Tutorials › Exam SRM - Statistics for Risk Modeling › Worked Analysis Of A Short Random Walk

Home  /  Tutorials  /  Exam SRM / Time-Series Foundations

Exam SRM · Time-Series Foundations · Tutorial 415 of 926

Worked Analysis Of A Short Random Walk

Practice reading random-walk sequences by separating changes from levels and checking what the observed path can—and cannot—show.

Beginner 10 min read

What You'll Learn

  • Calculate each increment from consecutive observed levels.
  • Verify a sequence using the sum of its increments.
  • Distinguish movement in levels from patterns in increments.
  • Calculate and interpret the mean and sample variance of observed increments.
  • Explain why a short path can be consistent with, but cannot establish, a random-walk model.
  • Avoid treating a rising sequence of levels as proof of a trend or a random walk.

Reading a Short Random-Walk Sequence

A random walk is often introduced through a model equation, but a data set presents a sequence of observed levels. To assess the sequence, first calculate how much the level changed at each step. Those increments are the quantities that the random-walk model describes; the levels are accumulated outcomes of earlier increments. Keeping the two separate prevents a common misreading: a steadily rising or falling path does not, by itself, show that each new change is steadily rising or falling.

This tutorial works through short sequences by calculating their increments, checking the arithmetic, and comparing the behavior of changes with the behavior of levels. A short observed path can be consistent with a random walk, but it cannot establish the model or prove that the increments are independent.

Definition: A random walk without drift has levels \(X_0,X_1,\ldots\) satisfying \(X_t=X_{t-1}+\varepsilon_t\), where the increments \(\varepsilon_t\) have mean zero and constant variance \(\sigma^2\), and are uncorrelated across time. If the model assumes independent increments, that independence is an additional assumption. The observed increment at time \(t\) is \(\Delta X_t=X_t-X_{t-1}\). A random walk with drift \(c\) instead satisfies \(X_t=X_{t-1}+c+\varepsilon_t\).

For a recorded sequence \(x_0,x_1,\ldots,x_n\), calculate \(n\) increments: \(x_1-x_0,\ x_2-x_1,\ldots,x_n-x_{n-1}\). The subscript \(0\) labels the starting observation; it does not represent an increment. A useful arithmetic check is that the sum of the increments must equal the final level minus the starting level.

$$ \sum_{t=1}^{n}(x_t-x_{t-1})=x_n-x_0 $$

The intermediate terms cancel: each level other than \(x_0\) and \(x_n\) appears once with a plus sign and once with a minus sign. This telescoping identity is true of every sequence, not just random walks. It checks the calculations, but it is not evidence that a random-walk model generated the data.

Why the Increments Matter

Under the zero-drift model, each new level is the previous level plus a new mean-zero increment. Consequently, the level at time \(t\) contains the accumulated effects of all increments up to that time. If the increments are uncorrelated and each has variance \(\sigma^2\), then

$$ X_t=X_0+\sum_{j=1}^{t}\varepsilon_j, \qquad E[X_t]=X_0, \qquad \operatorname{Var}(X_t)=t\sigma^2 $$

For the variance result, treat \(X_0\) as fixed. The variance of a sum is the sum of the variances plus twice the pairwise covariances. Here each variance is \(\sigma^2\) and the covariances between distinct increments are zero, so the total variance is \(t\sigma^2\). This explains why a random walk can wander widely in its levels even when its increments have no systematic upward or downward tendency. For independent increments the same variance conclusion holds; independence is stronger than the uncorrelatedness needed for this calculation.

In a short data sequence, the theoretical assumptions cannot be verified just by looking at the numbers. Instead, calculate the observed increments and describe what is compatible with the model. Check whether changes are being measured over consistent time steps; compare the increments rather than treating the level differences between nonconsecutive observations as single-step changes; and keep a model conclusion appropriately cautious.

Worked Example: Recovering Changes From Levels

A hypothetical index is observed at five equally spaced times, with levels \(40,43,41,45,44\). Calculate all one-step increments, check their sum, and state what the short path suggests about random-walk behavior.

Method. Subtract each level from the next level. The observations are \(x_0=40,\ x_1=43,\ x_2=41,\ x_3=45,\ x_4=44\).

Step 1: Calculate the increments. The four changes are \(43-40=3\), \(41-43=-2\), \(45-41=4\), and \(44-45=-1\). Thus the increment sequence is \(3,-2,4,-1\).

Step 2: Check the total change. The increment sum is \(3+(-2)+4+(-1)=4\). The endpoint difference is \(44-40=4\), so the calculation checks.

Step 3: Interpret cautiously. The levels end 4 units above their starting value, but the increments include both increases and decreases. Nothing in these four changes establishes independence or a zero-mean increment distribution. The observed path is compatible with a random walk; it does not prove that the model is correct.

Common mistake avoided: Calling the sequence “increasing” because its final level is higher than its initial level ignores the two negative increments. The path rises overall, but it does not rise at every step.

Summaries of a Short Increment Sequence

The increment mean and sample variance can summarize the observed changes, provided their limited role is clear. For increments \(d_1,\ldots,d_n\), their sample mean is \(\bar d=\sum_{t=1}^{n}d_t/n\), and their sample variance is \(s_d^2=\sum_{t=1}^{n}(d_t-\bar d)^2/(n-1)\), when \(n>1\). These are descriptive summaries of the observed increments. A sample mean different from zero does not by itself demonstrate drift, and a sample variance does not demonstrate that the increments are independent or uncorrelated.

