Board of Governors of the Federal Reserve System

Industrial Production Methodology[1]

Basic Concepts

The index of industrial production covers mining (North American Industry Classification System [NAICS] major groups 211–213), manufacturing (NAICS major groups 311–316; 321–327; 331–337; and 339; and industries 1133 and 5111), and electric and gas utilities (NAICS 2211,2).[2] Production has been defined as the "the process of creating economic values or utilities needed to satisfy human wants."[3]

Production in the mining industry includes the exploration and development of mineral properties and the extraction of minerals in the form of solids (such as coal and ores), liquids (such as crude oil), or gases (such as natural gas). Milling and other preparation customarily done at the mine (well) site is also included. Manufacturing, according to the NAICS, relates to the mechanical, physical, or chemical transformation of materials, substances, or components into new products. Production by utilities refers to electric power generation, transmission, and distribution for sale and to natural gas distribution.

Summary of Methods

Individual IP series are derived from (1) annual indexes of real industry output that are calculated from comprehensive information sources and (2) monthly and, in a few cases, quarterly production indicators that are available for inclusion in the monthly index within the regular six-month reporting window. The annual index determines the trend for a series from one year to the next, and the production indicator determines the monthly changes for a series within each year. Each series is seasonally adjusted, and the contribution of the change in an IP series for an industry to the monthly change in the overall IP index is based on the value added by that industry.

The annual indexes for individual IP series are derived from detailed industry data. For each six-digit NAICS industry in manufacturing, an annual chain-type measure of the real gross output of an industry is compiled. The value of the production is Census Bureau data on the industry's value added plus its cost of materials minus the cost of resales (when the data are available); the real output measure is obtained by deflating the value of production by an annually weighted chain-type price index compiled from detailed information on the composition of the industry's products.[4] Most of these price indexes are obtained from the Bureau of Economic Analysis (BEA), but some are calculated by the Federal Reserve. Because an individual IP series may be a combination of several six-digit NAICS industries, the annual indexes for many manufacturing IP series are constructed from a number of industry gross output measures; for these indexes, the contribution of each component industry to the annual index is based on the value added by that industry.

For many IP series, the production indicators are compiled from monthly (or quarterly) product data. The indicator may be the output of a product in physical terms (for example, tons of portland cement or barrels of distillate fuel oil); or the indicator may be data on the output of several types of a product (for example, unit counts of assemblies of crawlers, wheel loaders, skid steer loaders, and the like, with each having a fixed weight.) Alternatively, for selected series, the indicator is a chain-type quantity index that is compiled each month (or quarter) from very detailed data on the prices and quantities of specific products produced by an industry. This method is used for the monthly IP indexes for semiconductors, pharmaceuticals, books, and autos and light trucks.[5]

For non-energy mining, most annual and monthly indexes are developed from product data issued by the U.S. Geological Survey; the IP series on refined petroleum products and electric and gas utilities are developed from comprehensive monthly and annual data from the Department of Energy. For most IP series in these groups, the monthly data are measures of a product in physical terms, such as barrels of motor gasoline; for other series, the indicator is more complex. For example, coal production is the tonnage output of four geographic regions, weighted by the Btu content of the variety mined in each region.[6]

When high-frequency data on the physical quantity of production are unavailable, the Federal Reserve uses monthly data on the inputs to production, primarily monthly data on production-worker hours from the Bureau of Labor Statistics (BLS), as the production indicator (some indexes use the Federal Reserve's monthly data on electric power use for the period up to 1997). The production indicator is combined with a productivity factor calculated from the annual output index to obtain the monthly IP index.

Comparison Base and Index Calculation

The Federal Reserve reports the quantity of output as an index number—current output expressed as a percentage of production in a comparison base period. Currently, the base period is the year 2017.

The aggregation method for the current IP index from 1972 forward is a version of the Fisher-ideal index formula. The weights used to combine the individual industry output indexes are monthly "unit value added" measures ("or prices"), which are derived from annual data on industry value added. The formula for the change in monthly IP (or a monthly IP sub-aggregate) is the geometric mean of the change in output computed using current month weights and the change computed using weights for the previous month:


where I_m is an individual production index for a month and p_m is the unit value added in month m.

Classification System

The 296 individual series are indexed and combined into industry groups as indicated in the NAICS and are also aggregated in the major market groups of the index. The market groups are arranged according to the expected use of the output as final products (consumer goods and equipment); intermediate products (construction and business supplies, which are items that are used as inputs outside of the industrial sector); and materials (including fuel and power that are further processed or used in factories, mines and utilities).[7] All industries in the IP index have their output allocated to multiple market groups. The market group shares for the individual IP series are derived using relationships in the input-output tables issued by the BEA.[8]

Monthly and Annual Production Series

The main task in preparing the index is estimating the monthly changes in the real output of an industry. The underlying data used in these estimates are obtained, to the extent possible, from basic statistical series that express output in physical quantities (steel in tons, crude oil in barrels, and so on). The movements in such physical product series indicate the monthly changes in production without regard to movements in prices of the items involved. The monthly production of complex items like computers, machinery, and nuclear submarines, however, cannot be expressed meaningfully by reference to monthly movements in weight, length, or simple unit counts of production. Consequently, changes in the monthly output of such items are estimated on the basis of either: (1) the changes observed in relevant inputs, expressed in physical units, expended in the process of production; or (2) a chain-type index of real output developed from highly detailed and reasonably comprehensive price and quantity data. In the production index, the main monthly input data used are production-worker hours employed in industrial establishments; currently, the monthly movements in 130 series are based on input-type measures only. The main monthly series that are based on detailed price and quantity data include pharmaceuticals, semiconductors, books, autos, light trucks, and completed commercial aircraft.

