REVIEW 1 major objections 5 minor 3 references
Testing for multiple change-points in macroeconometrics: an empirical guide and recent developments
T0 review · 1 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read This review claims that a sequential testing workflow can reliably detect and estimate multiple change-points in the slope parameters of the main macroeconometric model classes.
desk verdict A careful, genuinely useful handbook review of multiple change-point methods for macroeconomists, with no new results but a few overbroad 'gap' claims that need tightening. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The load-bearing object is the sequential testing strategy for slope-parameter change-points. Its three components are tests of zero breaks against a fixed number $k$, against an unknown number up to a maximum $M$ (the double-maximum tests), and of $\ell$ versus $\ell + 1$ breaks; all are built from sup-F or sup-Wald statistics maximized over candidate partitions. A dynamic programming algorithm computes the objective in $O(T^3)$ operations regardless of the number of candidate breaks, which makes bootstrapping the full procedure feasible. The bootstrap, applied by resampling residuals under the null and either keeping regressors fixed or recursively reconstructing lagged dependent variables, is the device that extends validity to the non-covariance-stationary environments common in macro data. In time series, the procedure consistently estimates the break fractions $\tau_j^0$ but not the break dates themselves; in panels and factor models, the break-point $T_j^0$ can be consistently estimated.
What would settle it
A targeted literature search that surfaces a published sequential testing procedure for multiple breaks in the presence of unit roots, or a sequential multiple-break test for heterogeneous panels, would directly falsify the review's central gap claims; a simulation showing the recommended two-stage least squares bootstrap breaks down under autocorrelated errors would test the paper's own boundary.
Extended reading notes
Core claim
The central claim is that the literature now supplies a coherent, assumption-tailored toolkit for testing and estimating multiple discrete change-points in the slope parameters of the linear models macroeconomists use most. In univariate time series, the tools are the three test families for no breaks versus a known number, no breaks versus an unknown number up to a bound, and $\ell$ versus $\ell+1$ breaks, together with a dynamic programming algorithm that makes the search fast and a wild bootstrap that extends validity to data whose second moments change over time. With endogenous regressors, the same tests are built on two-stage least squares estimates after checking the first stage, and the review recommends this bootstrap two-stage least squares procedure because generalized method of moments based break detection has not been theoretically justified. In panels and factor models, extensions of the same logic work on transformed data or on the second moments of estimated factors, and these deliver consistent estimators of the break dates themselves, whereas time-series models only deliver consistent break fractions. The review also claims that modified information criteria, adaptive lasso, and mixed-integer programming are useful complements for breaks that are close together or near the sample edge.
Load-bearing premise
The guide's usefulness stands or falls on whether the gaps it identifies are really gaps: if any published method already handles sequential multiple-break testing under unit roots, in heterogeneous panels, or with generalized method of moments, the recommended workflows and scope statements would be incomplete.
Editorial extensions
If this is right
- Use the two-stage least squares sequential bootstrap procedure to detect breaks when regressors are endogenous; only after the break locations are fixed should generalized method of moments be used to estimate the slope parameters.
- Always use bootstrap critical values, even when asymptotic critical values exist, because finite-sample behavior is better in simulations.
- Modified BIC-type information criteria, adaptive lasso, and mixed-integer programming can detect breaks that sequential tests miss, in particular breaks close together or near the sample edges.
- In homogeneous panels with interactive fixed effects, sequentially testing on common-correlated-effects transformed data consistently estimates the number and location of breaks, and a dedicated software package is available.
- In factor models, tests based on the second moments of estimated factors can detect common breaks in loadings, and the likelihood-ratio version diverges faster under the alternative than the Wald version.
Reading between the lines
- The review's conjecture that a moving block bootstrap would make the sequential tests valid under autocorrelated errors, if proved, would fill a conspicuous gap and make the workflow directly applicable to multi-step local projection regressions.
- If the claimed gaps are real, the most valuable near-term targets are a sequential multiple-break procedure for unit-root regressors and one for heterogeneous panels; either would extend the guide's scope.
