WorksMathematics
Phase transitions and approximations of mean squared error for state-space models with fractional differencing
Abstract
We study trend estimation in state-space models in which the trend has a fractional stochastic difference of order and the observation errors form a short-range-dependent stationary process. Using finite-sequence fractional summation and differencing operators, we analyze the penalized least-squares estimator obtained by shrinking the fractional differences of the trend. We derive asymptotic mean squared error (MSE) approximations for all and identify a sharp phase transition at . When , the estimator is consistent and its optimally balanced MSE has order . At the boundary , we obtain a refined finite-sample approximation and show that the MSE decreases at the slower order . When , the MSE converges to an explicit positive limit, so consistent recovery of the trend is impossible under the considered scaling. We also describe a practical criterion for choosing the penalty parameter and differencing order, and numerical experiments illustrate the MSE approximations and the behavior of the selection procedure.
This paper is dedicated to the memory of our colleague, Professor Robert H. Shumway (1936–2025), whose vision, mentorship, and lifelong contributions continue to inspire us.
Introduction
Estimating the trend of a time series is a fundamental task in many applications like economic and financial forecasting [22, 30], climate modeling [24], biomedical monitoring [27], and public health surveillance [13]. In the time series analysis literature [3, 9, 12, 28], the following state-space model for the observations , , has received significant attention due to its simplicity and practical interpretability
where denotes the trend component, and is assumed to be a zero-mean stationary process with variance which is independent of .
Modeling the trend component in equation (1) can be broadly classified into two categories: deterministic and stochastic . Deterministic trends can be modeled either parametrically, when appropriate, or nonparametrically using techniques such as kernel smoothing, local polynomial regression, and sieve expansions. For a detailed review from this perspective, we refer readers to the monograph [14]. We point out that that, in the nonparametric modeling, it is typically assumed that the trend can be expressed as , where is a smooth function defined on . It is well known that if is times differentiable and its th derivative satisfies a Lipschitz condition of order , then the optimal mean squared error (MSE) of a nonparametric estimator of is of order , where [29].
Stochastic trend models are widely used by engineers, statisticians, and economists [2, 12, 22, 28] due to their modeling flexibility and ability to capture long-memory behavior in time series data. In particular, is modeled using fractional differencing, meaning that the th difference of the trend consists of mean-zero i.i.d. (or stationary) random variables, where is a real number representing the order of differencing [5, 19, 21, 25, 26]. To be more specific, one assumes that for some
where is the difference operator and are some mean zero i.i.d. random variables with variance . The model (2) and its variants have been widely used in the literature to model stochastic trends. For example, see [5, 7, 8, 12, 15, 18, 20, 21].
We point out that the system defined by (1) and (2) constitutes a state space model. In the literature, a common approach to obtaining a solution is to minimize the following objective:
where is an integer depending on . In particular, when is an integer, one typically chooses , as in [5]. A key technical advantage of minimizing (3)—as will be demonstrated in Section 2—is that it admits a closed-form solution, which is closely related to the Kalman filter [5, 19]. This connection allows us to analyze its asymptotic properties.
In this paper, we aim to investigate the theoretical performance of the solution obtained by minimizing (3) for the model (1) under the setup in (2) for all . Specifically, we study how the asymptotic behavior and phase transitions of the mean squared error (MSE) of the estimates depend on the parameters and , where is closely related to the signal-to-noise ratio; see Section 2 for a more precise formulation.
We now provide an overview of our main results. We show that, under mild conditions on , the asymptotic behavior of the mean squared error (MSE) exhibits a phase transition that can be fully characterized by for all . First, as shown in part (1) of Theorem 3.1, when , we obtain a consistent estimator whose MSE can be of order . Second, as stated in part (2) of Theorem 3.1, when , the estimator from (3) remains consistent, but with a much slower convergence rate of order . Finally, when , (3) fails to yield a consistent estimator under the scaling considered here. In this case, we explicitly compute the nontrivial limiting value of the MSE, as presented in part (3) of Theorem 3.1. In addition, when consistent estimators can be obtained by minimizing (3) (i.e., ), we also propose a data-driven approach to select the practically unknown parameters and , as detailed in Section 4. This method demonstrates superior performance in our numerical studies, as shown in Section 5.
