Pith. sign in

REVIEW 3 major objections 5 minor 38 references

Exhaustive Bayesian model averaging is computationally feasible for stochastic frontier models with up to about 30 candidate regressors, and using asymmetric errors changes model averaging and parameter recovery when inefficiency is strong

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

For stochastic frontier models, an exhaustive pre-screened Bayesian model search is feasible up to about 30 candidates, and non-Gaussian errors matter most when inefficiency dominates noise and signal strength is weak.

T0 review reviewed 2026-08-02 challenge →

load-bearing objection Useful, well-scoped computational paper that makes exhaustive SF-BMA/S feasible up to ~30 regressors; the Laplace approximation at the core needs validation across the model space, but the contribution is solid enough for peer review. the 3 major comments →

arxiv 2607.14274 v1 pith:DLGETCOO submitted 2026-07-15 econ.EM stat.CO

Model Uncertainty under Non-Gaussian Errors: Bayesian Model Averaging and Selection in Stochastic Frontier Models

Kamil Makie{\l}a This is my paper
classification econ.EM stat.CO
keywords Bayesian model averagingstochastic frontier analysismodel selectioncovariate selectionexhaustive searchnormal-exponential modelinefficiency-to-noise ratioLaplace approximation
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper claims that Bayesian model averaging and selection can be run exhaustively — checking all 2^k variable combinations — for stochastic frontier models with up to roughly 30 candidate regressors, using parallelization and a fast BIC-based pre-screening step. It further claims that accounting for the one-sided inefficiency error changes model-averaging outcomes and parameter-recovery accuracy relative to a conventional Gaussian-error model, particularly when inefficiency-to-noise ratio exceeds one and the signal of the data-generating process is weak. A Monte Carlo study supports this, showing almost uniform gains in posterior inclusion probabilities, Brier scores, and RMSE in those high-inefficiency, low-signal scenarios, while model selection (finding the true specification) is about equally good with either error assumption. The practical payoff is that applied efficiency analysts can get exact model weights over the full model space rather than stochastic approximations, and can know when the choice of error distribution matters.

Core claim

The central claim is that the normal-exponential stochastic frontier specification can be substituted into Bayesian model averaging/selection without giving up exhaustive search: by reparametrizing the error variance components for optimization and using Laplace approximation under a different parametrization to obtain integrated likelihoods, each model can be scored quickly enough to enumerate all 2^k models for k up to about 30. The simulation evidence then establishes that when inefficiency is large relative to noise (λ>1) and signal strength is not large (SS≤1), the stochastic-frontier BMA/S yields better posterior inclusion probabilities for true regressors, better Brier scores, and low

What carries the argument

Two reparametrizations of the normal-exponential stochastic frontier model. For numerical optimization the paper uses (β, log σ_ε², logit γ), where σ_ε² is total error variance and γ is the share of inefficiency variance, which keeps parameters identifiable and stable. For computing the integrated likelihood it uses (β, log σ_v, log σ_u), under which the posterior is more nearly Gaussian and Laplace approximation is accurate. The exhaustive search is made feasible by parallelizing over the 2^k models and by pre-screening with BIC from fast Gaussian-error regressions to keep only a top-M set of models for full stochastic-frontier scoring.

Load-bearing premise

The Laplace approximation of the integrated likelihood, computed under the log-variance parametrization, is accurate enough across all 2^k models that the resulting posterior model probabilities can be trusted; the paper cites earlier work for this rather than validating it here.