Physical product series (both monthly and quarterly) cover a bit more than one-half of industrial production by weight. Of the product series that are not estimated based on trends by the Federal Reserve, about 45 percent come from government agencies, such as the Departments of Energy and Agriculture; the rest come from a variety of private sources, primarily trade associations, such as the American Iron and Steel Institute and the American Forest and Paper Association. Ward's Communications provides data covering automobiles and trucks.

While the data on physical product are taken from a variety of public and private sources, the data on production-worker hours come from a survey by the BLS. The BLS employment survey covers only a portion of the month, the pay period containing the 12th of the month; for manufacturing firms, this pay period is typically only one week long.

A number of steps may be required to convert the monthly data into the final output indexes. Daily averages (based on the number of working days in the month) of physical product series are calculated and then converted into indexes, expressed as a percentage of the daily average in 2017.[9] Similarly, monthly averages are used for the production-worker series. (Daily averages are not used for these series because the employment series cover only one week in each month.)

After the basic data are expressed as index relatives, they may be multiplied by factors that correct them for cyclical or secular biases due to sampling or trends in productivity. These "benchmark adjustment factors" (BAFs) are typically derived by comparing the monthly data with more comprehensive annual benchmarks that primarily are taken from the Annual Survey of Manufactures (ASM), the Economic Census, and Mineral Yearbooks. The BAFs are re-estimated during annual revisions of the index but are studied during each year and may be modified if conditions have changed importantly. For instance, productivity per production-worker hour tends to rise faster in expansions than in recessions. Finally, seasonal factors are applied to the indexes to generate the seasonally adjusted indexes; seasonal adjustment methods are discussed below.

Weighting and Aggregation

In creating aggregate indexes, a suitable method for combining the individual indexes of output is needed. Changes in the number of items produced—barrels of oil, tons of steel, or automobiles—are not of equal significance. Clearly, the relative significance of the items must be taken into account in calculating the aggregates. Market prices are suitable for combining items at one stage of processing, but lead to double-counting when adding items from different stages, such as materials and products.

Because the industrial production index is based on estimates of real gross output at different stages of production—such as iron ore, raw steel, steel mill products, motor vehicle parts, and automobiles—value-added weights are used to combine the individual series; thus, double counting of the contributions of earlier stages to aggregate output is avoided. The census value added of an industry is used to represent the contribution of that industry to total industrial output. Implicitly then, constant-dollar value added for each series is assumed to move with gross output. This assumption is also relevant to the input-based series, because each input index is converted, through the use of the production factor coefficients, to an index of gross output before the individual series are aggregated into market and industry groups.

The value-added data are taken from the ASM and the Economic Census. For utilities, the weights are based on the income statements of utilities. Care is taken to allocate the NAICS-based census value added to physical product series that do not precisely follow the NAICS.

IP market group aggregates for products, as well as stage-of-processing aggregates, are also calculated using gross value weights. Because the market group for products is the final stage of production within the industrial sector and the output of each stage-of-processing group is not generally used as an input to an industry in the same group, final goods prices can be used without double counting. Compiling the IP index using gross-value weights facilitates comparison with other dollar-based data. The gross-product weights are derived from data collected in the Economic Census and ASM.

Monthly Compilation Practices

Preliminary estimates of industrial production are published about 15 days after the end of the relevant month. Beginning with the G.17 issued in April 2008, revisions may appear in the successive five months after the preliminary estimate. For each month the index is subject to estimation for six successive months. Thereafter, revisions of the indexes may occur in annual or benchmark revisions.

For the first estimate of a given month, the BLS's production-worker hours data are available along with about half of the physical product data.[10] By the time of the second estimate, the bulk of the product data and revised worker-hour data have been received. Revisions and most of the missing physical product data are received in time for the third and fourth estimates, but additional data may arrive in time for the fifth and sixth estimates. The table that summarizes the flow of information during a typical recent six-month IP window is shown in the Explanatory Note for the G.17. As the table indicates, in making the first estimate of IP, about a fourth of the index is estimated (as primary source data are not yet available for that month). As a result, a variety of information is used to supplement the basic data base, particularly for the first estimate. Consequently, the estimation of the index of industrial production sometimes involves the use of judgment and information about production other than the available monthly data base. The main sources of such editing, or adjustments, are as follows: missing observations, extreme preliminary values, and input data that are unrepresentative of output in a particular month. The last step involves the incorporation of an adjustment for hours worked in manufacturing by workers on assignment from temporary help supply (THS) firms, if applicable.

