SlideShare una empresa de Scribd logo
1 de 17
Descargar para leer sin conexión
Estimating the Uncertainty of the
Economic Forecast Using CBO’s
Bayesian Vector Autoregression Model
January 2023
1
For details about the analysis, see Congressional Budget Office, CBO’s Current View of the Economy in 2023 and 2024 and the Budgetary Implications (November 2022),
www.cbo.gov/publication/58757. In the figure, real GDP growth is calculated for a given quarter relative to four quarters earlier. Q = quarter.
CBO’s View of the Economy as of November 2022
In November 2022, the
Congressional Budget Office
was asked about its view of
the economy. From the fourth
quarter of 2022 to the fourth
quarter of 2023, CBO
estimated, there is a two-
thirds chance that growth in
economic output—specifically,
gross domestic product (GDP)
adjusted to remove the effects
of inflation, or real GDP—will
be between −2.0 percent and
1.8 percent.
2
For discussion of the first step, see Robert W. Arnold, How CBO Produces Its 10-Year Economic Forecast, Working Paper 2018-02 (Congressional Budget Office, February 2018),
www.cbo.gov/publication/53537. For discussion of the second step, see Congressional Budget Office, “Estimating the Uncertainty of the Economic Forecast Using CBO’s Expanded
Markov-Switching Model” (January 2023), www.cbo.gov/publication/58884. For related discussion, see Mark Lasky, The Congressional Budget Office’s Small-Scale Policy Model,
Working Paper 2022-08 (Congressional Budget Office, September 2022), www.cbo.gov/publication/57254.
The analysis of economic uncertainty was conducted in three main steps:
▪ Preliminary economic projections provided central estimates for each variable;
▪ 100 simulations of the rates of unemployment, inflation, and interest were jointly
estimated around the central estimates, reflecting asymmetric dynamics and relating
the variables through an expectations-augmented Phillips curve and an inertial
Taylor rule; and
▪ Forecasts conditional on those rates were estimated using symmetric distributions in
which economic output and other variables were synchronized with the simulations
of unemployment, inflation, and interest rates.
This document focuses on the third step, which used a Bayesian vector autoregression
(BVAR) model.
CBO’s Analytic Method for Estimating Uncertainty
3
The BVAR model draws on historical correlations between macroeconomic variables
to produce conditional forecasts.
Key conditions are the simulations of the rates of unemployment, inflation, and
interest, which are estimated using an expanded version of CBO’s Markov-switching
model with asymmetric dynamics in which the unemployment rate rises rapidly in
some periods and falls gradually in others and interest rates do not fall below zero.
The projections of economic output and other variables are synchronized—using
symmetric distributions—with the simulations of unemployment, inflation, and interest
rates. Additional variables that can be simulated with symmetric distributions can be
easily incorporated.
The historical correlations between macroeconomic variables used in the model may
be less predictive of future outcomes in the event of extreme changes in economic
conditions.
Historical Dynamics Reflected in the Modeling
4
Inputs include central forecasts from CBO’s large-scale macroeconometric model for
26 variables used to analyze effects of economic conditions on the federal budget.
Additional inputs are 100 simulations of six variables from CBO’s expanded Markov-
switching model: the unemployment rate, two inflation rates, and three interest rates.
For each calendar quarter, CBO uses those inputs to the BVAR model to project
values for the remaining 20 variables.
After 100 simulations are generated, the values are calibrated so that their average
equals CBO’s central forecast.
The parameters of the model are estimated using data from 1959 through 2022.
How CBO Uses the Model
5
The unemployment rate projection is taken as a set of conditions.
CBO uses the BVAR model to project the following:
▪ Payroll employment,
▪ The number of people in the labor force,
▪ Hours of work, and
▪ Wages and salaries.
Labor Market Variables Projected Using the Model
6
Two rates of inflation are taken as sets of conditions—the overall rate as measured
by the personal consumption expenditures price index and that rate excluding food
and energy prices.
CBO uses the BVAR model to project inflation as measured by:
▪ The GDP price index,
▪ The consumer price index for all urban consumers,
▪ The consumer price index for food at home, and
▪ The consumer price index for medical care.
Inflation Rates Projected Using the Model
7
Three interest rates are taken as sets of conditions—the federal funds rate (the rate
that financial institutions charge each other for overnight loans of their monetary
reserves), the 3-month Treasury bill rate, and the 10-year Treasury note rate.
CBO uses the BVAR model to project the following:
▪ The 5-year Treasury note rate,
▪ The corporate Aaa bond rate, and
▪ The corporate Baa bond rate.
Interest Rates Projected Using the Model
8
CBO uses the BVAR model to project the following:
Output Variables Projected Using the Model
▪ Real GDP
▪ Real personal consumption
expenditures
▪ Real nonresidential fixed
investment
▪ Real exports
▪ Real imports
▪ Total factor productivity
▪ Real potential GDP
▪ Nominal gross national product
▪ Nominal private nonresidential fixed
investment in equipment
9
See Richard K. Crump and others, A Large Bayesian VAR of the United States Economy, Staff Report 976 (Federal Reserve Bank of New York, August 2021),
http://newyorkfed.org/research/staff_reports/sr976.html. CBO’s modeling follows this approach closely, and this staff report provides more details than are included here.
CBO adapted its approach to conditional forecasting from that used by the staff of the
Federal Reserve Bank of New York.
Bayesian techniques are particularly well suited to estimating parameters in a large
system of equations given a limited amount of data.
The modeling is structured so that a projection of a variable at a given point in time is
