cbo's economic forecasting record: 2013 update

7/29/2019 CBO's Economic Forecasting Record: 2013 Update

1/49

Forecast Minus Actual Growth in Infation-Adjusted Output: Two-Year Forecasts

Percentage-Point Di erence in Annual Growth Rates

-4

1976 1978 1982 1984 1986 1988 1990 1992 20001996 2004 20081994 20021998 2006 20101980

6

4

2

0

-2

Administration

CBO

Blue Chip Consensus

CONGRESS OF THE UNITED STATES

CONGRESSIONAL BUDGET OFFICE

CBO

CBOs EconomicForecasting Record:

2013 Update

JANUARY 2013


2/49CBO

Note

Some of the figures have white vertical bars that indicate the duration of recessions.(A recession extends from the peak of a business cycle to its trough.)

Pub. No


3/49

Contents

C

Summary 1

Introduction 5Measuring the Quality of Forecasts 6Limitations of Forecast Evaluations 6

Some Sources of Forecasting Error 7Business Cycle Turning Points 7Changes in Productivity Trends 7Changes in Crude Oil Prices 10Revisions to Historical Data 10

CBOs Two-Year Forecasts 10Growth in Output 12BOX : COMPARISON OF TWO - YEAR FORECASTS BY CBO AND THE FEDERAL RESERVE 14Inflation 18Interest Rates 21

Wages and Salaries 27

CBOs Five-Year Forecasts 30

Growth in Output 32Inflation 32

Wages and Salaries 35

Appendix: Forecast and Historical Data 41

List of Tables and Figures 45

About This Document 46


4/49 C

Summary

For more than three decades, the Congressional Bud-get Office (CBO) has prepared economic forecasts thatunderlie the agencys projections for the federal budgetand cost estimates for proposed federal legislation. In par-ticular, forecasts of output, income, inflation, and interestrates play a significant role in the agencys budgetary anal-ysis; for example, projections of wages and salaries areused to forecast individual income tax receipts.

CBO regularly evaluates the quality of its economicforecasts by comparing them with the economys actualperformance and with forecasts by the Administrationand the Blue Chipconsensusan average of about50 private-sector forecasts. Such comparisons indicate theextent to which uncertainty and imperfect informationmay have caused CBO to miss patterns or turning points in the economy. They also identify areas where

CBO has tended to make larger errors or less accurateforecasts than other analysts.

How Does CBOs Forecasting Record Compare with Those of the Administration and the Blue Chip Consensus?CBOs forecasts generally have been comparable in qual-ity with those of the Administration and the Blue Chip consensus. When CBOs projections have proved inaccu-rate by large margins, the errors have tended to reflect dif-ficulties shared by other forecasters.

Do CBOs Forecasts Exhibit Notable Bias? A simple and widely used indicator of statistical bias isthe mean errorthe average tendency of a forecast to below or high over an entire period. In general, CBOs fore-casts and those by the Administration and theBlue Chip consensus have had similar mean errors. Specifically,CBOs evaluation finds this:

For CBOs forecasts that look two years ahead, themean errors have generally been very small. Theagencys forecasts have shown slight tendencies tooverestimate future interest rates and wages and sala-ries (seeSummary Figure 1).

For CBOs forecasts that look five years ahead, themean errors imply a slightly stronger tendency to over-estimate inflation compared with that of the agencystwo-year forecastswhich largely accounts for highermean errors for growth in nominal output and in

wages and salaries. In other respects, the mean errorsgenerally resemble those for forecasts that look twoyears ahead.

How Accurate Are CBOs Forecasts? Accuracy is the degree to which forecast values are dis-persed around actual outcomes. One widely used measureof accuracy is the root mean square error. By that mea-sure, the forecasts by CBO, the Administration, and theBlue Chipconsensus have been about equally accurateover two-year periods (seeSummary Figure 2) as well asover five-year periods. CBOs evaluation finds this:

Among two-year forecasts by CBO since the early 1980s, forecast values deviated from actual outcomesby 1.4 percentage points per year for real (inflation-adjusted) output growth and by 0.8 percentage points

per year for inflation in the consumer price index.

Among five-year forecasts by CBO since the early 1980s, forecast values deviated from actual outcomesby 1.2 percentage points per year for real outputgrowth and by 0.6 percentage points per year for infla-tion in the consumer price index.


5/49

2 CBOS ECONOMIC FORECASTING RECORD: 2013 UPDATE JANU

CBO

Summary Figure 1.

Mean Error for Two-Year Forecasts(Percentage points)

Sources: Congressional Budget Office; Office of Management and Budget; and Aspen Publishers, Blue Chip Economic Indicators .

Notes: The mean error is the arithmetic average of the forecasting errors. To compare forecast and actual data, annual averages werecomputed for growth rates, inflation rates, interest rates, and wages and salaries as a share of output.

Errors are forecast values minus actual values; therefore, a positive error is an overestimate.

CPI = consumer price index; GDP = gross domestic product; n.a. = not applicable (the Blue Chip consensus does not include aforecast of wages and salaries).

a. The Blue Chip consensus is the average of approximately 50 private-sector forecasts.

b. The gross national product price index was forecast before 1992; the GDP price index was forecast from 1992 onward.

c. Forecasts of Moody's Aaa corporate bond rate were used for the years in which the interest rate on 10-year Treasury notes was not

forecast: 1984 and 1985 for CBO's forecasts and 1984 through 1995 for the Blue Chip consensus forecasts.

Change in Wages and Salaries as aShare of Output (19802010)

Growth in Wages andSalaries (19802010)

Interest Rate on 10-YearTreasury Notes (19842010 )c

Real (Inflation-adjusted)Interest Rate on Three-Month

Treasury Bills (19822010)

Interest Rate on Three-MonthTreasury Bills (19822010)

Difference BetweenInflation in the CPI and the

GDP Price Index (19822010 )b

Inflation in the ConsumerPrice Index (1982-2010)

Growth in Nominal Output(19822010)

Growth in Real Output(19822010)

-0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

n.a.

n.a.

CBO

Blue Chip Consensus aAdministration


6/49

SUMMARY CBOS ECONOMIC FORECASTING RECORD: 201

C

Summary Figure 2.

Root Mean Square Error for Two-Year Forecasts(Percentage points)


Notes: The root mean square error is calculated by first squaring the errors, then taking the square root of the arithmetic average of thesquared errors. To compare forecast and actual data, annual averages were computed for growth rates, inflation rates, interest rates,and wages and salaries as a share of output.

Errors are forecast values minus actual values; therefore, a positive error is an overestimate.

CPI = consumer price index; GDP = gross domestic product; n.a. = not applicable (the Blue Chip consensus does not include aforecast of wages and salaries).



c. Forecasts of Moodys Aaa corporate bond rate were used for the years in which the interest rate on 10-year Treasury notes was notforecast: 1984 and 1985 for CBOs forecasts and 1984 through 1995 for the Blue Chip consensus forecasts.

Change in Wages and Salaries as aShare of Output (19802010)

Growth in Wages andSalaries (19802010)

Interest Rate on 10-YearTreasury Notes (19842010)

Real (Inflation-adjusted)Interest Rate on Three-Month

Treasury Bills (19822010)

Interest Rate on Three-MonthTreasury Bills (19822010)

Difference BetweenInflation in the CPI and the

GDP Price Index (19822010)

Inflation in the ConsumerPrice Index (1982-2010)

Growth in Nominal Output(19822010)

Growth in Real Output(19822010)

0 0.5 1 1.5 2 2.5

n.a.

n.a.

b

c

CBO

Blue Chip Consensus a

Administration


7/49


CBO

What Are Some Sources of Forecasting Errors?Sources of large forecasting errors have included the diffi-culty of predicting:

Turning points in the business cyclethe beginning and end of recessions;

Changes in trends in productivity; and

Changes in crude oil prices.

In addition, revisions to the historical data (on outputand income, for example) that forecasters use for eco-nomic projections can complicate the task of interpreting forecasting errors. CBO used current vintages of histori-cal data to compute the forecasting errors and statistics.

Had the revised data been available to forecasters, ratherthan the original information that was available when theforecasts were produced, the forecasts themselves wouldhave been different. Despite that complication, recently published data present a simple and consistent point of comparison for evaluating forecasts by CBO and others.