What would settle it

Take a small model space (k≈8) with strong inefficiency (λ=4) and weak signal (SS=0.25), and compute posterior model probabilities by Laplace approximation as in the paper. Recompute the same probabilities with a more exact method, e.g., direct numerical integration or MCMC over each model. If the top-model weights differ materially, the SF-BMA/S weights are not reliable.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • Applied SFA studies can replace stochastic-search BMA with exact exhaustive weights for up to about 30 candidate regressors, removing tuning of MCMC samplers.
  • When inefficiency-to-noise ratio is high and signal weak, Gaussian-error BMA will understate inclusion of true regressors and give less accurate parameter estimates; SF-BMA/S corrects this.
  • Model selection (choosing the single best specification) is not very sensitive to the error assumption; the main gains are in averaging and estimation.
  • Pre-screening with Gaussian-BIC is a cheap and safe way to reduce the model space for SF-BMA/S, with negligible loss of posterior probability mass even at top-M=100.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The speed gains from the BIC pre-screening step could be transferred to other non-Gaussian error structures (half-normal, truncated-normal, panel SFA) as long as the full SF model remains the final scorer; the paper only demonstrates the normal-exponential case.
  • Since the paper finds the two error assumptions agree on model selection, researchers primarily interested in choosing a model can keep using Gaussian BMA/S, but should switch to SF-BMA/S when the goal is efficiency/parameter inference under strong inefficiency.
  • A testable extension: the λ>1, SS≤1 region identified here suggests that applied studies with high estimated λ should re-run their BMA under an SF specification and compare PIPs; the paper's results predict material shifts.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper proposes computationally feasible methods for exhaustive Bayesian model averaging and selection in the normal-exponential stochastic frontier model (SF-BMA/S) for moderate-dimensional covariate selection (k up to about 30). The integrated likelihood in Eq. (3) is computed by Laplace approximation under parametrization A, and a fast pre-screening step based on BIC from Gaussian-error models is used to restrict the model space when k is large. A Monte Carlo study with 27 scenarios compares SF-BMA/S with conventional Gaussian-error BMA/S in terms of model recovery, model averaging, and parameter recovery. The paper reports that SF-BMA/S is generally at least as good as Gaussian BMA/S and is clearly preferable for BMA and parameter recovery when the inefficiency-to-noise ratio is high (λ>1) and signal strength is low (SS≤1). Two empirical applications illustrate the method for k=20 and k=27, including efficiency-score comparisons. Code is provided in MATLAB and Python.

Significance. If the results hold, the paper offers applied SFA researchers a practical exact-search alternative to stochastic-search BMA, with clear guidance on when accounting for asymmetric errors changes model-averaging conclusions. The computational contribution (parallel exhaustive search with BIC pre-screening) is potentially useful, and the simulation design with 27 scenarios per draw and correlated-regressor robustness checks is reasonably careful. The paper also states its limitations candidly, including numerical instability in poorly specified models. The main unresolved issue is that the load-bearing integrated-likelihood approximation is inherited from prior work rather than validated within this paper, and the Monte Carlo comparisons lack sampling uncertainty, which weakens the headline recommendations.

major comments (3)
  1. [§2.3, Eq. (3)] Posterior model probabilities in Eq. (2) are proportional to the integrated likelihoods p(y|M_j), computed via Laplace approximation under parametrization A, citing Makieła and Mazur (2022) rather than validating the approximation here. Since the Laplace approximation is a local Gaussian approximation around the posterior mode, it is most fragile in precisely the cases the paper itself flags (Section 5: 'almost flat regions' and poorly specified regressor sets). Approximation errors that vary across models directly bias the posterior model weights, PIPs, Brier scores, and the Section 5 recommendation to prefer SF over Gaussian when λ>1 and SS≤1. Please add a direct validation: compare the Laplace log integrated likelihood with a more accurate benchmark (e.g., adaptive quadrature or posterior simulation) for a sample of models spanning the model space, and report discrepancies as a functi
  2. [§3.3, Table 3] The pre-screening validation is limited to scenarios 19–27, where k=14 and the model space has 2^14=16,384 models. The empirical applications use k=20 and k=27 with topM=400/4000/20000, and the paper claims the strategy scales to k≈30. The current evidence does not establish that Gaussian-BIC pre-screening with these topM values captures enough posterior mass at such larger model spaces. Add simulations with k≥20 where the true model is known, reporting coverage of posterior mass, overlap with the exact ranking, and the impact of topM on the final BMA results. Also, Table 3 states 'averages based on 50 MC draws', whereas Section 3.1 states 100 Monte Carlo draws; clarify and reconcile this discrepancy.
  3. [§3.2–3.3, Tables 1–2] All Monte Carlo comparisons are reported as mean differences and win-scores without any measure of sampling uncertainty. With only 100 draws, small reported differences (e.g., overall ΔPIP_tr=0.0250, −ΔBS=0.0079 in Table 1) may be within Monte Carlo noise. Report standard errors or confidence intervals for the mean differences (e.g., bootstrap) and, for the scores, binomial confidence intervals or a sign test. This is necessary to support the statement that 'SF-BMA/S is almost always better' and the conditional recommendation in Section 5.
minor comments (5)
  1. [Eq. (14)] The definition of ΔSign includes a leading minus sign, but the text says positive values indicate an advantage of SF-BMA/S. This is inconsistent and should be corrected to a positive sign.
  2. [Table 4] The table header says 'Fast pre-screening with topM=400', while Section 2.3 states 'topM=4000 is set throughout the paper'. If 4000 was intended, the header is a typo; if 400 was actually used, this needs justification because it is much smaller than the default and the results are identical to the full-search results to the twelfth decimal.
  3. [Section 4, Figure numbering] The text says 'Figure 1 shows these posteriors' when referring to marginal posterior distributions of technical efficiency, but Figure 1 is already the runtime plot. This should be Figure 2.
  4. [Section 5] There is a typo: 'inefficiency-to-nose ratio' should be 'inefficiency-to-noise ratio'.
  5. [Table 3, note] Spearman's rank correlation is described as being calculated only for models that appear in both rankings. That is not a full rank correlation over the entire model space and may give an optimistic picture. Either compute the correlation on full rankings with appropriate handling of ties, or justify why the overlap-only calculation is informative.

