1. What is the functional autoregressive (FAR) process and its significance in high-frequency data analysis?
The functional autoregressive (FAR) process is a commonly used linear model to describe the auto-dependency in functional time series. It is similar to the classic real-valued time series analysis. Bosq (2000) investigated the causality and stationarity of FAR(1) processes and constructed one-step ahead predictors. Aue et al. (2015) proposed a prediction algorithm for FAR(p) models, which is accurate and easily implementable. The FAR process is significant in high-frequency data analysis as it helps in understanding the auto-dependency among observed curves, enabling researchers to make accurate predictions and gain insights into the underlying mechanisms of the data. It is a crucial tool in the functional data analysis approach, which has become increasingly popular in research frontiers due to advancements in data collection and storage.
read more
2. What is the Hilbert space used in the functional threshold autoregressive model?
The Hilbert space used in the functional threshold autoregressive model is denoted as H = L 2 ([0, 1]), which represents the space of square integrable functions on the interval [0, 1]. This space is equipped with the inner product x, y = 1 0 x(t)y(t) dt, and the norm x = ( 1 0 x(t) 2 dt) 1/2. The choice of the interval [0, 1] is made for convenience and does not impose any restrictions on generality. The Hilbert space H plays a crucial role in defining the functional time series and their properties in the model.
read more
3. What are the mild conditions for stationarity and ergodicity in fTAR models?
The mild conditions for stationarity and ergodicity in fTAR models involve the functional process Y k and exogenous variables X k,m, as well as the coefficients of fTAR/fTARX models. These conditions are easy to verify and sufficient to ensure stationarity and ergodicity. Theorem 1 provides a general sufficient criterion for strict stationarity and ergodicity of the fTAR process. Corollary 1 states that if max i p Y,i j=1 Ps i,j L < 1, then (3.5) holds. Corollary 2 states that if the operator Ps i,j L(H) is Hilbert-Schmidt with norm Ps i,j S < 1, then max i p Y,i j=1 Ps i,j L < 1, and hence (3.5) holds. These corollaries enhance the practicality of Theorem 1 by providing easily verifiable sufficient conditions for (3.5) to hold.
read more
4. What is the code length for encoding M in threshold and model order estimation?
The code length for encoding M in threshold and model order estimation is expressed as CL(M) = log 2 (r) + log 2 d + r-1 i=1 log 2 n i + r i=1 log 2 (p Y,i q + 1)q + r i=1 ((p Y,i q + 1)q log 2 n i) / 2. This formula represents the number of digits required to record the number of regimes, delay parameter, threshold estimators, and total number of parameters in each regime. The code length is calculated using logarithmic functions to efficiently represent the model parameters and their estimators.
read more