Daily Summary for 2026-10-27

Introduction to the show: ident: Statistics. The day in statistics: methods, theory and the inference under them.

Tom: It's the twenty-seventh of October, twenty twenty-six, and this is the day's research.

Jane: 32 new papers came out today.

Tom: I'm Tom.

Jane: And I'm Jane, here with Lu, Meng and our in-house large language model, Lalam.

Tom: We'll take the day in one pass, then pull out the papers we're staying with.

Jane: Start us off with Mirror Likelihood for Modeling Manifold Data with Application on Correlation Time Series.

The day's research: Tom: Today's focus is making data-driven models more trustworthy by quantifying their uncertainty.

Jane: Quantifying uncertainty is crucial for decision-making in complex systems, isn't it?

Lu: Researchers are looking at how to give polynomial chaos expansion surrogate models predictive intervals.

Meng: What technique introduced those intervals for AKMPC?

Lalam: That technique combines jackknife-based conformal prediction with regression-based polynomial chaos expansions.

Tom: It achieves these intervals without needing a separate hold-out dataset for calibration, which is efficient.

Jane: The linearity of the polynomial chaos model allows residuals to be calculated analytically, does it?

Tom: This approach provides target coverages when using low-accuracy models trained on small datasets.

Jane: Training data availability directly impacts how narrow and reliable these predictive intervals become.

Lu: That contrasts with other work focusing on sampling efficiency.

Meng: Specifically, the importance-tempered random-walk metropolis algorithm is mentioned there.

Lalam: It is described in Uniform ergodicity and spectral gap of importance-tempered MCMC with heavy-tailed targets.

Tom: That work shows tempering can improve sampler efficiency by reducing sensitivity to initialization.

Tom: The most important thing is selecting variables based on cost and harm instead of just statistical fit.

Jane: That approach uses a net benefit metric that penalizes the expense and burden of measuring a variable for clinical utility.

Lu: This method compares all predictor combinations using cross-validation, extending it via backward elimination for many predictors.

Meng: The R package NBvarsel implements this, focusing on maximizing patient benefit while minimizing unnecessary clinical burden and costs.

Lalam: Another work addresses uncertainty in generative models by combining density estimation with Gaussian mixture models for a new goodness-of-fit test.

Tom: That test signals genuine mismodeling when the standard point-null hypothesis test fails due to estimation error.

Jane: Graph neural networks benefit from adaptive aggregation techniques, introducing a learnable expectile level to adjust neighborhood aggregation.

Lu: This allows exploiting class-dependent scale differences for classification advantage over simple mean aggregation under feature corruption.

Meng: Bayesian hierarchical ordinal models in Western Australian oat trials correctly analyze ordinal growth stage data treated as continuous.

Lalam: That analysis accounted for the split-plot structure and spatial covariance, revealing deeper sowing depth raised tillering probability.

Tom: Agentic distributional regression estimates conditional densities by letting a language model agent write its own estimation program.

Jane: Simulation studies show ADR with Claude Opus 5.5 achieved the best average performance among methods considered on real datasets.

Lu: We also have work on Bias Correction and Inference for Panel Quantile Regression, developing a method to correct bias in estimating quantile regression parameters for panel data.

Meng: Mediated Disclosure introduces a mediator to show how communication can cause specific actions, including the sender's desired outcome.

Lalam: Time-Varying Quantile VAR Models Estimation and Inference study how conditional quantiles change over time in vector autoregression models.

Tom: Characterizing Maximum Nash Welfare Envy-Freeness and Proportionality shows maximum Nash welfare can be characterized by weaker fairness conditions than envy-freeness.

Jane: A Semiparametric Real-Time GARCH Model with Nonstationary Variance proposes a model combining long-run and short-run components for real-time forecasting.

Lu: Identifying and disentangling direct and spillover effects using multiple pre-treatment periods separates the direct effect of a treatment from the indirect spillover effect.

Meng: Personal preference or system role? Battery swapping among UK electric vehicle users analyzes whether users view swapping as a purchase or system technology based on preferences.

Lalam: We test how environmental changes influenced the invention and spread of agriculture over time, using climate change, sea level rise, and the agricultural revolution.

