Chapter Five : Time Series

A time series is a sequence of data points collected at successive, equally spaced points in
time. These data points are ordered chronologically, which allows for the analysis of trends,
patterns, and forecasting of future events. Common examples include daily stock prices,
hourly temperature readings, and monthly sales data.
Key characteristics
• Chronological order: Data points are arranged in the order they were recorded,
with each point associated with a specific timestamp.
• Regular intervals: In its most common form, measurements are taken at
consistent intervals, such as every minute, day, or year.
• Temporal dependency: The value of a data point often depends on previous
values in the series, a crucial aspect for analysis and forecasting.
Applications
• Forecasting: Predicting future values, such as the price of a stock next month or
the number of flu cases next week.
• Trend and pattern analysis: Identifying patterns, such as seasonality (repeated
fluctuations over time) or long-term trends in sales, weather, or stock market data.
• Business intelligence: Informing business planning, financial modeling, and
strategic decision-making.
• Scientific monitoring: Tracking data from sources like sensor networks, weather
stations, or medical devices over time.

5.1. Nature of Data

Time series data is a sequence of data points collected and ordered chronologically, with the
key characteristic that time is a crucial component. This temporal ordering means that past
data points can influence future ones, and the data exhibits patterns like trends, seasonality,
and cycles that can be analyzed for forecasting and other insights.
Key characteristics
• Chronological order: Data points are listed and analyzed in the order they were
collected, such as daily stock prices or hourly temperature readings.
• Dependence: A value in the series is often dependent on previous values, meaning
a value at time tis related to values at times t−1,t−2, etc..
• Patterns: Time series data commonly contains underlying patterns:
o Trends: The long-term upward or downward movement of the data over an
extended period.
o Seasonality: Predictable, repeating patterns within a fixed period, like yearly retail
sales spikes or monthly birth rates.
o Cycles: Fluctuations that are not of a fixed period, such as the longer-term business
cycles of expansion and recession.
o Noise: Unpredictable, random fluctuations that remain after accounting for trends,
seasonality, and cycles.
• Stationarity: A time series is stationary if its statistical properties (like mean
and variance) do not change over time, which is a necessary assumption for many models.
• Irregular intervals: While often collected at regular intervals, time series data
can also be collected at irregular times.
Why it's important
• Forecasting: The primary goal is often to predict future events by understanding
past patterns.
• Performance monitoring: It's used to track the performance of systems,
devices, or applications over time.
• Decision-making: It provides context for making more informed decisions by
revealing trends and patterns in areas like finance, economics, and business.
• Anomaly detection: It helps identify unusual data points that deviate from
expected patterns, which is crucial in areas like cyber security.
5.2. Trends and seasonality
In time series analysis, a trend is the long-term direction or general movement of data over
time, which can be upward (increasing), downward (decreasing), or stable. It represents the
underlying pattern, smoothing out short-term fluctuations and seasonality to reveal the
overall growth or decline of a dataset. For example, a retail store might see an upward trend
in sales over the holidays, while a company's costs might show a downward trend over time
due to efficiency improvements.
Types of trends
• Upward (positive) trend: Values generally increase over time.
• Downward (negative) trend: Values generally decrease over time.
• Stable (zero) trend: Values remain relatively constant over time.
• Monotonic trend: A trend that is consistently in one direction, either up or down.
• Deterministic trend: A trend with a clear, underlying cause, such as a consistent
increase in sales during the holiday season.
• Stochastic trend: A trend that is not as clearly defined and can be more random.
How trends are used
• Forecasting: Trends are a key component in predicting future values of a time series.
• Understanding patterns: They help identify long-term developments that may be
obscured by short-term variations.
• Strategic planning: Analyzing trends informs strategic decisions, like setting budgets
or improving operational efficiency.
• Anomaly detection: Deviations from a trend can help identify anomalies or outliers
in the data.
Methods for identifying trends
• Moving averages: A common method for smoothing out data to better see the
underlying trend.
• Regression analysis: Statistical techniques can be used to fit a trend line to the
data and estimate the slope.
• Specific tests: Statistical methods like the Mann-Kendall test can be used to detect
if a trend exists and is statistically significant

