Package 'qfa' reference manual

Title:	Quantile-Frequency Analysis (QFA) of Time Series and Spline Quantile Regression (SQR)
Description:	Implementation of quantile frequency analysis (QFA) for time series based on trigonometric quantile regression and of spline quantile regression (SQR) for estimating the coefficients in linear quantile regression models as smooth functions of the quantile level. References: [1] Li, T.-H. (2012). ''Quantile periodograms,'' J. of the American Statistical Association, 107, 765–776. <doi:10.1080/01621459.2012.682815> [2] Li, T.-H. (2014). Time Series with Mixed Spectra, CRC Press. <doi:10.1201/b15154> [3] Li, T.-H. (2025). ''Quantile Fourier transform, quantile series, and nonparametric estimation of quantile spectra,'' Communications in Statistics: Simulation and Computation, 1–22. <doi:10.1080/03610918.2025.2509820> [4] Li, T.-H. (2025). ''Quantile-crossing spectrum and spline autoregression estimation,'' Statistical Inference for Stochastic Processes, 28, 20. <doi:10.1007/s11203-025-09336-7> [5] Li, T.-H. (2025). ''Spline autoregression method for estimation of quantile spectrum,'' J. of Computational and Graphical Statistics, 1-15. <doi:10.1080/10618600.2025.2549452> [6] Li, T.-H., and Megiddo, N. (2026). ''Spline quantile regression,'' J. of Statistical Theory and Practice, 20, 30. <doi:10.1007/s42519-026-00545-8> [7] Li, T.-H. (2026). ''Spline quantile regression with cubic and linear smoothing splines,'' <doi:10.48550/arXiv.2603.22408>.
Authors:	Ta-Hsin Li [cre, aut]
Maintainer:	Ta-Hsin Li <[email protected]>
License:	GPL (>= 2)
Version:	5.0
Built:	2026-05-29 09:21:57 UTC
Source:	https://github.com/cran/qfa

Birth data

Description

A random sample of US birth data in 2022 (birth weight and covariates). Precare and Education should be treated as factors.

Usage

data(birthweight)
data(birthweight)

Format

An object of class data.frame with 50000 rows and 12 columns.

Source

Data file https://data.nber.org/nvss/natality/csv/2022/natality2022us.csv from NBER https://www.nber.org/research/data/vital-statistics-natality-birth-data. Original source from the National Center for Health Statistics https://www.cdc.gov/nchs/nvss/births.htm.

References

Koenker, R. (2005). Quantile Regression. Cambridge University Press.

Bootstrap of Spline Quantile Regression (SQR) Coefficients

Description

This function generates bootstrap samples for the SQR coefficients and their derivatives

Usage

boot.sqr(
  fit,
  nsim = 1000,
  blocklength = 1,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)
boot.sqr(
  fit,
  nsim = 1000,
  blocklength = 1,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)

Arguments

fit

output from sqr()

nsim

number of bootstrap samples (default = 1000)

blocklength

blocklength for block sampling scheme (default = 1)

n.cores

number of cores for parallel computing (default = 1)

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

seed

seed of random number generator (default = 1234567)

Value

A list with the following elements:

coefficients

array of bootstrap regression coefficients

derivatives

array of boostrap derivatives of regression coefficient

info

sequence convergence status

Engel's food expenditure data

Description

Engel's food expenditure data from the R package 'quantreg'.

Usage

data(engel)
data(engel)

Format

An object of class data.frame with 235 rows and 2 columns.

References

Koenker, R. (2005). Quantile Regression. Cambridge University Press.

Financial index

Description

Daily closing values of DJIA and FTSE 100 on trading days from 2001-11-27 to 2010-03-10.

Usage

data(finIndex)
data(finIndex)

Format

An object of class data.frame with 2048 rows and 3 columns.

Source

https://www.wsj.com/market-data/quotes/index/DJIA/historical-prices and https://www.wsj.com/market-data/quotes/index/UK/UKX/historical-prices

Periodogram (PER)

Description

This function computes the periodogram or periodogram matrix for univariate or multivariate time series.

