Interpolation of functionals of stochastic sequences with stationary increments from observations with noise

Basic Information

• Paper ID: 2510.22764
• Title: Interpolation of functionals of stochastic sequences with stationary increments from observations with noise
• Authors: Maksym Luz, Mykhailo Moklyachuk (Taras Shevchenko National University of Kyiv)
• Classification: math.ST stat.TH
• Published Journal: Applied Statistics. Actuarial and Financial Mathematics. No. 2, 131–148, 2012
• Paper Link: https://arxiv.org/abs/2510.22764

Abstract

This paper investigates the optimal estimation problem for linear functionals $A_N\xi = \sum_{k=0}^{N} a(k)\xi(k)$ that depend on unknown values of stochastic sequences $\xi(m)$ with stationary n-th order increments. The observational data consist of sequence values $\xi(k)$ at points $k = -1, -2, \ldots$ and noisy sequence values $\xi(k) + \eta(k)$ at points $k = N+1, N+2, \ldots$. Under spectral certainty conditions (where the spectral densities of sequences $\xi(m)$ and $\eta(m)$ are precisely known), formulas are proposed for computing the mean square error and spectral characteristics of optimal linear estimates. When spectral densities are imprecisely known and only a set of admissible spectral densities is given, the minimax (robust) estimation method is applied. For specific admissible density sets, formulas are provided for determining the least favorable spectral densities and minimax spectral characteristics.

Research Background and Motivation

1. Research Problem: The core problem addressed is optimal interpolation estimation of linear functionals of stochastic sequences with stationary increments under noisy observations.
2. Problem Significance:
   • Stochastic sequences with stationary increments represent an important generalization of generalized stationary stochastic sequences, including ARIMA sequences and periodic (seasonal) time series
   • In practical applications, observational data typically contain noise, necessitating estimation under incomplete information conditions
   • This problem has important applications in finance, signal processing, statistical forecasting, and related fields
3. Limitations of Existing Methods:
   • Classical Kolmogorov-Wiener methods are primarily applicable when spectral densities are known
   • Traditional methods lack robustness when spectral densities are unknown or uncertain
   • Systematic theoretical frameworks for handling noisy observations are lacking
4. Research Motivation: Establishing robust estimation theory under uncertain spectral density conditions to provide more reliable estimation methods for practical applications.

Core Contributions

1. Theoretical Framework: Establishes a complete theoretical framework for optimal linear interpolation of stochastic sequences with stationary n-th order increments under noisy observations
2. Estimation Formulas: Under spectral certainty conditions, provides explicit formulas for computing spectral characteristics and mean square errors of optimal linear estimates
3. Minimax Method: Extends the minimax (robust) estimation method to interpolation problems with noisy observations
4. Least Favorable Spectral Densities: Provides methods for determining least favorable spectral densities and minimax spectral characteristics for specific admissible spectral density classes
5. Filtering Relationships: Reveals intrinsic connections between interpolation and filtering problems

Methodology Details

Task Definition

Input:

• Observed sequence values $\xi(k)$ at points $k = -1, -2, \ldots$
• Observed noisy sequence values $\xi(k) + \eta(k)$ at points $k = N+1, N+2, \ldots$

Output: Optimal estimate of the linear functional $A_N\xi = \sum_{k=0}^{N} a(k)\xi(k)$

Constraints:

• $\xi(m)$ possesses stationary n-th order increments
• $\eta(k)$ is uncorrelated with $\xi(k)$
• The interval $\{0, 1, \ldots, N\}$ is the region of unknown values

Model Architecture

1. Spectral Representation of Stationary Increments

Definition: The n-th order random increment is defined as: $\xi^{(n)}(m,\mu) = (1-B_\mu)^n\xi(m) = \sum_{l=0}^{n} (-1)^l C_n^l \xi(m-l\mu)$

where $B_\mu$ is the shift operator with step size $\mu$.

Spectral Decomposition: Stationary n-th order increments possess the spectral representation: $\xi^{(n)}(m,\mu) = \int_{-\pi}^{\pi} e^{im\lambda} (1-e^{-i\mu\lambda})^n \frac{1}{(i\lambda)^n} dZ(\lambda)$

where $Z(\lambda)$ is an orthogonal random measure.