Tom: This implies production remains stuck at the lowest value stage, failing to cover costs when an exposure-capability framework is applied to hosting.

Jane: The support mechanisms weren't as strict as hoped; capability criteria were mere considerations in AKMPC work on polycrisis pressures.

Lu: Thus, hosting the green transition supply chain doesn't equate to capturing its value when using an exposure-capability framework.

Meng: What strategic options remain open for these dependent economies when analyzing hosting not capturing value?

Lalam: We need to examine those remaining strategic possibilities for these nations within the exposure-capability framework.

Tom: We develop a method to correct bias in estimating quantile regression parameters for panel data using Bias Correction and Inference for Panel Quantile Regression.

Jane: We introduce a mediator to show how communication can cause specific actions, including the sender's desired outcome through Mediated Disclosure.

Lu: Study how conditional quantiles change over time in vector autoregression models and develop tools for analyzing impulse responses using Time-Varying Quantile VAR Models Estimation and Inference.

Meng: Show that maximum Nash welfare can be characterized by weaker fairness conditions than envy-freeness when using Characterizing Maximum Nash Welfare Envy-Freeness and Proportionality.

Lalam: Propose a model to estimate volatility combining long-run and short-run components for real-time forecasting with A Semiparametric Real-Time GARCH Model with Nonstationary Variance.

Tom: Show how to separate the direct effect of a treatment from the indirect spillover effect using multiple pre-treatment periods.

Jane: Analyze whether UK EV users view battery swapping as a purchase or system technology based on their preferences using Personal preference or system role? Battery swapping among UK electric vehicle users analyze whether UK EV users view battery swapping as a purchase or a system technology based on their preferences.

Lu: Test how environmental changes influenced the invention and spread of agriculture over time using Climate change, sea level rise, and the agricultural revolution.

Meng: Use an exposure-capability framework to analyze whether a country's investment in a supply chain actually captures value when analyzing Hosting is not capturing: an exposure-capability framework and Hungary's electric-vehicle battery bet.

Lalam: Develop a method to correct bias in estimating quantile regression parameters for panel data using Bias Correction and Inference for Panel Quantile Regression.

Tom: Introduce a mediator to show how communication can cause specific actions, including the sender's desired outcome through Mediated Disclosure.

Lalam: We study how conditional quantiles change over time in vector autoregression models and develop tools for analyzing impulse responses using Time-Varying Quantile VAR Models Estimation and Inference.

Jane: Show that maximum Nash welfare can be characterized by weaker fairness conditions than envy-freeness when using Characterizing Maximum Nash Welfare Envy-Freeness and Proportionality.

Lu: Propose a model to estimate volatility combining long-run and short-run components for real-time forecasting with A Semiparametric Real-Time GARCH Model with Nonstationary Variance.

Meng: Show how to separate the direct effect of a treatment from the indirect spillover effect using multiple pre-treatment periods.

Lalam: Analyze whether UK EV users view battery swapping as a purchase or system technology based on their preferences using Personal preference or system role? Battery swapping among UK electric vehicle users analyze whether UK EV users view battery swapping as a purchase or a system technology based on their preferences.

Tom: Test how environmental changes influenced the invention and spread of agriculture over time using Climate change, sea level rise, and the agricultural revolution.

Jane: Use an exposure-capability framework to analyze whether a country's investment in a supply chain actually captures value when analyzing Hosting is not capturing: an exposure-capability framework and Hungary's electric-vehicle battery bet.

Lu: Develop a method to correct bias in estimating quantile regression parameters for panel data using Bias Correction and Inference for Panel Quantile Regression.

Meng: Introduce a mediator to show how communication can cause specific actions, including the sender's desired outcome through Mediated Disclosure.

Lucky paper: 2610.36448: Tom: Let's discuss Mirror Likelihood for Modeling Manifold Data with Application on Correlation Time Series. This paper tackles how manifold-valued data shows this is the key output there.

Kai: So the concrete result is they found a method to distinguish regimes, which is a functional piece of information about the system's dynamics.

Mira: It gives us an idea that even in highly complex lattice systems, there might be specific observable signatures for different dynamical behaviors.

Lev: I wonder if this concept of regime distinction has any bearing on how we look at state preparation in quantum computation.

