Robust Cauchy-Based Methods for Predictive Regressions
Listen
Radio episode about this paper
Transcript
Introduction to the show: ident: Economics. The day in economics: theory, mechanisms and the evidence beneath them.
Ruth: We're looking at the paper: "Robust Cauchy-Based Methods for Predictive Regressions".
Warren: Here's the gist of the paper. Robust Cauchy-Based Methods for Predictive Regressions develops two novel tests based on Cauchy estimation to address size distortions arising from endogeneity, near nonstationarity, heavy tails, and persistent volatility in errors The proposed methods are robust to heterogeneous and persistent volatility as well as to endogenous, persistent, and heavy-tailed regressors Both methods employ Cauchy estimation to address endogeneity, persistence, and heavy tails.
Ruth: The paper builds on the Cauchy estimation framework to propose two novel tests: one based on t-statistic group inference and another employing a hybrid approach that combines Cauchy and OLS estimation These methods are simple to implement and applicable to both continuous- and discrete-time models.
Warren: The two approaches differ in their treatment of heterogeneous volatility, with the first extending the t-statistic–based group inference of Ibragimov and M¨uller (2010) to asymptotically normal Cauchy estimators, while the second is a hybrid procedure that combines Cauchy and OLS estimation by using the Cauchy estimator for the coefficient and OLS residuals for variance estimation.
Ruth: The underlying model considers a linear predictive regression model yt = α + βx(t-1) + ut.
Paper summary: Ruth: So to wrap up this discussion on "Robust Cauchy-Based Methods for Predictive Regressions," the authors are presenting these two specific tests—the t-statistic group inference and the Cauchy-OLS hybrid—as a practical, theoretically sound alternative for making inferences in predictive regressions.
Warren: The implication is that if you're working with financial data that has these persistent, heavy-tailed, or endogenously correlated predictors, standard hypothesis tests can give you misleading results about whether a predictor actually works.
Iris: They are suggesting that by using these Cauchy-based methods, we can get inference procedures that are robust to those specific data challenges without having to resort to the more technically complex time-change transformations or nonparametric volatility corrections used in other literature.
Ruth: That’s right, they aim for simplicity in implementation while still addressing those challenging statistical environments. The paper shows these methods work well even when there are violations of the asymptotic exogeneity condition on volatility, which is a nice practical thing to know.
Warren: The final piece is that their empirical tests showed real predictive power for one specific ratio—the dividend-price ratio—which gives a tangible application to these statistical tools. It moves the discussion from just theory about robustness to showing what actually predicts returns in a relevant setting.
Conclusion: Ruth: So we've looked at these Cauchy based methods for predictive regressions, and now we're hitting the conclusion of this paper by looking at what they actually named it and who wrote it.
Warren: It’s called "Robust Cauchy-Based Methods for Predictive Regressions." And honestly, that title sounds a little like something you'd read in an old textbook.
Iris: It sounds technical because it is, but the point of the paper is that these specific methods are designed to handle real-world messy data without needing those extra complicated corrections.
Ruth: Exactly. The authors are showing how these two tests—the t-statistic group inference and that hybrid Cauchy OLS approach—are more reliable when your data has things like persistent volatility or heavy tails in the errors.
Warren: So, they're making sure that when we try to predict something, whether it’s stock returns or anything else, we aren't getting those big statistical surprises because of how weird the underlying data generating process is.
Iris: The implication for me is that for econometricians trying to build models on financial or complex time series data, this gives us a practical toolkit. It moves away from methods that assume everything is nice and clean and shows us how to proceed when it isn't.
Ruth: Right. And the empirical results they showed, where they tested it on stock returns, actually gave some pretty interesting hints about which predictors have real predictive power in practice.
Warren: Yeah, and that’s what I find compelling—it’s not just a theoretical exercise; it shows how these methods can point you toward a signal that actually works when you look at the actual numbers.
Iris: It changes the way we think about model validation, because we have to account for those non-standard error structures more seriously in our setups.
Ruth: So, if you’re thinking about building your next predictive model, this paper suggests that robustness against these kinds of data quirks is a very practical thing to bake into your testing procedure right from the start.
Warren: It gives us a way to run things without needing massive amounts of extra computation or super complex machinery just to get a reliable answer.
Iris: The real question now is how this scales up when you move from these simple linear models to more complicated structures where the volatility itself is changing over time in non-stationary ways.
