Robust Cauchy-Based Methods for Predictive Regressions

Anton Skrobotov; JiHyun Kim; Rustam Ibragimov

arxiv: 2511.09249 · v3 · submitted 2025-11-12 · 💰 econ.EM · math.ST· stat.TH

Robust Cauchy-Based Methods for Predictive Regressions

Rustam Ibragimov , JiHyun Kim , Anton Skrobotov This is my paper

Pith reviewed 2026-05-17 22:45 UTC · model grok-4.3

classification 💰 econ.EM math.STstat.TH

keywords predictive regressionsrobust inferenceCauchy estimationendogeneityheavy tailspersistent volatilitystock return predictability

0 comments

The pith

Cauchy-based methods deliver size-correct tests for predictive regressions under endogeneity and heavy tails.

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

This paper develops robust inference methods for predictive regressions that handle endogenously persistent or heavy-tailed regressors and persistent volatility in errors. Standard inference often suffers from size distortions in these situations, making it hard to trust tests of predictability such as whether dividend yields forecast stock returns. The authors propose two tests based on the Cauchy estimation framework: a t-statistic group inference approach and a hybrid method combining Cauchy and OLS estimation. These tests are straightforward to use and work for models in continuous or discrete time. Simulations confirm good performance in finite samples, and an application suggests the dividend-price ratio predicts excess returns while the earnings-price ratio does not.

Core claim

Building on the Cauchy estimation framework, this paper proposes two novel tests for predictive regressions: one based on t-statistic group inference and the other a hybrid approach combining Cauchy and OLS estimation. These methods mitigate size distortions that arise in standard procedures under endogeneity, near nonstationarity, heavy tails, and persistent volatility. The tests are simple to implement and apply to both continuous- and discrete-time models, with extensive simulations showing favorable finite-sample performance and an empirical study indicating predictive power for the dividend-price ratio but not the earnings-price ratio.

What carries the argument

The Cauchy estimation framework supporting group t-statistic inference and hybrid Cauchy-OLS tests for robust size control in predictive regressions.

Load-bearing premise

The validity of the Cauchy estimation framework and the hybrid or group inference procedures holds when regressors are endogenously persistent or heavy-tailed and errors show persistent volatility.

What would settle it

A set of simulations or empirical cases where the new tests exhibit rejection rates substantially different from the nominal significance level under conditions of endogenous persistence and heavy tails would falsify the robustness claim.

Figures

Figures reproduced from arXiv: 2511.09249 by Anton Skrobotov, JiHyun Kim, Rustam Ibragimov.

**Figure 1.** Figure 1: Simulated density of 𝒟2 in (6). 3.2 A Hybrid Test We now propose a simple hybrid test that remains consistent for a broad class of covariates. Under Assumptions 2.1–2.3, ∑︀𝑇 𝑡=1 |𝑥𝑡−1| √ 𝑇 (𝛽ˇ − 𝛽) = 1 √ 𝑇 ∑︁ 𝑇 𝑡=1 sign(𝑥𝑡−1)𝑢𝑡 →𝑑 ∫︁ 1 0 𝜎(𝑟) 𝑑𝑊(𝑟) = 𝜔𝑍, where 𝑍 ∼ N(0, 1) and 𝜔 2 = ∫︀ 1 0 𝜎 2 (𝑟) 𝑑𝑟. A key feature of the Cauchy estimator 𝛽ˇ is that its properly normalized limit distribution is invariant to… view at source ↗

read the original abstract

This paper develops robust inference methods for predictive regressions that address key challenges posed by endogenously persistent or heavy-tailed regressors, as well as persistent volatility in errors. Building on the Cauchy estimation framework, we propose two novel tests: one based on $t$-statistic group inference and the other employing a hybrid approach that combines Cauchy and OLS estimation. These methods effectively mitigate size distortions that commonly arise in standard inference procedures under endogeneity, near nonstationarity, heavy tails, and persistent volatility. The proposed tests are simple to implement and applicable to both continuous- and discrete-time models. Extensive simulation experiments demonstrate favorable finite-sample performance across a range of realistic settings. An empirical application examines the predictability of excess stock returns using the dividend-price and earnings-price ratios as predictors. The results suggest that the dividend-price ratio possesses predictive power, whereas the earnings-price ratio does not significantly forecast returns.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper introduces useful Cauchy-based robust tests for predictive regressions with good simulation support, but the hybrid method may lack full asymptotic coverage for joint persistence and volatility issues.