Circularity Check

0 steps flagged

No significant circularity: the paper's claims are supported by standard Bayesian identities, simulation experiments with known DGPs, and external (prior) validation of the Laplace approximation.

full rationale

The paper's model probabilities (Eq. 2) and integrated likelihoods (Eq. 3) are standard Bayesian definitions, not quantities derived from the outputs they are used to produce. The SF likelihood is a known convolution; the paper estimates it by numerical optimization, and the Laplace approximation used for integrated likelihoods is an approximation whose accuracy is the main technical assumption. This assumption is supported by a citation to Makieła and Mazur (2022), a self-citation, but the cited work is described as comparing the Laplace approximation to MC-based approaches (Pajor, 2017), i.e., external evidence, and the current paper's Monte Carlo recovery results provide indirect validation of the whole pipeline. The BIC pre-screening is not presented as a prediction of SF posterior weights; its usefulness is assessed directly by how much posterior mass the top-M models cover (Table 3), so it is not a fitted input renamed as an output. The simulation comparison uses a known DGP; showing that the correctly specified SF model outperforms a misspecified Gaussian model in high-inefficiency, low-signal scenarios is an empirical property of that design, not a result equivalent by construction to the inputs. The paper's own caveat that SF likelihoods can have almost flat regions and be numerically unstable for poorly specified regressor sets (Section 5) is a robustness limitation for the Laplace approximation, but it does not establish that the procedure's outputs are definitionally identical to its inputs. No equation reduces to another by construction, and no fitted parameter is relabeled as a prediction. Therefore no significant circularity is found.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 0 invented entities

The central claim rests on the SF error model, chosen priors, an unvalidated Laplace approximation, and a BIC pre-screen; no new theoretical entities are introduced.

free parameters (3)
  • Prior median efficiency r* = 0.75
    Hyperparameter for the inverse-gamma prior on sigma_u (via sigma_u^{-1} ~ Gamma(1, -ln r*)); chosen by hand and held fixed throughout; posterior model probabilities and efficiency inference depend on it.
  • Pre-screening pool size topM = 4000
    Maximum number of models retained by fast BIC pre-screening; chosen by hand (with alternatives 10000/20000 recommended for large k); the speed/accuracy trade-off claim depends on this value.
  • Prior scales on beta and sigma_v = C=100 I, g1=g2=1e-4
    Vague priors chosen by hand according to standard practice; not fitted to data but affect integrated likelihoods and therefore model weights.
axioms (5)
  • domain assumption The compound error is normal noise plus exponential inefficiency (normal-exponential SF model).
    The paper's SF-BMA/S is defined for this model (Eq. 4–5); if the true data-generating inefficiency is not exponential, the method's advantage is not established.
  • domain assumption Inverse-gamma/gamma priors and prior independence of beta, sigma_v, sigma_u.
    Priors from van den Broeck et al. (1994), used to compute integrated likelihoods; prior details can change BMA/S weights.
  • ad hoc to paper Laplace approximation around the posterior mode under parametrization A is accurate enough for integrated likelihoods across all models.
    Invoked in Section 2.3 without validation in this paper; errors in this approximation would bias posterior model probabilities.
  • ad hoc to paper BIC from Gaussian-error models is a sufficient pre-filter for the models that dominate SF posterior probability.
    Tested in Table 3 for selected scenarios and topM up to 1000, then extrapolated to topM=4000 and to other scenarios; if the pre-filter misses relevant models, exhaustive SF-BMA/S coverage fails.
  • domain assumption In simulation, data are generated from the normal-exponential SF model.
    The DGP in Section 3.1 matches the SF-BMA/S error structure; this makes the comparison favorable to SF-BMA/S by construction.