How Do CBOs Assumptions About Fiscal Policy Affect Forecasting Errors?CBO constructs its economic projections under theassumption that federal fiscal policy will follow current

law, thereby providing a benchmark for lawmakers asthey consider potential changes in the law. In contrast,the Administrations forecasts assume the adoption of policies reflected in the Presidents proposed budget.Forecasters in the private sector (represented in theBlue Chip consensus) form their own assumptions about thefuture stance of federal fiscal policy, which may anticipatechanges in law.

Differences between forecasts, and thus differences inforecasting errors, sometimes arise from differentassumptions about fiscal policy, particularly when policy-makers are considering major changes to current law. Forexample, in 2009 and 2010, different fiscal policy assumptions caused CBOs two-year forecasts of realoutput growth to diverge noticeably from those of the Administration and the Blue Chipconsensus.


8/49 C

CBOs Economic Forecasting Record:

2013 Update

IntroductionReleased on a regular basis since 1976, the CongressionalBudget Offices (CBOs) macroeconomic forecast is aninput for the agencys projections for the federal budgetand cost estimates for proposed federal legislation. Forexample, projections of wages and salaries feed into theforecast of individual income tax receipts.

CBO regularly evaluates the quality of its economicforecasts by comparing them with the economys actualperformance and with forecasts by the Administrationand the Blue Chipconsensus (an average of approximately 50 private-sector forecasts that is published periodically in the Blue Chip Economic Indicators ).1 Such comparisonshelp CBO improve its economic projections. Specifically,they indicate the extent to which uncertainty and imper-fect informationfactors that affect all forecastersmay have caused CBO to miss patterns or turning points in

the economy.2

They also identify areas where CBO hastended to make larger errors or less accurate forecaststhan other analystsperhaps implying that the agency has not effectively used available information. Compari-sons with the Blue Chipconsensus forecast are particu-larly helpful in that regard, because the variety of forecastsit embodies is produced from a broader blend of sourcesand methods than can be expected from any single fore-caster. Consequently, over time, theBlue Chipconsensusforecasts may provide better estimates than those by any single forecaster.3

Despite their value, comparisons of forecasting errorscan be misleading when forecasts are made for differentpurposes. In particular, forecasters in the private sector

attempt to predict the future stance of federal fiscalpolicy, and the Administrations forecasts assume theadoption of the fiscal policy reflected in the Presidentsproposed budget. CBO, however, is required to assumethat fiscal policy in the future will reflect the provisions incurrent law, an approach that derives from the agencysresponsibility to provide a benchmark for lawmakers asthey consider proposed changes in law. Forecasting errorsmay be driven by those different assumptions, particu-larly when policymakers are considering major changes inthe fiscal policy embedded in current law.4

This report evaluates CBOs macroeconomic forecastsover two-year and five-year periods. The forecastsincluded in this evaluation were originally published inthe early months of the years 1976 through 2010. (Two-year average forecasts published in early 2011 could notbe included because the latest full-year historical data donot extend beyond 2011 for most indicators.)

Relative to the forecasting record that CBO published in2010, this evaluation now includes two-year forecastsconducted in 2009 and 2010 and five-year forecasts con-ducted in 2006 and 2007. 5 Those additional forecastsdid not significantly alter findings from the previousforecasting record. In general, the evaluations indicate

1. The appendix to this report gives further details on the choice of historical time-series data and on the sources of forecast data forthe comparisons.

2. See David Reifschneider and Peter Tulip,Gauging the Uncertainty of the Economic Outlook from Historical Forecasting Errors , Financeand Economics Discussion Series Working Paper No. 2007-60(Washington, D.C.: Board of Governors of the Federal ReserveSystem, November 2007).

3. See, for example, Andy Bauer and others, Forecast Evaluation with Cross-Sectional Data: TheBlue ChipSurveys,Economic Review , vol. 88, no. 2 (Federal Reserve Bank of Atlanta, 2003),pp. 1731; Henry Townsend, A Comparison of Several Consen-sus Forecasts,Business Economics , vol. 31, no. 1 (January 1996);and Robert Clemen, Combining Forecasts: A Review and

Annotated Bibliography,International Journal of Forecasting ,vol. 5, no. 4 (1989), pp. 559583.

4. Different assumptions about monetary policy also can createdifferences between CBOs forecasts and other forecasts. CBOsassumptions about monetary policy reflect the economic environ-ment that CBO expects under the fiscal policy specified in currentlaw.

5. See Congressional Budget Office,CBOs Economic Forecasting Record: 2010 Update (July 2010).
http://www.cbo.gov/publication/21520http://www.cbo.gov/publication/21520http://www.cbo.gov/publication/21520http://www.cbo.gov/publication/21520


9/49


CBO

that the quality of CBOs two- and five-year forecasts issimilar to that of other organizations.

Measuring the Quality of Forecasts

Like CBOs earlier studies of its economic forecasts, thisevaluation focuses on two indicators of quality: statisticalbias and accuracy. Other characteristics of forecastqualitysuch as the efficiency with which a forecast usesavailable informationare harder to assess.6

Statistical Bias. Statistical bias indicates the tendency of a forecast to err in a certain direction. To measure statisticalbias, CBO used the mean errorthe arithmetic averageof the forecasting errors, which is the simplest and most

widely used measure. Because it is a simple average, how-ever, underestimates and overestimates offset one another.

As a result, the mean error imperfectly measures the qual-ity of a forecast: A small mean error would result if all of the errors were small or if large overestimates and under-estimates generally balanced one another. As an alterna-tive to the mean error measure, several studies by analystsoutside of CBO have used more elaborate techniques totest for bias in the agencys forecasts.7

Accuracy.The accuracy of a forecast is the degree to which forecast values are dispersed around actual out-comes. Narrower dispersion indicates greater accuracy.CBO used two measures of accuracy in its evaluation:

The mean absolute errorthe average of the forecastserrors without regard to arithmetic signdoes not allow underestimates and overestimates to offset each other, incontrast with the mean error. The root mean square erroralso shows the size of the error without regard to sign, butit gives greater weight to larger errors.8

Limitations of Forecast EvaluationsThere are several reasons for caution in drawing conclusions from this evaluation of CBOs forecasts:

Historical track records only weakly indicate the possi-ble direction or size of inaccuracies in the future. Tosome extent, that fact results from changes in proce-dures used to develop economic forecasts by CBO and

other analysts over the past three decades. Moreover,the forecasters included in theBlue Chipconsensushave varied over time.

When preparing forecasts, CBO, unlike private fore-casters and the Administration, does not assume any future changes in federal fiscal policy other than thoseprescribed in current law.9

The various Administration forecasts normally includethe projected economic effects of those Administra-

tions policy proposals. The various private forecastersincluded in the Blue Chipsurvey make their ownassumptions about fiscal policy, but the survey doesnot report them.

The common practice of revising statistical data couldmean that forecasters make predictions about one

6. For studies that have examined the relative efficiency of CBOseconomic forecasts, see Michael T. Belongia, Are Economic Fore-casts by Government Agencies Biased? Accurate?Review , vol. 70,no. 6 (Federal Reserve Bank of St. Louis, November/December1988), pp. 1523; and Stephen M. Miller, Forecasting FederalBudget Deficits: How Reliable Are U.S. Congressional BudgetOffice Projections? Applied Economics,vol. 23 (December 1991),pp. 17891799. Although both studies identify information thatmight have been used to make CBOs forecasts more accurate,they rely on statistics that are valid only when sample sizes arelarger than those used in the evaluations. Moreover, althoughstatistical tests can identify sources of inefficiency in a forecastafter the fact, they generally do not indicate how such informationcould be used to improve forecasts when they are being made.

7. One such alternative approach to testing a forecast for bias isbased on linear regression analysis of actual values against forecastvalues. For details of that method, see Jacob A. Mincer and VictorZarnowitz, The Evaluation of Economic Forecasts, in Jacob A.Mincer, ed., Economic Forecasts and Expectations: Analysis of Fore-casting Behavior and Performance (Cambridge, Mass.: NationalBureau of Economic Research, 1969). Studies that have used thatmethod to evaluate short-term forecasts published by CBO andthe Administration have not found statistically strong evidence of bias. See, for example, George A. Krause and James W. Douglas,Institutional Design Versus Reputational Effects on BureaucraticPerformance: Evidence from U.S. Government Macroeconomicand Fiscal Projections, Journal of Public Administration Researchand Theory,vol. 15, no. 2 (April 2005), pp. 281306; J. KevinCorder, Managing Uncertainty: The Bias and Efficiency of Fed-eral Macroeconomic Forecasts, Journal of Public AdministrationResearch and Theory,vol. 15, no 1 (January 2005), pp. 5570;and Belongia, Are Economic Forecasts by Government AgenciesBiased? Accurate? For a more elaborate study of bias thatincluded CBOs forecasts among a sizable sample, see Corder,Managing Uncertainty; and David Laster, Paul Bennett, and InSun Geoum, R a tional Bias in Macroeconomic Forecasts,Staff Report No. 21 (Federal Reserve Bank of New York, March 1997).