We now discuss several related works in the literature. In [5], the authors studied the problem when is an integer. They showed that for and divergent , the optimization problem in (3) yields a consistent estimator, and the mean squared error (MSE) can converge at a rate of . However, several challenges arise when extending their results to general . For example, one must appropriately choose the parameter in (3) and define fractional differences for general on finite sequences. Additionally, when , the estimator becomes inconsistent, and the analysis in [5] does not directly apply. Computing the non-vanishing asymptotic MSE in this case requires new tools. On the other hand, in [7, 8, 19, 20, 25], the authors consider a closely related fractionally integrated process instead of (2), and establish consistency of the parameter estimates under various assumptions on . However, the asymptotic behavior of the mean squared error (MSE) and its phase transitions have not been investigated within their framework.
Before concluding this section, we offer two remarks on possible future generalizations of our results and techniques. First, in the current paper, we study the performance of the estimators by minimizing (3), which is closely related to the likelihood function of (1) under the i.i.d. Gaussian assumption for [8]. An alternative to (3), especially when is not i.i.d., is to consider a weighted least squares estimator that incorporates the covariance structure of . Analogous results on the phase transitions of the asymptotic MSE have been established in [5] to some extent for . We believe our techniques can be extended to general ; see Remark 3.3 for further discussion. Second, although we focus on the stationary setup, we believe our results can be extended to cases where in (1) or in (2) is locally stationary, as considered in [10, 11]. We plan to pursue this direction in future work.
The remainder of the paper is organized as follows. Section 2 introduces the basic setup and presents the explicit solution to the optimization problem (3). Section 3 states our main results. In Section 4, we discuss strategies for selecting the practically unknown parameters and when consistent estimation is feasible. Numerical simulations are presented in Section 5, and technical proofs are provided in Appendices A and B.
Conventions. For two sequences of deterministic positive values and , we write if for some positive constant . We write if and . Moreover, we write if for some positive sequence . We use to denote the set of natural numbers (including zero), and to denote the set of positive natural numbers.
Some background, the assumptions and basic setup
In this section, we provide the basic setup, assumptions and some background on (2) and the closed form solution of (3). We start with introducing an important operator from [4]. For and any sequence , we denote the operator that for all
where is the generalized binomial coefficient. The operator defined in (4) possesses several important properties, as established in [4], Lemma 1, for example, for any real numbers ,
which implies that for
We point out that when , coincides with the difference operator [16]. Moreover, in matrix form, can be rewritten as
where is a lower triangular matrix with entries defined by for and .
Denote be the matrix obtained by deleting the first rows of . With the above definitions and the conventions and , we can rewrite equation (3) as follows
Minimizing the above quantity with respect to , we obtain the estimate
where is the identity matrix. For , we set , where denotes the ceiling of . Therefore, the main objective of this paper is to study .
Before introducing our assumptions, we first provide some discussion on the model (1). Following the discussion in [5], for , , satisfying (2), one can solve and express it in an alternative form as follows
where is the vector with all entries equal to one, and are constants determined by the initial conditions of the fractional-difference equation. Based on the above decomposition, we can proceed to discuss the overall signal-to-noise ratio (SNR) for the model (1), denoted as
where and . Due to stationarity, we have that . Moreover, by a discussion similar to equation (4) of [5], one can readily obtain that
where is defined using the , , in (9).
To avoid repetition, we now summarize the main assumptions of this paper as follows.
Assumption 2.1 Throughout the paper, we assume the following holds for the model (1).
We assume that is a zero-mean stationary process with . Furthermore, we assume that for some constant , its autocovariance function satisfies as .
We assume that satisfies (2) with some , where are i.i.d. mean-zero random variables, independent of , with variance , and for some constant , we have
*For , the right-hand side of (12) is defined as the limit as . Furthermore, we assume that in (11), . First, the assumptions imposed on ensure that the noise component forms a short-range dependent stationary process—a mild and commonly adopted condition in the literature [5, 8, 9, 12, 18, 19].