reviewed 2026-08-02 · how reviews work

0 comments
Cite this review

Pith. "Pith review of Model Uncertainty under Non-Gaussian Errors: Bayesian Model Averaging and Selection in Stochastic Frontier Models." pith.science (2026). https://pith.science/paper/DLGETCOO

@misc{pith2026260714274,
  author       = {Pith},
  title        = {Pith review of: Model Uncertainty under Non-Gaussian Errors: Bayesian Model Averaging and Selection in Stochastic Frontier Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DLGETCOO}},
  note         = {Machine review of arXiv:2607.14274}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

The paper investigates Bayesian Model Averaging and Selection (BMA/S) under non-standard stochastic assumptions, focusing on stochastic frontier analysis (SFA). We propose fast, reliable procedures for inference in the normal-exponential stochastic frontier model and examine whether accounting for asymmetric disturbances affects model averaging and/or selection outcomes relative to the conventional Gaussian-error BMA/S. Particular attention is given to moderate-dimensional covariate selection problems typical in SFA applications. We demonstrate that, with appropriate search strategies and parallelization techniques, exhaustive model search can be computationally feasible and, in some cases, more practical than stochastic search alternatives. A Monte Carlo simulation study is used to compare the proposed SF-BMA/S procedure with standard Gaussian-error BMA/S under varying levels of inefficiency-to-noise ratio and signal strength with respect to the data generating process. The results show that accounting for stochastic frontier structures may affect posterior inference and model averaging outcomes, especially in scenarios where efficiency analysis is most sensible.

Figures

Figures reproduced from arXiv: 2607.14274 by Kamil Makie{\l}a.

Figure 2
Figure 2. Figure 2: Marginal posterior distributions of technical efficiency for selected U.S. states 1.64 1.58 1.51 1.64 1.92 2.55 3.92 6.65 12.33 24.54 50.02 102.82 212.7 439.82 914.19 1890.48 0 200 400 600 800 1000 1200 1400 1600 1800 2000 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 seconds number of candidate variables (k) [PITH_FULL_IMAGE:figures/full_fig_p021_2.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 19 canonical work pages

  1. [1]

    When, where and how to estimate persistent and transient efficiency in stochastic frontier panel data models

    Badunenko, O. and Kumbhakar, S.C. (2016), “When, where and how to estimate persistent and transient efficiency in stochastic frontier panel data models”, European Journal of Operational Research, Vol. 255 No. 1, pp. 272–287, doi: 10.1016/j.ejor.2016.04.049

    work page doi:10.1016/j.ejor.2016.04.049 2016

  2. [2]

    Economic Growth in a Cross Section of Countries

    Barro, R.J. (1991), “Economic Growth in a Cross Section of Countries”, The Quarterly Journal of Economics, Vol. 106 No. 2, p. 407, doi: 10.2307/2937943

    doi:10.2307/2937943 1991

  3. [3]

    Estimation of a Production Frontier Model: With Application to the Pastoral Zone of Eastern Australia

    Battese, G.E. and Corra, G.S. (1977), “Estimation of a Production Frontier Model: With Application to the Pastoral Zone of Eastern Australia”, Australian Journal of Agricultural Economics, Vol. 21 No. 3, pp. 169–179, doi: 10.1111/j.1467-8489.1977.tb00204.x

    arXiv 1977

  4. [4]

    Verification of Forecasts Expressed in Terms of Probability

    Brier, G.W. (1950), “Verification of Forecasts Expressed in Terms of Probability”, Monthly Weather Review , Vol. 78 No. 1, pp. 1 –3, doi: 10.1175/1520 - 0493(1950)078<0001:VOFEIT>2.0.CO;2

    doi:10.1175/1520 1950

  5. [5]

    Stochastic frontier models: a Bayesian perspective

    van den Broeck, J., Koop, G., Osiewalski, J. and Steel, M.F.J. (1994), “Stochastic frontier models: a Bayesian perspective”, Journal of Econometrics , Vol. 61 No. 2, pp. 273 –303, doi: 10.1016/0304-4076(94)90087-6

    work page doi:10.1016/0304-4076(94)90087-6 1994

  6. [6]