Seasonality in time series is a predictable pattern that repeats over a fixed period, such as
daily, weekly, monthly, or yearly, caused by seasonal factors like weather or holidays. It's a
fluctuation with a known, fixed period, unlike a cyclic pattern which has an irregular
duration of two or more years. Identifying seasonality is crucial for accurate forecasting
because it allows analysts to understand normal variations and make more reliable
predictions.
Key characteristics of seasonality
• Fixed period: The pattern repeats at regular, predictable intervals. For example, ice
cream sales consistently rise in the summer and fall in the winter.
• Calendar-related: The fixed period is usually tied to the calendar, such as quarters,
months, or days of the week.
• Cause: Seasonality is often driven by external factors that are periodic in nature.
o Example: Increased retail sales at the end of the year due to holidays.
o Example: Higher traffic during weekday commute hours.
• Distinction from cycles: Seasonal patterns have a fixed duration, while cyclic
patterns have irregular durations that are not tied to a calendar.
• Impact on data: It causes data to not have a constant mean and variance over time,
meaning the data is not stationary.
Why it's important
• Improved forecasting: By accounting for seasonality, forecasting models
become more accurate. For instance, a model will be more reliable if it predicts higher ice
cream sales in the summer.
• Anomaly detection: Knowing the typical seasonal pattern helps in identifying
when an event is unusual. For example, a sudden drop in summer sales could signal a
problem.
• Informed decision-making: Organizations can use seasonal data for better
planning, such as stocking up on inventory before a busy period or scheduling more staff
during peak times.
How to analyze seasonality
• Visual inspection: Plotting the data over time can reveal repeating patterns.
• Autocorrelation functions (ACF): This statistical tool measures how a time
series is related to past values. Regular peaks in the ACF plot at specific time lags indicate
seasonality.
• Seasonal decomposition: This technique breaks down the time series into its
components (trend, seasonality, and residual) to better measure and visualize the seasonal
effects.
5.3. Stationarity
Stationarity in a time series is a statistical property where the statistical characteristics of the
data, such as the mean and variance, remain constant over time. This means the series does
not have a trend or seasonality and will revert to a long-term mean if it is temporarily
disturbed. Non-stationary data, in contrast, has statistical properties that change over time,
which can lead to misleading conclusions if not addressed.
Characteristics of a stationary time series
• Constant Mean: The average value of the series remains consistent throughout its
duration.
• Constant Variance: The spread or volatility of the data points does not change
over time.
• No Trend: The series does not exhibit a long-term upward or downward
movement.
• No Seasonality: There are no predictable, repeating patterns at fixed intervals.
• Autocorrelation: The correlation between different points in the series is
consistent and does not change depending on the time period.
• Mean-Reverting: If the series deviates from its mean, it is expected to return to it
over time.
Why is stationarity important?
• Reliable Models: Stationary data is crucial for building reliable forecasting
models, as the patterns observed are consistent and not just a snapshot of a changing system.
• Accurate Forecasts: Models built on stationary data are more likely to produce
accurate and dependable predictions.
• Valid Inference: It is more difficult to draw reliable conclusions from non-
stationary data because its statistical properties are constantly changing.
How to achieve stationarity
• Transformation: You can use mathematical transformations, such as taking the
logarithm or square root of the data, to stabilize variance.
• Differencing: You can subtract each data point from the previous one to remove
trends. This process can be repeated multiple times if necessary.

5.4. Box-Jenkin Methodology

Box - Jenkins Analysis refers to a systematic method of identifying, fitting, checking, and
using integrated autoregressive, moving average (ARIMA) time series models. The method
is appropriate for time series of medium to long length (at least 50 observations).