Second, (12) is simply a reformulation of the second term in (11) using the Box-Cox transform. It ensures that , which, without loss of generality, facilitates a nontrivial discussion of the signal-to-noise ratio (SNR). In fact, we emphasize that the assumptions and scalings imposed here are mild and primarily serve to ensure that the signal-to-noise ratio (SNR) defined in (10) lies in a nontrivial regime, i.e., . On the one hand, if SNR approaches zero—indicating that the signal from the trend is completely dominated by noise—then reliable estimation of the trend from becomes unlikely. This scenario arises when . In particular, when the latter term is bounded, the assumption rules out this degenerate case. On the other hand, if SNR tends to infinity, becomes nearly identical to the underlying trend, making it sufficient to use directly as an estimate. This occurs when , i.e., either approaches zero or or diverges. Our assumptions are designed to exclude such extreme cases.
Finally, we note that the i.i.d. assumption on is not essential. It can be readily relaxed to allow to follow a stationary process. We will pursue this direction in future work.
Asymptotic properties of the mean squared error
Main results
We study the MSE of the estimator in (8). The results follow from the bias-variance decomposition in (26) and Proposition (16). We first prepare some notations.
Let denote the spectral density of that
For , we define
where denotes the beta function. For , also define
For , set
and let
Moreover, when , we define the refined approximation
The main result of the paper is summarized as follows.
Theorem 3.1 Suppose Assumption (4) holds and . Recall in (3). Then the following statements hold as .
(1). If , , and , then
(2). If , and , then for defined in (18)
If, in addition, when , then
where is defined in (12).
(3). If and , then
Theorem (13) establishes the asymptotic results for the MSE of the estimators. Based on these results, we identify several phase transitions. In particular, minimizing (3) yields consistent estimators when , and inconsistent ones when . We elaborate on these results as follows.
First, when , the two leading terms in (19) are balanced by choosing
This choice yields an MSE of order . The rate matches the usual nonparametric rate for a deterministic trend of smoothness [29]. Thus, in this asymptotic sense, the random trend in (2) has effective smoothness . Second, when , under Assumption (4), we find that , which satisfies the additional condition in part (2) of Theorem (13). Consequently, the right-hand side of (21) also vanishes, but at a slower rate of compared to the case when . In addition, when , under Assumption (4), we find that , so the right-hand side of (22) does not vanish, resulting in an inconsistent estimator.
Before concluding this section, we provide two remarks as follows.
Remark 3.2 We highlight a connection with the perspective in [31], which assumes a differential operator of order
where is a standard Brownian motion and each is normally distributed. Wahba [31] showed that when is a positive integer, one can obtain an efficient estimator by minimizing the following objective function
It is well known in [31] that minimizing the above objective function yields the smoothing spline estimator when .
The connection between penalizing finite-order differences in (3) and finite-order differentials in (24) was first identified in [5] for . We note that our techniques developed in this paper can potentially be applied to analyze (23) and extend the results of [31] by studying the MSE properties for without requiring , provided the stochastic integral is well defined in the Itô sense. However, the differential formulation becomes invalid in the Itô framework when , in which case the problem must be addressed using tools from fractional or Malliavin calculus. In contrast, the difference operator remains well defined even at , although it yields a much slower convergence rate. Since this is beyond the scope of the current paper, we will explore this direction in future work.
Remark 3.3 We note that, in addition to minimizing (3)—which leads to the estimator in (8)—an alternative approach is to consider the weighted least squares method, as discussed in Section 2.2 of [5]. Specifically, let . Instead of minimizing (7), one may consider minimizing the following objective
This yields the following estimator
One can follow and slightly generalize the techniques developed in the proof of Theorem to establish the phase transitions and MSE approximations for . In fact, under the assumption that follows an AR process, the consistency and convergence rate of the MSE for have been established for in Theorem 4 of [5]. We believe that analogues of Theorem can also be derived for for all , with distinct transitions occurring at , , and . We plan to pursue this direction in future work.
Finally, in practice, is unknown and must be estimated using an estimator with a sufficiently fast convergence rate. This can be effectively achieved when follows an ARMA model; see the discussion in Section 2.2 of [5] for more details.
Bias-variance argument: proof of Theorem 3.1
In this section, we present the proof of Theorem along with several key technical components. Our analysis is based on the bias-variance decomposition. Let . We begin with the following decomposition
The following proposition gives the bias and variance approximations. Its proof will be given in Appendix fig:A.
