Subsample-Based Estimation under Dynamic Contamination
Yukai Yang, Rickard Sandberg · 2026
Read Subsample-Based Estimation under Dynamic Contamination by Yukai Yang, Rickard Sandberg (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-17676.txt.
Introduction
Yang and Sandberg study a failure of subsample-based estimation in dynamic models. Even knowing where contamination sits, removing tainted points does not restore the clean goal. Contamination spreads through the residual filter, distorting the rule, so subsample-based methods are wrong for the clean-data value. The authors call this a clash between point-by-point subsampling and residual spread, arising when contamination spreads through steps feeding the rule. They propose a spread-friendly change of index sets, leaving the method the same in the limit.
Catalog record
| CREATOR | Yukai Yang, Rickard Sandberg |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 118237 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.17676
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Subsample-Based Estimation under Dynamic Contamination Yukai Yang, Rickard Sandberg Abstract Subsample-based estimation is a standard tool for achieving robustness to outliers in econometric models. This paper shows that, in dynamic time series settings, such procedures are fundamentally invalid under contamination, even under oracle knowledge of contamination locations. The key issue is that contamination propagates through the model’s residual filter and distorts the estimation criterion itself. As a result, removing contaminated observations does not, in general, restore the uncontaminated objective or ensure consistency. We characterise this failure as a structural incompatibility between pointwise subsampling and residual propagation. To address it, we propose a propagation-compatible transformation of index sets, formalised through a patch removal operator that removes the residual footprint of contamination. Under suitable conditions, the proposed operator leaves the estimator asymptotically unchanged under the uncontaminated model, while restoring consistency for the clean-data parameter under contamination. The results apply to a broad class of residual-based estimators and show that valid subsample-based estimation in dynamic models requires explicit control of residual propagation. Keywords : Outliers; Residual propagation; Robustness; Consistency; Asymptotic theory. 1 Introduction Outliers are a pervasive feature of economic and financial time series. Their main statistical effect is not merely to create a few anomalous observations, but to obstruct accurate estimation of the underlying clean process. From an econometric perspective, the object of interest is the latent data generating mechanism rather than the contaminated observations themselves. If the clean process does not exhibit a given feature, but the observed data appear to do so because of outliers, inference based on the contaminated series is misleading. The relevant question is therefore not whether such features are detected in the observed data, but whether they can be attributed to the underlying clean process. This, in turn, requires that the clean model be consistently estimated in the presence of contamination. This requirement is standard, but its implications differ sharply between cross-sectional and dynamic settings. In cross-sectional models, contamination is local in the sense that its effect is tied to the contaminated observation, and robust procedures operate by down-weighting or removing individual observations. In dynamic time series models, however, this abstraction breaks down, as contamination propagates through the temporal structure and, in particular, through the residual filter that defines the estimation criterion, inducing non-local distortions in the residual sequence and hence in the estimation criterion. As a consequence, robustness cannot be achieved by selecting subsets of observations alone, and the classical subsampling paradigm becomes structurally incompatible with the estimation problem, even under perfect detection. As a result, standard subsample-based procedures do not, in general, restore consistency in dynamic models, even under oracle knowledge of contamination locations. The failure is structural rather than algorithmic, arising from the interaction between subsampling and the model-implied residual filter, since approaches based on improved detection, residual down-weighting, or ad hoc trimming operate at the level of individual observations or residual magnitudes rather than on the mechanism by which contamination enters the criterion. The relevant object for estimation is therefore not contamination at the level of the observed series, but contamination at the level of the residual-based criterion, where local contamination induces non-local distortions. Consequently, the objective function is not centred at the true parameter, so that consistency cannot be restored by pointwise robustification alone. The economic relevance of this issue is immediate. In linear models, outliers may induce spurious statistical significance, so that parameters that are insignificant under the clean model appear significant in contaminated samples, or vice versa ( 21 ; 37 ; 11 ; 14 ) . In nonlinear settings, outliers may create false evidence of volatility clustering ( 18 ; 9 ) , structural breaks, regime switching and other forms of nonlinearity ( 3 ; 41 ) , even when the clean process exhibits no such features, or distort or mask such features when they are present. Even features such as persistence or unit-root-like behaviour may be induced by contamination rather than by the underlying data generating mechanism. This broader view is consistent with the classical time-series outlier literature, which has long noted that outliers may mimic or obscure level shifts and variance changes ( 6 ; 38 ; 19 ) . Such features reflect properties of the underlying stochastic mechanism and should not be confused with nonstationarity, whereas outliers represent external contamination arising from atypical or transient events ( 11 ) . The econometric objective is therefore to recover the structure of the clean process rather than the distortions induced by contamination. Estimation procedures must remain consistent for the clean model even when the observed data are contaminated. Without this property, hypothesis testing, model interpretation, and specification analysis become unreliable, because the estimated model reflects contamination rather than the underlying data generating mechanism. The existing literature recognises that outliers in time series may have dynamic effects. Early work, beginning with 17 and later surveyed by 30 and 29 , studies additive and innovative outliers and their effects on residuals, diagnostics, and influence functions in dependent data. Classical model-based approaches include 10 and 11 , who analyse outlier effects and develop procedures for parameter estimation and joint estimation of model and outlier effects in ARMA-type settings. More recent contributions, particularly 24 , develop a unified asymptotic theory for outlier detection procedures, formalising their behaviour through the notion of gauge and establishing calibration results for the fraction of observations flagged as outliers. These procedures are closely related to residual-based screening, down-weighting, and trimming methods, including M-type estimators ( 21 ; 43 ) , least trimmed squares (LTS) and related high-breakdown estimators ( 34 ; 42 ) , and forward search methods ( 20 ; 4 ) . While this provides a rigorous basis for detection and tuning, the procedures still operate through identification and removal, down-weighting, or trimming of aberrant observations based on residual diagnostics, a problem long recognised as subtle in time series settings; see, for example, 27 . The asymptotic properties of these associated estimators are well understood under clean-model conditions, including for M-estimation and trimming-based procedures ( 21 ; 12 ; 25 ; 42 ) . However, this literature focuses on detection, robustification, or estimator redesign, rather than on the structural effect of dynamic contamination on the estimation criterion itself. Even when contaminated observations are correctly identified or detection is asymptotically well calibrated, the resulting procedures do not address how contamination enters the residual-based criterion. This includes robust estimation for multivariate dynamic models, such as VARMA settings ( 5 ) , and approaches based on modified residual constructions that limit the effect of individual outliers ( 31 ) . These contributions do not recognise that, as a consequence, subsample-based estimation may fail to remain consistent even under oracle knowledge of contamination locations. By contrast, the present paper identifies this incompatibility and proposes a propagation-compatible correction that can be combined with a broad class of residual-based procedures. Importantly, the analysis does not rely on modelling the contamination mechanism itself, but instead operates through high-level conditions on its magnitude, rate, and propagation through the residual filter. The contribution of this paper is threefold. First, we establish that subsample-based estimation is generically inconsistent in dynamic time series models under contamination, even under oracle knowledge of contamination locations, in the