Three Questions for Every Time-Series Observation
A time series records values in time order. To read one correctly, keep three questions separate: When was the observation recorded? What was its level? How did it change from the preceding time? The time index answers the first question, the level answers the second, and a difference between levels answers the third. Confusing these quantities can reverse the interpretation of a trend or attach a change to the wrong period.
For instance, if a hypothetical monthly measure is 42 in January and 47 in February, then 47 is the February level. The change from January to February is \(47-42=5\). The value 47 is not itself a change, and the change of 5 is not the level in either month.
The index might represent years, quarters, days, or another sequence of equally spaced times. The level retains the units of the measured quantity: if \(x_t\) is in thousands of dollars, then \(x_t\) is a level in thousands of dollars. Its first difference has the same units. A positive difference means the level increased over that interval; a negative difference means it decreased; zero means the two recorded levels are equal.
The notation \(\Delta x_t\) requires \(x_t\) and \(x_{t-1}\). If the first recorded index is \(t=1\), the series alone does not provide \(\Delta x_1\), because it does not include \(x_0\). The first difference that can be calculated from those observations is \(\Delta x_2=x_2-x_1\).
Calculating and Labeling First Differences
For observations at consecutive, equally spaced indices, subtract the earlier level from the later one. Keep the order consistent: current level minus preceding level. The index on \(\Delta x_t\) is the time at the end of the change interval, not the time at its beginning.
A difference is an absolute change in the original measurement units. It is not automatically a percentage change, nor does it describe what happened at every instant within an interval. If observations are quarterly, for example, \(\Delta x_t\) compares two consecutive quarterly observations; it does not reveal the path between them.
Worked Example: Quarterly Reserve Levels
A hypothetical insurer records a reserve measure in thousands of dollars at five consecutive quarter-ends. The observed levels are \(x_1=84\), \(x_2=91\), \(x_3=87\), \(x_4=96\), and \(x_5=102\). Find the level at \(t=3\), all available first differences, and the total change from \(t=1\) to \(t=5\).
Method. Read the level directly from the observation at the requested index. For each difference, subtract the previous quarter’s level from the current quarter’s level.
Level at \(t=3\). The recorded level is \(x_3=87\) thousand dollars.
Difference ending at \(t=2\). \(\Delta x_2=x_2-x_1=91-84=7\) thousand dollars.
Difference ending at \(t=3\). \(\Delta x_3=x_3-x_2=87-91=-4\) thousand dollars.
Difference ending at \(t=4\). \(\Delta x_4=x_4-x_3=96-87=9\) thousand dollars.
Difference ending at \(t=5\). \(\Delta x_5=x_5-x_4=102-96=6\) thousand dollars.
Total change. Subtract the first level from the last: \(x_5-x_1=102-84=18\) thousand dollars. Checking by adding the consecutive differences gives \(7+(-4)+9+6=18\) thousand dollars. Both calculations agree.
Common mistake avoided: \(x_3=87\) is the level at the third quarter, whereas \(\Delta x_3=-4\) is the change from the second quarter to the third. Reporting 87 as the change would confuse a level with a difference.
Why Consecutive Changes Add to the Total Change
The relationship between a sequence of first differences and an overall change follows directly from their definition. For observations at every integer index from \(a\) through \(b\), write each difference as the later level minus the preceding level:
Every intermediate level appears once with a positive sign and once with a negative sign, so those terms cancel. Only the ending level \(x_b\) and starting level \(x_a\) remain. This identity provides a useful check on calculations: the sum of all consecutive differences between two observed endpoints must equal the difference between those endpoints.
The sum describes the net change over the full span. It does not mean the level moved in one direction throughout. In the reserve example, the level first rose, then fell, and then rose twice; the net increase was still 18 thousand dollars.
Worked Example: Monthly Activity and the Interval Ending at Each Index
A hypothetical service team counts new requests at the end of each month. The recorded levels are 250 in January, 240 in February, 258 in March, and 252 in April. Calculate the changes, identify the March level and the change ending in March, and find the net change from January to April.
Method. The levels are request counts. Subtract the previous month’s count from the current month’s count for each change.
Let January through April have indices \(t=1,2,3,4\), respectively. Then \(x_1=250\), \(x_2=240\), \(x_3=258\), and \(x_4=252\).
February change. \(\Delta x_2=x_2-x_1=240-250=-10\) requests.
March change. \(\Delta x_3=x_3-x_2=258-240=18\) requests.
April change. \(\Delta x_4=x_4-x_3=252-258=-6\) requests.
Requested March quantities. The March level is \(x_3=258\) requests. The change ending in March is \(\Delta x_3=18\) requests, an increase from February.
Net January-to-April change. \(x_4-x_1=252-250=2\) requests. As a check, the consecutive changes sum to \(-10+18-6=2\) requests.
