Data Seasonality with FRED

Compelling Question

How does seasonality affect data?

Description

FRED^® provides access to rich data from more than 100 sources. Some of the data experience seasonal upswings and downswings, and sources report the data after accounting for those seasonal patterns. This article describes seasonality for new data users and can serve as a reference for advanced data users.

Introduction

Economic activity increases and decreases over time for a variety of reasons—some unexpected and others predictable. Economic data reflect those changing conditions, and it’s important to understand whether the upswings and downswings are long-lasting or short-lived.

For example, some types of economic activity slowly become obsolete—such as renting DVDs to watch movies at home—and the data reflect that trend by showing a long-term decline in output or employment. At the same time, overall economic activity, measured through gross domestic product, periodically experiences medium-length periods of expansion and contraction known as business cycles. And, specific sectors of economic activity, such as farming or retail sales, record short-lived highs and lows that follow a yearly calendar schedule. These are called seasonal patterns. Data with these patterns are said to experience seasonality.

Seasonality in the data can mislead or confuse an otherwise straightforward interpretation of monthly or quarterly rates of change. In other words, data users can misinterpret short-term seasonal highs or lows as the start of longer-lasting cyclical booms or crashes. Researchers can use a mathematical process called seasonal adjustment to remove the seasonal patterns in the data.

Data in FRED are always labeled as either “not seasonally adjusted” or “seasonally adjusted.” Those labels reflect whether the source reports the raw data or the data after they’re adjusted to remove the variability from seasonal patterns.

This article describes how to identify seasonal patterns in the data and how to best tell the story behind the numbers, whether they are seasonally adjusted or not.

Seasonal Patterns in Data

There are three types of patterns in economic data: trends, cycles, and seasonal patterns. They can be broadly defined by their period, regularity, and predictability.

Trends are broad, long-term changes in the data. They are visible in data plots as gradual increases or decreases in data values over many quarters or years. Trends aren’t regular in shape or length, making them difficult to predict. See this February 2024 FRED Blog post for a discussion of long-term trends in employment at video-rental establishments, for example.

Cycles are medium-length changes in the data. They are visible in data plots as fluctuations in data values over several months or quarters. As with trends, cycles aren’t regular in shape or length and are difficult to predict. The March 2023 Page One Economics article by Scott Wolla covers the topic of business cycles and recessions in greater depth.

Seasonal patterns are short-term changes in the data; they reflect seasonal changes in economic activity within a calendar year. They are visible in data plots as recurring fluctuations in monthly or quarterly data values. Seasonal patterns are regular in shape and length, making them usually easy to predict. Keep in mind that it isn’t possible to observe seasonal patterns in annual data.

Seasonal patterns in economic activity are easily recognizable as seesaw patterns in data plots. Figure 1 above shows two different series of employment data. The smooth blue line shows the trend and cycles in employment numbers after seasonal upswings and downswings have been removed. The jagged red line shows the raw data, without any adjustment for seasonality.

Different types of economic activity show different seasonal patterns. Figure 2 above shows that grape and strawberry prices fall during the third quarter of each year, when harvest season temporarily boosts the supply of fresh fruit, which was described in a 2020 FRED Blog post.

In contrast, the monthly employment numbers discussed earlier have a different seasonal pattern: They peak in November and December because that is when retailers and other employers are the busiest and short-term help is in high demand. See this December 2022 FRED Blog post for a discussion of seasonal patterns in postal employment, for example.

Adjusting for Seasonality in Data

The presence of seasonal patterns in data can make their short-term upswings and downswings very dramatic and confuse data users when analyzing trends and cycles. To help tell the story behind the numbers, some economic data sources report their data series after adjusting for seasonal patterns.

There are different mathematical processes to calculate the value of seasonal factors in data collected over time. The U.S. Census Bureau has developed many of the most-popular seasonal-adjustment methods, so we’ll describe the outline of the process it follows as a general rule. For more on the seasonal-adjustment process originally developed by the U.S. Census Bureau, see “Seasonal Adjustment Questions and Answers.”

The process starts with removing a crude, preliminary estimate of the long-term trend and cyclical components of a series. In the case of overall economic activity, that includes the average rate of economic growth over several decades and the contractions in economic activity, known as recessions, that generally last several quarters. The seesaw patterns left in this roughly de-trended data are a first approximation of the seasonal patterns. This process is then repeated several times, with each repetition—also known as an iteration—refining the scope of the adjustment so that one-of-a-kind events do not distort the estimation of trends, cycles, and seasonal factors.

Figure 3 above shows the difference between two quarterly data series on electricity and gas production. By plotting the difference between the raw data and the data after seasonal adjustment, the seasonal factors stand out to the naked eye: There are two seasonal highs in electricity and gas production, in Q1/winter and Q3/summer, and two seasonal lows, in Q2/spring and Q4/fall. See my November 2021 Page One Economics primer for more on data units in FRED; and, see this September 2020 FRED Blog post for a discussion of seasonal patterns in electricity consumption and production.

Data-collection organizations, such as the U.S. Bureau of Labor Statistics and U.S. Bureau of Economic Analysis, continually refine the process they use to report data; that is, they periodically revise the seasonal adjustment of their data to best reflect changing patterns in economic activity. Better data facilitate better decisionmaking. For more on data revisions, see my August 2022 Page One Economics primer, “Data Revisions with FRED.”

When To Use Seasonally Adjusted Data

Data in FRED are always labeled as either “not seasonally adjusted” or “seasonally adjusted.” The choice of data units depends on the type of story you want to tell. For example, it is easier to describe trends in fruit prices by using seasonally adjusted data because they smooth out the impact of short-term spikes and drops resulting from regular harvesting cycles. Seasonally adjusted data are usually preferred when researching economic activity and formulating economic policy.

On the other hand, data not seasonally adjusted are needed to understand, for example, the periodic surges in hiring during the October-to-December holiday season. The raw data during those three months can be compared with data from the same period in recent years to describe the relative strength of an underlying retail sales spike.

Summary

Economic data that experience regular upswings and downswings matched to the calendar year show seasonal patterns, or seasonality: This is noticeable in data graphs as a seesaw pattern that repeats year after year. Because those patterns can make data trends and cycles harder to see, organizations that report data frequently use a mathematical process called seasonal adjustment to smooth out the data. FRED labels data as either “not seasonally adjusted” or “seasonally adjusted”; the choice of data units used for storytelling depends on the type of story one is trying to tell.