Proposition 3.4 Under the respective asymptotic regimes in Theorem 3.1, the bias and variance terms in (26) satisfy
(1). For the bias part , we have that
(2). For the variance part , we have that
Proof of Theorem 3.1. We apply Proposition 3.4 separately in the three regimes. Suppose first that . Proposition 3.4 gives
Since both leading terms are nonnegative, we can therefore obtain that
which proves (19).
Next, let . Proposition 3.4 yields
Adding these two expressions proves (20). To show (21), by Lemma A.3 and Theorem A.1 of the appendix, we have that
Moreover,
under the additional condition . The remainder in (20) is also negligible relative to , since . This establishes (21).
Finally, when , Proposition [?] gives
and
Since , both leading terms are finite and nonnegative. This proves (22). ■
We now provide a few remarks below.
Remark 3.5 When , the nontrivial regime under Assumption 2.1 is . In this regime, both terms in (22) remain non-negligible, so the MSE does not vanish. Thus minimizing (3) cannot yield a consistent trend estimator in this range of .
Remark 3.6 In Proposition [?] and Theorem [31], we analyze the standard MSE, defined as the Euclidean distance between the true random trend and its estimate . It is also of interest to consider the weighted MSE, i.e., the quadratic form [6] , where is as in (25). If the spectral density of is bounded away from zero and infinity — which holds, for example, for the usual ARMA model — then the theory of Toeplitz matrices [17] implies that all eigenvalues of are bounded between two positive constants. In this case, there exist constants and such that
Consequently, all asymptotic convergence rate results established for the standard MSE also hold when the distance is measured by the weighted MSE. We will pursue this direction in the future work.
Practical selection of the parameters
In this section, we develop a practical, data-driven rule for selecting the penalty parameter and the differencing order when . For , the signal component is asymptotically negligible relative to the observation noise, so consistent estimation of is not feasible in this setting. The procedure is motivated by the Mallows-type criterion in Section 3 of [5].
Recall the estimator in (8). For each , let
where . Let be a finite candidate grid. In implementation, the candidate orders are chosen on a user-specified grid in , and, for each candidate , the candidate penalty parameters are chosen on a logarithmic grid over a range determined by the sample size .
Let be an estimator of the spectral density of , obtained, for example, from a preliminary trend estimate and a fitted short-range-dependent error model. Such a residual spectral-density estimator is standard; see, for example, [5], Section 3 and [3]. Following [5], our primary criterion is
where we recall (16) and (17). We then select
In our implementation, is obtained by fitting autoregressive models to preliminary residuals and selecting the order by BIC.
For large , the weights in the correction term concentrate near frequency zero. Thus, when is sufficiently smooth near zero, the full-spectrum criterion in (27) has the simpler approximation
The full-spectrum criterion is preferred in implementation; (29) is useful as a computational simplification and for asymptotic interpretation.
Before concluding this section, we summarize the implementation of the selection procedure. Let . Here, is the half-width of the local fitting window, so that the interior window contains observations.
For each , define
Fit a local linear regression to the observations by minimizing , over and , and let be the resulting intercept estimate . This automatically provides the appropriate boundary modification.
Form the preliminary residuals
Fit models to over a prespecified collection of orders. Select the order using BIC, and let denote the spectral density associated with the fitted model.
Evaluate the criterion in (27) at every , and let
Finally, report in (8).
Numerical studies
In this section, we conduct Monte Carlo experiments using the exact finite-sequence fractional operators in (4) and the estimator in (8). Throughout, we set
The innovation sequences in (2) and in (1) are Gaussian sequences, and is specified according to (12). The penalty parameter is set to its model value , as in (3).
In what follows, we consider different combinations of and , where and , and generate 500 independent data sets for each combination. In each replication, we construct and compute from (8). The replication-specific per-coordinate squared error is
The empirical MSE reported below is the average of these 500 values.
First, we compare the empirical MSE with the theoretical approximation derived in Theorem 3.1. In this experiment, the differencing order, penalty parameter, and error spectrum are known. Thus, the comparison directly assesses the finite-sample accuracy of the asymptotic risk approximation. The results are reported in Figure 5.1.
Figure 5.1. Empirical MSE (points) and theoretical approximation (dashes).