more likely to be influenced by recent data than by older data. The structure prevents
the estimation from explaining historical data well but having poor ability to forecast
beyond the data used for estimation—which would be the case if the estimation
process overfit the parameters.
The approach is flexible, and the staff of the Federal Reserve Bank of New York
found that it generated reasonable conditional forecasts.
How the Model Works
10
CBO used the following equation:
yt = c + B1yt - 1 + ⋯ + Bp yt - p + ϵt , ϵt ~N(0,Σ)
where
▪ yt is a vector of m economic variables at time t (t = 1, … , T),
▪ Bs (s = 1, … , p) is a (m × m) matrix of parameters of lagged variables, and
▪ ϵt is an error term distributed by a normal distribution with a covariance matrix Σ.
The model has many parameters to be estimated when the number of variables is large.
Because m equals 26 and p equals 6 in the model, the set of Bs (s = 1, … , p) has 4,056
parameters (26 × 26 × 6). In this case, the traditional vector autoregression techniques are
vulnerable to overfitting and tend to show poor out-of-sample forecasting accuracy.
A Bayesian procedure addresses the overfitting issue by automatically selecting the degree
of shrinkage, using tighter priors when the number of unknown coefficients relative to
available data is high and looser priors otherwise.
The Bayesian Vector Autoregression Model
11
For discussion of the Minnesota prior, see Robert Litterman, Techniques of Forecasting Using Vector Autoregressions, Working Paper 115 (Federal Reserve of Minneapolis,
November 1979), www.minneapolisfed.org/research/working-papers/techniques-of-forecasting-using-vector-autoregressions.
CBO used the Minnesota prior, under which each variable follows an independent random
walk process with potential drift. The prior sets the mean of each Bs as
where (Bs)ij is the row i, column j element of Bs.
The Minnesota prior sets tighter distributions for the parameters corresponding to smaller
lags. Variances and covariances between elements of B are set as
where
▪ λ determines the general tightness of the prior distribution of B,
▪ Σik is the row i, column k element of Σ, and
▪ ψj is an estimate of the variation of variable j in yt.
CBO estimated λ using the hierarchical Bayesian approach.
The Bayesian Prior Distribution
12
The posterior distribution of the parameters was calculated via Bayes’ rule as
p(θ│YT )∝ p(θ)p(YT |θ)
where
▪ θ is the vector of the parameters in the BVAR (θ = (c, B1, … , Bp, Σ));
▪ YT is the vector of the historical values of all the variables or YT = (y1, … , yT);
▪ p(θ) is the prior distribution (the Minnesota prior); and
▪ p(YT |θ) is the likelihood function of the BVAR.
CBO used a Markov chain Monte Carlo (MCMC) algorithm to generate the draws of θ
(or θ(g) for g = 1, … , G) from the posterior distribution.
The Posterior Distribution and Draws of Parameters
13
CBO projected all the variables in the BVAR using Bayesian inference by computing
the predictive density, defined as
p(yT + 1, … ,yT + h│YT ) = ∫ p(yT + 1, … ,yT + h | YT, θ)p(θ│YT)dθ
where h is the forecasting horizon. In practice, the future value is projected for each
MCMC draw of the parameters (θ(g)) using the structure of the BVAR. The whole set
of projected values is the predictive density.
CBO also generated conditional forecasts using conditional predictive densities,
defined as
p(yT + 1, … ,yT + h│YT, Ch) = ∫ p(yT + 1, … ,yT + h | YT, Ch, θ)p(θ│YT )dθ
where Ch is a set of given conditions for a scenario. The conditions can be imposed
on any of the variables in the BVAR for any time.
Forecasting
14
To implement conditional forecasting, CBO cast the BVAR model into a linear state-
space model:
where
▪ yt
* is conditional on future values of some variables in yt (t = T + 1, … , T + H),
▪ xt is (yt, yt - 1,…, yt – p + 1, c)',
▪ Gt is a matrix identifying the conditioned future values in xt,
▪ F is a matrix representing the dynamics of xt, and
▪ ut = (ϵt, 0, … , 0)'.
Then, CBO applied the Kalman filter and smoother to generate conditional forecasts.
The approach is equivalent to estimating unobservable variables (or missing values)
while treating the conditions as observable variables (or nonmissing values).
Conditional Forecasting
15
For discussion of CBO’s historical forecasting errors, see Congressional Budget Office, CBO’s Economic Forecasting Record: 2021 Update (December 2021),
www.cbo.gov/publication/57579.
The variance of the paths generated by conditional forecasting is related to the way
economic conditions have changed over time and to the simulations of the six
variables from CBO’s expanded Markov-switching model that are taken as sets of
conditions.
To create simulations for the purpose of communicating uncertainty about the central
estimates in the economic forecast, CBO calibrated their variance. Specifically, CBO
used the paths generated by conditional forecasting to create 100 simulations of the
20 variables discussed above (including real GDP growth), which incorporated
correlations with the six variables from CBO’s expanded Markov-switching model.
The variance of real GDP growth was calibrated by considering alternative ways to
average multiple paths to form a single simulation.
For this analysis, CBO used the simple average of two paths to form each simulation
because the range of the middle two-thirds of the distribution of those simulations
matched the middle two-thirds of CBO’s historical forecasting errors over two years.
Calibrating the Variance of Simulations
16
This document was prepared to enhance the transparency of CBO’s work and to
encourage external review of that work. In keeping with CBO’s mandate to provide
objective, impartial analysis, the document makes no recommendations.
Byoung Hark Yoo prepared the document with guidance from Sebastien Gay. Robert
Arnold, Mark Lasky, and Michael McGrane provided comments.
Mark Hadley and Jeffrey Kling reviewed the document. Christine Browne edited it
and R. L. Rebach created the graphics. The document is available at
www.cbo.gov/publication/58883.
CBO seeks feedback to make its work as useful as possible. Please send comments
to communications@cbo.gov.
About This Document