8. The root mean square error is calculated by first squaring theerrors and then taking the square root of the arithmetic average of the squared errors. Squaring the errors places greater weight onlarger errors.

9. The purpose of current-law assumptions in CBOs economicforecasts is explored in Congressional Budget Office,What Is a Current-Law Economic Baseline? (June 2005).


10/49

JANUARY 2013 CBOS ECONOMIC FORECASTING RECORD: 201

C

concept of an economic variable and the statisticalagencies that compile those data ultimately report on a materially different concept. For example, in 1999,the Bureau of Economic Analysis (BEA) redefined

business and government spending on computer software as investment, which led to significant revisionsto historical estimates of investment, particularly dur-ing much of the 1990s.10

Some Sources of Forecasting ErrorThe physicist Niels Bohr is credited with saying thatPrediction is very difficult, especially if its about thefuture. There are indeed many ways that economic fore-casts can go wrong. Some key sources of error include thedifficulties of predicting turning points in the businesscycle, changes in productivity trends, and changes incrude oil prices. As well, revisions to historical data used by forecasters can complicate the interpretation of forecasting error.

Business Cycle Turning PointsPeaks and troughs (or turning points) in the businesscycle mark the beginning and end of recessions, which areperiods of significant contraction in economic activity.Forecasts by CBO, the Administration, and theBlue Chip consensus have made large overpredictions of real (infla-

tion-adjusted) output growth before each recession since1976, with the exception of the 1980 recession (seeFigure 1). Forecasting errors tend to be large around busi-ness cycle peaks (when a recession begins) for a numberof reasons:

Recessions are sometimes prompted by events orshocks that cannot be reasonably predicted by fore-casters. For example, in August 1990, the Iraqi inva-sion of Kuwait led to a spike in oil prices and a drop inconsumer confidence, which probably contributed to

the recession that followed.Economists cannot be sure that a recession has begununtil sufficient data are available. For example, theBusiness Cycle Dating Committee of the NationalBureau of Economic Research (NBER) did notannounce the December 2007 business cycle peak

until 11 months later. For that reason, forecasters may miss a recession even after it has started.

Business cycle turning points often occur during peri-ods of high uncertainty. For example, in January 2008,one month after the business cycle peak, CBOreported, The economic outlook this year is particu-larly vulnerable to uncertainty about the degree to

which the problems in the housing and financial mar-kets will spill over to affect other sectors of the econ-omy. Growth in 2008 could be weaker than CBOexpects if the turmoil in the financial markets leads toa more severe economywide curtailment of lending than CBO anticipates. Under such uncertain condi-tions, widely different outcomes can appear equally probable, making it difficult to gauge whether an eco-nomic downturn is imminent.

Changes in Productivity TrendsForecasts of productivity growth play a critical role inforecasting potential output, which is CBOs estimate of the amount of output that the economy would produce with a high rate of use of its capital and labor resources. As such, CBOs forecast of potential output shows how much the economy can sustainably grow during periodsof expansion and determines the trajectory of grossdomestic product (GDP) in the later years of the agencys10-year forecasts.

Labor productivity is the average real output per hour of work; by definition, real output equals labor productivity times the total number of labor hours worked. Somesources of growth in labor productivity include:

Capital accumulation (that is, more tools, equipment,structures, and infrastructure),

Education and skills development (also called invest-ment in human capital), and

Innovation (that is, the greater efficiency achievedthrough better tools, systems, or methods).

When forecasting productivity growth, CBO considershistorical trends in capital accumulation and the effects of public policy on incentives to invest. Shifts in such trendsmay be difficult to identify until several years after thefact. Consequently, forecasters may make incorrectassumptions about the trajectory of productivity growthand, therefore, potential output growth.

10. Previously, business and government spending on software wasconsidered to be the purchasing of an intermediate goodaninput in the production process and not a component of grossdomestic product.


11/49


CBO

Figure 1.

Errors in Forecasting the Two-Year Growth of Real Output NearBusiness Cycle Peaks(Percentage points)

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; and Departmentof Commerce, Bureau of Economic Analysis (BEA).

Notes: Errors are forecast values minus actual values; therefore, a positive error is an overestimate. Date labels refer to the initial year of thetwo-year period.

Errors are shown for forecasts conducted near business cycle peaks in January 1980, July 1981, July 1990, March 2001, andDecember 2007, as defined by the Business Cycle Dating Committee of the National Bureau of Economic Research.

Real (inflation-adjusted) output is either real gross domestic product (GDP) or real gross national product (GNP). GNP differs from

GDP primarily by including the capital income that residents earn from investments abroad and excluding the capital income thatnonresidents earn from domestic investment. GNP was forecast before 1992; GDP was forecast from 1992 onward. Errors are basedon the most recent data reported by BEA.

All forecasts were issued in the first half of the initial year of the period or in December of the preceding year.

a. As a point of comparison, the mean absolute error is one indicator of the accuracy of forecasts over the 19822010 period, excludingthose produced near a business cycle peak. The measure is the average of forecasting errors without regard to arithmetic sign.

b. As a point of comparison, the root mean square error is one indicator of the accuracy of forecasts over the 19822010 period, excludingthose produced near a business cycle peak. The measure is calculated by first squaring the errors, then taking the square root of thearithmetic average of the squared errors.

Since the early 1970s, forecasting errors reveal twounexpected shifts in productivity trends (seeFigure 2):

Following the 19731975 recession, labor productiv-ity growth in the nonfarm business sector did notreturn to the previous postwar trend rate of about2 percent per year. Over the next two decades, pro-ductivity grew more slowly, by about 1 percent peryear. Partly because most forecasters in the 1970sassumed that the productivity trend of the previousdecades would prevail, their forecasts of real output inthe mid- to late 1970s turned out to be too optimistic.Partly for the same reason, forecasters repeatedly underestimated inflation in the late 1970s.

In the late 1990s, growth in labor productivity inthe nonfarm business sector accelerated to nearly

3 percent per year. In part because most forecastersunderestimated, in several consecutive years, the trendrate of productivity growth, their predictions of theeconomys growth rate were too low and their predic-tions of inflation were too high.11 As the economy

1980 1981 1990 2001 2008 Mean AbsoluteError in

Other Years

Root MeanSquare Error in

Other Years

-1

0

1

2

3

4

5

a b

CBO

Blue Chip Consensus

Administration

11. See Spencer Krane, An Evaluation of Real GDP Forecasts: 19962001, Economic Perspectives , vol. 27, no. 1 (Federal Reserve Bank of Chicago, 2003), pp. 221; and Scott Schuh, An Evaluation of Recent Macroeconomic Forecast Errors,New England Economic Review (Federal Reserve Bank of Boston, January/February 2001),pp. 3556.


12/49


C

Figure 2.

Labor Productivity and Hours

Sources: Congressional Budget Office; Department of Labor, Bureau of Labor Statistics.

Notes: Data show labor productivity and hours in the nonfarm business sector.

Data are annual and are plotted through 2011.

continued to perform above expectations, analysts putmore effort into investigating the possible causes of the increase in productivity growth. Those investiga-tions initially focused on the possible contribution of technological progress that improved and quickenedthe flow of information among producers andbetween producers and consumers. Using revised data on production and inputs to production, CBO now estimates that an increase in the amount of capital(buildings, equipment, and software) per worker

sometimes called capital deepeningwas the primary source of the faster growth in productivity in the late1990s.12

In addition to misestimating labor productivity, making incorrect assumptions about growth in labor hours may also cause large forecasting inaccuracies. In the early

120

100

80

60

40

1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 20100

125

100

75

50

25

1945 1950 1955 1960 1965 1970 1975 1980 1985 1990 1995 2000 2005 20100

Real Output per Hour

Labor Hours

Trend Growth,19471973

(2.6 percent)


(1.4 percent)


(2.9 percent)

Trend

Actual

(Index, 2005 = 100, log scale)

(Index, 2005 = 100, log scale)

12. See Congressional Budget Office, Labor Productivity: Develop-ments Since 1995 (March 2007).


13/49


14/49


C

Figure 3.

