Constructing A Random Walk

From Increments to a Random Walk

A time series can change from one time point to the next, and its level at a later time reflects the accumulation of all those changes. A random walk is a simple model for this construction: begin at a specified value, then add one increment at each step. The increments may be positive, negative, or zero, so the path can rise, fall, or remain level.

The key calculation is recursive. At each time, the current value equals the preceding value plus the increment for that step. This makes it important to keep three quantities distinct: the starting value, the increments, and the resulting sequence of levels. The starting value is not itself an increment, and an increment is a change rather than a level.

Definition: A random walk $X_0,X_1,\ldots,X_n$ is constructed from a starting value $X_0$ and successive increments $\varepsilon_1,\varepsilon_2,\ldots,\varepsilon_n$ by the recursion $$ X_t=X_{t-1}+\varepsilon_t,\qquad t=1,\ldots,n. $$ Here $X_t$ is the level at time $t$, and $\varepsilon_t$ is the change from time $t-1$ to time $t$. A random walk is stochastic when its increments are random variables; a particular constructed path uses their realized values.

The term “random” refers to the uncertain increments in the model, not to a requirement that every observed change look unpredictable. For a concrete path calculation, the increments may simply be supplied as numbers. The recursion then determines each level without needing to calculate probabilities or distributional properties.

Some specifications include a constant drift $c$ as well as a random shock $\varepsilon_t$. In that case, the recursion is

X_t=X_{t-1}+c+\varepsilon_t.

The total change in period $t$ is then $c+\varepsilon_t$, not just $\varepsilon_t$. If a problem gives a sequence of total increments directly, add those increments as given; do not add a drift again unless the model specifies that the listed values are shocks excluding drift.

Why the Accumulated-Sum Formula Works

Repeated use of the recursion shows that a level contains the starting value and every increment up to that time. The first few steps make the pattern visible:

\begin{aligned} X_1 &= X_0+\varepsilon_1,\\ X_2 &= X_1+\varepsilon_2 = X_0+\varepsilon_1+\varepsilon_2,\\ X_3 &= X_2+\varepsilon_3 = X_0+\varepsilon_1+\varepsilon_2+\varepsilon_3. \end{aligned}

In general, this gives $X_t=X_0+\sum_{j=1}^{t}\varepsilon_j$. The formula can also be justified by telescoping the successive changes. Rearranging the recursion gives $\varepsilon_j=X_j-X_{j-1}$, so

\sum_{j=1}^{t}\varepsilon_j =\sum_{j=1}^{t}(X_j-X_{j-1}) =(X_1-X_0)+(X_2-X_1)+\cdots+(X_t-X_{t-1}) =X_t-X_0.

Every intermediate level other than the first and last appears once with a plus sign and once with a minus sign, so those terms cancel. Therefore $X_t=X_0+\sum_{j=1}^{t}\varepsilon_j$. This result is a useful check, but the recursive calculation is often safer for avoiding an indexing error.

When there is a constant drift $c$, the same argument accumulates the drift once per step:

X_t=X_0+tc+\sum_{j=1}^{t}\varepsilon_j.

Constructing a Path One Step at a Time

A reliable calculation starts by identifying the time index of the initial value and the meaning of each supplied change. Then apply each increment once, in order. A short table is especially helpful when the increments have mixed signs.

Record the starting value.
Write $X_0$ at time zero. Do not include it among the increments.

Identify each period's total change.
Use the stated increment directly, or combine drift and shock if the model defines the change as $c+\varepsilon_t$.

Update the level in order.
For each time $t$, calculate $X_t=X_{t-1}+$ that period's total change.

Check the accumulated change.
Compare the final level with the starting value plus the sum of all total changes.

Worked Example: Accumulating Mixed Increments

A hypothetical index begins at $X_0=12$. Its next four increments are $3,-2,4,-5$. Construct the levels $X_1$ through $X_4$, and verify the final level using the accumulated-sum formula.

Method. Apply each listed increment once, in its stated order. There is no separate drift in this model.

At time 1, add the first increment: $X_1=X_0+\varepsilon_1=12+3=15$.

At time 2, add the second increment to the current level: $X_2=X_1+\varepsilon_2=15+(-2)=13$.

At time 3: $X_3=X_2+\varepsilon_3=13+4=17$.

At time 4: $X_4=X_3+\varepsilon_4=17+(-5)=12$. Thus the path is $12,15,13,17,12$, with $X_0$ shown first.

Check. The total change is $3-2+4-5=0$, so the accumulated-sum formula gives $X_4=12+0=12$, agreeing with the recursive calculation. The common mistake avoided here is subtracting a negative increment twice: the second increment is $-2$, so adding it means $15+(-2)=13$.

Worked Example: Including Drift and Shocks

A hypothetical weekly index starts at $X_0=100$. Its model has a drift of $c=1.5$ index points per week and three realized shocks $\varepsilon_1=-3$, $\varepsilon_2=2$, and $\varepsilon_3=-0.5$. Find each weekly level and the total change over the three weeks.

