statistical approaches to balancing national accounts brent r. moulton oecd, working party on...

14
Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

Upload: ashley-sawyer

Post on 27-Mar-2015

214 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

Statistical Approaches to Balancing National Accounts

Brent R. MoultonOECD, Working Party on National Accounts,

ParisOctober 5, 2007

Page 2: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 2

Initial Estimates Need to Be Adjusted

“It is impossible to establish by direct estimation a system of national accounts free of statistical discrepancies, residual errors, unidentified items, balancing entries and the like since the information available is in some degree incomplete, inconsistent and unreliable. Accordingly, the task of measurement is not finished when the initial estimates have been made and remains incomplete until final estimates have been obtained which satisfy the constraints that hold between their true values.”

—Richard StoneJournal of the Royal Statistical Society,

1982

Page 3: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 3

Initial Estimates not Sufficient

Source data may not be available for all components of the accounts—some estimates are derived residually or by assumption.

In other cases, more than one estimate may be available, based on national accounting identities. Examples: Total commodity supply = total commodity

use; GDP via production, expenditure, and income

approaches; Net lending or borrowing from capital

account versus financial account.

Page 4: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 4

Source Data Are Subject to Errors

Survey data are subject to sampling and non-sampling errors.

Administrative data may be more comprehensive, but are not designed to match national accounting concepts.

Mixture of enterprise and establishment-based data may require bridging.

Data from different sources may not use the same classifications.

Estimates for some components may be extrapolated from earlier periods.

Page 5: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 5

Simple Methods

In an input-output table, the RAS method (bi-proportional adjustment) updates inter-industry multipliers to be consistent with given row and column totals. It is simple to compute and preserves zero and non-negative flows.

A number of related techniques have also been developed, based on linear or quadratic programming or Theil’s entropy approach.

Page 6: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 6

Stone, Champernowne, and Meade

In 1942 (Review of Economic Studies), Stone, Champernowne, and Meade recognized that measures of reliability could be used to determine which flows should be adjusted. Explicitly recognizes errors in measurement. Larger adjustments made to flows with largest

errors; little adjustment to flows with reliable initial estimates.

In absence of standard errors, margins of error may be set judgmentally.

Flexible approach; allows some constraints to hold exactly, others to be subject to error.

Least squares method for solution.

Page 7: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 7

Byron Method

Method proposed by Stone et al. required a large amount of computation.

In 1978 (Journal of the Royal Statistical Society, A), Byron proposed a conjugate gradient algorithm that is computationally efficient, even for very large matrices.

Byron’s method led to applications, for example: van der Ploeg, J. Royal Stat. Soc., 1982; Barker, van der Pleog, and Weale, Rev.

Income and Wealth, 1984.

Page 8: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 8

Balancing U.S. Industry Accounts

Most important obstacle to implementation of approaches of Stone, et al., was lack of objective information on reliability of initial data.

Research by Baoline Chen of BEA:“A Balanced System of Industry Accounts for the U.S. and Structural Distribution of Statistical Discrepancy,” 2006.

Proposed an efficient generalized least squares (GLS) method.

Systematically gathered information on coefficients of variation.

Page 9: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 9

Data Problems to Be Addressed

For benchmark input-output accounts, gross output (GO) compiled mostly from economic census.

Initial estimate of intermediate consumption (IC) generally based on a business expense survey.

Initial estimate of gross operating surplus/mixed income (GOS) largely based on administrative (tax return) data.

Inconsistencies between gross value added calculated using: Production approach (GVA = GO — IC) and Income approach (GVA = Compensation + Taxes less

subsidies on production and imports + GOS). Least reliable initial estimates were IC and GOS.

Page 10: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 10

Sampling and Non-Sampling Errors

Census Bureau and Statistics of Income Division of the Internal Revenue Service provided coefficients of variation for published estimates.

Surveys are subject to non-sampling error. BEA analysts make adjustments for identifiable non-sampling errors in order to reduce bias. However, these adjustments may be subject to misallocation errors.

Page 11: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 11

Adjustments to Source Data

A number of adjustments must made to source data: Conceptual adjustments Misreporting adjustments (for under-

reporting or misreporting on tax returns)

Double counting adjustments Current-cost accounting of inventories

and consumption of fixed capital Imputations

Page 12: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 12

Statistical Discrepancy

The estimates based on the expenditure approach and the income approach differ. The difference is shown in the U.S. national income and product accounts as a statistical discrepancy.

The balanced industry accounts are consistent with the reconciled expenditure-based estimate and adjust initial estimates of IC and GOS to be consistent.

Chen applied her approach to historical data (the 1997 benchmark industry accounts).

Page 13: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 13

Implementation

As part of BEA’s integration efforts, the GLS method was applied to the reconciliation of the 2002 benchmark use table.

Paper by Howells, Morgan, Rassier, and Roesch of BEA:“Implementing a Reconciliation and Balancing Model in the U.S. Industry Accounts,” 16th International Conference on Input-Output Techniques, 2007, http://www.iioa.at/conferences-IO.html

Page 14: Statistical Approaches to Balancing National Accounts Brent R. Moulton OECD, Working Party on National Accounts, Paris October 5, 2007

www.bea.gov 14

Results and Next Steps

Input-output tables were released on September 21, 2007.

Method was computationally efficient. Allowed less experienced staff to do

balancing work. Didn’t eliminate need for judgmental

adjustments, but allowed quick identification of the most important discrepancies.

BEA plans to refine the model and potentially expand its use.