Lucky paper: 2610.36488: Tom: So we're looking at Second-order Sinkhorn Geometry and sqrt n-Inference today, which is about giving a formal Riemannian structure to the space of probability measures.

Jane: It starts by saying the Sinkhorn divergence, while computationally tractable, isn't really a metric and doesn't naturally have that formal Riemannian structure that we need for things like interpolation.

Lu: The paper addresses this by constructing a formal Riemannian tangent space and a metric tensor that quadratically approximates the Sinkhorn divergence for a suitable class of finite-smoothness cost functions.

Meng: What does proving Hadamard differentiability of the Sinkhorn metric tensor actually let us do?

Tom: It lets us derive the Levi-Civita covariant derivative in this geometry, which then lets us obtain the geodesic and parallel transport equations as ordinary ordinary differential equations on a fixed Reproducing Kernel Hilbert Space.

Jane: That sounds like a lot of machinery, but the real payoff is linearizing those solutions using standard tools from Banach space ODE theory.

Lu: The paper shows that under standard regularity conditions, these linearizations lead to sqrt n-limiting Gaussian processes centered at population values for estimators like the exponential map and the logarithmic map.

Tom: They also show that this nonparametric bootstrap is consistent for all those canonical geometric operations, which is a big deal for robust estimation.

Jane: That means we get rigorous convergence properties now because they are using these new sqrt n-limit theorems and bootstrap consistency results in a transport-based geometry on the space of probability measures.

Meng: So, if we are dealing with complex data structures where traditional transport methods fail, this framework provides a way to rigorously analyze those estimators with stronger convergence properties.

Lu: The paper is very ambitious because it claims these are the first sqrt n-limit theorems and bootstrap consistency results for canonical geometric operations in a transport-based geometry on the space of probability measures.

Lucky paper: 2610.36519: Tom: We're looking at Effect-Size Estimation and Confounding-Aware Permutations for PERMANOVA today, which is paper number three of five in this rundown.

Jane: So, PERMANOVA is used to test associations between covariates and high-dimensional data with arbitrary distributions, like ecology or microbial community profiling.

Lu: The conventional effect size measure R2 is usually used to indicate how much the high-dimensional measurement covaries with respect to a separate, univariate measure.

Meng: But as this paper shows, that conventional effect size is inflated under the null, meaning it doesn't accurately reflect the variance explained.

Lalam: Standard permutation schemes also yield inflated false positives when you have non-independent samples like repeated measure designs.

Tom: That brings us to their main contribution: they introduce confounding-aware permutation schemes for both invariant and variant covariates in the paper, Effect-Size Estimation and Confounding-Aware Permutations for PERMANOVA.

Jane: They also develop a causally-grounded, distance-based effect size estimator derived from the Hájek average treatment effect estimator.

Lu: That new estimator provides interpretable effect sizes in the units of the chosen dissimilarity metric, which is pretty useful when you're working with complex structures.

Meng: The simulation results using microbiome profiles demonstrate that these new approaches control false positives while maintaining power, outperforming standard PERMANOVA implementations.

Tom: So they are showing that controlling false positives doesn't require sacrificing statistical power in these high-dimensional settings.

Jane: That moves the discussion beyond just adjusting the permutation process to also developing a better effect size measure itself for PERMANOVA.

Lu: The method uses an estimator derived from the Hájek average treatment effect estimator, which is based on distance-based measures instead of relying solely on correlation structure.

Meng: How does this distance-based approach specifically help with non-independent samples that standard methods struggle with?

Lalam: The paper introduces a causally grounded effect size estimator derived from the Hájek average treatment effect estimator, which addresses the issue of non-independence by incorporating a distance metric.

Tom: So we're talking about moving away from just looking at covariance structures and incorporating causal inference ideas into the effect size calculation for PERMANOVA.

Jane: That sounds like a significant shift in how we interpret the results when dealing with complex, high-dimensional data sets.

Lucky paper: 2610.36690: Tom: We're looking at Coupled unbiased MCMC for estimating Shapley effects under dependent inputs today, which is a big hurdle in global sensitivity analysis.

Jane: It sounds like the paper tackles the fundamental problem of allocating output variance when input variables aren't independent.