Más contenido relacionado

Similar a Estimating the Uncertainty of the Economic Forecast Using CBO’s Bayesian Vector Autoregression Model

Financial Market Assumptions & Models for Pension Plans
Financial Market Assumptions & Models for Pension PlansFinancial Market Assumptions & Models for Pension Plans
Financial Market Assumptions & Models for Pension Plans
Ankur Dadhania
 
Review Parameters Model Building & Interpretation and Model Tunin.docx
Review Parameters Model Building & Interpretation and Model Tunin.docxReview Parameters Model Building & Interpretation and Model Tunin.docx
Review Parameters Model Building & Interpretation and Model Tunin.docx
carlstromcurtis
 

Similar a Estimating the Uncertainty of the Economic Forecast Using CBO’s Bayesian Vector Autoregression Model (20)

Assessing the Budgetary Implications of Economic Uncertainty With CBO’s Incom...
Assessing the Budgetary Implications of Economic Uncertainty With CBO’s Incom...Assessing the Budgetary Implications of Economic Uncertainty With CBO’s Incom...
Assessing the Budgetary Implications of Economic Uncertainty With CBO’s Incom...
 
Telecom customer churn prediction
Telecom customer churn predictionTelecom customer churn prediction
Telecom customer churn prediction
 
Scenario generation and stochastic programming models for asset liabiltiy man...
Scenario generation and stochastic programming models for asset liabiltiy man...Scenario generation and stochastic programming models for asset liabiltiy man...
Scenario generation and stochastic programming models for asset liabiltiy man...
 
A dynamic Nelson-Siegel model with forward-looking indicators for the yield c...
A dynamic Nelson-Siegel model with forward-looking indicators for the yield c...A dynamic Nelson-Siegel model with forward-looking indicators for the yield c...
A dynamic Nelson-Siegel model with forward-looking indicators for the yield c...
 
Time series analysis- Part 2
Time series analysis- Part 2Time series analysis- Part 2
Time series analysis- Part 2
 
Advanced microeconometric project
Advanced microeconometric projectAdvanced microeconometric project
Advanced microeconometric project
 
The prognosis of Basel III costs in Albanian Economy
The prognosis of Basel III costs in Albanian EconomyThe prognosis of Basel III costs in Albanian Economy
The prognosis of Basel III costs in Albanian Economy
 
Financial Market Assumptions & Models for Pension Plans
Financial Market Assumptions & Models for Pension PlansFinancial Market Assumptions & Models for Pension Plans
Financial Market Assumptions & Models for Pension Plans
 
Customer Lifetime Value Modeling
Customer Lifetime Value ModelingCustomer Lifetime Value Modeling
Customer Lifetime Value Modeling
 
Multifactorial Heath-Jarrow-Morton model using principal component analysis
Multifactorial Heath-Jarrow-Morton model using principal component analysisMultifactorial Heath-Jarrow-Morton model using principal component analysis
Multifactorial Heath-Jarrow-Morton model using principal component analysis
 
Review Parameters Model Building & Interpretation and Model Tunin.docx
Review Parameters Model Building & Interpretation and Model Tunin.docxReview Parameters Model Building & Interpretation and Model Tunin.docx
Review Parameters Model Building & Interpretation and Model Tunin.docx
 
Ignou 10 no. project
Ignou 10 no. projectIgnou 10 no. project
Ignou 10 no. project
 
Econometrics
EconometricsEconometrics
Econometrics
 
Determinants of bank's interest margin in the aftermath of the crisis: the ef...
Determinants of bank's interest margin in the aftermath of the crisis: the ef...Determinants of bank's interest margin in the aftermath of the crisis: the ef...
Determinants of bank's interest margin in the aftermath of the crisis: the ef...
 