Usage

per(y)
per(y)

Arguments

y

vector or matrix of time series s (if matrix, nrow(y) = length of time series)

Value

vector or array of periodogram

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.per <- per(y)
plot(y.per)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.per <- per(y)
plot(y.per)

Quantile Autocovariance Function (QACF)

Description

This function computes quantile autocovariance function (QACF) from time series or quantile discrete Fourier transform (QDFT).

Usage

qacf(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)
qacf(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

y.qdft

matrix or array of pre-calculated QDFT (default = NULL: compute from y and tau); if y.qdft is supplied, y and tau can be left unspecified

n.cores

number of cores for parallel computing of QDFT if y.qdft = NULL (default = 1)

cl

pre-existing cluster for repeated parallel computing of QDFT (default = NULL)

Value

matrix or array of quantile autocovariance function

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qacf <- qacf(y,tau)
# compute from QDFT 
y.qdft <- qdft(y,tau) 
y.qacf <- qacf(y.qdft=y.qdft)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qacf <- qacf(y,tau)
# compute from QDFT 
y.qdft <- qdft(y,tau) 
y.qacf <- qacf(y.qdft=y.qdft)

Quantile-Crossing Series (QCSER)

Description

This function creates the quantile-crossing series (QCSER) for univariate or multivariate time series.

Usage

qcser(y, tau, normalize = FALSE)
qcser(y, tau, normalize = FALSE)

Arguments

y

vector or matrix of time series

tau

sequence of quantile levels in (0,1)

normalize

TRUE or FALSE (default): normalize QCSER to have unit variance

Value

A matrix or array of quantile-crossing series

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qser <- qcser(y,tau)
dim(y.qser)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qser <- qcser(y,tau)
dim(y.qser)

Quantile Discrete Fourier Transform (QDFT)

Description

This function computes quantile discrete Fourier transform (QDFT) for univariate or multivariate time series.

Usage

qdft(y, tau, n.cores = 1, cl = NULL)
qdft(y, tau, n.cores = 1, cl = NULL)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

n.cores

number of cores for parallel computing (default = 1)

cl

pre-existing cluster for repeated parallel computing (default = NULL)

Value

matrix or array of quantile discrete Fourier transform of y

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y,tau)
# Make a cluster for repeated use
n.cores <- 2
cl <- parallel::makeCluster(n.cores)
parallel::clusterExport(cl, c("tqr.fit"))
doParallel::registerDoParallel(cl)
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.qdft <- qdft(y1,tau,n.cores=n.cores,cl=cl)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.qdft <- qdft(y2,tau,n.cores=n.cores,cl=cl)
parallel::stopCluster(cl)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y,tau)
# Make a cluster for repeated use
n.cores <- 2
cl <- parallel::makeCluster(n.cores)
parallel::clusterExport(cl, c("tqr.fit"))
doParallel::registerDoParallel(cl)
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.qdft <- qdft(y1,tau,n.cores=n.cores,cl=cl)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y.qdft <- qdft(y2,tau,n.cores=n.cores,cl=cl)
parallel::stopCluster(cl)

Quantile Autocovariance Function (QACF)

Description

This function computes quantile autocovariance function (QACF) from QDFT.

Usage

qdft2qacf(y.qdft, return.qser = FALSE)
qdft2qacf(y.qdft, return.qser = FALSE)