2. Functional Decomposition

The target functional is decomposed as: $A_N\xi = B_N\xi - V_N\xi$

where:

• $B_N\xi = \sum_{k=0}^{N} b_\mu(k)\xi^{(n)}(k,\mu)$ is the increment functional
• $V_N\xi = \sum_{k=-1}^{-\mu n} v_\mu(k)\xi(k)$ is the known observation functional

3. Optimal Estimate Construction

Using orthogonal projection methods in Hilbert space, $B_N\xi$ is projected onto the observation space: $\hat{B}_N\xi = \int_{-\pi}^{\pi} h_\mu^{(1)}(\lambda) dZ_{\xi_\mu^{(n)}}(\lambda) + \int_{-\pi}^{\pi} h_\mu^{(2)}(\lambda) dZ_{\xi_\mu^{(n)} + \eta_\mu^{(n)}}(\lambda)$

Technical Innovations

1. Spectral Characteristic Determination: Optimal spectral characteristics are determined by solving linear equation systems:
   • $(F_\mu)^{-1}$ and $(G_\mu)^{-1}$ matrices are constructed from Fourier coefficients of spectral densities
   • Coefficients satisfy orthogonality conditions ensuring optimality of estimates
2. Matrix Equation Method: Transforms continuous spectral problems into discrete matrix equations: $G_\mu^c c_\mu = G_\mu^e e_\mu$$b_{\mu,N} + G_\mu^e c_\mu = G_\mu^c e_\mu + F_\mu^c e_\mu$
3. Minimax Framework: When spectral densities are uncertain, seeks saddle points $(h^0, f^0, g^0)$ satisfying: $\Delta(h; f^0, g^0) \geq \Delta(h^0; f^0, g^0) \geq \Delta(h^0; f, g)$

Experimental Setup

Theoretical Verification Framework

This paper is primarily theoretical research, verifying method correctness through mathematical derivations, including:

1. Condition Verification: Verifies that spectral densities satisfy integrability conditions (15) and (16)
2. Matrix Invertibility: Ensures invertibility of relevant matrices
3. Convergence Analysis: Analyzes convergence of Fourier series

Special Case Analysis

1. White Noise Case: When $\eta^{(n)}(k,\mu)$ is white noise, matrix $G_\mu^e = \sigma^2 I$
2. Finite Support Case: Simplified formulas when coefficient sequences have finite support
3. Specific Spectral Density Classes: Analyzes properties of classes $D_{0,\mu}^-$ and $D_{M,\mu}^-$

Experimental Results

Main Theoretical Results

Theorem 3.1 (Existence of Optimal Estimates)

Under conditions (15) and (16), the optimal linear estimate $\hat{B}_N\xi$ exists, with spectral characteristics given by formulas (17) and (18), and mean square error computed by formula (19).

Theorem 3.2 (Functional Estimation)

The optimal estimate of the target functional $A_N\xi$ is: $\hat{A}_N\xi = -\sum_{k=-1}^{-\mu n} v_\mu(k)\xi(k) + \int_{-\pi}^{\pi} h_\mu^{(a,1)}(\lambda) dZ_{\xi_\mu^{(n)}}(\lambda) + \int_{-\pi}^{\pi} h_\mu^{(a,2)}(\lambda) dZ_{\xi_\mu^{(n)} + \eta_\mu^{(n)}}(\lambda)$

Theorem 6.1 (Least Favorable Spectral Density)

In the class $D_{0,\mu}^- \times D_{0,\mu}^-$, the least favorable spectral density has the form: $f^0(\lambda) = \frac{\lambda^{2n}}{|1-e^{i\lambda\mu}|^{2n} \sum_{k=-\infty}^{\infty} f_\mu^0(|k|)e^{i\lambda k}}$

Special Case Results

1. Under White Noise Conditions: When noise is white and coefficient sequences are positive, the least favorable density simplifies to: $f^0(\lambda) = \left|\frac{(i\lambda)^n}{(1-e^{-i\lambda\mu})^n} \sum_{k=0}^{N} \gamma_\mu(k)e^{-i\lambda k}\right|^2$
2. Finite Constraint Classes: In class $D_{M,\mu}^-$, the least favorable density is obtained by solving finite-dimensional linear equation systems.