Petroleum Prices and Consumer Inflation

Sources: Congressional Budget Office; Department of Labor, Bureau of Labor Statistics; Department of Commerce, Bureau of EconomicAnalysis.

Notes: Data are annual and are plotted through 2011.

CPI-U = consumer price index for all urban consumers.

a. The index for the price of petroleum imports is deflated by an index for consumer prices that excludes prices for food and energy.

b. In the CPI, major components of energy prices include motor fuel (which is primarily composed of petroleum products), electricity, andnatural gas purchased from utilities.

here involves various economic outcomes, including growth in output (in both real and nominal terms),inflation, the difference between inflation in the consumerprice index and the GDP price index, interest rates on3-month Treasury bills and 10-year Treasury notes, andchanges in wages and salaries (a significant part of taxable

income).16 (Box 1 on page 14 presents a comparison of CBOs forecasts of real output growth and inflation overtwo-year periods with those of the Federal Reserve.)

0

0.4

0.8

1.2

1.6

2.0

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

1965 1970 1975 1980 1985 1990 1995 2000 2005 2010

0

4

8

12

-2

2

6

10

14

Price of Petroleum Imports a

Consumer Price Index With and Without Energy Prices b

(Index, 2005 = 1)

(Perce ntage cha nge)

CPI-U

CPI-U Without

Energy Prices

16. Tables showing the errors of each forecast are available assupple-mental material on CBOs Web site ( www.cbo.gov ).
http://www.cbo.gov/publication/43846http://www.cbo.gov/publication/43846http://www.cbo.gov/http://www.cbo.gov/http://www.cbo.gov/publication/43846http://www.cbo.gov/publication/43846


15/49


CBO

Figure 4.

Forecasts by CBO and Revisions to Values for Real Gross Domestic Product (Percentage change from year ago)

Sources: Congressional Budget Office; Department of Commerce, Bureau of Economic Analysis.

Notes: Solid lines represent historical data that were available at the time each forecast was conducted. Dashed lines represent forecast data.

Real gross domestic product is the output of the economy adjusted to remove the effects of inflation.

Data are quarterly and are plotted through the fourth quarter of 2011.

Growth in Output Two-year forecasts of output growth by CBO, the

Administration, and the Blue Chipconsensus have moved

closely together over the past 30 years. As measured by the root mean square error, the projected two-year aver-age growth rate of output (both real and nominal) by allthree sets of forecasts deviated from the actual growth rateby roughly 1 percentage points between 1982 and2010. In large part, forecasting errors for output growthreveal forecasters difficulty in anticipating business cycleturning points and changing trends in productivity growth.

Growth in Real Output. Forecasting errors over the period

from 1976 to 1982 reflected the unusual economicdevelopments of the time:

Low productivity growth relative to the previoustrend,

High rates of inflation exacerbated by sudden andunexpected movements in petroleum prices, and

The Federal Reserves monetary policy, which resistedthose inflationary pressures and induced the two reces-sions that occurred between 1980 and 1982.17

In the late 1970s, CBO and the Administration, likemost forecasters, had expected productivity growth tomove back up to its earlier post-World War II trend,

which contributed to slight overpredictions of the growthin real output. Early in 1980, CBO and the Administra-tion anticipated the coming recession and producedrelatively accurate forecasts that year. However, forecastsconducted in the next two years did not anticipate theadvent and depth of the 19811982 recession, causing overpredictions of the growth in real output (seeFigure 5on page 16).

In 1983 and 1984, economic activity recovered strongly from the 19811982 recession, with real output growing faster than expected by CBO, the Administration, andthe Blue Chipconsensus. In forecasts conducted during the 19831989 expansion, CBO and the Blue Chip consensus underpredicted real output growth by roughly 1 percentage point, on average; in the Adminis-trations forecasts, underpredictions were notably lower,particularly during the latter half of the decade.

-6

4

-2

0

2

4

6

2006 2007 2008 2009 2010 2011

Current Actuals

January 2010Actuals and

ForecastJanuary 2009Actuals and

Forecast

17. The credit controls imposed in March 1980 contributed to theseverity of the recession in that year. See Stacey L. Schreft, CreditControls: 1980, Economic Review (Federal Reserve Bank of Richmond, November/December 1990), pp. 2555.


16/49


C

Table 1.

Summary Measures of Performance for Two-Year Forecasts(Percentage points)


Notes: Errors are forecast values minus actual values; therefore, a positive error is an overestimate.

CPI = consumer price index; GDP = gross domestic product; n.a. = not applicable.



c. Forecasts of Moody's Aaa corporate bond rate were used for the years in which the interest rate on 10-year Treasury notes was not

forecast: 1984 and 1985 for CBO's forecasts and 1984 through 1995 for the Blue Chip consensus forecasts.

Mean error -0.1 0.1 -0.1Mean absolute error 1.1 1.2 1.1Root mean square error 1.4 1.6 1.4

Mean error 0.1 0.4 0.3Mean absolute error 1.1 1.2 1.1Root mean square error 1. 5 1.7 1.5

Mean error 0.2 0.1 0.2Mean absolute error 0.7 0.7 0.7Root mean square error 0.8 0.9 0.9

Mean error -0.1 -0.2 -0.1Mean absolute error 0.3 0.4 0.4Root mean square error 0.4 0. 5 0.4




Mean error 0. 5 0.7 n.a.Mean absolute error 1.4 1.6 n.a.Root mean square error 1.9 2.1 n.a.

Mean error 0.2 0.2 n.a.Mean absolute error 0.8 0.8 n.a.Root mean square error 1.0 1.0 n.a.

Blue Chip

Difference Between Inflation in the CPI an d the GDP Price In dex (19822010) b

Interest Rate on Three-Month Treasury Bills (19822010)

Real Interest Rate on Three-Month Treasury Bills (19822010)

Interest Rate on 10-Year Treasury Notes (19842010) c

Growth in Wages an d Salaries (19802010)

Change in Wages an d Salaries as a Share of Output (19802010)

CBO Administration Consensus a

Growth in Real Output (19822010)

Growth in Nominal Output (19822010)

Inflation in the Consumer Price In dex (19822010)


17/49


CBO

Continued

Box 1.

Comparison of Two-Year Forecasts by CBO and the Federal ReserveLike those by the Administration and theBlue Chip consensus, forecasts by the Federal Reserve provide aninformative point of comparison when evaluating theCongressional Budget Offices (CBOs) forecasts. Butthe Federal Reserve does not immediately release itstwo-year forecasts of interest rates or of wages andsalaries, and it does not publish five-year forecasts.Therefore, CBOs principal analysis for this reportdid not include the Federal Reserves forecasts.However, the Federal Reserve has published timely two-year forecasts of real output growth and inflation

rates, allowing for a comparison of forecasts of thosevariables.

Since 1979, the staff of the Board of Governors of theFederal Reserve System has regularly prepareddetailed two-year macroeconomic forecasts for theFederal Open Market Committee (FOMC), the body

responsible for conducting monetary policy. Thoseforecasts are released to the public on a delayedscheduletypically five years later. In conjunction

with certain meetings of the FOMC, members of thecommitteethe Board of Governors and the presi-dents of the regional Federal Reserve Banksalsocompile their own forecasts for selected economicindicators; the range and central tendency of thoseforecasts have been published in the minutes of themeetings in recent years. CBOs comparison with theforecasts by the staff of the board covers 1979

through 2007; a comparison with the central ten-dency of the FOMC members forecasts of realoutput and inflation in consumer prices covers 2008through 2010. All of the Federal Reserves forecastsused in this analysis were issued in January or Febru-ary of the initial year of the forecast period or inDecember of the preceding year.

Growth in Real Output: Forecast Minus Actual

(Percentage points)

Sources: Congressional Budget Office; Board of Governors of the Federal Reserve System; and Department of Commerce, Bureau ofEconomic Analysis (BEA).

Notes: Errors are shown for forecasts of the average annual growth rate of real (inflation-adjusted) output over two-year periods.Date labels refer to the initial year of the two-year period.

Real output is either real gross domestic product (GDP) or real gross national product (GNP). Real GNP differs from real GDPprimarily by including the capital income that residents earn from investments abroad and excluding the capital income thatnonresidents earn from domestic investment. Real GNP was forecast before 1992; real GDP was forecast from 1992 onward.Errors are based on the most recent data reported by BEA.

