Subject
Probability, inference, estimators, regression, causal methods, and uncertainty.
A preprint reports that a partially collapsed particle-learning sampler mixed well in synthetic models and showed better autocorrelation-based mixing than JAGS in one Norway hospitalization analysis, with the authors warning against broad generalization.
Scientific publication4 min read
In one 10-variable quarterly exercise, non-centred stochastic volatility scored higher on density forecasts than centred stochastic volatility, but point accuracy was nearly unchanged.
Scientific publication4 min read
A theoretical study finds that characteristic polynomials and trace statistics of self-normalized random matrices settle into a family of Poisson-type limits, with a Gaussian form at the boundary.
Scientific publication4 min read
A theoretical analysis finds that exact calibration does not make two-sided p-values unique. Symmetry forces several methods to agree, while numerical examples show large differences in asymmetric models.
Scientific publication4 min read
A new mathematical preprint shows that the Fourier coefficients of the canonical critical Gaussian multiplicative chaos on the circle tend to zero at high frequencies almost surely. The proof uses auxiliary random fields, concentration estimates and a covariance correction, but it does not establish a decay rate for the original coefficients.
Scientific publication4 min read
A new mathematical preprint gives a complete test for atomless distance distributions and describes the shape of spaces that can realize them.
Scientific publication4 min read
A theoretical analysis links a leading cutoff proportion to an exact integer threshold by tracking how expected payoffs change from one candidate to the next.
Scientific publication5 min read
A mathematical preprint reports a common pattern linking limits of weighted averages with the hidden distribution of their random weights.
Scientific publication4 min read
A mathematical filtering framework produced lower uncertainty and lower threshold probabilities in a simulated learning trajectory, but it has not been validated with child data.
Scientific publication4 min read
SCAN is designed to detect multiple distributional shifts in long, serially dependent time series. Its strongest reported results came on longer simulations, while ARFIMA long-memory scenarios produced less favorable and more threshold-sensitive results.
Scientific publication4 min read
A methods study found higher reported sampling efficiency for BPS and Boomerang in RI-CLPM simulations, but OMRF comparisons were mixed and one posterior gap remained unresolved.
Scientific publication5 min read
A new methods preprint describes a model that groups observations, estimates cluster-specific relationships and flags unusual cases while handling missing responses and covariates.
Scientific publication4 min read
A theorem-based analysis finds that a noising-denoising construction can approach a dynamic Schrödinger bridge at rates tied to the reference diffusion's spectrum.
Scientific publication4 min read
A statistical preprint adds spatial information to factor analysis and tests the approach on simulations, one brain-tissue section and African vegetation records.
Scientific publication5 min read
A preprint proposes a Bayesian quantile-regression model that separates regular observations from extreme ones. Its theory and selected tests point to lower sensitivity to outliers, while a calibration exercise found that narrower intervals can lose coverage when the likelihood is misspecified.
Scientific publication5 min read
An arXiv preprint gives a formal designability guarantee for a narrow class of RNA secondary structures, while stressing that the result is confined to a simplified pairing model.
Scientific publication4 min read
A mathematical preprint finds that strong decay of long connections is sufficient for recurrence in specified one- and two-dimensional graph models, but leaves phase boundaries unresolved.
Scientific publication3 min read
A statistical methods preprint reports a one-step log-quantile calculation, low variability and nearly no bias in selected simulations, with model-specific results in Google returns and hurricane-loss data.
Scientific publication5 min read
A preprint reports that fixed OBABO step size and friction cannot guarantee uniform ballistic cold-start mixing in total variation across the full smooth strongly convex class.
Scientific publication4 min read
An arXiv preprint reports that BERGM Elastic Net had the lowest mean squared errors among the methods compared in a 50-replication simulation. It also had the lowest false-positive and false-discovery rates, but a lower true-positive rate at the selected reporting rule.
