Time Series Analysis Questions for SSC CGL

What is a Time Series

A time series is a sequence of data points collected at regular intervals over time, arranged chronologically. Think of it like a patient's daily temperature readings or monthly rainfall data. The key feature is that observations are ordered by time, and consecutive observations are often dependent on each other. In SSC CGL, you'll see time series in economic data (GDP growth), weather patterns, stock prices, and population trends.

• Data points collected at fixed time intervals (daily, monthly, yearly, etc.)
• Order matters — rearranging data destroys the time series structure
• Used for forecasting, trend analysis, and understanding patterns
• Real-world examples: stock prices, rainfall, inflation rates, exam scores over years

Components of a Time Series

Every time series can be decomposed into four main components. Think of it as breaking down a song into melody, rhythm, harmony, and noise. The trend is the long-term direction, seasonality is the repeating pattern within a year, cyclical is the longer-term business cycle (3-10 years), and irregular/random is unpredictable noise. SSC CGL questions often ask you to identify which component is present in a given scenario.

• Trend (T): Long-term upward or downward movement over years
• Seasonal (S): Regular, repeating pattern within a year (e.g., ice cream sales peak in summer)
• Cyclical (C): Longer-term oscillations (3-10 years), like business cycles
• Irregular/Random (I): Unpredictable fluctuations due to shocks or events
• Additive model: Y = T + S + C + I (when components are independent)
• Multiplicative model: Y = T × S × C × I (when components interact)

Trend Analysis and Fitting

Identifying and estimating the trend is the first step in time series analysis. The trend tells you the overall direction of the data. SSC CGL tests your ability to recognize linear vs. non-linear trends and apply simple fitting methods. Linear trend is most common in exams — it assumes the data increases or decreases at a constant rate over time.

• Linear trend: Straight-line pattern, fitted using least squares method
• Non-linear trend: Polynomial, exponential, or logarithmic patterns
• Method of least squares: Minimizes sum of squared deviations from the fitted line
• Trend equation used for forecasting future values
• Always plot data first to visually identify trend direction and type

Seasonal Decomposition

Seasonality is a repeating pattern within a fixed period (usually 12 months for annual data). SSC CGL questions ask you to identify seasonal indices and understand how to remove seasonality from data. A seasonal index above 100 means that period is above average; below 100 means below average. This is crucial for forecasting because ignoring seasonality leads to poor predictions.

• Seasonal index: Ratio of actual value to trend value (in multiplicative model)
• Average seasonal index for a season = sum of indices for that season / number of years
• Seasonal indices must sum to 1200 for 12 months (or 100 × number of seasons)
• Deseasonalized data = Actual / Seasonal Index (removes seasonal effect)
• Use deseasonalized data to identify true trend without seasonal noise

Forecasting Methods

SSC CGL tests two main forecasting techniques: moving averages and exponential smoothing. Moving average is simple — take the average of the last n values. Exponential smoothing gives more weight to recent observations. Both are used when trend and seasonality are not strong, or after removing them. Choose the method based on data characteristics and forecast horizon.

• Simple Moving Average (SMA): Average of last n observations, smooth out noise
• Weighted Moving Average: Assign higher weights to recent data
• Exponential Smoothing: Uses smoothing constant (0 < α < 1) to weight recent vs. old data
• Higher α (closer to 1): More weight on recent data, responds quickly to changes
• Lower α (closer to 0): More weight on historical data, smoother forecast
• Choose n or α based on data volatility and forecast accuracy on test data

Stationarity and Differencing

A stationary time series has constant mean, variance, and autocorrelation over time — it doesn't have a trend or seasonality. Many forecasting methods require stationarity. If data is non-stationary (has trend), use differencing: subtract each value from the previous one. This removes the trend and makes the series stationary. SSC CGL may ask you to recognize non-stationary data and suggest differencing as a solution.

• Stationary series: Mean and variance don't change over time, no trend or seasonality
• Non-stationary series: Has trend, seasonality, or changing variance
• First differencing: D_t = Y_t - Y_(t-1) removes linear trend
• Seasonal differencing: D_t = Y_t - Y_(t-s) removes seasonal pattern (s = seasonal period)
• Check stationarity visually (plot) or using tests like ACF plot
• Differencing makes data suitable for methods like exponential smoothing

• Confusing seasonal and cyclical components — seasonal is fixed (12 months), cyclical is longer and irregular (3-10 years). A question about 'repeating every 6 months' is seasonal, not cyclical.
• Using additive model when multiplicative is correct — if seasonal effect grows with trend level (e.g., ice cream sales variation increases in hot years), use multiplicative Y = T × S × C × I.
• Forgetting to code time variable correctly — when fitting trend, code time as 0, 1, 2... or center it around mean. Don't use actual years like 2020, 2021.
• Choosing wrong moving average period — too small (n=2) gives noisy forecast, too large (n=12) lags behind actual changes. Match n to seasonal period or use cross-validation.
• Misinterpreting seasonal index — index of 120 means 20% above average, not 120% of average. Always remember SI is expressed as percentage of trend.
• Applying forecasting methods to non-stationary data without differencing first — moving averages and exponential smoothing assume stationarity. Always check and difference if needed.

• TSCI: Trend, Seasonal, Cyclical, Irregular — the four components. Remember: Trend is the 'direction', Seasonal is 'yearly pattern', Cyclical is 'business waves', Irregular is 'noise'.
• SMA vs. ES: SMA is Simple (equal weights), ES is Exponential (recent data heavier). Think 'Simple = Same weight, Exponential = Extra weight to recent'.
• Additive = Add (independent), Multiplicative = Multiply (dependent). If components interact, multiply; if separate, add.
• Differencing = Subtract to Stabilize. D_t = Y_t - Y_(t-1) removes trend; D_t = Y_t - Y_(t-12) removes seasonality.
• α (alpha) in exponential smoothing: High α (0.7-0.9) = Responsive (recent data rules), Low α (0.1-0.3) = Stable (history rules).

• SSC CGL Tier-II Paper-IV (Statistics) typically has 1-2 questions on time series. One question usually asks to identify components from a scenario (e.g., 'Sales peak every December' = seasonal). Another may ask to calculate moving average or seasonal index.
• Recent papers (2022-2024) show preference for practical scenarios over pure theory. Expect questions like 'Monthly rainfall data shows pattern repeating every 12 months — identify component' or 'Calculate 4-month moving average for forecasting'.
• Trend fitting questions are common. You may get 4-5 data points, need to code time variable, calculate trend equation, and forecast one future value. Practice this calculation under 2-3 minutes.
• Seasonal index calculation is tested frequently. Know how to calculate SI for each period, average them by season, and verify they sum to 1200 (for 12 months). This is a high-scoring topic if you practice.
• Exponential smoothing questions are less common but appear in some papers. If asked, you'll be given α value and asked to forecast 1-2 steps ahead. Formula is straightforward; focus on correct substitution.
• Time management: Allocate 4-5 minutes per time series question. Read scenario carefully to identify which component is being tested. Use rough calculations (no calculator needed for most SSC CGL questions).