From 2008 onward, growth rates were measured on a fourth-quarter-to-fourth-quarter basis.


a. Before 2008, forecasts were prepared by the staff of the Board of Governors of the Federal Reserve System. Between 2008and 2010, the shaded band encompasses the central tendency of Federal Reserve forecasts. The central tendency reflects theforecasts of the members of the Board of Governors and the presidents of the Federal Reserve Banks without the three highestand three lowest projections

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010-3-2-1

012345

CBO Federal Reserve a


18/49


C

Box 1. Continued

Comparison of Two-Year Forecasts by CBO and the Federal ReserveInflation in Consumer Prices: Forecast Minus Actual

(Percentage points)

Sources: Congressional Budget Office; Board of Governors of the Federal Reserve System; Department of Labor, Bureau of LaborStatistics (BLS); Department of Commerce, Bureau of Economic Analysis (BEA).

Notes: Errors are shown for forecasts of the average annual growth rate of the consumer price index (CPI) or the personalconsumption expenditures (PCE) price index over two-year periods. Date labels refer to the initial year of the two-yearperiod.

Before 2008, the CPI was forecast. For most years, the CPI-U (CPI for all urban consumers) was forecast. However, theCPI-W (CPI for urban wage earners and clerical workers) was forecast by CBO from 1976 through 1978 and f rom 1986through 1989. From 2008 onward, the PCE price index was forecast. Errors are based on the most recent data reported byBEA.

From 1986 onward, growth rates were measured on a fourth-quarter-to-fourth-quarter basis.


a. Before 2008, forecasts were prepared by the staff of the Board of Governors of the Federal Reserve System. Between 2008and 2010, the shaded band encompasses the central tendency of Federal Reserve forecasts. The central tendency reflects the

forecasts of the members of the Board of Governors and the presidents of the Federal Reserve Banks without the three highestand three lowest projections.

CBO and the Federal Reserve largely have had similarforecasts of the growth of real output over two-yearperiods (see the figure on the preceding page). Nota-ble divergences occurred during the early 1980s andin 2010. Before the 1980 recession, CBO produced a relatively accurate forecast of real output growth,

while the Federal Reserve overestimated the depth of the coming recession. However, in early 1981 and1982, CBO did not anticipate the advent or depth of

the 19811982 recession, while the Federal Reserveaccurately forecast the downturn and subsequentrecovery. In 2010, CBOs forecast correctly antici-pated a continued slow economic recovery following the 20072009 recession; however, that forecastassumed additional fiscal restraint from expiring tax provisions that were subsequently extended.1 In con-

trast, the central tendency of the Federal Reservesforecasts proved too optimistic.

In general, CBO and the Federal Reserve also hadsimilar forecasts of inflation (see the figure above).However, forecasts conducted between 2001 and2005 represent an exception. In early 2001, CBOsforecast overpredicted growth in consumer prices,largely because of the unexpected 2001 recession,

while the Federal Reserves expectations showed littleerror. Between 2003 and 2005, both forecastersunderpredicted inflation rates, but the errors by theFederal Reserve were somewhat larger.

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

CBO

Federal Reserve a

-5-4-3-2-10123

1. In early 2010, current law included the scheduled expirationof several tax provisions at the end of December 2010. Mostof those provisions were originally enacted in the EconomicGrowth and Tax Relief Reconciliation Act of 2001 and the

Jobs and Growth Tax Relief Reconciliation Act of 2003.


19/49


CBO

Figure 5.

Growth in Real Output: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; Department of

Commerce, Bureau of Economic Analysis (BEA).

Notes: Actual and forecast data show the average annual growth rate of real (inflation-adjusted) output over two-year periods. Date labels

refer to the initial year of the two-year period.

Real output is either real gross domestic product (GDP) or real gross national product (GNP). Real GNP differs from real GDP primarily

by including the capital income that residents earn from investments abroad and excluding the capital income that nonresidents earn

from domestic investment. Real GNP was forecast before 1992; real GDP was forecast from 1992 onward. Actual values show the most

recent data reported by BEA.


-2

0

2

4

6

8

Comparison of CBO Forecast and Actual Growth

Forecast Minus Actual(Perce ntage points)

(Average a nnual perce ntage cha nge)

CBO

Actual

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010-4

-2

0

2

4

6

CBOAdministration

Underestimate

Overestimate

ConsensusBlue Chip


20/49


C

The unexpected 19901991 recession resulted in over-predictions of real output growth in 1990. Even so, theerrors by CBO and the Blue Chip consensus that year

were actually smaller than the root mean square error forthe overall 19822010 period. In contrast, the Adminis-tration overpredicted growth by 2 percentage points in its1990 forecast.

In every year between 1992 and 1999, all of the forecastsunderpredicted two-year growth in real output, withvery large errors made between 1996 and 1999. Aboutone-fourth of the extent of those errors resulted from sub-sequent revisions that BEA made to the national incomeand product accounts (NIPAs), which included impor-tant definitional changes. Those data revisions aside, thesignificant underpredictions made between 1996 and

1999 reflect several important economic developmentsthat analysts did not anticipatein particular, the invest-ment boom of the late 1990s, which increased the capitalstock and thereby boosted labor productivity and realoutput more than many forecasters had expected.

Forecasts conducted in 2001 did not anticipate the rela-tively mild recession in that year. As a result, CBO andthe Blue Chipconsensus overpredicted real output growthby about 1 percentage points, and the Administrationoverpredicted growth by about 1 percentage points. Following the recession, economic activity underwent an

unusually slow recovery and weak expansion. During thattime, productivity continued to grow at the strong post-1995 rate while labor hours grew very little. In forecastsconducted between 2004 and 2006, expectations for realoutput growth proved too optimistic; however, errors by the Administration and the Blue Chipconsensus wereslightly smaller than those by CBO. Perhaps contributing to the overpredictions, rising energy prices (unanticipatedby many forecasters) dampened growth in real GDP by roughly a quarter of a percentage point in 2004, less thanhalf of a percentage point in 2005, and about a quarter of a percentage point during the first half of 2006.18

Forecasts conducted in 2007 and 2008 failed to antici-pate the growing imbalances in the housing and financialmarkets. During the early 2000s, real output growth waspartly supported by a boom in residential construction,

which was fueled by a growing bubble in house prices. By 2007, a downturn in the housing market was apparent,

and tensions in financial markets began to emerge.Despite those tensions, forecasts conducted in early 2008assumed that a recession would be avoided. For example,in January 2008, CBO reported, If a severe creditcrunch did occur, it would drive the economy intorecession by significantly curbing financial activity andconsumer spending. However, CBO assumes in its fore-cast that the Federal Reserve will implement policies toprevent such a crunch and that the financial sector iscapable of absorbing most of the losses it faces.19 Thoseassumptions did not hold true: In 2008, forecasts by CBO, the Administration, and the Blue Chipconsensusoverpredicted real output growth by at least 4 percentagepoints.

In 2009 and 2010, CBO produced relatively accurate

forecasts of the economic recovery. Relative to CBO, the Administration and the Blue Chipconsensus expected a faster economic recovery and overpredicted real outputgrowth in their 2009 and 2010 forecasts. During thoseyears, differences in fiscal policy assumptions causedCBOs forecast to diverge from outside forecasts. In early 2009, participants in the Blue Chipconsensus reportedthat they expected additional fiscal stimulus, whichimplied stronger output growth than under then-currentlaw.20 In early 2010, CBOs forecast assumed additionalfiscal restraint from expiring tax provisions that weresubsequently extended.21

Growth in Nominal Output. Differences in forecasting errors between real and nominal output growth indicateinaccuracies in projections of inflation in the GDP priceindex.22 (For information about the difference betweenthe GDP price index and the consumer price index (CPI), see the Inflation section below.) During the1980s and 1990s, CBO, the Administration, and theBlue Chipconsensus tended to overpredict inflation,

18. See Congressional Budget Office,The Economic Effects of Recent Increases in Energy Prices (July 2006), p. 6.

19. See Congressional Budget Office,The Budget and Economic Outlook: Fiscal Years 2008 to 2018 (January 2008), p. 23.

20. See Congressional Budget Office,The Budget and Economic Outlook: Fiscal Years 2009 to 2019 (January 2009), pp. 1011.

21. In early 2010, current law included the scheduled expiration of several tax provisions at the end of December 2010. Most of thoseprovisions were originally enacted in the Economic Growth andTax Relief Reconciliation Act of 2001 and the Jobs and GrowthTax Relief Reconciliation Act of 2003.