Arguments

y.qdft

matrix or array of QDFT from qdft()

return.qser

if TRUE, return quantile series (QSER) along with QACF

Value

matrix or array of quantile autocovariance function if return.sqer = FALSE (default), else a list with the following elements:

qacf

matirx or array of quantile autocovariance function

qser

matrix or array of quantile series

Examples

# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qacf <- qdft2qacf(y.qdft)
plot(c(0:9),y.qacf[c(1:10),1],type='h',xlab="LAG",ylab="QACF")
y.qser <- qdft2qacf(y.qdft,return.qser=TRUE)$qser
plot(y.qser[,1],type='l',xlab="TIME",ylab="QSER")
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qacf <- qdft2qacf(y.qdft)
plot(c(0:9),y.qacf[1,2,c(1:10),1],type='h',xlab="LAG",ylab="QACF")
# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qacf <- qdft2qacf(y.qdft)
plot(c(0:9),y.qacf[c(1:10),1],type='h',xlab="LAG",ylab="QACF")
y.qser <- qdft2qacf(y.qdft,return.qser=TRUE)$qser
plot(y.qser[,1],type='l',xlab="TIME",ylab="QSER")
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qacf <- qdft2qacf(y.qdft)
plot(c(0:9),y.qacf[1,2,c(1:10),1],type='h',xlab="LAG",ylab="QACF")

Quantile Periodogram (QPER)

Description

This function computes quantile periodogram (QPER) from QDFT.

Usage

qdft2qper(y.qdft)
qdft2qper(y.qdft)

Arguments

y.qdft

matrix or array of QDFT from qdft()

Value

matrix or array of quantile periodogram

Examples

# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qper <- qdft2qper(y.qdft)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
qfa.plot(ff[sel.f],tau,Re(y.qper[sel.f,]))
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qper <- qdft2qper(y.qdft)
qfa.plot(ff[sel.f],tau,Re(y.qper[1,1,sel.f,]))
qfa.plot(ff[sel.f],tau,Re(y.qper[1,2,sel.f,]))
# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qper <- qdft2qper(y.qdft)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
qfa.plot(ff[sel.f],tau,Re(y.qper[sel.f,]))
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qper <- qdft2qper(y.qdft)
qfa.plot(ff[sel.f],tau,Re(y.qper[1,1,sel.f,]))
qfa.plot(ff[sel.f],tau,Re(y.qper[1,2,sel.f,]))

Quantile Series (QSER)

Description

This function computes quantile series (QSER) from QDFT.

Usage

qdft2qser(y.qdft)
qdft2qser(y.qdft)

Arguments

y.qdft

matrix or array of QDFT from qdft()

Value

matrix or array of quantile series

Examples

# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qser <- qdft2qser(y.qdft)
plot(y.qser[,1],type='l',xlab="TIME",ylab="QSER")
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qser <- qdft2qser(y.qdft)
plot(y.qser[1,,1],type='l',xlab="TIME",ylab="QSER")
# single time series
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qdft <- qdft(y1,tau)
y.qser <- qdft2qser(y.qdft)
plot(y.qser[,1],type='l',xlab="TIME",ylab="QSER")
# multiple time series
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y.qdft <- qdft(cbind(y1,y2),tau)
y.qser <- qdft2qser(y.qdft)
plot(y.qser[1,,1],type='l',xlab="TIME",ylab="QSER")

Quantile-Frequency Plot

Description

This function creates an image plot of quantile spectrum.

Usage

qfa.plot(
  freq,
  tau,
  rqper,
  rg.qper = range(rqper),
  rg.tau = range(tau),
  rg.freq = c(0, 0.5),
  color = colorRamps::matlab.like2(1024),
  ylab = "QUANTILE LEVEL",
  xlab = "FREQUENCY",
  tlab = NULL,
  set.par = TRUE,
  legend.plot = TRUE,
  line = 0.5
)
qfa.plot(
  freq,
  tau,
  rqper,
  rg.qper = range(rqper),
  rg.tau = range(tau),
  rg.freq = c(0, 0.5),
  color = colorRamps::matlab.like2(1024),
  ylab = "QUANTILE LEVEL",
  xlab = "FREQUENCY",
  tlab = NULL,
  set.par = TRUE,
  legend.plot = TRUE,
  line = 0.5
)