Related Work

Historical Development

1. Classical Theory: Kolmogorov (1992), Wiener (1966) classical prediction theory
2. Stationary Increment Theory: Yaglom (1955), Pinsker (1955) pioneering work
3. Minimax Methods: Grenander (1957) first proposed minimax extrapolation methods

Modern Development

1. Convex Optimization Methods: Franke (1985), Kassam & Poor (1985) robust methods
2. Vector Processes: Rozanov (1967), Masyutka & Moklyachuk (2012) vector generalizations
3. Periodically Correlated: Dubovets'ka & Moklyachuk (2013) research on periodically correlated sequences

Innovations in This Paper

Compared to existing work, the main innovations are:

• Systematic treatment of interpolation problems with noisy observations
• Establishing connections between interpolation and filtering problems
• Providing explicit solutions for multiple spectral density classes

Conclusions and Discussion

Main Conclusions

1. Complete Theoretical Framework: Establishes a comprehensive interpolation theory for stochastic sequences with stationary increments under noisy observations
2. Explicit Formulas: Provides explicit computational formulas for spectral characteristics and mean square errors of optimal estimates
3. Robustness: The minimax method provides robust estimation strategies when spectral densities are uncertain
4. Practicality: Theoretical results can be directly applied to ARIMA models, seasonal time series, and other practical problems

Limitations

1. Computational Complexity: Involves inversion of infinite-dimensional matrices; practical computation requires truncation approximations
2. Condition Restrictions: Requires spectral densities to satisfy specific integrability conditions
3. Linear Limitation: Only considers linear estimates; nonlinear estimates may perform better
4. Noise Assumptions: Assumes noise is uncorrelated with signals; actual scenarios may involve correlation

Future Directions

1. Computational Algorithms: Develop efficient numerical computation algorithms
2. Nonlinear Extensions: Extend to estimation of nonlinear functionals
3. Correlated Noise: Consider noise correlated with signals
4. Multidimensional Extensions: Extend to multidimensional random fields

In-Depth Evaluation

Strengths

1. Theoretical Rigor: Mathematical derivations are rigorous and theorem proofs are complete
2. Methodological Innovation: Creatively applies minimax methods to noisy interpolation problems
3. Complete Results: Provides a comprehensive theoretical system from existence to construction
4. Practical Value: Theoretical results provide guidance for practical applications

Weaknesses

1. Lack of Numerical Experiments: No numerical examples provided to verify theoretical results
2. Computational Complexity: Insufficient analysis of computational complexity for practical implementation
3. Convergence Analysis: Limited analysis of convergence for truncation approximations
4. Comparative Studies: Lacks performance comparisons with other methods

Impact

1. Academic Contribution: Provides new theoretical tools for stochastic process estimation theory
2. Application Potential: Has promising applications in financial time series and signal processing
3. Theoretical Foundation: Provides theoretical foundation for subsequent related research
4. Methodological Significance: The application of minimax methods has methodological importance

Applicable Scenarios

1. Financial Modeling: Parameter estimation and forecasting for ARIMA models
2. Signal Processing: Interpolation and filtering of noisy signals
3. Economic Forecasting: Analysis of economic time series with trends and seasonality
4. Engineering Applications: State estimation in control systems

References

This paper cites important literature in the field, including:

• Kolmogorov (1992): Foundational work in stochastic process theory
• Yaglom (1955, 1987): Classical theory of stationary increment processes
• Grenander (1957): Pioneering work on minimax methods
• Moklyachuk series: Systematic development of robust estimation theory

Overall Evaluation: This is an excellent theoretically strong paper that makes important contributions to stochastic process estimation theory. Although lacking numerical verification, it features a complete theoretical framework and methodological innovation, significantly advancing development in this field.