    Panel Data Econometrics in R : The plm Package

    Croissant, Y. and Millo, G. (2008), “Panel Data Econometrics in R : The plm Package”, Journal of Statistical Software, Vol. 27 No. 2, doi: 10.18637/jss.v027.i02

    doi:10.18637/jss.v027.i02 2008

  7. [7]

    Benchmark priors for Bayesian model averaging

    Fernández, C., Ley, E. and Steel, M.F.J. (2001), “Benchmark priors for Bayesian model averaging”, Journal of Econometrics , Vol. 100 No. 2, pp. 381 –427, doi: 10.1016/S0304 - 4076(00)00076-2

    work page doi:10.1016/s0304 2001

  8. [8]

    Variable Selection via Gibbs Sampling

    George, E.I. and McCulloch, R.E. (1993), “Variable Selection via Gibbs Sampling”, Journal of the American Statistical Association , Vol. 88 No. 423, pp. 881 –889, doi: 10.1080/01621459.1993.10476353

    arXiv 1993

  9. [9]

    Strictly Proper Scoring Rules, Prediction, and Estimation

    Gneiting, T. and Raftery, A.E. (2007), “Strictly Proper Scoring Rules, Prediction, and Estimation”, Journal of the American Statistical Association , Vol. 102 No. 477, pp. 359 –378, doi: 10.1198/016214506000001437

    doi:10.1198/016214506000001437 2007

  10. [10]

    Distinguishing between heterogeneity and inefficiency: stochastic frontier analysis of the World Health Organization’s panel data on national health care systems

    Greene, W. (2004), “Distinguishing between heterogeneity and inefficiency: stochastic frontier analysis of the World Health Organization’s panel data on national health care systems”, Health Economics, Vol. 13 No. 10, pp. 959–980, doi: 10.1002/hec.938

    work page doi:10.1002/hec.938 2004

  11. [11]

    Fixed and Random Effects in Stochastic Frontier Models

    Greene, W. (2005), “Fixed and Random Effects in Stochastic Frontier Models”, Journal of Productivity Analysis, Vol. 23 No. 1, pp. 7–32, doi: 10.1007/s11123-004-8545-1

    work page doi:10.1007/s11123-004-8545-1 2005

  12. [12]

    (2018), Econometric Analysis, 8th ed., Prentice Hall / Pearson, Upper Saddle River, NJ

    Greene, W. (2018), Econometric Analysis, 8th ed., Prentice Hall / Pearson, Upper Saddle River, NJ

    2018

  13. [13]

    Bayesian model averaging: a tutorial (with comments by M. Clyde, David Draper and E. I. George, and a rejoinder by the authors

    Hoeting, J.A., Madigan, D., Raftery, A.E. and Volinsky, C.T. (1999), “Bayesian model averaging: a tutorial (with comments by M. Clyde, David Draper and E. I. George, and a rejoinder by the authors”, Statistical Science, Vol. 14 No. 4, doi: 10.1214/ss/1009212519

    arXiv 1999

  14. [14]

    The efficiency of health production: re‐estimating the WHO panel data using parametric and non‐parametric approaches to provide additional information

    Hollingsworth, B. and Wildman, J. (2003), “The efficiency of health production: re‐estimating the WHO panel data using parametric and non‐parametric approaches to provide additional information”, Health Economics, Vol. 12 No. 6, pp. 493–504, doi: 10.1002/hec.751

    work page doi:10.1002/hec.751 2003

  15. [15]

    On the estimation of technical inefficiency in the stochastic frontier production function model

    Jondrow, J., Knox Lovell, C.A., Materov, I.S. and Schmidt, P. (1982), “On the estimation of technical inefficiency in the stochastic frontier production function model”, Journal of Econometrics, Vol. 19 No. 2–3, pp. 233–238, doi: 10.1016/0304-4076(82)90004-5

    work page doi:10.1016/0304-4076(82)90004-5 1982

  16. [16]

    A zero inefficiency stochastic frontier model

    Kumbhakar, S.C., Parmeter, C.F. and Tsionas, E.G. (2013), “A zero inefficiency stochastic frontier model”, Journal of Econometrics , Vol. 172 No. 1, pp. 66 –76, doi: 10.1016/j.jeconom.2012.08.021

    work page doi:10.1016/j.jeconom.2012.08.021 2013

  17. [17]

    On the effect of prior assumptions in Bayesian model averaging with applications to growth regression