Worked Example: Comparing Level Stability With Increment Behavior

A hypothetical monthly measure has levels \(30,28,31,29,32,30\) at six equally spaced times. Calculate the five increments, their mean and sample variance, and assess what can be said from this path.

Step 1: Find each change. The increments are \(28-30=-2\), \(31-28=3\), \(29-31=-2\), \(32-29=3\), and \(30-32=-2\). Thus they are \(-2,3,-2,3,-2\).

Step 2: Calculate the mean increment. Their sum is \(-2+3-2+3-2=0\), so \(\bar d=0/5=0\).

Step 3: Calculate the sample variance. Since the mean is zero, the squared deviations are \(4,9,4,9,4\). Their sum is \(30\), and \(s_d^2=30/(5-1)=7.5\).

Step 4: Interpret the result. The starting and ending levels are both 30, while the intervening levels range from 28 to 32. The zero mean increment is consistent with no net change over this particular path. However, the increments alternate between \(-2\) and \(3\), rather than looking like an obviously unstructured sequence. Five increments are far too few to establish a dependence pattern or rule out a random walk.

Common mistake avoided: A zero observed increment mean is not proof that the process has zero drift. It is a summary of these five changes only; the assumed distribution of future changes cannot be determined from this mean alone.

Separate a Rising Path From a Rising Increment Pattern

A sequence of levels can rise overall even when the increments vary in size and include a decline. Conversely, increments that are all positive in a short sample do not establish that the underlying process has a positive drift. The distinction is especially important when a question describes a path informally as “trending.” Identify what is increasing: the levels, the increments, or a summary of the increments. Those are different claims.

Worked Example: A Rising Sequence Does Not Mean Constant Growth

A hypothetical account balance, measured in consistent units at six equally spaced observations, is \(200,202,205,204,208,210\). Calculate the increments, their mean and sample variance, and explain whether the sequence demonstrates a positive-drift random walk.

Step 1: Calculate the increments. The changes are \(202-200=2\), \(205-202=3\), \(204-205=-1\), \(208-204=4\), and \(210-208=2\). Thus the increments are \(2,3,-1,4,2\).

Step 2: Check the total and mean. The changes sum to \(2+3-1+4+2=10\), which equals \(210-200=10\). There are five increments, so their mean is \(10/5=2\).

Step 3: Calculate the sample variance. The deviations from 2 are \(0,1,-3,2,0\). Their squares sum to \(0+1+9+4+0=14\). Therefore, \(s_d^2=14/(5-1)=3.5\).

Step 4: Interpret the evidence. The final level is 10 units above the initial level, and the average observed increment is 2. But one increment is negative, and the increments are not all equal to 2. The sequence is compatible with a path having positive average change, but five observations do not establish a positive-drift model. A finite-sample average can differ from the population mean.

Common mistake avoided: The average increment is not the same as every increment. Treating the value 2 as a fixed increase at every step would predict levels \(200,202,204,206,208,210\), which do not match the observed sequence.

What a Short Path Can Establish

A careful analysis of a short sequence can establish its observed step-by-step changes, its net change, and descriptive summaries such as the mean increment. It can also identify visible features that deserve caution, such as a repeating alternation or a long run of changes with the same sign. Those features are descriptions of a small sample, not conclusive tests of the model.

A random-walk interpretation requires assumptions about the increment process, including its mean and dependence over time. A handful of levels cannot verify those assumptions. Therefore, use language such as “consistent with a random walk” when the calculations do not contradict the idea, rather than claiming that the sequence proves it. The same caution applies to apparent trends: an upward endpoint difference is an observed fact, not by itself evidence that future increments have positive mean.

Key takeaway: Calculate every one-step increment from consecutive levels, verify that the increments sum to the endpoint difference, and interpret the changes separately from the accumulated levels. A short path can illustrate random-walk behavior, but cannot prove the assumptions that define the model.

Check Your Understanding

For each sequence, calculate the requested changes and use the results carefully.

  1. A hypothetical sensor records levels \(7,9,8,11\) at equally spaced times. Find all increments, their sum, and the endpoint difference. (Answer: Increments \(2,-1,3\); sum \(4\); endpoint difference \(11-7=4\).)
  2. A reserve measure starts at 50 and then has three successive changes of \(-3,4,-2\). Find the four levels and the net change. (Answer: Levels \(50,47,51,49\); net change \(-1\).)
  3. A hypothetical index has levels \(15,14,18,17,20\). Find its increments, their mean, and sample variance. (Answer: Increments \(-1,4,-1,3\); mean \(1.25\); sample variance \(20.75/3\approx6.9167\).)
  4. Levels \(100,101,100,102\) are observed at equally spaced times. Find the mean increment and state whether the result proves a positive drift. (Answer: Increments \(1,-1,2\), mean \(2/3\); it does not prove positive drift.)
  5. A sequence starts at 25 and has increments \(-2,3,-4,-1\). Find all successive levels and verify the total change. (Answer: Levels \(25,23,26,22,21\); increment sum \(-4\), equal to endpoint difference \(21-25=-4\).)

Previous

Next

Back to Tutorials
Clinical Biostats