Missing Observations

The Federal Reserve compiles the production index from a variety of sources that become available at different times. For the first estimate, production-worker hours, weekly physical product data, and some of monthly physical product data are available; for the second estimate, much more of the monthly physical product data become available. Most of the remaining physical product data become available by month six and revisions to all types of sources are received as well. Given the flow of data, most of the estimates for missing observations are for the current IP month. These estimated values are based on relevant data that are available, particularly the corresponding production-worker hour data, changes in the employment of THS workers, reports of strikes or unusual conditions, trade and industry news, and analysis of near-term productivity trends, cycles, and seasonal patterns in production.[11]

Extreme and Unrepresentative Values

Occasionally preliminary data appear to be extremely inconsistent with past experience, other current data, and information about current business conditions. Judgmental adjustments are sometimes made to such outliers, and experience has shown that revised data typically support the direction, if not the degree, of most of these adjustments.

Adjustments are also occasionally made because unusual conditions make the input-based indexes unrepresentative of actual production. The production-worker hour data cover only the pay period that includes the 12th of the month, not the entire month. If the survey reference period is not representative of the month because of a strike or severe storm, for example, then an adjustment is made so that the production index reflects the average level of activity in the month.

Seasonal Adjustment

The seasonal adjustment of the index dates back to the 1920s, when the Federal Reserve adopted the ratio-to-moving-average method. This method has been refined over time and gained wide acceptance. The Census Bureau has developed computer programs, including the X-11, X-12, and X-13 versions, that have permitted its application to a vast number of series, including the production index. All individual series are now adjusted with the Census X-13-ARIMA-SEATS seasonal adjustment method using the computer program developed at the Census Bureau. This is supplemented by a prior de-trending procedure (designed to reduce the influence on the estimated seasonal factors of sharp business cycle swings) and by interventions and alternative holiday adjustment when appropriate.

The intervention method developed by Box and Tiao is applied to not seasonally adjusted production data to allow for replacement of "outlier" data patterns that bias the X-13's computations of seasonal factors. The outliers could include unusual one-time data anomalies; they could also include periods when extremely sharp business cycle swings affect the estimated seasonal factors. Once outliers are identified, the intervention program fits a nonlinear regression equation to the unadjusted data. Once the model has been fit, with the iterative use of least squares, the outlying data points are replaced by their estimated values. The production series, adjusted for holidays, outliers, and forecasts, is then processed through the ratio-to-moving-average part of the standard X-13 program to yield seasonal adjustment factors, which may be combined with holiday factors to calculate the final seasonally adjusted series. During annual revisions, the seasonal adjustment factors may be revised for a number of years.

---

[1] This summary is derived from recent articles from the Federal Reserve Bulletin and from chapters in Industrial Production—1986 Edition.
Return to text

[2] Manufacturing consists of those industries included in the North American Industry Classification System, or NAICS, definition of manufacturing plus those industries—logging (NAICS 1133) and newspaper, periodical, book, and directory publishing (NAICS 5111)—that have traditionally been considered to be manufacturing and included in the industrial sector.
Return to text

[3] Donald W. Miffed, Economics Dictionary (American Elsevier, 1976). p. 209.
Return to text

[4] The definition of the nominal value of production was changed in the 2004 annual revision and implemented from 1997 through 2002. Previously, it was computed as value added plus the cost of materials. The cost of resales is the amount spent to purchase goods that are then resold without any material transformation. Data on the cost of resales for detailed NAICS industries are unavailable, however, for 2004, so the traditional calculation is being employed after 2002 until those data become available.
Return to text

[5] The method was introduced for the monthly measurement of semiconductors in the 1998 annual revision, for computers and motor vehicles in the 1999 annual revision, and for pharmaceuticals in the 2000 annual revision. For semiconductors, computers, and pharmaceuticals, the method consists of (1) estimating the value of U.S. production for the industry from monthly and quarterly data on highly detailed unit counts and values of individual products produced by industry and (2) deflating the value of production by a chain-type matched-model price index constructed, for the most part, from the same data.

For motor vehicles, detailed monthly data on the production of each vehicle model are aggregated using annual prices as weights. For a few other series in the IP index, the production indicator is obtained by deflating detailed data on the value of production or shipments from a trade source by a corresponding BLS producer price index.
Return to text

[6] This method was introduced in the 1998 revision.
Return to text

[7] The detailed composition of the industry and market structures of the index is shown in the "Source and Description Information" tables found on the Board's website and included in the attachments.
Return to text

[8] This refinement was introduced in the 2002 annual revision. Prior to that, an individual IP series was assigned to the market grouping based on its highest share.
Return to text

[9] The number of working days in the month varies by industry. Some industries, such as wood pulp, work around the clock, seven days per week; other industries work five or six days per week.
Return to text

[10] For some series based on monthly physical product data, available weekly physical product data are used as a proxy for the monthly series in the initial estimate.
Return to text

[11] For example, an "initial" estimate of a given IP series for which the source data are not yet available is based upon the monthly change in the corresponding production-worker hour series, seasonally adjusted, coupled with the recent estimated trend in productivity for that series. This estimate may be refined further taking into account the other relevant factors already mentioned.
Return to text