    Ley, E. and Steel, M.F.J. (2009), “On the effect of prior assumptions in Bayesian model averaging with applications to growth regression”, Journal of Applied Econometrics, Vol. 24 No. 4, pp. 651–674, doi: 10.1002/jae.1057

    work page doi:10.1002/jae.1057 2009

  18. [18]

    Bayesian model averaging by combining deep learning models to improve lake water level prediction

    Li, G., Liu, Z., Zhang, J., Han, H. and Shu, Z. (2024), “Bayesian model averaging by combining deep learning models to improve lake water level prediction”, Science of The Total Environment, Vol. 906, p. 167718, doi: 10.1016/j.scitotenv.2023.167718

    arXiv 2024

  19. [19]

    Bayesian Stochastic Frontier Analysis of Economic Growth and Productivity Change in the EU, USA, Japan and Switzerland

    Makieła, K. (2014), “Bayesian Stochastic Frontier Analysis of Economic Growth and Productivity Change in the EU, USA, Japan and Switzerland”, Central European Journal of Economic Modelling and Econometrics , Vol. 6 No. 3, pp. 193 –216, doi: https://doi.org/10.24425/cejeme.2014.119239

    arXiv 2014

  20. [20]

    Bayesian Inference and Gibbs Sampling in Generalized True Random - Effects Models

    Makieła, K. (2017), “Bayesian Inference and Gibbs Sampling in Generalized True Random - Effects Models”, Central European Journal of Economic Modelling and Econometrics , Vol. 9 No. 1, pp. 69–95, doi: https://doi.org/10.24425/cejeme.2017.122200

    arXiv 2017

  21. [21]

    MATLAB Function for Bayesian Stochastic Frontier Analysis

    Makieła, K. (2020), “MATLAB Function for Bayesian Stochastic Frontier Analysis”

    2020

  22. [22]

    Are European Farms Equally Efficient? What Do Regional FADN Data on Crop Farms Tell Us?

    Makieła, K., Marzec, J., Pisulewski, A. and Mazur, B. (2025), “Are European Farms Equally Efficient? What Do Regional FADN Data on Crop Farms Tell Us?”, Journal of Agricultural and Applied Economics, Vol. 57 No. 1, pp. 157–181, doi: 10.1017/aae.2025.1

    work page doi:10.1017/aae.2025.1 2025

  23. [23]

    Bayesian Model Averaging and Prior Sensitivity in Stochastic Frontier Analysis

    Makieła, K. and Mazur, B. (2020), “Bayesian Model Averaging and Prior Sensitivity in Stochastic Frontier Analysis”, Econometrics, Vol. 8 No. 2, p. 13, doi: 10.3390/econometrics8020013

    work page doi:10.3390/econometrics8020013 2020

  24. [24]

    Model uncertainty and efficiency measurement in stochastic frontier analysis with generalized errors

    Makieła, K. and Mazur, B. (2022), “Model uncertainty and efficiency measurement in stochastic frontier analysis with generalized errors”, Journal of Productivity Analysis , Vol. 58 No. 1, pp. 35–54, doi: 10.1007/s11123-022-00639-y

    work page doi:10.1007/s11123-022-00639-y 2022

  25. [25]

    Cost Efficiency Analysis of Electricity Distribution Sector under Model Uncertainty

    Makieła, K. and Osiewalski, J. (2018), “Cost Efficiency Analysis of Electricity Distribution Sector under Model Uncertainty”, The Energy Journal , Vol. 39 No. 4, pp. 31 –56, doi: 10.5547/01956574.39.4.kmak

    work page doi:10.5547/01956574.39.4.kmak 2018

  26. [26]

    (2002), Subset Selection in Regression , Chapman and Hall/CRC, doi: 10.1201/9781420035933

    Miller, A. (2002), Subset Selection in Regression , Chapman and Hall/CRC, doi: 10.1201/9781420035933

    work page doi:10.1201/9781420035933 2002

  27. [27]

    Bayesian Variable Selection in Linear Regression

    Mitchell, T.J. and Beauchamp, J.J. (1988), “Bayesian Variable Selection in Linear Regression”, Journal of the American Statistical Association , Vol. 83 No. 404, pp. 1023 –1032, doi: 10.1080/01621459.1988.10478694

    arXiv 1988

  28. [28]