22. Gross national product and its price index were forecast by CBO,the Administration, and the Blue Chipconsensus before 1992;GDP and its price index were forecast from 1992 onward.
http://www.cbo.gov/publication/17984http://www.cbo.gov/publication/17984http://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/89xx/doc8917/01-23-2008_budgetoutlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/89xx/doc8917/01-23-2008_budgetoutlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/99xx/doc9957/01-07-outlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/99xx/doc9957/01-07-outlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/99xx/doc9957/01-07-outlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/99xx/doc9957/01-07-outlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/89xx/doc8917/01-23-2008_budgetoutlook.pdfhttp://www.cbo.gov/sites/default/files/cbofiles/ftpdocs/89xx/doc8917/01-23-2008_budgetoutlook.pdfhttp://www.cbo.gov/publication/17984http://www.cbo.gov/publication/17984


21/49

18 CBOS ECONOMIC FORECASTING RECORD: 2013 UPDATE JANUARY

CBO

which partially offset underpredictions of real outputgrowth. Consequently, forecasts of nominal outputgrowth appear to have less bias over that period than doforecasts of real output growth.

During much of the 2000s, CBO, the Administration,and the Blue Chipconsensus tended to underpredictinflation rates, which generally offset overpredictions of real output growth. However, forecasts conducted in2008 provided a notable exception. Early that year, allthree sets of forecasts assumed that the economy wouldavoid a recession and therefore overpredicted both infla-tion and real output growth. As a result, forecasts con-ducted in that year overpredicted nominal output growthby 4 to 5 percentage points (see Figure 6).

Inflation

The errors in inflation forecasts generally have reflectedturbulence in crude oil prices and variation in the stateof the economy. For example, rapidly rising oil pricescontributed to forecasters sizable underpredictions of inflation during the late 1970s and mid-2000s. During the early 1980s, the deep recession dramatically andunexpectedly reduced the rate of inflation, but forecastersonly gradually recognized the extent of that reductionand consequently made large overpredictions of pricegrowth during much of the decade.

The evaluation focuses on two measures of inflation thatare important for projecting federal outlays and revenues.One is the consumer price index, which measures infla-tion in the prices of a fixed basket of consumer goodsand services.23 Forecasts of federal outlays depend onexpected inflation in that index. For example, the CPI isused to annually adjust payments to Social Security bene-ficiaries. Federal revenues also depend on inflation inconsumer prices, because elements of the individualincome tax, such as tax brackets, have been indexed to theCPI since the mid-1980s. All else being equal, higher

inflation in the CPI implies faster growth in outlays andslower growth in revenues.

The second measure is the difference between the rate of inflation in the CPI and the rate of inflation in the priceindex for GDP. The GDP price index is a summary mea-sure of the prices of all goods and services that make upgross domestic product. Its growth is a critical determi-nant in forecasting the growth of nominal GDP and,therefore, the growth of income subject to federal taxes. All else being equal, higher inflation in the GDP priceindex implies faster growth in revenues. Consequently,if the GDP price index was forecast to grow more slowly than the CPI, the projected deficit would be larger than if the reverse was forecast.

Inflation in the CPI. During the late 1970s, CBO and the Administration made similarly large errors in forecastsof CPI inflation (seeFigure 7 on page 20). Primarily because of the spike in crude oil prices in 1979 and 1980,forecasts conducted in 1978 and 1979 underpredictedinflation by about 4 percentage points, on average.

In forecasts conducted between 1982 and 1986, CBO,the Administration, and the Blue Chipconsensus over-predicted inflation in the CPI by about 1 percentagepoints, on average. That tendency largely stemmed fromthe fact that the 19811982 recession led to an unantici-pated sharp and lasting reduction in the rate of inflation. As well, the forecasters did not expect the drop in crudeoil prices that occurred in early 1986.

Between 1987 and 2003, CBO, the Administration, andthe Blue Chipconsensus made relatively small errors inforecasts of inflation in the CPI, with a root mean squareerror of roughly one-half of a percentage point. Inflationforecasts probably benefited from the relatively benigneconomic environment during most of that period, incontrast to the turbulence of the late 1970s and early 1980s. Growth in the CPI remained within a narrow range, particularly after 1990.

Between 2004 and 2007, the forecasts were persistently optimistic about inflation in the CPI largely because of the unexpected rise in crude oil prices, underpredicting two-year inflation rates by about 1 percentage point, onaverage.

In 2008, CBO, the Administration, and the Blue Chip consensus did not anticipate the 20072009 recession

23. In most years, the inflation forecasts are for the CPI-U, whichmeasures inflation in the prices paid by all urban consumers. Inthe period from 1976 to 1978 and from 1986 to 1989, CBO fore-cast the CPI-W, which measures inflation in the prices paid by urban wage earners and clerical workers, while the Administrationforecast the CPI-W through 1991. For evaluation purposes, thedistinction between the two measures was consequential mainly in1984, when inflation in the CPI-U and CPI-W diverged by 0.9 percentage points.


22/49


C

Figure 6.

Growth in Nominal Output: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; Department ofCommerce, Bureau of Economic Analysis (BEA).

Notes: Actual and forecast data show the average annual growth rate of nominal output over two-year periods. Date labels refer to the initialyear of the two-year period.

Nominal output is either gross domestic product (GDP) or gross national product (GNP). GNP differs from GDP primarily by includingthe capital income that residents earn from investments abroad and excluding the capital income that nonresidents earn f romdomestic investment. GNP was forecast before 1992; GDP was forecast from 1992 onward. Actual values show the most recent datareported by BEA.


-2

0

2

4

6

8

10

12

14




CBO

Actual

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010-4

-2

0

2

4

6

CBO

Administration

Underestimate

Overestimate

ConsensusBlue Chip


23/49


CBO

Figure 7.

Inflation in the Consumer Price Index: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; Department ofLabor, Bureau of Labor Statistics (BLS).

Notes: Actual and forecast data show the average annual growth rate of the consumer price index over two-year periods. Date labels refer tothe initial year of the two-year period.

Before 1978, BLS published only one consumer price index series, now known as the CPI-W. In January 1978, the bureau beganpublishing the CPI-U. For most years since 1979, the CPI-U was forecast. However, the CPI-W was forecast by CBO from 1986 through1989 and by the Administration through 1991.


CPI-U = consumer price index for all urban consumers; CPI-W = consumer price index for urban wage earners and clerical workers.

-6

-4

-2

0

2

4

Comparison of CBO Forecast and Actual Inflation


(Average a nnual rate of i nflation)

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

CBO

Administration

Underestimate

Overestimate

0

2

4

6

8

10

12

14

CBO

Actual

ConsensusBlue Chip


24/49


C

and the downward pressure that the recession wouldplace on consumer price growth. As a result, CBO andthe Administration overpredicted inflation slightly, asdid the Blue Chipconsensus to a greater extent. Forecasts

conducted in 2009 proved relatively accurate, and fore-casts in 2010 slightly underpredicted inflation.

Difference Between Inflation Measures. For forecastsconducted between 1978 and 1980, CBO and the

Administration underpredicted the difference in inflationmeasures by more than 2 percentage points, on average.In 1979 and 1980, the difference between inflation in theCPI and the gross national product (GNP) price index spiked to levels unprecedented during the postwar period(seeFigure 8). A significant part of the divergence can beexplained by the oil price shock; a surge in oil prices has a larger effect on the CPI than on the GNP price index because petroleum products represent a much larger shareof the goods and services consumed in this country thanof the goods and services produced. Even so, the gapbetween the two inflation measures was unusually wide

with the effect of energy prices removed.

In forecasts conducted through early 1999, the tendency to underpredict the difference between the inflation mea-sures largely reflected the methodological change to thenational income and product accounts that year, whenBEA added business and government purchases of soft-

ware to investment and, therefore, to GDP. Because theprice index for software purchases grew much less rapidly than other prices, on average, the change in the classifica-tion of software spending caused a downward revision of the historical data for the growth of the GDP price index.Hence, the forecasts made before 2000 were based on a pattern of historical growth in the GDP price index that

was higher than is currently reported. That differenceprobably accounted for about 0.2 percentage pointsortwo-thirdsof the apparent bias in forecasts for thatperiod.

Between 2002 and 2007, the difference between theinflation measures narrowed, with the CPI growing atroughly the same rate as the GDP price index. Thatchange in relationship reflects an increase in inflation forgoods and services measured by the GDP price index but not by the CPI, including some investment goods(particularly those involved in business, residential, andgovernment structures) and military compensation.Those developments were not generally anticipated, so

forecasts conducted between 2001 and 2006 overesti-mated the difference between inflation measures.