Lu: The authors propose a permutation-based framework using coupled unbiased Markov chain Monte Carlo to estimate these Shapley effects.

Tom: They combine two coupling mechanisms: a joint coupling along a random permutation to reduce variance, and a meeting-based coupling to remove finite-time MCMC bias through a telescoping correction.

Meng: That sounds computationally intensive; what are the specific conditions under which they prove the resulting estimators are unbiased and have finite variance?

Jane: The paper states that under suitable conditions, the paper focuses on d at least eleven.

Kai: The method relies on reducing classification to counting roots of an explicit polynomial for dimensions eleven and above.

Mira: So the core technical contribution is proving this additional root exists in every odd dimension d greater than or equal to eleven.

Lev: It's a significant step because it moves the classification from a complex geometric problem to a more tractable counting exercise for those high dimensions.

Lucky paper: 2610.37071: Tom: Okay, we're moving into our final segment of this round, and we have the paper on the Analysis of Uncertain GARCH Model with the application of crude oil future prices.

Jane: It looks like a very focused piece, dealing with how time series data exhibits variation and persistence in uncertain environments.

Lu: The core idea is that existing studies usually focus only on conditional mean structures and ignore variance and persistent dynamics.

Meng: So this paper is aiming to address that gap by developing an uncertain generalized autoregressive conditional heteroskedasticity or UGARCH model.

Lalam: It incorporates lagged conditional variance terms into the UGARCH framework.

Tom: They first prove sufficient conditions to ensure mean stationarity of the variance process before they propose recursive constrained least squares for parameter estimation.

Jane: And they use cross-validation to determine the orders of both the mean and variance equations, which is a key step for model specification.

Lu: They also employ uncertain hypothesis tests to determine if a time series exhibits a UGARCH effect and to assess the goodness of fit.

Meng: The numerical simulations demonstrated that this UGARCH achieves smaller conditional-variance fitting errors compared to the standard UARCH under the considered settings.

Tom: They apply this method empirically to West Texas Intermediate crude oil futures closing prices, which are known for exhibiting time-varying volatility.

Jane: That means they support short-term point and interval forecasts using the recursively obtained next-period conditional variance and the expected value of the mean equation.

Lu: The paper also discusses how they construct these point and interval forecasts from that next-period conditional variance, which is quite a specific detail.

Meng: The authors show that UGARCH achieves smaller fitting errors than UARCH under the settings they defined for those crude oil futures prices.

Tom: It's interesting to see how they use recursive constrained least squares when parameter estimation for these models is involved.

Jane: Does this work apply to other types of time series data besides commodity prices?

Lu: The paper focuses specifically on the empirical analysis applied to West Texas Intermediate crude oil futures closing prices.

Meng: What are the specific numerical results they show regarding those fitting errors when comparing UGARCH and UARCH?

Tom: They show that the UGARCH achieves smaller conditional-variance fitting errors than UARCH under the considered settings.

Jane: So, when we look at real-world financial data with volatility clustering, this model offers a more accurate estimation of the variance component.

Lu: The authors proved sufficient conditions for mean stationarity of the variance process before proposing their recursive constrained least squares method.

Meng: That means they have to establish those mathematical requirements upfront before they can even start estimating parameters.

Tom: They also use uncertain hypothesis tests to determine if a time series exhibits a UGARCH effect and assess its goodness of fit.

Jane: That helps distinguish between models that are just mean-focused versus those that properly account for the persistent dynamics in the variance.

Lu: The recursive constrained least squares is used to estimate the model parameters once those conditions are met, which is a specific optimization technique.

Meng: The paper uses cross validation to determine the orders of the mean and variance equations, which gives them a way to select those orders automatically.

Tom: It's a method for automated model order selection within this framework.

Jane: So, when we talk about forecasting, what kind of intervals are constructed from that next-period conditional variance?

Lu: Point and interval forecasts are constructed from the recursively obtained next-period conditional variance and the expected value of the mean equation.

Meng: That combination gives them a forecast based on both the mean and how volatile things are expected to be in the very near future.

Tom: The performance metrics they report are based on these forecasts, showing that they support short-term point and interval forecasts.

Jane: It’s a practical application for anyone dealing with financial data where uncertainty about future volatility matters for risk assessment.