Presentation1.pptx
Presentation1.pptxPresentation1.pptx
Presentation1.pptx
 
International Journal of Computational Engineering Research (IJCER)
International Journal of Computational Engineering Research (IJCER) International Journal of Computational Engineering Research (IJCER)
International Journal of Computational Engineering Research (IJCER)
 
COMPARISON OF BANKRUPTCY PREDICTION MODELS WITH PUBLIC RECORDS AND FIRMOGRAPHICS
COMPARISON OF BANKRUPTCY PREDICTION MODELS WITH PUBLIC RECORDS AND FIRMOGRAPHICSCOMPARISON OF BANKRUPTCY PREDICTION MODELS WITH PUBLIC RECORDS AND FIRMOGRAPHICS
COMPARISON OF BANKRUPTCY PREDICTION MODELS WITH PUBLIC RECORDS AND FIRMOGRAPHICS
 
GDP Prediction and Forecasting using Machine Learning
GDP Prediction and Forecasting using Machine LearningGDP Prediction and Forecasting using Machine Learning
GDP Prediction and Forecasting using Machine Learning
 
EAD Parameter : A stochastic way to model the Credit Conversion Factor
EAD Parameter : A stochastic way to model the Credit Conversion FactorEAD Parameter : A stochastic way to model the Credit Conversion Factor
EAD Parameter : A stochastic way to model the Credit Conversion Factor
 
Econometrics_1.pptx
Econometrics_1.pptxEconometrics_1.pptx
Econometrics_1.pptx
 

Más de Congressional Budget Office

Más de Congressional Budget Office (20)

The Outlook for the Budget and the Economy
The Outlook for the Budget and the EconomyThe Outlook for the Budget and the Economy
The Outlook for the Budget and the Economy
 
The U.S. Budget and Economic Outlook (Presentation)
The U.S. Budget and Economic Outlook (Presentation)The U.S. Budget and Economic Outlook (Presentation)
The U.S. Budget and Economic Outlook (Presentation)
 
CBO’s Recent Appeals for New Research on Health-Related Topics
CBO’s Recent Appeals for New Research on Health-Related TopicsCBO’s Recent Appeals for New Research on Health-Related Topics
CBO’s Recent Appeals for New Research on Health-Related Topics
 
The Federal Budget and Health Care Policy
The Federal Budget and Health Care PolicyThe Federal Budget and Health Care Policy
The Federal Budget and Health Care Policy
 
How the Congressional Budget Office Assists Lawmakers
How the Congressional Budget Office Assists LawmakersHow the Congressional Budget Office Assists Lawmakers
How the Congressional Budget Office Assists Lawmakers
 
Introduction to the Role of the Congressional Budget Office
Introduction to the Role of the Congressional Budget OfficeIntroduction to the Role of the Congressional Budget Office
Introduction to the Role of the Congressional Budget Office
 
CBO’s Work on Health Care and a Call for New Research
CBO’s Work on Health Care and a Call for New ResearchCBO’s Work on Health Care and a Call for New Research
CBO’s Work on Health Care and a Call for New Research
 
The Federal Perspective on Coverage of Medications to Treat Obesity: Consider...
The Federal Perspective on Coverage of Medications to Treat Obesity: Consider...The Federal Perspective on Coverage of Medications to Treat Obesity: Consider...
The Federal Perspective on Coverage of Medications to Treat Obesity: Consider...
 
CBO’s Approach to Estimating the Budgetary Effects of the No Surprises Act of...
CBO’s Approach to Estimating the Budgetary Effects of the No Surprises Act of...CBO’s Approach to Estimating the Budgetary Effects of the No Surprises Act of...
CBO’s Approach to Estimating the Budgetary Effects of the No Surprises Act of...
 
The Budget and Economic Outlook for 2024 to 2034: Press Briefing Slides
The Budget and Economic Outlook for 2024 to 2034: Press Briefing SlidesThe Budget and Economic Outlook for 2024 to 2034: Press Briefing Slides
The Budget and Economic Outlook for 2024 to 2034: Press Briefing Slides
 
The Economic Outlook for 2024 to 2034 in 19 Slides
The Economic Outlook for 2024 to 2034 in 19 SlidesThe Economic Outlook for 2024 to 2034 in 19 Slides
The Economic Outlook for 2024 to 2034 in 19 Slides
 
Using Multiple Data Sources to Learn About the Race and Ethnicity of Taxpayers
Using Multiple Data Sources to Learn About the Race and Ethnicity of TaxpayersUsing Multiple Data Sources to Learn About the Race and Ethnicity of Taxpayers
Using Multiple Data Sources to Learn About the Race and Ethnicity of Taxpayers
 