Interest Rates

On average, between 1982 and 2010, CBO, the Administration, and the Blue Chipconsensus tended tooverpredict interest rates. Forecasts by the Administrationappear less biased over the period because large negativeforecasting errors in the late 1980s offset positive errorsduring other periods (particularly the 2000s). Notably,forecasts made after the 19901991, 2001, and 20072009 recessions underestimated the duration of the eas-ing of monetary policy, which largely accounts for thetendency to overpredict interest rates.

CBO forecasts interest rates on Treasury securities toproject payments on the federal debt and other compo-nents of the budget. Those forecasts focus on two key ratesthe rate on 3-month Treasury bills and that on10-year Treasury notes. All else being equal, higher inter-est rates result in larger interest payments and fastergrowth in federal debt held by the public. Forecasts of interest rates depend on a variety of factors, including these:

Monetary policy . During periods of low inflation andhigh unemployment, for example, the Federal Reserveattempts to stimulate demand by lowering short-terminterest rates, which in turn can lower the cost of borrowing over longer periods of time.

Inflation. Expectations of inflation are embedded ininterest rates. Interest rates generally rise, for example,

when participants in financial markets expect a higherrate of inflation in the future. Moreover, the FederalReserve has responded to increasing inflationary pressures by taking actions to raise interest rates.

The issuance of debt securities . The federal governmentissues Treasury securities to finance budget deficits. Allelse being equal, an increase in the supply of thosesecurities would tend to increase interest rates.

Turmoil in the financial system. In periods when inves-tors have been increasingly concerned about the safety of their investments, they have sought to hold moreU.S. Treasury securities. Such an increase in demandlowers interest rates on those securities.


25/49


CBO

Figure 8.

Difference Between Inflation in the CPI and the GDP Price Index: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; Department ofLabor, Bureau of Labor Statistics (BLS); Department of Commerce, Bureau of Economic Analysis (BEA).

Notes: Actual and forecast data show the difference between average annual inflation measures (the CPI minus the GDP price index) overtwo-year periods. Date labels refer to the initial year of the two-year period.

The gross national product price index was forecast before 1992; the GDP price index was forecast from 1992 onward. Actual valuesshow the most recent data reported by BEA.

Before 1978, BLS published only one CPI series, now known as the CPI-W. In January 1978, the bureau beganpublishing the CPI-U. For most years since 1979, the CPI-U was forecast. However, the CPI-W was forecast by CBO from 1986 through1989 and by the Administration through 1991.


CPI = consumer price index; GDP = gross domestic product; CPI-U = consumer price index for all urban consumers;CPI-W = consumer price index for urban wage earners and clerical workers.

-1

0

1

2

3

4Comparison of CBO Forecast and Actual Difference


(Perce ntage points)

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Underestimate

Overestimate

Actual

-4

-3

-2

-1

0

1

CBO

Administration

ConsensusBlue Chip

CBO


26/49


C

CBO has evaluated forecasts of the interest rate on three-month Treasury bills in both nominal and real terms. Thenominal rate of interest is the rate quoted in the second-ary market.24 The real interest rate used here equals the

nominal rate minus predicted inflation.Interest Rate on Three-Month Treasury Bills. In 1978and 1979, both CBO and the Administration under-predicted the two-year average nominal interest rate onthree-month Treasury bills by about 2 percentagepoints, on average (seeFigure 9). That tendency stemmed from underpredicting inflation rates during that period. In fact, CBO and the Administration over-predicted real interest rates by more than 1 percentagepoints, on average, in those years (seeFigure 10). Thoseoverpredictions may have stemmed from the agenciesoverly optimistic forecasts of real output growth during the period.

During the early 1980s, actions by the Federal Reserveraised interest rates, which contributed to two consecu-tive recessions and ultimately a sharp and lasting cut inthe rate of inflation. In 1980 and 1981, many forecastersdid not fully anticipate that prolonged period of restraintin monetary policy or its impact on price growth. In1980, CBO and the Administration underpredicted bothnominal and real interest rates, suggesting that monetary

policy proved tighter than expected over the following two years. In 1981, CBOs forecast overpredicted nomi-nal interest rates but underpredicted real interest rates,reflecting an overprediction of inflation; the Administra-tions forecast underpredicted both nominal and realinterest rates.

Between 1982 and 2010, forecasts of nominal interestrates on three-month Treasury bills displayed notableupward bias. On average, forecasts by CBO and theBlue Chip consensus overpredicted nominal interest rates by 0.6 percentage points; the Administration overpredictednominal interest rates by 0.3 percentage points. In part,that bias stemmed from forecasters tendency to overpre-dict inflation rates between 1982 and 1998. Forecastersdifficulty in anticipating business cycle turning pointsalso accounts for overpredictions around the 19901991,2001, and 20072009 recessions.

Since 1990, forecasts made after recessions have tended tounderestimate the duration of the easing of monetary policy. For example, forecasts conducted in early 1991and 1992 expected interest rates to begin rising as the

economy recovered from the 19901991 recession. Therecovery, however, was unexpectedly weak, and inflationremained low, so the Federal Reserve continued to easemonetary policy for several years, pushing down thenominal interest rate on three-month Treasury bills from7.8 percent in the first half of 1990 to roughly 3 percentin 1993.

In forecasts conducted during the 20002010 period,CBO, the Administration, and the Blue Chipconsensusoverpredicted real interest rates by about 1 percentagepoints, on average. Much of that bias can be attributed to

the 2001 and 20072009 recessions and to the surpris-ingly sluggish recovery in economic activity following those downturns. Indeed, despite the Federal Reservesprolonged easing of monetary policy following the 2001recession, real output growth during the mid-2000sremained weaker than expected.

Interest Rate on 10-Year Treasury Notes. Between 1984and 2010, CBO and the Blue Chipoverpredicted thenominal interest rate on 10-year Treasury notes by 0.4 percentage points, on average (seeFigure 11 onpage 26).25 Forecasts by the Administration appear lessbiased (overpredicting the rate by 0.1 percentage point,on average), because large negative forecasting errors inthe late 1980s and early 1990s offset positive errors dur-ing other periods (particularly the 2000s). As measuredby the root mean square error, forecasts by CBO and theBlue Chipconsensus deviated from actual interest rates by about 0.7 percentage points, on average; forecasts by the Administration deviated from actual interest rates by about 0.9 percentage points, on average.

Between 2000 and 2008, CBO, the Administration, and

the Blue Chipconsensus persistently overpredicted thenominal interest rate on 10-year Treasury notes (by 0.7 percentage points, on average). To some extent, the2001 and 20072009 recessions explain overpredictions

24. The rate on newly issued bills was forecast by the Administrationthrough 2000 and by the Blue Chipconsensus from 1982 to 1985and from 1992 to 1997.

25. For simplicity of exposition, this evaluation refers to 10-yearTreasury notes. However, forecasts of the Moodys Aaa corporatebond rate were used in years when forecasts of 10-year Treasury notes were not made. Those years are 1984 and 1985 for CBOsforecasts and 1984 through 1995 for the Blue Chipconsensusforecasts.


27/49


CBO

Figure 9.

Interest Rate on Three-Month Treasury Bills: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; Federal Reserve;Department of the Treasury, Bureau of the Public Debt; Haver Analytics.

Notes: Actual and forecast data show the geometric average of the secondary-market interest rate over two-year periods.

The rate on newly issued bills was forecast by the Administration through 2000 and by the Blue Chip consensus from 1982 to 1985 andfrom 1992 to 1997.


by forecasts conducted before and during those down-turns. To some extent, optimistic forecasts of real outputgrowth probably account for overpredictions during theexpansionary period of the mid-2000s. Given thatoutlook for faster growth in the economy, forecastersprobably expected the Federal Reserve to try to temper

that growth and the inflationary pressures that could haveresulted from it.

By early 2009, forecasters had revised their expectationsfor interest rates downward in the wake of the recession.For that reason, forecasts of the rate for 10-year Treasury

Comparison of CBO Forecast and Actual Interest Rate


(Perce nt)

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Underestimate

Overestimate

CBO

Actual

-4

4CBO

Administration

0

2

4

6

8

10

12

14

-3

-2

-1

0

1

2

3ConsensusBlue Chip


28/49


C

Figure 10.

Real Interest Rate on Three-Month Treasury Bills: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; the Federal

Reserve; Department of Labor, Bureau of Labor Statistics (BLS); Department of the Treasury, Bureau of the Public Debt;Haver Analytics.

Notes: Actual and forecast data show the geometric average of the secondary-market interest rate deflated by growth in the consumer priceindex over two-year periods.

