Seasonally Adjusting Data

How seasonally adjusting data helps researchers conduct economic analysis; what seasonal adjustment method is preferred by most economists.

The Economic Problem

Many Economic Data Series Exhibit Strong Seasonal Movements
Economists and business people study many data series to help get an idea of where the economy is heading. They look at the job count, new home permits, home sales and retail sales, to name a few. Inspecting these data at one point in time doesn't always tell a complete story. Studying them over time helps show the extent to which a particular industry or area of the economy is contributing to or detracting from economic growth.

One problem with interpreting data over time is that many data series exhibit movements that recur every year in the same month or quarter. For example, housing permits increase every spring when the weather improves, while toy sales usually peak in December. This dynamic makes it hard for economists to interpret the underlying trend in some data series. For instance, were sales better this December or was it just the usual holiday runup? Economists want to know if sales were better than the normal seasonal increase. To understand what the data are really saying about economic growth, statisticians and economists remove such predictable fluctuations—or seasonality—from the data.

Chart 1 illustrates seasonality in data. The chart plots the raw (that is, nonseasonally adjusted) employment data for Texas. (Because it is one of the most widely followed series for the Texas economy, Texas employment data will be used as the example in the remainder of this article.) As the chart shows, the data series exhibits distinctive fluctuating patterns—the most obvious pattern being the sharp downward spike each January as temporary holiday hiring comes to a halt. Another downward spike in July results from the end of the school year as most teachers take a summer break. Because of the month-to-month seasonal variations in the data, it is difficult to isolate the actual trend in monthly employment.

Chart 1

Comparing Data Annually Gets Around the Seasonal Problem, but with Drawbacks
One way analysts avoid the problem of seasonal fluctuation is to compare monthly or quarterly data on a year-over-year basis. In other words, the current month's data point is compared with the data point from the same month in the prior year. A growth rate (or percentage change) is then calculated to get a comparative measure for how fast employment rose or fell over the 12-month period. As shown in Chart 2, this method reduces the fluctuations and can help reveal a trend in the data. However, using the year-over-year method has drawbacks. It relies heavily on data that is 12 months old for the calculations. Comparison from month to month helps economists determine significant changes in the business cycle soon after they occur. Thus, a more sophisticated seasonal adjustment method is called for.

Chart 2

Technical Solution

The X12 Procedure Isolates and Removes Seasonal Factors
Most statisticians, economists and government agencies that report data use a method called the X12 procedure to adjust data for seasonal patterns. The X12 procedure and its predecessor X11, which is still widely used, were developed by the U.S. Census Bureau. [1] When applied to a data series, the X12 process first estimates effects that occur in the same month every year with similar magnitude and direction. These estimates are the “seasonal” components of the data series. In addition, the procedure estimates the “trend-cycle” and “irregular” components. The trend-cycle component is the series' long-term tendency to grow or decline and can fluctuate because of the economic trends or other long-term cyclical factors. The irregular component comes from unseasonable weather, natural disasters, strikes or sampling error. The goal of the seasonal adjustment procedure is to separate out the seasonal component, leaving the trend-cycle and irregular components.

The Census Bureau provides X12 and X11 computer programs free of charge. Additionally, these programs are included in most statistical software.[2]

Many national data series are already seasonally adjusted before they are released. For example, data such as U.S. building permits, housing starts, retail sales, gross domestic product (GDP), the producer price index (PPI) and the consumer price Index (CPI) are released in seasonally adjusted form, simplifying data analysis for economists and business analysts who track the national economy. Some regional data series are not released in seasonally adjusted form, however, and must be adjusted before they can be used in economic analysis.

Real-World Example

Seasonally Adjusted Data Show the Economic Trend
Now let's look at a real-world example to see the effect seasonal adjustment has on a data series. Chart 3 plots both nonseasonally adjusted Texas employment data and the seasonally adjusted series available from the Dallas Fed. As demonstrated in the chart, the seasonally adjusted series is much smoother and shows a trend in employment. It shows the strong growth in Texas employment in 1999 and 2000 and the ensuing downturn in jobs during the state's recent recession.

Chart 3

Seasonally adjusted data are especially useful when trying to determine a significant change in the economy's direction. As Chart 3 indicates, the seasonally adjusted series shows the turning point—the month when jobs actually started falling during the most recent recession. Using the unadjusted data, it's nearly impossible to tell when the decline actually occurred. Similarly, looking at the year-over-year method in Chart 2, the data indicate that the employment decline in Texas began in September 2001. However, if we use the seasonally adjusted data and calculate annualized percentage changes from month to month, we can see that the decline actually began in April 2001 (Chart 4). This example illustrates the importance of month-to-month comparisons of seasonally adjusted data. For most purposes, it is preferred to the year-over-year method because it gives a more accurate and timely view of economic activity.

Chart 4

Notes The X12 procedure is an enhanced version of the X11 method. Enhancements include a more versatile user interface and new tools that help overcome modeling problems, thereby enlarging the range of economic data series that can be adequately seasonally adjusted. In many cases, the enhancements are not needed and the X11 procedure is used. For more information on the X12 or X11 procedure, refer to the statistics software manuals or visit the Census Bureau's X12 web site [off-site]. Glossary at a Glance Irregular component: The component of a data series that comes from unseasonable weather, natural disasters, strikes or sampling error. NSA: Not seasonally adjusted. SA: Seasonally adjusted. SAAR: Seasonally adjusted annual rate. Seasonal adjustment: A statistical smoothing process that estimates and separates seasonal components in a data series. Seasonal component: Movements in a data series that recur every year in the same month or quarter, with similar magnitude and direction. Trend-cycle component: A data series' long-term tendency to grow or decline. X11 procedure: A widely used predecessor to the X12 procedure, developed by the U.S. Census Bureau. X12 procedure: A seasonal-adjustment method developed by the U.S. Census Bureau and used by most economists and government agencies. The procedure estimates and removes the seasonal factors from a data series. Year-over-year method: A method for smoothing raw data, whereby the current month's data is compared with data from the same month a year earlier.