Common mistake avoided: The March change is not the change from March to April. That later change ends in April and is \(\Delta x_4=-6\). The subscript identifies the endpoint of the interval.
Gaps in the Time Index
The subtraction \(x_t-x_{t-1}\) is a first difference only when both consecutive observations are available. If the series skips an index, subtracting the nearest recorded earlier level still gives a total change between those observations, but it does not give a one-period first difference. The elapsed time matters when interpreting the result.
For a gap spanning two periods, the endpoint difference describes the net change over those two periods. Dividing it by two gives an average change per period across the span, but that average does not establish what the change was in either individual period. The missing observation cannot be recovered from the endpoints alone.
Worked Example: A Missing Quarterly Observation
A hypothetical portfolio measure, in thousands of dollars, is recorded at indices \(t=0,1,3,4\): \(x_0=500\), \(x_1=520\), \(x_3=548\), and \(x_4=541\). Find the change from \(t=0\) to \(t=1\), the total change from \(t=1\) to \(t=3\), the average change per quarter over that two-quarter span, and the change from \(t=3\) to \(t=4\). State what cannot be determined.
Method. Use the first-difference formula when the indices are consecutive. When they are not consecutive, calculate the endpoint difference and label the full span.
From \(t=0\) to \(t=1\). These are consecutive indices, so \(\Delta x_1=x_1-x_0=520-500=20\) thousand dollars.
From \(t=1\) to \(t=3\). The total change is \(x_3-x_1=548-520=28\) thousand dollars over two quarters. The average change per quarter across that span is \(28\div2=14\) thousand dollars per quarter.
From \(t=3\) to \(t=4\). These indices are consecutive, so \(\Delta x_4=x_4-x_3=541-548=-7\) thousand dollars.
What cannot be determined. The observation \(x_2\) is missing. Therefore, neither \(\Delta x_2=x_2-x_1\) nor \(\Delta x_3=x_3-x_2\) can be calculated. The total increase of 28 thousand dollars does not show how that increase was split between the two quarters.
Check. The change from \(t=0\) to \(t=4\) is \(541-500=41\) thousand dollars. The changes over the observed consecutive spans and the two-quarter gap add to \(20+28-7=41\) thousand dollars, as required.
Common mistake avoided: Calling \(548-520=28\) the one-quarter change ending at \(t=3\) ignores the missing time index. It is a two-quarter endpoint change, not an observed first difference.
Common Interpretation Traps
Before reporting a level or change, check the index, subtraction order, interval, and units. These checks address the most common errors in short time-series calculations:
- Reversing the subtraction. The first difference is current minus previous, \(x_t-x_{t-1}\). Reversing the order changes the sign and the interpretation.
- Using a change as a level. A value such as \(x_4=96\) describes the level at \(t=4\); \(\Delta x_4=9\) describes its change from \(t=3\).
- Attaching the change to the wrong time. \(\Delta x_t\) describes the transition from \(t-1\) to \(t\), not the transition from \(t\) to \(t+1\).
- Ignoring units or elapsed time. A difference in thousands of dollars remains in thousands of dollars. A two-period endpoint change should not be labeled a one-period change.
- Inferring a path from endpoints. Knowing the beginning and ending levels does not reveal intermediate levels or whether the series increased at every step.
Check Your Understanding
For each problem, distinguish the level from the change and label the time span carefully.
- A hypothetical weekly count has levels \(x_1=31\), \(x_2=29\), \(x_3=35\), and \(x_4=34\). State the level at \(t=3\), the change ending at \(t=3\), and the net change from \(t=1\) to \(t=4\). (Answer: \(x_3=35\); \(\Delta x_3=+6\); net change \(=+3\).)
- Daily readings are \(x_0=18.0\), \(x_1=18.5\), \(x_2=17.9\), and \(x_3=19.2\). Find the changes ending at \(t=2\) and \(t=3\), then check their sum against the change from \(t=1\) to \(t=3\). (Answer: \(\Delta x_2=-0.6\), \(\Delta x_3=+1.3\), and both the sum and endpoint change are \(+0.7\).)
- A hypothetical quarterly measure is observed at \(t=2,3,5,6\), with levels \(70,75,82,79\), respectively. Find the change from \(t=3\) to \(t=5\), its average per quarter, and the change ending at \(t=6\). Can the individual change from \(t=4\) to \(t=5\) be determined? (Answer: total change \(=+7\) over two quarters; average \(=+3.5\) per quarter; \(\Delta x_6=-3\); the individual change from \(t=4\) to \(t=5\) cannot be determined.)
- A hypothetical monthly index has levels \(x_1=96\), \(x_2=101\), \(x_3=99\), and \(x_4=108\). A report calls \(99\) the change ending at \(t=3\). Correct the report and calculate the actual change ending at \(t=3\) and the total change from \(t=1\) to \(t=4\). (Answer: \(99\) is the level \(x_3\); \(\Delta x_3=-2\); total change \(=+12\).)