Week 5:
Intro to Time Series
and Project 1 Overview

Agenda

Motivation for analyzing time series data

Time series analysis

Forecasting and prediction

Project 1 overview

Cautionary note

• We will teach you enough to be dangerous

• We are introducing you to the tip of the analytical iceberg. Learn more before you put this into practice.

• Incorrect analysis can lead to worse conclusions than no analysis.

Intro to Time Series Analysis

The problem:
We don’t know what will
happen in the future

The solution:
We can use past data to
make a guess about the future*

What is time series data?

Time series data is a sequence of data points collected over time.

It typically consists of observations taken at regular intervals (e.g. every hour, every day, every week, etc.) and can be used to study trends and patterns over time.

Example: Store sales volume over time.

How is time series data used in business

Time Series Analysis

• learn about trends

• compare current period to past periods

Forecasting

• make an educated guess about future values

• develop business strategy around forecasts

Time series decomposition

The statistical method often used with time series data is called time series decomposition.

This method breaks down time series data into several components, each representing an underlying pattern within the data.

Isolating these components and analyzing them separately is important for the following reasons:

• to understand what contributes to the observed trends
• to make better forecasts

Time series decomposition

Decomposition statistically deconstructs a time series into several components:

1. seasonality

2. trend

3. cyclical

4. residual or “noise”

Hyndman & Athanasopoulos (2021)

Seasonality: This reflects the regular, predictable patterns within a fixed period, such as a year, month, or day. Seasonal effects are patterns that repeat at known intervals and are attributed to seasonal factors like time of year, holidays, or days of the week.

Trend: This represents the long-term progression of the series, showing how the data moves up and down over a long period. The trend is the slowest changing component and reflects the underlying growth or decline in the dataset.

Cyclical: These are fluctuations that occur in a non-fixed, irregular pattern, often influenced by economic and business cycles. These cycles are longer than a year, and are not predictable in a seasonal sense. Not all time series decomposition methods explicitly separate out cyclical from the trend.

Residual or Noise: This consists of random, unpredictable variations in the time series. They results from short-term, unanticipated and non-recurring factors that cannot be classified as seasonal or cyclical.

Example

Structural breaks

Analysts may want to identify changes in the time series that are not just noise.

Structural breaks refer to points in time when there is a sudden and lasting change in the behavior of the time series data.

We refer to the “behavior” of any dataset as the “data generating mechanism” (or data generating process).

Statistically identifying these changes can be the goal of the analysis or an exploratory exercise.

Examples of structural breaks: Economic policy changes, technological innovations, natural disasters, or shifts in consumer behavior.

Synthetic time series with three data generating mechanisms

Exercise

Briefly describe one reason for a structural break in a time series.

Forecasting examples

We might want to forecast. . .

• power demand to decide whether to build new power plant

• call volumes to schedule staff in call center

• inventory requirements to meet demands

Exercise

How could you use forecasting to improve decision-making in ag business or environmental and natural resource management?

Provide an example.

Forecasting overview

Use historical data to develop a model, then use the model to predict the future

Model quality depends on past data and assumptions

Forecasts are uncertain; we can quantify some of that uncertainty

Forecasting steps

1. Problem definition

2. Gathering data and institutional knowledge

3. Preliminary (exploratory) analysis

4. Choosing and fitting models

5. Using and evaluating a forecast model

Source: Hyndman & Athanasopoulos (2021)

Summary

Time series data is a common structure of data

Certain techniques are designed to analyze time series data

Forecasting is a common need

Introductions in this course (see Hyndman & Athanasopoulos (2021))

Project 1

Project 1 overview

Groups of 2

Choose an ag biz or enre management question to answer with time series data

Collect time series data

Analyze trends

Generate a forecast

Present results in a recorded video

Example: Corn and Combines

https://www.farm-equipment.com/articles/5142-historical-corn-prices-provide-look-at-future-equipment-sales

Context: Equipment dealers sell more when ag prices rise. Most people grow corn near dealer x.

Question: Will corn prices rise or fall in the next year and by how much?

Approach: Analyze time series of corn prices (decomp). Forecast corn PPI one year.

Present: Provide context (plot raw data) and describe decision. Explain your insights. Use effective graphics. Tell your data story.

Finding data

Many sources of public data (e.g., St Louis FED, tidyquant, and more on course website)

Choose data appropriate to answer the question

Instructors and TA can help

Analysis

Must include some:

• time series analysis (unit 1 week 2)

• forecasting (unit 2 week 3)

Process

Identify an interesting question

Collect data

Develop analysis strategy, explore, refine

Develop presentation

Get Started

References

Hyndman, R.J., & Athanasopoulos, G. (2021) Forecasting: principles and practice, 3rd edition, OTexts: Melbourne, Australia. https://otexts.com/fpp3/. Accessed on 02-14-2023.