- The paper's comparison suggests an empirical extension: applied studies could benchmark the sequential workflow against modified information criteria and lasso methods on the same macroeconomic datasets to see which detects adjacent or edge breaks more often.
- Because factor-model tests operate on second moments of estimated factors, one could adapt them to detect joint breaks in loadings and in factor-augmented forecasting equations, the setting the single-break test in the review addresses.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This manuscript is a practitioner-oriented survey of methods for detecting and estimating multiple change-points in slope parameters of time-series models, panel data models, factor models, and nonlinear models, with emphasis on macroeconometric applications. It organizes the literature around sequential testing procedures, discusses information criteria, lasso, and mixed-integer programming alternatives, and explicitly flags conjectures and open gaps. The chapter contains no new derivations, simulations, or code.
Significance. If accurate, the survey would be a valuable and much-needed guide for empirical macroeconomists: it is organized around slope-parameter breaks rather than mean shifts, covers endogenous regressors, unit roots, panels, factor models, and recent penalized methods, and it gives concrete practitioner advice such as the Bonferroni correction, bootstrap under the null, and first-stage break checks for 2SLS. It is also transparent in labeling conjectures and open problems. However, because it is a review with no derivations or reproducible code, its accuracy rests entirely on correct citation and scoping of the primary literature; the unit-root gap claim discussed below fails that check as written.
major comments (1)
- [Section 2, Unit roots] The sentence 'The asymptotic properties of a sequential testing procedure for breaks in the presence of unit roots is to our knowledge not yet available in the literature' is too broad as written. Published work develops sequential procedures for multiple breaks in regressions with integrated variables and/or integrated noise; Kejriwal and Perron (2010, 'Testing for multiple structural changes in cointegrated regression models') is one example, and it falls within the chapter's stated scope of slope-parameter break inference in macro-relevant time-series models. If the intended claim is restricted to the predictive-regression setting with a unit-root regressor and martingale-difference errors discussed immediately above, the sentence should be rewritten to state that restriction and to acknowledge the cointegrated and integrated-noise literature. Because the chapter's usefulness depends on the accuracy of its 'gap' claims, this should be corrected before publication.
minor comments (5)
- [Section 2, OLS] The statement that the Bai-Perron dynamic programming algorithm computes the tests in O(T^3) operations 'regardless of the number of candidate breaks' is not supported by Bai and Perron (2003a), which describe an O(mT^2) algorithm for m breaks; please correct or justify the complexity claim.
- [Section 3, first paragraph] The sentence beginning 'Once can also consistently estimate...' contains a typo: 'Once' should be 'One'.
- [Section 5, Factor-augmented forecasting equations] The phrase 'apart form the intercept' should read 'apart from the intercept', and the phrase 'denotes the the ith row' should read 'denotes the ith row'.
- [References] The reference for Corradi and Swanson (2014) contains the typo 'Testinag' and should read 'Testing'.
- [Section 2, Partial structural change] The proposed two-step strategy for determining which parameters change is presented without a formal consistency result; if this is a new suggestion, it should be labeled as a conjecture or supported by a citation.
Circularity Check
No significant circularity: a review with no derivations; self-citations are to published proofs and are not load-bearing in a circular sense.