The rate on newly issued bills was forecast by the Administration through 2000 and by the Blue Chip consensus from 1982 to 1985 andfrom 1992 to 1997.

Before 1978, BLS published only one consumer price index series, now known as the CPI-W. In January 1978, the bureau beganpublishing the CPI-U. For most years since 1979, the CPI-U was forecast. However, the CPI-W was forecast by CBO from 1986 through1989 and by the Administration through 1991.


CPI-U = consumer price index for all urban consumers; CPI-W = consumer price index for urban wage earners and clerical workers.



1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Underestimate

Overestimate

CBO

Actual

(Perce nt)

-4

-2

0

2

4

6

-3

-2

-1

0

1

2

3

4

CBO

Administration

-4

ConsensusBlue Chip


29/49


CBO

Figure 11.

Interest Rate on 10-Year Treasury Notes: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Aspen Publishers, Blue Chip Economic Indicators ; and the FederalReserve.

Notes: Actual and forecast data show the geometric average of the interest rate over two-year periods. Date labels refer to the initial year ofthe two-year period.

Forecasts of Moodys Aaa corporate bond rate were used for the years in which the interest rate on 10-year Treasury notes was notforecast: 1984 and 1985 for CBOs forecasts and 1984 through 1995 for the Blue Chip consensus forecasts.



-2

-1

0

1

2


(Perce nt)

CBO

Administration

CBO

Actual

0

2

4

6

8

10

12

1976 19 78 1980 1 98 2 1984 19 86 1988 1990 19 92 1994 1 99 6 1998 20 00 2002 2004 20 06 2008 2 01 0

1 976 1978 19 80 1982 1 984 198 6 1988 1 990 1992 19 94 1996 1 998 200 0 2002 2 00 4 2006 20 08 2010

Underestimate

Overestimate

ConsensusBlue Chip


30/49

JANUARY 2013 CBOS ECONOMIC FORECASTING RECORD: 2013 UP

C

notes in 2009 were relatively accurate, deviating from actual interest rates by less than one-quarter of a percentage point.

In early 2010, long-term interest rates were expected torise, on average, during the economic recovery; however,rates continued to decline over the next two years (partic-ularly in 2011). CBOs 2010 forecast overpredicted the10-year Treasury note rate by 0.7 percentage points; fore-casts by the Administration and theBlue Chipconsensusoverpredicted that rate by about 1.2 percentage points.

Wages and SalariesParticularly since 2001, CBO and the Administrationhave tended to overpredict growth in wages and salariesand the change in wages and salaries as a percentage of

GDP.26

To some extent, the fact that forecasters did notanticipate the 2001 and 20072009 recessions accountedfor that tendency. However, both agencies also were sur-prised by the unusually sluggish recovery in wages andsalaries relative to output following the two recessions.

Projections of federal revenues importantly depend onforecasts of wages and salaries, which are a major compo-nent of taxable income.27 Errors in forecasts of wages andsalaries may result from inaccurate forecasts of variousitems:

Gross domestic product . Wages and salaries generally grow with overall economic activity and inflation. A forecast that fails to anticipate a downturn in outputgrowth would probably overpredict growth in wagesand salaries as well.

The statistical discrepancy between GDP and gross domestic income (GDI, the income earned in the pro-duction of GDP). In principle, GDP and GDI shouldbe equal, but in practice, they differ because BEA usesdifferent primary sources to estimate product on theone hand and income on the other. To forecast GDI,forecasters must also project the statistical discrepancy,

which is difficult because the discrepancy stems from

imperfect data collection and estimation processes.Unexpected swings in the discrepancy may artificially inflate or deflate wages and salaries relative to GDP.

Income shares . Income shares refer to the percentage of each type of income in GDI.28 Unexpected shifts inthe composition of income may cause sizable errors inforecasts of wages and salaries.

Because theBlue Chipconsensus does not report forecastsof wages and salaries, the evaluation here discusses only forecasts conducted by CBO and the Administration.

Growth in Wages and Salaries. Between 1980 and 2010,the projected growth in wages and salaries exceededactual growth by 0.5 percentage points for CBO and by

0.7 percentage points for the Administration, on average(seeFigure 12). As measured by the root mean squareerror, forecasts by both agencies deviated from actualgrowth by about 2 percentage points during that period.The directions of the errors in forecasting the growth of wages and salaries were similar to those for the errors inforecasts of nominal output, indicating that the errorsstemmed in part from errors in predicting the growth of both real output and prices.

Change in Wages and Salaries as a Share of Output.To isolate the errors that were unique to the forecasts of

wages and salaries, evaluating those forecasts as a share of output is helpful (seeFigure 13). Historically,two patterns have been notable:

Wages and salaries as a share of output typically movein a cyclical pattern, falling during periods of highunemployment and rising when labor markets tighten.

Since the early 1970s, the share has followed a downward trend. In part, that trend has stemmed from thefact that employers and employees have preferred tosubstitute untaxed noncash, or fringe, benefits (suchas employer-paid health insurance premiums andpension contributions) for taxable wages and salaries.29

26. Reported data refer to wage and salary disbursements rather thanaccruals.

27. In past editions of this report, CBO included an analysis of itsforecast of the sum of wages and salaries and corporate book profits. That sum has been dropped from the analysis becauselegislative changes to the tax rules affecting corporations can affectbook profits and have increasingly done so, which makes it diffi-cult to identify the economic forecasting errors. Wages and salariesare less directly affected by legislation.

28. Gross domestic income includes wages and salaries, domesticeconomic profits, employee benefits, proprietors income, rentalincome, net interest payments, taxes on production and imports,the surplus of government enterprises, business current transferpayments, and depreciationall minus subsidies.

29. Further details about contributions to defined-benefit pensionplans are outlined in Congressional Budget Office,T he Budget and Economic Outlook: An Update (August 2005), Box 2-2,pp. 3233.


31/49


CBO

Figure 12.

Growth in Wages and Salaries: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Department of Commerce, Bureau of Economic Analysis (BEA).

Notes: Actual and forecast data show the average annual growth rate of wage and salary disbursements over two-year periods. Date labelsrefer to the initial year of the two-year period.

Actual values show the most recent data reported by BEA. The Blue Chip consensus does not include forecasts of wages and salaries.


Between 1980 and 2010, forecasts of the two-year changein the wage and salary share displayed a very slight biason average (overpredictions of less than one-quarter of a percentage point).30 As measured by the root mean squareerror, forecasts by both CBO and the Administration

deviated from the actual change in the share by about1 percentage point.



1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

Underestimate

Overestimate

CBO


CBO

Administration

-2

0

2

4

6

8

10

12

-4

-2

0

2

4

6

8

Actual

30. For forecasts conducted before 1992, wages and salaries werecomputed as a percentage of GNP; from 1992 onward, they werecomputed as a percentage of GDP.


32/49


C

Figure 13.

Change in Wages and Salaries as a Share of Output: Two-Year Forecasts

Sources: Congressional Budget Office; Office of Management and Budget; Department of Commerce, Bureau of Economic Analysis (BEA).Notes: Actual and forecast data show the change in wage and salary disbursements as a percentage of output over two-year periods. Date

labels refer to the initial year of the two-year period.

Output is either gross domestic product (GDP) or gross national product (GNP). GNP differs from GDP primarily by including thecapital income that residents earn from investments abroad and excluding the capital income that nonresidents earn f rom domesticinvestment. GNP was forecast before 1992; GDP was forecast from 1992 onward.

Actual values show the most recent data reported by BEA. The Blue Chip consensus does not include forecasts of wages and salaries.


Comparison of CBO Forecast and Actual Change


1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010

CBO

(Perce ntage points)

Actual

-3

-2

-1

0

1

2

Underestimate

Overestimate

-2

-1

0

1

2

3

CBO

Administration


33/49


CBO

During the first half of the 1980s, wages and salaries fellmarkedly as a percentage of GNP. In large part, thatdecline can be attributed to the 1980 and 19811982recessions. CBO and the Administration correctly antici-

pated a decline in the wage and salary share but actually overpredicted the decline in most years.

Following a slight rebound in the wage and salary shareduring the mid-1980s, it generally declined through thefirst half of the 1990s. To a large extent, that declinederived from a large and unexpected increase in the statis-tical discrepancy, indicating that the measure of totaloutput grew faster than the measure of total income.That shift in the discrepancy probably explains over-predictions made by both agencies during the period.

In the late 1990s, wa

cbo's economic forecasting record: 2013 update

Documents