The theoretical approximations agree closely with the empirical MSEs, and their accuracy generally improves as increases, supporting the asymptotic results in Theorem 3.1. Moreover, the figure illustrates the phase transition predicted by the theory. When , the risk decreases polynomially with . At the boundary , the decrease is substantially slower. When , the risk approaches a positive limiting value. Across the displayed sample sizes, the theoretical approximations capture these distinct finite-sample behaviors.
Second, we examine the practical selection procedure described in Section 4. We consider data-generating trend orders under three error specifications: white noise; an AR(1) process satisfying ; and an ARMA(1, 1) process satisfying , where is a Gaussian white noise sequence. As outlined at the end of Section 4, for each simulated data set, we first obtain a preliminary local-linear trend estimate. We then fit autoregressive models of orders to the resulting residuals, select the order using BIC, and use the spectral density of the selected autoregressive model as in (27). The candidate penalty values are when , and when . For each replication, let denote the resulting fitted pair. We compute the empirical MSE of in (8) and compare it with the theoretical risk approximation evaluated at the same fitted pair. The latter uses (18) when and the right-hand side of (19) when . The known simulation spectrum is used only to evaluate this theoretical benchmark; the parameter-selection procedure itself uses the residual-based BIC–AR estimate .
Figure 5.2 shows that, after parameter selection, the plug-in theoretical approximation continues to track the empirical risk reasonably well for different values of . This comparison provides a numerical check that the practical residual-based implementation selects penalties in a range for which the risk approximation remains informative.
Figure 5.2. Empirical MSE (points) and theoretical approximation evaluated at the fitted pair (dashes). Error bars are Monte Carlo 95% confidence intervals.
Proof of Proposition 3.4
Some preliminary results
In this section, we establish auxiliary results used in the proof of Proposition 3.4. The following theorem provides a precise approximation for the bias and variance components in (3.14) whose proof will be given in Appendix B.2. Recall and from (3.4) and (3.5). Recall the spectral density function in (3.1) and the bias-variance decomposition in (3.14).
Theorem A.1 Suppose the assumptions of Proposition 3.4 hold. Then we have that for all
where and are defined in (3.4) and (3.5), respectively. Moreover, we have that
The following lemma establishes the asymptotic behavior of the Riemann-type sums in (A.1) and (A.2).
Lemma A.2 Let
and let be continuously differentiable on . Define
Then, for every , and
Proof. For , define
Since
we have
Recall (A.3). This yields that . Consequently, we see that
where the equality follows from the change of variables .
It remains to justify the replacement of the sum by the integral. The mesh points have spacing . We use the following elementary Riemann-sum error bound for functions of bounded variation (see Chapter 7 of [1]). If has bounded variation on , then
Consequently,
We now verify that is uniformly bounded. The function is increasing on . Moreover, is decreasing, while increases on and decreases on . Hence, for either choice of , there exists some constant ,
uniformly in . Since is continuously differentiable on , it has bounded variation. The product rule for total variation (see Chapter 6 of [1]) therefore implies
where does not depend on or . Consequently, together with (A.4), we can conclude the proof. ■
Technical proof of the Proposition 3.4
Recall (26). Theorem A.1 gives
and
For , for the variance term, we apply Lemma A.2 with
Since , , , and for every fixed , dominated convergence gives
Similarly, for the bias term, we apply Lemma A.2 with
Using (12) that
we obtain
Because , the remainder is negligible relative to , and the remainder is negligible relative to . Thus, both additive remainders are absorbed into the respective terms. This concludes the proof for .
For , the identity gives
Likewise, we have
These are precisely the bias and variance expressions in the proposition. The further asymptotic simplifications of these sums are recorded separately in Lemma A.3.
Finally, let , and define
Since , the functions and are uniformly continuously differentiable on . By the first-order Euler–Maclaurin formula (see Chapter 7 of [1]), we have that
and
Since according to (2.9), we obtain from (A.2) that
where we used the fact that the remainder is . Similarly,
This proves the part and completes the proof of Proposition 3.4.
Boundary asymptotics at
The case is the boundary between the consistent and inconsistent regimes and requires a separate asymptotic argument. We record the required scaled-sum approximations below which is used in the proof of Theorem 3.1.