Perspectives on the Navy’s 2024 Shipbuilding Plan
Perspectives on the Navy’s 2024 Shipbuilding PlanPerspectives on the Navy’s 2024 Shipbuilding Plan
Perspectives on the Navy’s 2024 Shipbuilding Plan
 
The 2024 Outlook for Navy Shipbuilding: Familiar Plans and Higher Costs
The 2024 Outlook for Navy Shipbuilding: Familiar Plans and Higher CostsThe 2024 Outlook for Navy Shipbuilding: Familiar Plans and Higher Costs
The 2024 Outlook for Navy Shipbuilding: Familiar Plans and Higher Costs
 
CBO’s Use of GitHub
CBO’s Use of GitHubCBO’s Use of GitHub
CBO’s Use of GitHub
 
Exploring the Effects of Medicaid During Childhood on the Economy and the Budget
Exploring the Effects of Medicaid During Childhood on the Economy and the BudgetExploring the Effects of Medicaid During Childhood on the Economy and the Budget
Exploring the Effects of Medicaid During Childhood on the Economy and the Budget
 
CBO’s Role and Most Recent Long-Term Budget Projections
CBO’s Role and Most Recent Long-Term Budget ProjectionsCBO’s Role and Most Recent Long-Term Budget Projections
CBO’s Role and Most Recent Long-Term Budget Projections
 
Atlas of Military Compensation
Atlas of Military CompensationAtlas of Military Compensation
Atlas of Military Compensation
 
Approaches to Estimating the Noncyclical Rate of Unemployment
 Approaches to Estimating the Noncyclical Rate of Unemployment Approaches to Estimating the Noncyclical Rate of Unemployment
Approaches to Estimating the Noncyclical Rate of Unemployment
 
CBO’s Analysis of the Long-Term Budgetary Outlook
CBO’s Analysis of the Long-Term Budgetary OutlookCBO’s Analysis of the Long-Term Budgetary Outlook
CBO’s Analysis of the Long-Term Budgetary Outlook
 

Último

Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
gajnagarg
 
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
gajnagarg
 
Unique Value Prop slide deck________.pdf
Unique Value Prop slide deck________.pdfUnique Value Prop slide deck________.pdf
Unique Value Prop slide deck________.pdf
ScottMeyers35
 
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
Cara Menggugurkan Kandungan 087776558899
 

Último (20)

Call Girls Basheerbagh ( 8250092165 ) Cheap rates call girls | Get low budget
Call Girls Basheerbagh ( 8250092165 ) Cheap rates call girls | Get low budgetCall Girls Basheerbagh ( 8250092165 ) Cheap rates call girls | Get low budget
Call Girls Basheerbagh ( 8250092165 ) Cheap rates call girls | Get low budget
 
Pakistani Call girls in Sharjah 0505086370 Sharjah Call girls
Pakistani Call girls in Sharjah 0505086370 Sharjah Call girlsPakistani Call girls in Sharjah 0505086370 Sharjah Call girls
Pakistani Call girls in Sharjah 0505086370 Sharjah Call girls
 
Finance strategies for adaptation. Presentation for CANCC
Finance strategies for adaptation. Presentation for CANCCFinance strategies for adaptation. Presentation for CANCC
Finance strategies for adaptation. Presentation for CANCC
 
Call Girls Koregaon Park - 8250092165 Our call girls are sure to provide you ...
Call Girls Koregaon Park - 8250092165 Our call girls are sure to provide you ...Call Girls Koregaon Park - 8250092165 Our call girls are sure to provide you ...
Call Girls Koregaon Park - 8250092165 Our call girls are sure to provide you ...
 
Kolkata Call Girls Halisahar 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl ...
Kolkata Call Girls Halisahar  💯Call Us 🔝 8005736733 🔝 💃  Top Class Call Girl ...Kolkata Call Girls Halisahar  💯Call Us 🔝 8005736733 🔝 💃  Top Class Call Girl ...
Kolkata Call Girls Halisahar 💯Call Us 🔝 8005736733 🔝 💃 Top Class Call Girl ...
 
Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Morena [ 7014168258 ] Call Me For Genuine Models We...
 
NGO working for orphan children’s education
NGO working for orphan children’s educationNGO working for orphan children’s education
NGO working for orphan children’s education
 
Delivery in 20 Mins Call Girls Malappuram { 9332606886 } VVIP NISHA Call Girl...
Delivery in 20 Mins Call Girls Malappuram { 9332606886 } VVIP NISHA Call Girl...Delivery in 20 Mins Call Girls Malappuram { 9332606886 } VVIP NISHA Call Girl...
Delivery in 20 Mins Call Girls Malappuram { 9332606886 } VVIP NISHA Call Girl...
 