Scientific publication5 min read
A mathematical preprint reports a positive sampling-noise baseline and suggests that widespread small shifts may be easier to detect than low-rank changes in modeled settings.
Scientific publication5 min read
A methods preprint introduces a multiple-quantile test for structural changes in multivariate volatility. In simulations it stayed near its 5% target under the null, while an application to weekly Fama-French factor returns rejected covariance stability across four periods.
Scientific publication3 min read
A mathematical preprint gives finite-sample Gaussian and bootstrap bounds for U-statistic norms, with dimension-explicit results and a minimax rate for dense Kendall’s tau alternatives.
Scientific publication4 min read
A statistical methods preprint reports a theoretical link between observation frequency and prediction rates in scalar-on-function and function-on-function linear regression. It pairs a pooling-ridge estimator with matching bounds and tests the approach in simulations, wheat spectra and child-growth data.
Scientific publication5 min read
A penalised model selected a sparse cross-firm volatility network and had lower out-of-sample errors than preset networks in a 20-firm UK sample.
Scientific publication3 min read
A mathematical preprint derives upper bounds for gradient-trained, over-parametrized neural-network regression on exponentially β-mixing data and replaces ambient dimension with manifold dimension under an exact support condition.
Scientific publication4 min read
A preprint models Opportunity Zone designation and neighborhood spillovers in California, reporting positive estimated contrasts for designated tracts but uncertain results as policy expands.
Scientific publication4 min read
A new method combines independent frequency simulation with subsampling to estimate dependence that the first stage misses. Its full consistency is theoretical and conditional on stated assumptions, while simulations found close agreement with reported exact values in three synthetic models.
Scientific publication4 min read
A modeling paper combining theory, simulations and regional onset reconstructions says pooled mobility can retain aggregate invasion signals while making transport-mode and route attribution harder.
Scientific publication5 min read
A mathematical analysis of mean-field coupled maps reports finite-time control of empirical measures, propagation of chaos and Gaussian fluctuation limits under stated assumptions.
Scientific publication4 min read
A mathematical study finds distinct limiting behavior for rearranged density and copula estimates, with a separate boundary result for histograms.
Scientific publication3 min read
A theoretical preprint maps uncertainty about discrete outcomes into uncertainty about partially identified model parameters, then tests the approach in simulation.
Scientific publication4 min read
A new econometrics preprint reports that K-averaging can deliver the fastest theoretical convergence, while coefficient differences and weak trends limit the advantage in applications.
Scientific publication4 min read
An arXiv preprint presents a closed-form random-rotation test for autocorrelation. Its strongest reported results came at lag 3 and across selected lags, while moving-average and heavy-tail tests exposed important limits.
Scientific publication4 min read
A proposed deep-neural-network method explicitly models observed covariates, improving recovery of image-related effects in simulations and restoring much of an age-confounded MRI model’s performance.
Scientific publication4 min read
A preprint reports that SPM stayed unbiased in an out-of-phase simulation, with efficiency comparable to a correctly specified spline method.
Scientific publication5 min read
A mathematical analysis of Laplace-noised observations finds that privacy and recovery pull in opposite directions, with a sufficient low-noise consistency condition and an impossibility result at square-root-of-sample-size noise.
Scientific publication4 min read
A mathematical preprint reports that the zeros of permanental characteristic polynomials built from Gaussian random matrices converge, with probability one, to a semicircle distribution rotated onto the imaginary axis. The result addresses a conjecture but remains asymptotic and limited to Gaussian ensembles.
Scientific publication3 min read
The framework aims to turn quantitative estimates into cautious qualitative conclusions, with an explicit inconclusive outcome when the evidence does not separate the available possibilities.
Scientific publication5 min read
A statistical preprint develops corrections for GMM when its moment conditions contain related local specification errors. In a baseline simulation, empirical Bayes had the lowest RMSE among the compared procedures, while an application to schooling data showed that corrected estimates varied less across control specifications than conventional TSLS.
Scientific publication4 min read