Lemma A.3 Suppose Assumption 2.1 holds and , , and . Then
and
Proof. Recall (32). For , we have
and .
We first consider the variance sum. Apply Lemma A.2 with (34), we have that
Since and , we have that , . Moreover, for every fixed ,
Because is continuous and strictly positive on , there exists a constant such that
Therefore,
The dominating function is integrable on . Hence, by dominated convergence, we have
It follows that
For the bias part, applying Lemma A.2 with (35) and multiplying the resulting identity by , we obtain
Since , it remains to evaluate the integral. By an argument similar to (36), we have
Finally, by direct calculation, we have
Consequently, as , we have
This completes the proof. ■
Proof of Theorem A.1
Technical preparation
In this section, we collect and prove auxiliary results that will be used in the proof of Theorem A.1. We first collect some results from linear algebra. For a symmetric matrix , we denote its eigenvalues by , and its operator norm by .
Lemma B.1 Let and be two real symmetric matrices so that has rank at most . Then for the following inequalities hold: (a) , and (b) .
Proof. See [23], Theorem 4.3.6. ■
Lemma B.2 Let and be two nonnegative definite matrices with the property . Assume that all the eigenvalues of and are bounded above by a constant , and the values of the integer and the constant do not depend on . Then we have
Proof. The result follows directly from Lemma B.1. ■
We next collect and prove auxiliary results concerning the matrix introduced below (6). The analysis relies on its connection with Toeplitz and Hankel matrices. Recall that a matrix is Toeplitz if . If is given by , then is called the symbol of . We write for the corresponding infinite-dimensional Toeplitz matrix and for its leading principal submatrix.
The finite-sample boundary correction arising in the analysis of has Hankel structure. Accordingly, a matrix is called a Hankel matrix if . If is given by , then is called the symbol of . We write for the corresponding infinite-dimensional Hankel matrix and for its leading principal submatrix.
With the above preparation, now we connect with the Toeplitz matrix. For , recall that the th entry of the matrix is given by
Equivalently, according to the generalized binomial expansion, we can write
More precisely, let denote the infinite lower triangular Toeplitz operator with symbol , where . Then is the leading principal submatrix of . Recall from (17) that . Although the corresponding infinite product has symbol , the finite-dimensional matrix is not exactly , because truncation at the sample boundary produces a Hankel-type correction. We formally establish the approximation results as follows.
To state the result, let denote the flip operator defined by
for every .
Lemma B.3 Let be the Toeplitz matrix with symbol , where for every and . Define
and let , where we recall (16) and denote
Then we have that
Proof. First, suppose that . By the definitions of the Toeplitz and Hankel entries and the identity , we have
Indeed, the second equality is the discrete sine-transform quadrature formula. Since is an even trigonometric polynomial of degree at most , this formula follows by expanding in its Fourier series and applying discrete cosine orthogonality on the grid . The condition ensures that no aliasing terms arise in this calculation. Next, suppose that . Applying the same discrete sine-transform identity to the indices and gives
For the penultimate equality, note that the two sine factors acquire the same sign under the replacement and , so their product is unchanged. Because has Fourier coefficients equal to zero outside , whenever . Moreover, the th entry of is , which is zero whenever . Thus, the first identity applies when , both Hankel terms vanish when , and the second identity applies when . Combining these cases entrywise yields
It remains to replace by . Put . For the Hankel term, is a finite compression of the Hankel operator associated with the symbol . This operator is obtained by composing multiplication by , which has operator norm , with an orthogonal projection and a reflection, both of which have operator norm one. Therefore,
For the sine-basis term, let be the orthogonal discrete sine-transform matrix with th entry . By the definition of , is obtained by conjugating the diagonal matrix with diagonal entries by . Orthogonal conjugation preserves the operator norm. Hence,
Moreover, is an orthogonal and symmetric matrix; that is, . Therefore, conjugation by preserves the operator norm, and hence
Furthermore, the th entry of is , where . Thus, only Fourier coefficients with enter . Since retains all Fourier coefficients of with , the matrices and have identical entries. Therefore,
Combining this equality with the identity established above for , the difference in the statement of the lemma is the sum of the differences between the corresponding and terms. The triangle inequality, together with the preceding bounds, shows that its operator norm is bounded by . This proves the result. ■
The following results provide the key asymptotic expressions used in the proof of Theorem A.1.