Method. The model says that each period's total change is $c+\varepsilon_t$. Compute those changes first: $1.5+(-3)=-1.5$, $1.5+2=3.5$, and $1.5+(-0.5)=1.0$.

The first level is $X_1=100+(-1.5)=98.5$. The second is $X_2=98.5+3.5=102.0$. The third is $X_3=102.0+1.0=103.0$. Thus the levels are $100,98.5,102.0,103.0$.

Check. Across three weeks, the drift contributes $3(1.5)=4.5$. The shocks sum to $-3+2-0.5=-1.5$. The total change is $4.5+(-1.5)=3.0$, and $X_0+3.0=103.0$, as calculated. The common mistake is treating the shocks as the full increments and omitting drift; that would incorrectly produce a final level of $100-1.5=98.5$.

Worked Example: Recovering a Missing Increment

A hypothetical daily balance follows a random walk with starting value $X_0=250$. The first two increments are $-4$ and $7$. The observed level after the third increment is $X_3=242$, but that third increment is not recorded. A fourth increment is $-3$. Find the missing increment, the intervening levels, and $X_4$.

Method. First construct the levels for the known increments. The first is $X_1=250+(-4)=246$, and the second is $X_2=246+7=253$.

Let the missing third increment be $\varepsilon_3$. The recursion at time 3 gives $242=253+\varepsilon_3$, so $\varepsilon_3=242-253=-11$. The missing change is a decrease of 11.

Now apply the fourth increment: $X_4=X_3+\varepsilon_4=242+(-3)=239$. The full path is $250,246,253,242,239$, and the increments are $-4,7,-11,-3$.

Check. The increments sum to $-4+7-11-3=-11$. Adding this net change to $250$ gives $239$, which matches $X_4$. The common mistake is to use $X_0$ rather than $X_2$ when solving for the third increment; each increment changes the immediately preceding level.

Reading the Construction Correctly

A random walk's sequence of levels is not the same thing as its sequence of increments. For example, if the levels are $20,23,21$, the increments are $3$ and $-2$, not $23$ and $21$. In general, a change is found by subtracting consecutive levels:

\varepsilon_t=X_t-X_{t-1}.

This relation works in both directions: increments build the levels, and consecutive levels reveal the increments. It also provides a practical consistency check. If a claimed increment does not equal the difference between its two neighboring levels, at least one entry in the path is wrong.

Pay close attention to indexing. With a starting value $X_0$, the first listed increment produces $X_1$; $n$ increments produce $X_n$. The starting value is not an extra period. If the initial observation is labelled $X_1$ instead, follow the question's labels consistently, but the same add-one-change-at-a-time rule applies.

These calculations matter because many time-series questions describe observed changes but ask about levels, or provide levels and ask about changes. A clear record of the starting point and each update prevents sign errors and misplaced time steps. For a stochastic model, a realized path is constructed in exactly this way; the fact that increments are random does not change the arithmetic once their realized values are known.

Common Mistakes

Confusing a level with an increment. The level is the value at a time; the increment is the change since the preceding time.
Starting from the wrong value. Each update uses $X_{t-1}$, not the original $X_0$ repeatedly without including earlier changes.
Misreading a negative increment. Adding $-4$ means subtracting 4; it does not mean subtracting a negative number.
Adding drift twice or not at all. Determine whether supplied values are total increments or shocks to which drift must be added.
Counting the starting value as a step. Starting at $X_0$, the first increment gives $X_1$, and $n$ increments give $X_n$.

Key takeaway: Start at $X_0$, add the total change for each period exactly once, and check that the final level equals the starting value plus all accumulated changes.

Check Your Understanding

For each problem, identify whether the stated changes are total increments or shocks, then construct or check the requested levels.

A hypothetical reserve index starts at $X_0=18$, and its total increments are $-2,5,-1$. Find $X_1,X_2,X_3$. (Answer: $X_1=16,\ X_2=21,\ X_3=20$)
A hypothetical weekly score starts at $X_0=50$, has drift $c=0.8$, and has shocks $1.2,-2,0.5$. Find the three total increments and the final level. (Answer: Total increments $2,-1.2,1.3$; $X_3=52.1$)
A process starts at $X_0=7$, has first increment $4$, and is at $X_2=9$. Find the second increment. If the third increment is $6$, find $X_3$. (Answer: $\varepsilon_2=-2;\ X_3=15$)
A hypothetical index starts at $X_0=120$. Its model has drift $c=-0.5$ per period and shocks $2,-1$. Find $X_1$ and $X_2$, then verify the net change using the sum of drift and shocks. (Answer: $X_1=121.5,\ X_2=120.0$; net change $0$)
A recorded path is $X_0=30,\ X_1=26,\ X_2=29,\ X_3=24$. Find its three increments and verify the final value from their sum. (Answer: Increments $-4,3,-5$; their sum is $-6$, and $30+(-6)=24$)

What You'll Learn