sequence, and a sequence has a texture that no snapshot can show.

From set to sequence

A collection of separate draws is unordered; the only thing that matters is which values appear and how often. A sequence is different. The values are tied to moments, and the relation between neighbours becomes a fact about the record. Shuffle a sequence and you keep its values but destroy its texture, which is exactly why shuffling is a useful contrast when a reader wants to know whether neighbouring values carry information about one another.

Indexing time

Write the record as a list of values, each tagged with the moment at which it was observed. The index may be taken as a count of steps rather than a calendar date, and that abstraction is deliberate: the mathematics rarely cares whether the steps are seconds or seasons, only that they are evenly spaced and arranged in sequence. The list is the primary object of the later steps, and everything else is a way of summarising it.

A record plotted against a time index: the eye follows runs, swings and sudden turns that a frequency count alone would miss.

Runs, swings and the memory question

Read a plotted record from left to right and you will notice patterns. Values above the centre may arrive in runs rather than singly; the line may swing slowly from high ground to low and back instead of darting at random. These features are the grammar of a sequence, and they raise a precise question. Does the value at one moment tell you anything about the value at the next? If the answer is no, the record is a pure shuffle and carries no memory. If the answer is yes, the record has a structure to be measured.

Time dependence described simply

The simplest summary of memory is the tendency of neighbouring values to be similar or dissimilar. A record whose neighbours resemble one another is called positively associated in time; one whose neighbours tend to alternate is called negatively associated. A reader does not need a full model to detect either tendency. Counting how often a sign change occurs, or how long typical runs last, already gives a rough and honest picture that a later, finer analysis can refine.

Why the snapshot view is not enough

Many classical results assume that observations are separate draws. Applied to a record with memory, those results can mislead badly, because they understate the uncertainty of a summary. The next two papers do the work of repairing that gap: first by naming the stability a record must show before averages settle, and then by measuring how quickly memory fades.

An abacus with rows of beads used for counting and sequence exercises — A counting frame makes the indexing plain: each row is a step, and each step is a moment tied to the one before it.

The shuffle as a measuring rod

One of the most useful habits in series work is to shuffle a record deliberately and compare the shuffle with the original. The shuffle keeps every value and destroys every relation between neighbours. If a summary of the original sits close to the same summary computed from the shuffle, the summary was describing the values alone and not their sequence. If the two diverge, the sequence itself carries information, and the size of the divergence measures how much.

The comparison asks little and answers a real question. It also guards against a common error, which is to read a pattern into a record that a shuffle would have produced just as readily. A reader who has shuffled a record once and watched the pattern dissolve is a reader who will not be fooled by a coincidence twice.