Time Series Analysis for SSC CGL — Components, Models, and Trend Methods

Concept

A time series is a sequence of observations recorded at successive, equally-spaced points in time. Think of your electricity bill — it fluctuates every month because of multiple forces pulling it in different directions simultaneously: a long-run rise in tariff rates, a summer spike when ACs run full blast, a broader economic cycle, and the occasional random fault in your meter. Time series analysis is the formal process of separating and measuring each of those forces.

Here is the core idea: any observed value in a time series is not a single clean number — it is the combined result of four distinct forces acting at once. Your job in an exam question is to identify which force is being tested, isolate it, and compute it.

The four components (साख-स्थर, as they are sometimes called in Hindi statistics classrooms):

• Trend (T) — the long-run direction of the data. A steel plant's output rising by roughly 5% per year over two decades is a trend. It is smooth, persistent, and directional.
• Seasonal (S) — fluctuations that repeat within a fixed period, usually a year. Hotels fill up in December. Fertiliser sales spike before the Kharif season. The pattern repeats calendar-year after calendar-year.
• Cyclical (C) — wavelike swings that stretch across multiple years, usually linked to business cycles. These are longer than a year and less predictable than seasonal patterns.
• Irregular (I) — everything left over: a flood, a factory fire, a policy shock. Random, non-repeating, and unexplainable by the other three.

The analogy that works for most students: imagine a river (Trend) flowing downstream, with predictable tides (Seasonal) on top of it, longer oceanic swells (Cyclical) underneath, and random waves from a passing boat (Irregular) thrown in. The water level you measure at any moment is all four superimposed.

In SSC CGL Statistics, this topic generates direct formula-substitution questions, moving average computations, and model identification questions. None of them are conceptually hard — they are execution traps. Stay sharp on which years go into which average, and which formula belongs to which model.

Deep Dive

The Two Compositional Models

Time series theory uses two standard models to describe how the four components combine.

Additive Model: $Y = T + S + C + I$

Use this when the amplitude of seasonal swings stays roughly constant regardless of where the trend is. If December always contributes an extra ₹500 irrespective of whether the base trend is ₹2,000 or ₹5,000, you are in additive territory.

Multiplicative Model: $Y = T \times S \times C \times I$

Use this when seasonal swings grow proportionally with the trend — December contributes 25% extra whether the base is ₹2,000 or ₹5,000. Most real-world economic data (sales, production, income) is multiplicative. In SSC CGL questions, unless told otherwise, assume multiplicative.

In the multiplicative model, S, C, and I are expressed as indices (ratios, not absolute values). An S of 1.2 means 20% above the average seasonal level. A C of 0.9 means 10% below the trend-seasonal baseline. T is in the original unit (tonnes, ₹, etc.).

Methods of Measuring Trend

1. Free-Hand Curve Method Draw a smooth curve through the data by eye. Completely subjective, no calculations. Only mentioned in theory questions; never used in numerical problems.

2. Method of Semi-Averages Split the data into two equal halves, compute the mean of each half, plot these two points, and join them with a straight line. Objective but crude — gives only a straight-line trend. Step count: 4 steps. Fast for rough estimates, never used when least squares is an option.

3. Method of Moving Averages Replace each value with the average of a group of consecutive values centred around it. This smooths out short-term fluctuations.

For an odd-period moving average (3-year, 5-year): The average is centred exactly at the middle year.

• 3-year MA for year $t$ = $\frac{Y_{t-1} + Y_t + Y_{t+1}}\{3\}$

For an even-period moving average (4-quarter, 12-month): The average falls between two time points, so you must centre it by taking a 2-point moving average of the moving averages.

• First: compute overlapping 4-quarter totals
• Then: average pairs of consecutive totals and divide by 8 (or equivalently, take the average of two overlapping 4-quarter averages)

This centring step is where most exam errors happen. If the question asks for a 3-year MA, no centring needed. If it asks for a 4-quarter MA, check whether centring is required.

4. Method of Least Squares This is the gold standard. It fits a mathematical equation — usually a straight line — to the data by minimising the sum of squared deviations: $\sum (Y - \hat{Y})^2 \text{ is minimised}$

The resulting linear trend equation is: $\hat{Y} = a + bX$

where:

• $a$ = trend value at the origin (X = 0)
• $b$ = annual increment (slope)
• $X$ = number of years from the origin

The origin year is stated in the problem. To find the trend value for any year, compute X = (target year − origin year) and substitute.

This method is objective, mathematical, and gives the best-fit line. It is the answer whenever any question asks "which method minimises squared deviations."

Seasonal Variation: Link Relative Method

The link relative method (chain base method) measures seasonal variation by expressing each period's value as a percentage of the preceding period's value:

$\text{Link Relative} = \frac{\text{Current period value}}{\text{Previous period value}} \times 100$

Seasonal indices are then computed from these link relatives. This is the exam answer for "which method measures seasonal variation using ratios of successive values."

Decomposing a Multiplicative Series

If you are given Y, T, S, and I and asked for C: $C = \frac{Y}{T \times S \times I}$

If you are given Y, T, S, and C and asked for I: $I = \frac{Y}{T \times S \times C}$

Rearrange the fundamental equation algebraically. No new formula needed — just isolate the unknown component.