full rationale
This manuscript is a survey and practitioner's guide, not a derivation, so there is no claimed chain of equations whose 'prediction' reduces to its inputs. Recommendations are explicitly delegated to published theorems rather than derived in the chapter: for example, the recommended 2SLS sequential bootstrap workflow is justified by 'Boldea, Cornea-Madeira, and Hall (2019) prove the bootstrap validity of the fixed regressor or recursive bootstrap with residuals resampled using a wild bootstrap,' which is an external, peer-reviewed result with stated assumptions, not an unverified assertion that presupposes this chapter's conclusions. The negative-gap statements, such as 'The asymptotic properties of a sequential testing procedure for breaks in the presence of unit roots is to our knowledge not yet available in the literature' and 'To our knowledge, a sequential testing approach to infer multiple breaks has not been proposed in this setting,' are literature-coverage claims; even if one of them were inaccurate, the error would be one of completeness or correctness, not circularity. The manuscript also openly marks its own conjectures, such as the moving-block bootstrap validity and modified BIC consistency under autocorrelated errors, rather than presenting them as established results. Heavy self-citation is present (e.g., Boldea, Cornea-Madeira, and Hall 2019; Hall, Han, and Boldea 2012; Boldea, Drepper, and Gan 2020; Hall, Osborn, and Sakkas 2013), but no load-bearing argument reduces to a self-citation that is itself unverified, and no specific circular reduction can be exhibited by quoting the paper. Accordingly, the circularity score is 0.
Assumptions & free parameters
assumptions (4)
- domain assumption The asymptotic distributions and consistency results of the cited papers (Bai and Perron 1998; Perron and Qu 2006; Hall, Han, and Boldea 2012; Boldea, Cornea-Madeira, and Hall 2019; and others) are correctly summarized.
- domain assumption Macroeconomic series relevant to the guide violate covariance stationarity under the null, for example due to the Great Moderation and the 2008 crisis.
- domain assumption The small break versus big break classification in factor models, with the O(N^{-1/2}) and O(1/min(N,T)) thresholds, applies to typical macro datasets such as FRED-MD.
- domain assumption The chapter's negative claims about open gaps are correct, for example that multiple-break sequential testing is unavailable in unit-root regressions and in GMM settings.
Cite this review
Pith. "Pith review of Testing for multiple change-points in macroeconometrics: an empirical guide and recent developments." pith.science (2026). https://pith.science/paper/QPE6KPHG
@misc{pith2026250722204,
author = {Pith},
title = {Pith review of: Testing for multiple change-points in macroeconometrics: an empirical guide and recent developments},
year = {2026},
howpublished = {\url{https://pith.science/paper/QPE6KPHG}},
note = {Machine review of arXiv:2507.22204}
}
read the original abstract
We review recent developments in detecting and estimating multiple change-points in time series models with exogenous and endogenous regressors, panel data models, and factor models. This review differs from others in multiple ways: (1) it focuses on inference about the change-points in slope parameters, rather than in the mean of the dependent variable - the latter being common in the statistical literature; (2) it focuses on detecting - via sequential testing and other methods - multiple change-points, and only discusses one change-point when methods for multiple change-points are not available; (3) it is meant as a practitioner's guide for empirical macroeconomists first, and as a result, it focuses only on the methods derived under the most general assumptions relevant to macroeconomic applications.
Reference graph
Works this paper leans on
-
[1]
Aminikhanghahi, S., and Cook, D. J. (2017). ‘A survey of methods for time series change point detec- tion’, Knowledge and Information Systems , 51: 339–367. Andrews, D. W. K. (1993). ‘Tests for parameter instability and structural change with unknown change point’, Econometrica, 61: 821–856. (2003). ‘Tests for parameter instability and structural change w...
work page Pith review arXiv 2017
-
[419]
Sen, A., and Hall, A. R. (1999). ‘Two further aspects of some new tests for structural stability’, Structural Change and Economic Dynamics , 10: 431–443. Sowell, F. (1996). ‘Optimal tests of parameter variation in the Generalized Method of Moments frame- work’, Econometrica, 64: 1085–1108. Stock, J. H., and Watson, M. W. (2002a). ‘Forecasting using princi...
work page Pith review arXiv 1999
-
[631]
Jord` a,`O., Schularick, M., and Taylor, A. M. (2020). ‘The effects of quasi-random monetary experi- ments’, Journal of Monetary Economics , 112: 22–40. Jord` a,`O., and Taylor, A. M. (2025). ‘Local projections’, Journal of Economic Literature , 63: 59–110. Jørgensen, P. L., and Lansing, K. J. (2025). ‘Anchored inflation expectations and the slope of the ...
arXiv 2020
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.