read the letter

The punchline here is that the paper introduces two Cauchy-based tests for predictive regressions that handle common issues like persistent regressors and volatile errors, with simulations backing up better size control than standard approaches. They propose a group inference method using t-statistics and a hybrid that combines Cauchy and OLS estimation. These are presented as simple to implement for both continuous- and discrete-time models. The paper does a solid job with the Monte Carlo experiments. They test across various realistic scenarios involving endogeneity, near nonstationarity, heavy tails, and persistent volatility, and the results look favorable for finite samples. The application to stock returns predictability using dividend-price and earnings-price ratios is a nice touch, suggesting the former has power while the latter does not. On the downside, the asymptotic theory for the hybrid approach under the joint presence of endogenous persistence and GARCH volatility might be incomplete. The claim is that size distortions are mitigated, but if the volatility process creates dependence that affects the Cauchy component's distribution, the standard normal critical values could fail in exactly the cases that matter most for applications. I didn't find an explicit proof covering that combination in detail. This work is for applied econometricians in finance who need robust tools for predictability tests. Readers dealing with real data problems in these models will appreciate the practical focus and the evidence from simulations. It should go to peer review. The topic is important, the methods are novel enough, and the empirical results add value, though revisions on the theory side would help.

Referee Report

2 major / 2 minor

Summary. The paper develops robust inference methods for predictive regressions using a Cauchy estimation framework to handle endogenously persistent or heavy-tailed regressors and persistent volatility in errors. It proposes two tests: one based on t-statistic group inference and another using a hybrid Cauchy-OLS approach. These are claimed to mitigate size distortions common in standard procedures, with simplicity of implementation for both continuous- and discrete-time models. Support comes from simulation experiments showing favorable finite-sample performance and an empirical application to predictability of excess stock returns using dividend-price and earnings-price ratios.

Significance. If the size-control claims hold, the methods would provide practical, easy-to-implement tools for inference in predictive regressions, a setting where OLS-based tests frequently exhibit distortions under endogeneity, near-unit roots, heavy tails, and volatility clustering. The simulation evidence across realistic settings and the empirical application to stock returns are clear strengths that demonstrate applicability. The extension to both continuous- and discrete-time models further increases potential impact in finance and macroeconometrics.

major comments (2)

[§3.2] §3.2 (Hybrid Cauchy-OLS procedure): The central size-control claim for the hybrid estimator under simultaneous endogenous persistence (near unit root) and persistent volatility (GARCH-type errors) lacks an explicit asymptotic derivation. The paper does not show that volatility clustering does not induce dependence between the Cauchy-weighted score and regressor innovations that would invalidate standard-normal critical values for the t-statistic.
[§4] §4 (Simulation design): The reported Monte Carlo experiments do not include the joint case of endogenously persistent regressors combined with persistent volatility, which is the regime where the hybrid method's robustness is most needed to support the abstract's claim of mitigating size distortions; without these designs the finite-sample evidence is incomplete for the load-bearing assertion.

minor comments (2)

[Abstract] The abstract states that the methods are 'simple to implement' but does not provide explicit algorithmic steps or pseudocode for the group-inference critical-value construction.
[§2] Notation for the Cauchy weighting function and the group-inference partitioning should be defined earlier and used consistently in the empirical section to improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the careful reading and constructive comments, which help clarify the scope and presentation of our results. We address each major comment below and indicate the revisions we will make.

read point-by-point responses

Referee: [§3.2] §3.2 (Hybrid Cauchy-OLS procedure): The central size-control claim for the hybrid estimator under simultaneous endogenous persistence (near unit root) and persistent volatility (GARCH-type errors) lacks an explicit asymptotic derivation. The paper does not show that volatility clustering does not induce dependence between the Cauchy-weighted score and regressor innovations that would invalidate standard-normal critical values for the t-statistic.

Authors: We acknowledge that the current version does not contain a self-contained asymptotic derivation establishing that the hybrid Cauchy-OLS t-statistic remains asymptotically standard normal when near-unit-root endogeneity and GARCH volatility are present simultaneously. The paper motivates the hybrid procedure by combining the robustness properties of Cauchy weighting (which down-weights large innovations) with OLS efficiency, and shows consistency and asymptotic normality under either feature separately. In the revision we will add a brief discussion of the joint case, drawing on the fact that the Cauchy score is bounded and the GARCH volatility process is stationary, to argue that the dependence between the weighted score and regressor innovations remains asymptotically negligible for the purpose of standard-normal critical values. If a full joint proof proves lengthy, we will also report additional simulation evidence under the joint design as supporting finite-sample justification. revision: partial
Referee: [§4] §4 (Simulation design): The reported Monte Carlo experiments do not include the joint case of endogenously persistent regressors combined with persistent volatility, which is the regime where the hybrid method's robustness is most needed to support the abstract's claim of mitigating size distortions; without these designs the finite-sample evidence is incomplete for the load-bearing assertion.