Vivek @ Cheap Call Girls In Kamla Nagar | Book 8448380779 Extreme Call Girls ...
Vivek @ Cheap Call Girls In Kamla Nagar | Book 8448380779 Extreme Call Girls ...Vivek @ Cheap Call Girls In Kamla Nagar | Book 8448380779 Extreme Call Girls ...
Vivek @ Cheap Call Girls In Kamla Nagar | Book 8448380779 Extreme Call Girls ...
 
Genuine Call Girls in Salem 9332606886 HOT & SEXY Models beautiful and charm...
Genuine Call Girls in Salem  9332606886 HOT & SEXY Models beautiful and charm...Genuine Call Girls in Salem  9332606886 HOT & SEXY Models beautiful and charm...
Genuine Call Girls in Salem 9332606886 HOT & SEXY Models beautiful and charm...
 
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
Top profile Call Girls In Haldia [ 7014168258 ] Call Me For Genuine Models We...
 
2024: The FAR, Federal Acquisition Regulations, Part 31
2024: The FAR, Federal Acquisition Regulations, Part 312024: The FAR, Federal Acquisition Regulations, Part 31
2024: The FAR, Federal Acquisition Regulations, Part 31
 
Cheap Call Girls In Hyderabad Phone No 📞 9352988975 📞 Elite Escort Service Av...
Cheap Call Girls In Hyderabad Phone No 📞 9352988975 📞 Elite Escort Service Av...Cheap Call Girls In Hyderabad Phone No 📞 9352988975 📞 Elite Escort Service Av...
Cheap Call Girls In Hyderabad Phone No 📞 9352988975 📞 Elite Escort Service Av...
 
Unique Value Prop slide deck________.pdf
Unique Value Prop slide deck________.pdfUnique Value Prop slide deck________.pdf
Unique Value Prop slide deck________.pdf
 
Call Girls Mehsana / 8250092165 Genuine Call girls with real Photos and Number
Call Girls Mehsana / 8250092165 Genuine Call girls with real Photos and NumberCall Girls Mehsana / 8250092165 Genuine Call girls with real Photos and Number
Call Girls Mehsana / 8250092165 Genuine Call girls with real Photos and Number
 
2024 UN Civil Society Conference in Support of the Summit of the Future.
2024 UN Civil Society Conference in Support of the Summit of the Future.2024 UN Civil Society Conference in Support of the Summit of the Future.
2024 UN Civil Society Conference in Support of the Summit of the Future.
 
Lorain Road Business District Revitalization Plan Final Presentation
Lorain Road Business District Revitalization Plan Final PresentationLorain Road Business District Revitalization Plan Final Presentation
Lorain Road Business District Revitalization Plan Final Presentation
 
Scaling up coastal adaptation in Maldives through the NAP process
Scaling up coastal adaptation in Maldives through the NAP processScaling up coastal adaptation in Maldives through the NAP process
Scaling up coastal adaptation in Maldives through the NAP process
 
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
Cara Gugurkan Pembuahan Secara Alami Dan Cepat ABORSI KANDUNGAN 087776558899
 
Dating Call Girls inBaloda Bazar Bhatapara 9332606886Call Girls Advance Cash...
Dating Call Girls inBaloda Bazar Bhatapara  9332606886Call Girls Advance Cash...Dating Call Girls inBaloda Bazar Bhatapara  9332606886Call Girls Advance Cash...
Dating Call Girls inBaloda Bazar Bhatapara 9332606886Call Girls Advance Cash...
 

Estimating the Uncertainty of the Economic Forecast Using CBO’s Bayesian Vector Autoregression Model