Rustam Ibragimov, Jihyun Kim, Anton Skrobotov
Imperial College Business School · New Economic School · School of Economics, Sungkyunkwan University · HSE University
econ.EM, math.ST, stat.TH
Submitted: 2025-11-12
Updated: 2026-11-02
License: http://creativecommons.org/licenses/by/4.0/
Importance score: 83/100
The gist: Robust Cauchy-Based Methods for Predictive Regressions develops two novel tests based on Cauchy estimation to address size distortions arising from endogeneity, near nonstationarity, heavy tails…
Key concepts
- Cauchy Estimation
- This is a specific statistical technique used here to estimate model coefficients. It treats the coefficient as an instrumental variable (IV) estimator, which helps correct biases caused by endogeneity and heavy tails in the data.
- Size Distortions
- These are errors in statistical tests that occur when assumptions about the data distribution are violated, such as having heavy tails or persistent volatility. The proposed methods are designed to make inferences robust against these distortions.
- Hybrid Test ($ au(eta^eta)$)
- This is a second proposed test that combines two estimation methods: using the Cauchy estimator for the coefficient and using Ordinary Least Squares (OLS) residuals to estimate variance. This combination aims to improve efficiency compared to using only one method.
Terminology
Summary
Robust Cauchy-Based Methods for Predictive Regressions develops two novel tests based on Cauchy estimation to address size distortions arising from endogeneity, near nonstationarity, heavy tails, and persistent volatility in errors The proposed methods are robust to heterogeneous and persistent volatility as well as to endogenous, persistent, and heavy-tailed regressors Both methods employ Cauchy estimation to address endogeneity, persistence, and heavy tails. p. 0 Robust Cauchy-Based Methods for Predictive Regressions∗
Rustam Ibragimov𝑎,𝑏, Jihyun Kim𝑑
, Anton Skrobotov𝑒
𝑎
Imperial College Business School
𝑏 New Economic School
𝑑 School of Economics, Sungkyunkwan University
𝑒 HSE University
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.Where the paper says it
Methodology and Framework
The paper builds on the Cauchy estimation framework to propose two novel tests: one based on t-statistic group inference and another employing a hybrid approach that combines Cauchy and OLS estimation These methods are simple to implement and applicable to both continuous- and discrete-time models. The two approaches differ in their treatment of heterogeneous volatility, with the first extending the t-statistic–based group inference of Ibragimov and M¨uller (2010) to asymptotically normal Cauchy estimators, while the second is a hybrid procedure that combines Cauchy and OLS estimation by using the Cauchy estimator for the coefficient and OLS residuals for variance estimation. p. 2 (2024), to address endogeneity, persistence, and
heavy tails. The two approaches differ in their treatment of heterogeneous volatility: the first
extends the 𝑡-statistic–based group inference of Ibragimov and M¨uller (2010) to asymptotically
normal Cauchy estimators, while the second is a hybrid procedure that combines Cauchy and
OLS estimation by using the Cauchy estimator for the coefficient and OLS residuals for variance
estimation. , the time series of price–dividend ratios). As
usual, we consider the linear predictive regression model
𝑦𝑡 = 𝛼 + 𝛽𝑥𝑡−1 + 𝑢𝑡
, 𝑡 = 1, .Where the paper says it
Where the paper says it
Theoretical Properties and Asymptotics
The Cauchy estimator is defined as βˇ = (∑T t=1 xt-1)−1 (∑T t=1 sign(xt-1) yt), which can be interpreted as an instrumental variable (IV) estimator. Under Assumptions 2. p. 1 (2015), among others, develop inference methods that perform well
in the presence of persistent and endogenous regressors. However, these approaches typically
rely on assumptions such as homoskedasticity or weak dependence in volatility and may not
remain valid under empirically relevant features such as persistent or nonstationary volatility,
structural breaks, or regime switching. Under Assumptions 2. 2, the properly normalized Cauchy estimator satisfies
(︃∑︁
𝑇
𝑡=1
|𝑥𝑡−1|/
√
𝑇
)︃
(𝛽ˇ − 𝛽) ⇒
∫︁ 1
0
𝜎(𝑟) 𝑑𝑊(𝑟),
by standard results on the convergence of stochastic integrals (see Hansen, 1992; Kurtz and
Protter, 1991; Ibragimov et al. 2, the properly normalized Cauchy estimator satisfies
(︃∑︁
𝑇
𝑡=1
|𝑥𝑡−1|/
√
𝑇
)︃
(𝛽ˇ − 𝛽) ⇒
∫︁ 1
0
𝜎(𝑟) 𝑑𝑊(𝑟),
by standard results on the convergence of stochastic integrals (see Hansen, 1992; Kurtz and
Protter, 1991; Ibragimov et al. The limit ∫︀ 1
0
𝜎(𝑟) 𝑑𝑊(𝑟) is in general a non-Gaussian