Memory Tricks & Shortcuts

Fast-Solving Framework

When you see a time series question in the exam hall, run this decision tree:

Step 1 — Identify question type:

• "Which component / model?" → Recall definitions, use elimination (TSCI mnemonic, multiplicative vs additive rule)
• "Compute moving average" → Identify period (odd/even), select correct window of values, average them
• "Find trend value using least squares" → Compute X = target year − origin year, substitute in $\hat{Y} = a + bX$
• "Find missing component" → Write $Y = T \times S \times C \times I$, isolate unknown by dividing Y by the product of known components

Step 2 — Watch for traps:

• 3-year MA: centre year uses its immediate neighbours. Do not include 4 values.
• 4-quarter MA: check if centring is needed. If only one 4-quarter average is possible, report it directly.
• Least squares X: always computed from the stated origin year, not from position in a table.
• Multiplicative model: S, C, I are indices (around 1.0), not raw values.

Step 3 — Calculate and match: Most questions require at most two arithmetic operations. If your working runs to five lines, you have likely misidentified the question type — go back to Step 1.

Solved PYQs

Why this question: Tests the most fundamental moving average computation — getting the window boundaries right is the only challenge.

Solving path: Data: 2019=4, 2020=6, 2021=8, 2022=10, 2023=12. For 3-year MA centred at 2021, the window is [2020, 2021, 2022] = [6, 8, 10]. Average = 24/3 = 8. The trap: some students include 2019 and compute (4+6+8)/3 = 6 — that is the MA for 2020, not 2021. Always centre the window on the year asked.

Why this question: Tests conceptual identification of the least squares method — a pure definition question answered in under 5 seconds with the elimination trick.

Solving path: The phrase "minimising the sum of squares of deviations" is the definition of least squares. Eliminate moving averages (smooths, doesn't minimise squared deviations), semi-averages (rough, no minimisation), and free-hand curve (subjective). Answer: Method of Least Squares.

Why this question: Tests the multiplicative model formula directly — the most commonly tested compositional model in SSC CGL Statistics.

Solving path: Two models to remember. Multiplicative: $Y = T \times S \times C \times I$ (all four multiplied). Additive: $Y = T + S + C + I$ (all four added). The question states "multiplicative model", so the answer is $Y = T \times S \times C \times I$. Scan the options: only option A matches exactly. Options B and C are hybrids that do not correspond to any standard model — eliminate immediately.

Why this question: Tests least squares forecasting — the most direct application of the trend equation. Straightforward substitution once X is correctly computed.

Solving path: Origin year: 2020 (X=0). Target year: 2023. X = 2023 − 2020 = 3. Trend equation: $\hat{Y} = 50 + 3X = 50 + 3(3) = 50 + 9 = 59$ thousand tonnes. The trap: using X=4 because 2023 is "the 4th year after 2019" — irrelevant. Only the stated origin (2020) matters.

Why this question: Tests multiplicative model component isolation — you are given three components and must find the fourth by rearranging $Y = T \times S \times C \times I$.

Solving path: $Y = T \times S \times C \times I = 200 \times 1.2 \times 0.9 \times 1$. Compute step by step: $200 \times 1.2 = 240$, then $240 \times 0.9 = 216$, then $216 \times 1 = 216$. Answer: 216. Trap: multiplying only T and S (getting 240) and selecting that option — always include all four components even when I = 1.

Why this question: Tests the link relative method — a definition question that trips students who confuse it with cyclical measurement.

Solving path: The link relative method expresses each period's value as a percentage of the preceding period's value, producing seasonal indices. It is used for seasonal variation, not cyclical (which has no standard "link" method). The keyword in the method name — "link" — signals chaining successive periods, which is a within-year repetitive pattern (seasonal). Eliminate Irregular (random), Cyclical (multi-year), and Secular Trend (long-run direction). Answer: Seasonal Variation.

Common Mistakes

• Using the wrong window for moving averages. For a 3-year MA of year $t$, the window is $[t-1, t, t+1]$. A frequent error is shifting the window to $[t, t+1, t+2]$, which gives the MA for year $t+1$. Always centre on the year asked.

• Confusing additive and multiplicative model structures. In the additive model, S, C, and I are absolute deviations (can be negative). In the multiplicative model, they are indices around 1.0 (S = 1.2 means 20% above average, never a raw sales figure). Plugging a raw-unit seasonal value into a multiplicative formula will give a wildly wrong answer.

• Computing X from rank instead of year difference. When the trend equation says "X = 0 for year 2020", X for 2023 is 3, not 4. Students who number years sequentially (2019=1, 2020=2, 2021=3 ...) and then use rank 5 for 2023 make this error consistently.

• Skipping the centring step in even-period moving averages. A 4-quarter moving average falls between quarters. If a question asks for the centred MA at a specific quarter, you must average two overlapping 4-quarter averages. Reporting the single 4-quarter average without centring is the standard trap for even-period MA questions.

• Equating cyclical and seasonal variation. Seasonal variation repeats within one year (monthly, quarterly patterns). Cyclical variation spans multiple years and is linked to business cycles. In exam questions, the time horizon in the question's description distinguishes the two: "every summer" → seasonal; "over a 5-year economic expansion" → cyclical.

• Forgetting that I = 1 still needs to appear in multiplication. When I = 1, it does not change the product, but writing $Y = T \times S \times C$ without including I in your reasoning sometimes causes you to misidentify which component you are solving for when the question asks you to isolate a specific one.

Related Topics

• measures of central tendency
• regression analysis
• correlation
• index numbers
• measures of dispersion
• probability distributions

Practice on SarkariRise