  • 1. Estimating the Uncertainty of the Economic Forecast Using CBO’s Bayesian Vector Autoregression Model January 2023
  • 2. 1 For details about the analysis, see Congressional Budget Office, CBO’s Current View of the Economy in 2023 and 2024 and the Budgetary Implications (November 2022), www.cbo.gov/publication/58757. In the figure, real GDP growth is calculated for a given quarter relative to four quarters earlier. Q = quarter. CBO’s View of the Economy as of November 2022 In November 2022, the Congressional Budget Office was asked about its view of the economy. From the fourth quarter of 2022 to the fourth quarter of 2023, CBO estimated, there is a two- thirds chance that growth in economic output—specifically, gross domestic product (GDP) adjusted to remove the effects of inflation, or real GDP—will be between −2.0 percent and 1.8 percent.
  • 3. 2 For discussion of the first step, see Robert W. Arnold, How CBO Produces Its 10-Year Economic Forecast, Working Paper 2018-02 (Congressional Budget Office, February 2018), www.cbo.gov/publication/53537. For discussion of the second step, see Congressional Budget Office, “Estimating the Uncertainty of the Economic Forecast Using CBO’s Expanded Markov-Switching Model” (January 2023), www.cbo.gov/publication/58884. For related discussion, see Mark Lasky, The Congressional Budget Office’s Small-Scale Policy Model, Working Paper 2022-08 (Congressional Budget Office, September 2022), www.cbo.gov/publication/57254. The analysis of economic uncertainty was conducted in three main steps: ▪ Preliminary economic projections provided central estimates for each variable; ▪ 100 simulations of the rates of unemployment, inflation, and interest were jointly estimated around the central estimates, reflecting asymmetric dynamics and relating the variables through an expectations-augmented Phillips curve and an inertial Taylor rule; and ▪ Forecasts conditional on those rates were estimated using symmetric distributions in which economic output and other variables were synchronized with the simulations of unemployment, inflation, and interest rates. This document focuses on the third step, which used a Bayesian vector autoregression (BVAR) model. CBO’s Analytic Method for Estimating Uncertainty
  • 4. 3 The BVAR model draws on historical correlations between macroeconomic variables to produce conditional forecasts. Key conditions are the simulations of the rates of unemployment, inflation, and interest, which are estimated using an expanded version of CBO’s Markov-switching model with asymmetric dynamics in which the unemployment rate rises rapidly in some periods and falls gradually in others and interest rates do not fall below zero. The projections of economic output and other variables are synchronized—using symmetric distributions—with the simulations of unemployment, inflation, and interest rates. Additional variables that can be simulated with symmetric distributions can be easily incorporated. The historical correlations between macroeconomic variables used in the model may be less predictive of future outcomes in the event of extreme changes in economic conditions. Historical Dynamics Reflected in the Modeling
  • 5. 4 Inputs include central forecasts from CBO’s large-scale macroeconometric model for 26 variables used to analyze effects of economic conditions on the federal budget. Additional inputs are 100 simulations of six variables from CBO’s expanded Markov- switching model: the unemployment rate, two inflation rates, and three interest rates. For each calendar quarter, CBO uses those inputs to the BVAR model to project values for the remaining 20 variables. After 100 simulations are generated, the values are calibrated so that their average equals CBO’s central forecast. The parameters of the model are estimated using data from 1959 through 2022. How CBO Uses the Model
  • 6. 5 The unemployment rate projection is taken as a set of conditions. CBO uses the BVAR model to project the following: ▪ Payroll employment, ▪ The number of people in the labor force, ▪ Hours of work, and ▪ Wages and salaries. Labor Market Variables Projected Using the Model
  • 7. 6 Two rates of inflation are taken as sets of conditions—the overall rate as measured by the personal consumption expenditures price index and that rate excluding food and energy prices. CBO uses the BVAR model to project inflation as measured by: ▪ The GDP price index, ▪ The consumer price index for all urban consumers, ▪ The consumer price index for food at home, and ▪ The consumer price index for medical care. Inflation Rates Projected Using the Model
  • 8. 7 Three interest rates are taken as sets of conditions—the federal funds rate (the rate that financial institutions charge each other for overnight loans of their monetary reserves), the 3-month Treasury bill rate, and the 10-year Treasury note rate. CBO uses the BVAR model to project the following: ▪ The 5-year Treasury note rate, ▪ The corporate Aaa bond rate, and ▪ The corporate Baa bond rate. Interest Rates Projected Using the Model
  • 9. 8 CBO uses the BVAR model to project the following: Output Variables Projected Using the Model ▪ Real GDP ▪ Real personal consumption expenditures ▪ Real nonresidential fixed investment ▪ Real exports ▪ Real imports ▪ Total factor productivity ▪ Real potential GDP ▪ Nominal gross national product ▪ Nominal private nonresidential fixed investment in equipment
  • 10. 9 See Richard K. Crump and others, A Large Bayesian VAR of the United States Economy, Staff Report 976 (Federal Reserve Bank of New York, August 2021), http://newyorkfed.org/research/staff_reports/sr976.html. CBO’s modeling follows this approach closely, and this staff report provides more details than are included here. CBO adapted its approach to conditional forecasting from that used by the staff of the Federal Reserve Bank of New York. Bayesian techniques are particularly well suited to estimating parameters in a large system of equations given a limited amount of data. The modeling is structured so that a projection of a variable at a given point in time is more likely to be influenced by recent data than by older data. The structure prevents the estimation from explaining historical data well but having poor ability to forecast beyond the data used for estimation—which would be the case if the estimation process overfit the parameters. The approach is flexible, and the staff of the Federal Reserve Bank of New York found that it generated reasonable conditional forecasts. How the Model Works