Authors: We agree that the joint design is the most relevant stress test for the hybrid procedure. The existing Monte Carlo section examines the two sources of distortion in isolation to isolate their individual effects. We will add a new set of experiments that simultaneously impose endogenous near-unit-root regressors and GARCH(1,1) errors with high persistence in volatility. These results will be reported alongside the existing designs and will directly document size and power of both the t-statistic group and hybrid tests under the combined conditions highlighted in the abstract. revision: yes

Circularity Check

0 steps flagged

No significant circularity in derivation chain

full rationale

The paper introduces novel t-statistic group inference and hybrid Cauchy-OLS procedures for predictive regressions, building on an existing Cauchy estimation framework. Claims of size control under endogeneity, near nonstationarity, heavy tails, and persistent volatility are supported by proposed asymptotic arguments, extensive Monte Carlo simulations across continuous- and discrete-time models, and an empirical application to stock return predictability. No load-bearing step reduces by construction to fitted inputs, self-referential definitions, or unverified self-citation chains; the central results are presented as independent methodological contributions with external validation via simulations rather than tautological equivalences.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review provides no information on free parameters, axioms, or invented entities used in the derivations or simulations.

pith-pipeline@v0.9.0 · 5451 in / 1050 out tokens · 22081 ms · 2026-05-17T22:45:27.253066+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

The Cauchy estimator ˇβ = (∑ |x_{t-1}|)^{-1} ∑ sign(x_{t-1}) y_t ... under Assumptions 2.1–2.3 yields ∫ σ(r) dW(r) limit

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

and Hochberg, Y

Benjamini, Y. and Hochberg, Y. (1995), ‘Controlling the false discovery rate: A practical and powerful approach to multiple testing’,Journal of the Royal Statistical Society: Series B (Methodological)57(1), 289–300. Billingsley, P. (1986),Convergence of Probability Measures, John Wiley & Sons. Breitung, J. and Demetrescu, M. (2015), ‘Instrumental variable...

work page 1995
[2]

and Yogo, M

Campbell, J. and Yogo, M. (2006), ‘Efficient tests of stock return predictability’,Journal of Financial Econometrics81, 27–60. Cavaliere, G. (2004), ‘Testing stationarity under a permanent variance shift’,Economics Letters 82, 403–408. Cavaliere, G. and Taylor, A. R. (2007), ‘Testing for unit roots in time series models with non-stationary volatility’,Jou...

work page 2006
[3]

and M¨ uller, U

Dou, L. and M¨ uller, U. K. (2021), ‘Generalized local-to-unity models’,Econometrica 89(4), 1825–1854. Dufour, J.-M. and Torr` es, O. (2000), ‘Markovian processes, two-sided autoregressions and finite-sample inference for stationary and nonstationary autoregressive processes’,Journal of Econometrics99, 255–289. Elliott, G. and Stock, J. H. (1994), ‘Infere...

work page 2021

[1] [1]

and Hochberg, Y

Benjamini, Y. and Hochberg, Y. (1995), ‘Controlling the false discovery rate: A practical and powerful approach to multiple testing’,Journal of the Royal Statistical Society: Series B (Methodological)57(1), 289–300. Billingsley, P. (1986),Convergence of Probability Measures, John Wiley & Sons. Breitung, J. and Demetrescu, M. (2015), ‘Instrumental variable...

work page 1995

[2] [2]

and Yogo, M

Campbell, J. and Yogo, M. (2006), ‘Efficient tests of stock return predictability’,Journal of Financial Econometrics81, 27–60. Cavaliere, G. (2004), ‘Testing stationarity under a permanent variance shift’,Economics Letters 82, 403–408. Cavaliere, G. and Taylor, A. R. (2007), ‘Testing for unit roots in time series models with non-stationary volatility’,Jou...

work page 2006

[3] [3]

and M¨ uller, U

Dou, L. and M¨ uller, U. K. (2021), ‘Generalized local-to-unity models’,Econometrica 89(4), 1825–1854. Dufour, J.-M. and Torr` es, O. (2000), ‘Markovian processes, two-sided autoregressions and finite-sample inference for stationary and nonstationary autoregressive processes’,Journal of Econometrics99, 255–289. Elliott, G. and Stock, J. H. (1994), ‘Infere...

work page 2021