Lemma B.4 Let and . Suppose that is of bounded variation, and let be defined as in Lemma B.3. Then, as ,
and
Proof. It suffices to prove the first assertion when is bounded, nonnegative, and decreasing. The result for a general function of bounded variation then follows by applying the Jordan decomposition and using linearity. Indeed, let
and define
Then and are bounded, nonnegative, and decreasing on , and . Since both and the right-hand side of the asserted approximation are linear in , the general result follows by applying the decreasing-case result separately to and .
Write , and let , where is a nonnegative integer and . We invoke the finite-rank comparison for fractional-difference operators established in [4]. In the notation used here, this result implies that, for every , can be compared with
up to a symmetric perturbation of rank at most , where we recall (17).
We first prove the approximation involving . The matrix is diagonalized by the discrete sine basis whose entries are , with eigenvalues [17]. Since is diagonalized by the same basis, its eigenvalues are . For a fixed , the comparison from [4], Theorem 6(a) and Theorem 7(b), (d), followed by the interlacing theorem in Theorem B.1, gives the lower bound
for .
To identify the right-hand side, choose . Since , we have
Moreover, is decreasing in , is decreasing in , and is increasing in , because is decreasing on . Thus, the relevant eigenvalues are ordered as required for the interlacing argument. Summing the preceding inequality over gives
The omitted terms are uniformly bounded because is bounded. The companion comparison for in [4], together with the same interlacing argument, yields the reverse bound
This proves the first assertion.
It remains to prove the bias-trace approximation. Since , the same argument, with , applies to the two resolvent powers and , and gives
Finally, the elementary identity
implies that
This proves the second assertion. ■
Proof of Theorem A.1
For notational simplicity, let be the matrix obtained from by deleting its first rows, and define
Both and are symmetric matrices. Throughout this proof, denotes the spectral density in (13). Recall (8). We can write
Since and , the variance component is
The matrices and differ by a matrix of rank at most . Indeed, the resolvent identity gives
Hence, this difference has rank at most . Consequently, has rank at most . Both resolvents are contractions because and are positive semidefinite, while because is bounded [17]. Thus, the trace of the resulting bounded-rank perturbation is , and hence
Let . By Lemma B.3, we may write
where
Assumption 2.1(1) implies that . To bound the Hankel term, write
where has ones on its th anti-diagonal and zeros elsewhere. Since , we have
Because is orthogonal,
Moreover,
Since , the trace inequality shows that each Hankel term contributes . Also,
Consequently,
Furthermore, (1) of Assumption 2.1 implies . Hence the spectral density is continuously differentiable on , and therefore has bounded variation. Thus, the first assertion of Lemma B.4 applies with and gives
which proves (A.1).
We next consider the bias component. Recall (2.6) and and
where the zero block has columns. Consequently, . It follows that
Using , we obtain
Let . Since , and has rank at most , Lemma eq:B.2 and the uniform bound imply that
By Lemma B.4, we have
Consequently,
which proves (A.2).
References
- [1]T. M. Apostol. Mathematical Analysis: A Modern Approach to Advanced Calculus. Addison–Wesley, Reading, MA, 2 edition, 1974.
- [2]G. E. P. Box, G. M. Jenkins, G. C. Reinsel, and G. M. Ljung. Time Series Analysis: Forecasting and Control. Wiley, Hoboken, NJ, 5th edition, 2015.
- [3]P. J. Brockwell and R. A. Davis. Introduction to Time Series and Forecasting. Springer Texts in Statistics. Springer, New York, 3rd edition, 2016.
- [4]P. Burman. Approximate singular values of the fractional difference and summation operators. Linear Algebra and its Applications, 416(2):677–687, 2006.DOI
- [5]P. Burman and R. H. Shumway. Estimation of trend in state-space models: Asymptotic mean square error and rate of convergence. The Annals of Statistics, 37(6B):3715–3742, 2009.
- [6]R. J. Carroll and D. Ruppert. Transformation and Weighting in Regression. Monographs on Statistics and Applied Probability. Chapman and Hall, New York, NY, 1988.