  • 11. 10 CBO used the following equation: yt = c + B1yt - 1 + ⋯ + Bp yt - p + ϵt , ϵt ~N(0,Σ) where ▪ yt is a vector of m economic variables at time t (t = 1, … , T), ▪ Bs (s = 1, … , p) is a (m × m) matrix of parameters of lagged variables, and ▪ ϵt is an error term distributed by a normal distribution with a covariance matrix Σ. The model has many parameters to be estimated when the number of variables is large. Because m equals 26 and p equals 6 in the model, the set of Bs (s = 1, … , p) has 4,056 parameters (26 × 26 × 6). In this case, the traditional vector autoregression techniques are vulnerable to overfitting and tend to show poor out-of-sample forecasting accuracy. A Bayesian procedure addresses the overfitting issue by automatically selecting the degree of shrinkage, using tighter priors when the number of unknown coefficients relative to available data is high and looser priors otherwise. The Bayesian Vector Autoregression Model
  • 12. 11 For discussion of the Minnesota prior, see Robert Litterman, Techniques of Forecasting Using Vector Autoregressions, Working Paper 115 (Federal Reserve of Minneapolis, November 1979), www.minneapolisfed.org/research/working-papers/techniques-of-forecasting-using-vector-autoregressions. CBO used the Minnesota prior, under which each variable follows an independent random walk process with potential drift. The prior sets the mean of each Bs as where (Bs)ij is the row i, column j element of Bs. The Minnesota prior sets tighter distributions for the parameters corresponding to smaller lags. Variances and covariances between elements of B are set as where ▪ λ determines the general tightness of the prior distribution of B, ▪ Σik is the row i, column k element of Σ, and ▪ ψj is an estimate of the variation of variable j in yt. CBO estimated λ using the hierarchical Bayesian approach. The Bayesian Prior Distribution
  • 13. 12 The posterior distribution of the parameters was calculated via Bayes’ rule as p(θ│YT )∝ p(θ)p(YT |θ) where ▪ θ is the vector of the parameters in the BVAR (θ = (c, B1, … , Bp, Σ)); ▪ YT is the vector of the historical values of all the variables or YT = (y1, … , yT); ▪ p(θ) is the prior distribution (the Minnesota prior); and ▪ p(YT |θ) is the likelihood function of the BVAR. CBO used a Markov chain Monte Carlo (MCMC) algorithm to generate the draws of θ (or θ(g) for g = 1, … , G) from the posterior distribution. The Posterior Distribution and Draws of Parameters
  • 14. 13 CBO projected all the variables in the BVAR using Bayesian inference by computing the predictive density, defined as p(yT + 1, … ,yT + h│YT ) = ∫ p(yT + 1, … ,yT + h | YT, θ)p(θ│YT)dθ where h is the forecasting horizon. In practice, the future value is projected for each MCMC draw of the parameters (θ(g)) using the structure of the BVAR. The whole set of projected values is the predictive density. CBO also generated conditional forecasts using conditional predictive densities, defined as p(yT + 1, … ,yT + h│YT, Ch) = ∫ p(yT + 1, … ,yT + h | YT, Ch, θ)p(θ│YT )dθ where Ch is a set of given conditions for a scenario. The conditions can be imposed on any of the variables in the BVAR for any time. Forecasting
  • 15. 14 To implement conditional forecasting, CBO cast the BVAR model into a linear state- space model: where ▪ yt * is conditional on future values of some variables in yt (t = T + 1, … , T + H), ▪ xt is (yt, yt - 1,…, yt – p + 1, c)', ▪ Gt is a matrix identifying the conditioned future values in xt, ▪ F is a matrix representing the dynamics of xt, and ▪ ut = (ϵt, 0, … , 0)'. Then, CBO applied the Kalman filter and smoother to generate conditional forecasts. The approach is equivalent to estimating unobservable variables (or missing values) while treating the conditions as observable variables (or nonmissing values). Conditional Forecasting
  • 16. 15 For discussion of CBO’s historical forecasting errors, see Congressional Budget Office, CBO’s Economic Forecasting Record: 2021 Update (December 2021), www.cbo.gov/publication/57579. The variance of the paths generated by conditional forecasting is related to the way economic conditions have changed over time and to the simulations of the six variables from CBO’s expanded Markov-switching model that are taken as sets of conditions. To create simulations for the purpose of communicating uncertainty about the central estimates in the economic forecast, CBO calibrated their variance. Specifically, CBO used the paths generated by conditional forecasting to create 100 simulations of the 20 variables discussed above (including real GDP growth), which incorporated correlations with the six variables from CBO’s expanded Markov-switching model. The variance of real GDP growth was calibrated by considering alternative ways to average multiple paths to form a single simulation. For this analysis, CBO used the simple average of two paths to form each simulation because the range of the middle two-thirds of the distribution of those simulations matched the middle two-thirds of CBO’s historical forecasting errors over two years. Calibrating the Variance of Simulations
  • 17. 16 This document was prepared to enhance the transparency of CBO’s work and to encourage external review of that work. In keeping with CBO’s mandate to provide objective, impartial analysis, the document makes no recommendations. Byoung Hark Yoo prepared the document with guidance from Sebastien Gay. Robert Arnold, Mark Lasky, and Michael McGrane provided comments. Mark Hadley and Jeffrey Kling reviewed the document. Christine Browne edited it and R. L. Rebach created the graphics. The document is available at www.cbo.gov/publication/58883. CBO seeks feedback to make its work as useful as possible. Please send comments to communications@cbo.gov. About This Document