UNIVERSITY OF MALAKAND

DEPARTMENT OF STATISTICS

ASSIGNMENT : TIME SERIES ANALYSIS

SUBMITTED BY ROIDAR KHAN

ENROLLMENT NO 509

SUBMITTED TO MS KAINAT

1
 Question no 1

Discuss the difference between time series and cross sectional data with
the help of suitable example.

Answer

Time Series Data

Definition: Time series data tracks the same variable(s) over a period of time. It’s
like a timeline where each data point is associated with a specific time stamp.

Key Feature: The data is time-dependent and ordered chronologically.

Focus: It helps analyze trends, patterns, and changes over time.

Example: Daily temperature readings, monthly sales data, or annual GDP growth.

2. Cross-Sectional Data

Definition: Cross-sectional data captures multiple variables or entities at a single

point in time. It’s like a snapshot of different subjects at the same moment.

Key Feature: The data is not time-dependent and focuses on comparisons.

Focus: It helps analyze differences, distributions, and relationships between

entities.

Example: Average income of people in different cities in 2023, or test scores of

students in a class.

Cricket Examples to Illustrate the Difference

Now that we’ve established the difference, let’s apply these concepts to cricket to
make it more engaging and relatable.

Time Series Data in Cricket: Tracking a Player’s Career

2
Time series data in cricket is like watching a player’s career unfold over the years.
It tracks performance metrics over time, revealing trends, peaks, and slumps.
Think of it as the storyline of a player’s journey.

Example: Virat Kohli’s Batting

Average Over the Years

Imagine we’re analyzing Virat Kohli’s batting average in ODIs from 2010 to 2023.
Your dataset might look like this:

YEARS Average
2010 35.00
2012 42.50
2013 42.80
…….. …….
2023 50.00

Cross-Sectional Data in Cricket: Comparing Players in a Tournament

Cross-sectional data in cricket is like taking a group photo of players in a

tournament. It captures performance metrics for multiple players or teams at a
single point in time. Think of it as a snapshot of the current state.

Example: Batting Averages of Top Batsmen in the 2023 World Cup

Now, imagine we’re analyzing the batting averages of the top 5 batsmen in the
2023 Cricket World Cup. Your dataset might look like this:

player Batting avrage

Virat Kohli 75.00
Rohit Sharma 70.00
Kane Williamson 65.00

3
 Babar Azam 60.00
David Warner 50.00

ANSWR NO 2

The main objectives of time series analysis are:

1. Forecasting: Predict future values based on past data.

2. Trend Analysis: Identify long-term movements or patterns.

3. Seasonality Detection: Find repeating patterns at fixed intervals.

4. Noise Reduction: Separate random fluctuations from meaningful data.

5. Pattern Recognition: Understand underlying structures in the data.

6. Decision Making: Support planning and strategy with data-driven insights.

Steps of Time Series Data Analysis:

1. Data Collection:

 Gather historical data recorded at regular intervals (e.g., daily, monthly,

yearly).

 Ensure the data is clean, complete, and consistent.

2. Data Visualization:

 Plot the data (e.g., line graph) to visualize trends, seasonality, and patterns.

 Identify outliers or anomalies.

3. Decomposition:

 Break down the time series into its components:

o Trend: Long-term movement.

o Seasonality: Repeating patterns.

4
 o Irregular (Noise): Random fluctuations.

4. Stationarity Check:

 Ensure the data is stationary (mean, variance, and autocorrelation are

constant over time).

 Use statistical tests (e.g., Augmented Dickey-Fuller test) or transformations

(e.g., differencing, log transformation).

5. Model Selection:

 Choose a suitable model based on the data characteristics:

o ARIMA: For non-seasonal data.

o SARIMA: For seasonal data.

o Exponential Smoothing: For data with trends and seasonality.

o Machine Learning Models: For complex patterns (e.g., LSTM, Random

Forest).

6. Model Fitting:

 Fit the selected model to the data.

 Estimate model parameters (e.g., p, d, q for ARIMA).

7. Model Validation:

 Test the model’s accuracy using:

o Training and Testing Sets: Split the data into training (e.g., 80%) and
testing (e.g., 20%).

o Error Metrics: Use MAE, RMSE, or MAPE to evaluate performance.

ANSWER NO 3

5
In time series analysis, stationarity is a key concept that determines whether the
statistical properties of a series remain constant over time. There are two types of
stationarity: weak stationarity and strong stationarity. Here's an explanation of
both and their differences:

1. Weak Stationarity (Covariance Stationarity):

A time series is weakly stationary if it satisfies the following conditions:

1. Constant Mean: The mean of the series does not change over time.

o E(Yt)=μE(Yt)=μ (same for all tt).

2. Constant Variance: The variance of the series is constant over time.

o Var(Yt)=σ2Var(Yt)=σ2 (same for all tt).

3. Constant Autocovariance: The covariance between two points depends

only on the time lag (kk) and not on the actual time (tt).

o Cov(Yt,Yt+k)=γkCov(Yt,Yt+k)=γk (depends only on kk).

2. Strong Stationarity (Strict Stationarity):

A time series is strongly stationary if the entire distribution of the series is

invariant over time. This means:

 The joint distribution of any set of time points (Yt1,Yt2,...,Ytn)(Yt1,Yt2,...,Ytn

) is the same as the joint distribution of (Yt1+k,Yt2+k,...,Ytn+k)(Yt1+k,Yt2+k
,...,Ytn+k) for any kk.

Difference Between Weak and Strong Stationarity:

Aspect Weak Stationarity Strong Stationarity

Definition Focuses on mean, variance, and Focuses on the entire

6
Aspect Weak Stationarity Strong Stationarity

autocovariance. distribution.

Moments Only first and second moments All moments (mean, variance,
Considered (mean, variance). skewness, etc.).

Ease of Testing Easier to test (e.g., ADF test). Harder to test.

Commonly used in time series

Practical Use Rarely used in practice.
modeling.

ARIMA models assume weak Rarely assumed in real-world

Example
stationarity. applications.

ANSWER NO 4

White Noise Process:

 Definition: A sequence of random variables with:

1. Constant Mean: E(Yt)=μE(Yt)=μ (often μ=0μ=0).

2. Constant Variance: Var(Yt)=σ2Var(Yt)=σ2.

3. No Autocorrelation: Cov(Yt,Yt+k)=0Cov(Yt,Yt+k)=0 for k≠0k =0.

 Properties:

o Unpredictable: No patterns, trends, or seasonality.

o Residuals of good models should resemble white noise.

PART B

We need stationarity in time series analysis because it simplifies modeling and

ensures reliable results. Here’s why stationarity is important:

7
1. Simplifies Modeling:

 Stationary data has constant statistical properties (mean, variance,

autocorrelation), making it easier to model.

 Many time series models (e.g., ARIMA) assume stationarity.

2. Improves Forecast Accuracy:

 Stationary data allows models to capture consistent patterns, leading to

more accurate predictions.

 Non-stationary data (e.g., with trends or seasonality) can mislead models.

3. Valid Statistical Inference:

 Stationarity ensures that statistical tests and confidence intervals are valid.

Non-stationary data can lead to spurious results (e.g., false correlations

ANSWER 5

Definition of Random Walk:

A time series {Yt}{Yt} is called a Random Walk if it satisfies the following equation:

Yt=Yt−1+ϵtYt=Yt−1+ϵt

where:

 YtYt: Value at time tt.

 Yt−1Yt−1: Value at the previous time step.

ϵtϵt: Random noise (usually white noise with mean 0 and constant var