martingale, becoming Gaussian only if 𝑊 and 𝜎 are independent.Where the paper says it
Where the paper says it
Where the paper says it
Robust Inference Procedures
The paper proposes two practical tests that serve as robust alternatives to existing methods. The first approach relies on t-statistic-based inference using group estimates of β, where the test statistic is tq(βˇ) = √q¯ sβ, and this approach is conservative under H0 but consistent under HA when (xt) is stationary with a finite first moment. p. 5 3 Robust Inference for Predictive Regressions
3. 1 Robust 𝑡-Statistic Approach
The first approach relies on 𝑡-statistic-based inference using group estimates of 𝛽, as proposed by Ibragimov and M¨uller (2010) (see also Ibragimov and M¨uller, 2016; Section 3. 7
We define the hybrid test statistic as
𝜏 (𝛽ˇ) = 𝛾ˇ
𝜔ˆ
,
where 𝛾ˇ = (∑︀𝑇
𝑡=1 |𝑥𝑡−1|/
√
𝑇)𝛽ˇ as in (2), and
𝜔ˆ
2 =
1
𝑇
∑︁
𝑇
𝑡=1
𝑢ˆ
2
𝑡
, 𝑢ˆ𝑡 = 𝑦𝑡 − 𝛽𝑥ˆ
𝑡−1. Here, 𝜔ˆ
2
estimates 𝜔
2 =
∫︀ 1
0
𝜎
2
(𝑟) 𝑑𝑟 using OLS residuals.Where the paper says it
Where the paper says it
Empirical Performance and Conclusion
Simulation experiments demonstrate that the proposed tests perform well in finite samples, maintaining correct size and competitive power under a wide range of data-generating processes, including those characterized by stochastic volatility, structural breaks, and regime switching. In an empirical application to stock return predictability using the dividend–price and earnings–price ratios as predictors, the results suggest that the dividend–price ratio possesses predictive power while the earnings–price ratio does not significantly forecast returns. p. 0 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. Overall, our procedures complement
existing Cauchy-based methods and provide a practical alternative for robust inference under
a broad range of empirically relevant environments. In addition to the continuous-time framework, we evaluate the performance of our methods
in discrete-time predictive regressions and compare them with the IVX approach of Kostakis
et al.Where the paper says it
Where the paper says it
Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.
Improvements for AI systems
-
Robust inference for predictive regressions in finance can be improved by using a hybrid test statistic that combines Cauchy estimation with Ordinary Least Squares residuals for variance estimation, which is described as
a hybrid procedure that combines Cauchy and OLS estimation by using the Cauchy estimator for the coefficient and OLS residuals for variance estimation.
-
The improved system can perform robust inference under challenging data conditions such as endogeneity, near nonstationarity, heavy tails, and persistent volatility in errors. This robustness is achieved by using two novel tests: one based on t-statistic group inference and another employing a hybrid approach that
effectively mitigate size distortions that commonly arise in standard inference procedures under endogeneity, near nonstationarity, heavy tails, and persistent volatility.
-
The AI system can accurately test the predictability of financial variables like stock returns using predictors such as dividend-price ratios, as empirical applications show that
the dividend–price ratio possesses predictive power
while others do not significantly forecast returns. -
The system can handle models with multiple predictors by extending univariate robust inference procedures to a multivariate setting, allowing for testing the joint predictability of covariates using Bonferroni correction on the p-values from univariate Cauchy estimators.
-
The system can perform robust joint hypothesis testing for multiple predictors by using a Wald-type statistic that converges to chi squared under the null hypothesis, which is motivated by
the convergence of (sum over t=1 to T sign(x1,t−1)y t) (sigma hat squared sum over t=1 to T sign(x1,t−1)z't−1) (sum over t=1 to T sign(x1,t−1)y t)
under specific assumptions. -
The system can handle models with an intercept by employing a modified Cauchy estimator derived from first-differencing the regression model, which is described as
the modified Cauchy estimator
that has key properties related to martingale difference sequences when applied to even-indexed data.
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.
Related papers
- Stationarity with Occasionally Binding Constraints
- Backtesting High-Dimensional Expected Shortfall
- Higher-Order Neyman Orthogonality in Moment-Condition Models
- Online Bayesian Inference for a Dirichlet Process Mixture of Binary Logit Models
- Debiased Conditional Independence Tests for Causal Effects
- Sharp Testable Implications of Encouragement Designs