- [7]N. H. Chan and W. Palma. State space modeling of long-memory processes. The Annals of Statistics, 26(2):719–740, 1998.
- [8]N. H. Chan and W. Palma. Estimation of long-memory time series models: A survey of different likelihood-based methods. In Econometric Analysis of Financial and Economic Time Series, pages 89–121. Emerald Group Publishing Limited, 2006.
- [9]J. J. Commandeur and S. J. Koopman. An introduction to state space time series analysis. Oxford university press, 2007.
- [10]X. Ding and Z. Zhou. Autoregressive approximations to nonstationary time series with inference and applications. The Annals of Statistics, 51(3):1207–1231, 2023.DOI
- [11]X. Ding and Z. Zhou. On the partial autocorrelation function for locally stationary time series: characterization, estimation and inference. Biometrika, 112(2):asaf016, 2025.
- [12]J. Durbin and S. J. Koopman. Time Series Analysis by State Space Methods. Oxford Statistical Science Series. Oxford University Press, Oxford, 2nd edition, 2012.
- [13]K. B. Ensor, J. C. Schedler, T. Sun, R. Schneider, A. Mulenga, J. Wu, L. B. Stadler, and L. Hopkins. Online trend estimation and detection of trend deviations in sub-sewershed time series of sars-cov-2 rna measured in wastewater. Scientific Reports, 14(1):5575, 2024.
- [14]J. Fan and Q. Yao. Nonlinear Time Series: Nonparametric and Parametric Methods. Springer, 2003.
- [15]S. Grassi and P. S. de Magistris. When long memory meets the kalman filter: A comparative study. Computational Statistics & Data Analysis, 76:301–319, 2014.
- [16]H. L. Gray and N. F. Zhang. On a new definition of the fractional difference. Mathematics of Computation, 50(182):513–529, 1988.
- [17]U. Grenander and G. Szegö. Toeplitz Forms and Their Applications, volume 2 of California Monographs in Mathematical Sciences. University of California Press, Berkeley, 1958.
- [18]N. Haldrup and M. Ø. Nielsen. Estimation of fractional integration in the presence of data noise. Computational Statistics & Data Analysis, 51(6):3100–3114, 2007.
- [19]T. Hartl. Robust trend estimation for strongly persistent data with unobserved memory. Journal of Business & Economic Statistics, 2025.
- [20]T. Hartl and R. Jucknewitz. Approximate state space modelling of unobserved fractional components. Econometric Reviews, 41(1):75–98, 2022.
- [21]U. Hassler. Time series analysis with long memory in view. John Wiley & Sons, 2018.
- [22]M. Hatanaka. Time–Series–Based Econometrics: Unit Roots and Co–integrations. Advanced Texts in Econometrics. Oxford University Press, Oxford, 1996.
- [23]R. A. Horn and C. R. Johnson. Matrix Analysis. Cambridge University Press, Cambridge, 1985.
- [24]M. Mudelsee. Trend analysis of climate time series: A review of methods. Earth-Science Reviews, 190:310–322, 2019.DOI
- [25]M. Ø. Nielsen. Asymptotics for the conditional-sum-of-squares estimator in multivariate fractional time-series models. Journal of Time Series Analysis, 36(2):154–188, 2015.
- [26]W. Palma. Long-memory time series: theory and methods. John Wiley & Sons, 2007.
- [27]A. Patharkar, F. Cai, F. Al-Hindawi, and T. Wu. Predictive modeling of biomedical temporal data in healthcare applications: review and future directions. Frontiers in Physiology, 15:1386760, 2024.
- [28]R. H. Shumway and D. S. Stoffer. Time Series Analysis and Its Applications: With R Examples. Springer Texts in Statistics. Springer, 5th edition, 2025.
- [29]C. Stone. Optimal global rates of convergence for nonparametric regression. Annals of Statistics, 10:1040–1053, 1982.
- [30]R. S. Tsay. Analysis of Financial Time Series. Wiley Series in Probability and Statistics. Wiley, Hoboken, NJ, 3rd edition, 2010.
- [31]G. Wahba. Improper priors, spline smoothing and the problem of guarding against model errors in regression. Journal of the Royal Statistical Society Series B: Statistical Methodology, 40(3):364–372, 1978.DOI