    Why Has Productivity Growth Declined? Productivity and Public Investment

    Munnell, A.H. (1990), “Why Has Productivity Growth Declined? Productivity and Public Investment”, New England Economic Review, pp. 3–22

    1990

  29. [29]

    Regression Models under Competing Covariance Structures: A Bayesian Perspective

    Osiewalski, J. and Steel, M.F.J. (1993), “Regression Models under Competing Covariance Structures: A Bayesian Perspective”, Annales d’Economie et de Statistique, No. 32, pp. 65–79

    1993

  30. [30]

    Efficiency Analysis: A Primer on Recent Advances

    Parmeter, C.F. and Kumbhakar, S.C. (2014), “Efficiency Analysis: A Primer on Recent Advances”, Foundations and Trends® in Econometrics , Vol. 7 No. 3 –4, pp. 191 –385, doi: 10.1561/0800000023

    work page doi:10.1561/0800000023 2014

  31. [31]

    Bayesian Model Averaging for Linear Regression Models

    Raftery, A.E., Madigan, D. and Hoeting, J.A. (1997), “Bayesian Model Averaging for Linear Regression Models”, Journal of the American Statistical Association, Vol. 92 No. 437, pp. 179– 191, doi: 10.1080/01621459.1997.10473615

    arXiv 1997

  32. [32]

    BMA: An R Package for Bayesian Model Averaging

    Raftery, A.E., Painter, I.S. and Volinsky, C.T. (2005), “BMA: An R Package for Bayesian Model Averaging”, R News, Vol. 5 No. 2, pp. 2–8

    2005

  33. [33]

    Nonparametric regression using Bayesian variable selection

    Smith, M. and Kohn, R. (1996), “Nonparametric regression using Bayesian variable selection”, Journal of Econometrics, Vol. 75 No. 2, pp. 317–343, doi: 10.1016/0304-4076(95)01763-1

    work page doi:10.1016/0304-4076(95)01763-1 1996

  34. [34]

    Model Averaging and Its Use in Economics

    Steel, M.F.J. (2020), “Model Averaging and Its Use in Economics”, Journal of Economic Literature, Vol. 58 No. 3, pp. 644–719, doi: 10.1257/jel.20191385

    work page doi:10.1257/jel.20191385 2020

  35. [35]

    Bayesian Model Averaging in Causal Instrumental Variable Models

    Steiner, G. and Steel, M. (2026), “Bayesian Model Averaging in Causal Instrumental Variable Models”, Journal of Applied Econometrics, doi: 10.1002/jae.70070

    work page doi:10.1002/jae.70070 2026

  36. [36]

    Bayesian model averaging with non-conjugate priors

    Tasiopoulos, A.E., Tsionas, E.G. and Vlastakis, N.D. (2026), “Bayesian model averaging with non-conjugate priors”, Journal of Econometrics , p. 106256, doi: 10.1016/j.jeconom.2026.106256

    arXiv 2026

  37. [37]

    Maximum likelihood estimation of stochastic frontier models by the Fourier transform

    Tsionas, E.G. (2012), “Maximum likelihood estimation of stochastic frontier models by the Fourier transform”, Journal of Econometrics , Vol. 170 No. 1, pp. 234 –248, doi: 10.1016/j.jeconom.2012.04.001

    work page doi:10.1016/j.jeconom.2012.04.001 2012

  38. [38]

    Zellner, A. (1971). An Introduction to Bayesian Inference in Econometrics . New York: John Wiley & Sons. Tables and Figures Table 1: Mean differences between accuracy measures in SF-BMA/S and Gaussian BMA/S Sc. 𝒌𝟐 𝝀 SS 𝚫𝑻𝑷𝑹̅̅̅̅̅̅̅̅ −𝚫𝑭𝑷𝑹̅̅̅̅̅̅̅̅ 𝚫𝑬𝑹 𝚫𝑷𝑰𝑷̅̅̅̅̅̅̅̅𝒕𝒓 −𝚫𝑷𝑰𝑷̅̅̅̅̅̅̅̅𝒇𝒔 −𝚫𝑩𝑺̅̅̅̅̅̅ −𝚫𝑹𝑴𝑺𝑬̅̅̅̅̅̅̅̅̅̅̅ 𝚫𝑺𝒊𝒈𝒏̅̅̅̅̅̅̅̅̅ 1 4 0.5 0.25 0 0 0.01 0.0013 -0....

    1971

This paper was first reviewed by deepseek-v4-flash on August 2, 2026.