Arguments

freq

sequence of frequencies in (0,0.5) at which quantile spectrum is evaluated

tau

sequence of quantile levels in (0,1) at which quantile spectrum is evaluated

rqper

real-valued matrix of quantile spectrum evaluated on the freq x tau grid

rg.qper

zlim for qper (default = range(qper))

rg.tau

ylim for tau (default = range(tau))

rg.freq

xlim for freq (default = c(0,0.5))

color

colors (default = colorRamps::matlab.like2(1024))

ylab

label of y-axis (default = "QUANTILE LEVEL")

xlab

label of x-axis (default = "FREQUENCY")

tlab

title of plot (default = NULL)

set.par

if TRUE, par() is set internally (single image)

legend.plot

if TRUE, legend plot is added

line

distance between the image and the axis (default = 0.5)

Value

no return value

Kullback-Leibler Divergence of Quantile Spectral Estimate

Description

This function computes Kullback-Leibler divergence (KLD) of quantile spectral estimate.

Usage

qkl.divergence(y.qper, qspec, sel.f = NULL, sel.tau = NULL)
qkl.divergence(y.qper, qspec, sel.f = NULL, sel.tau = NULL)

Arguments

y.qper

matrix or array of quantile spectral estimate from, e.g., qspec.lw()

qspec

matrix of array of true quantile spectrum (same dimension as y.qper)

sel.f

index of selected frequencies for computation (default = NULL: all frequencies)

sel.tau

index of selected quantile levels for computation (default = NULL: all quantile levels)

Value

real number of Kullback-Leibler divergence

Quantile Periodogram (QPER)

Description

This function computes quantile periodogram (QPER) from time series or quantile discrete Fourier transform (QDFT).

Usage

qper(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)
qper(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

y.qdft

matrix or array of pre-calculated QDFT (default = NULL: compute from y and tau); if y.qdft is supplied, y and tau can be left unspecified

n.cores

number of cores for parallel computing of QDFT if y.qdft = NULL (default = 1)

cl

pre-existing cluster for repeated parallel computing of QDFT (default = NULL)

Value

matrix or array of quantile periodogram

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qper <- qper(y,tau)  
# compute from QDFT 
y.qdft <- qdft(y,tau) 
y.qper <- qper(y.qdft=y.qdft)  
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qper <- qper(y,tau)  
# compute from QDFT 
y.qdft <- qdft(y,tau) 
y.qper <- qper(y.qdft=y.qdft)

Quantile Periodogram Type II (QPER2)

Description

This function computes type-II quantile periodogram for univariate time series.

Usage

qper2(y, freq, tau, weights = NULL, n.cores = 1, cl = NULL)
qper2(y, freq, tau, weights = NULL, n.cores = 1, cl = NULL)

Arguments

y

univariate time series

freq

sequence of frequencies in [0,1)

tau

sequence of quantile levels in (0,1)

weights

sequence of weights in quantile regression (default = NULL: weights equal to 1)

n.cores

number of cores for parallel computing (default = 1)

cl

pre-existing cluster for repeated parallel computing (default = NULL)

Value

matrix of quantile periodogram evaluated on freq * tau grid

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qper2 <- qper2(y,ff,tau)
qfa.plot(ff[sel.f],tau,Re(y.qper2[sel.f,]))
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qper2 <- qper2(y,ff,tau)
qfa.plot(ff[sel.f],tau,Re(y.qper2[sel.f,]))

Quantile Series (QSER)

Description

This function computes quantile series (QSER) from time series or quantile discrete Fourier transform (QDFT).

Usage

qser(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)
qser(y, tau, y.qdft = NULL, n.cores = 1, cl = NULL)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

y.qdft

matrix or array of pre-calculated QDFT (default = NULL: compute from y and tau); if y.qdft is supplied, y and tau can be left unspecified

n.cores

number of cores for parallel computing of QDFT if y.qdft = NULL (default = 1)

cl

pre-existing cluster for repeated parallel computing of QDFT (default = NULL)

Value

matrix or array of quantile series

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qser <- qser(y,tau)  
# compute from QDFT
y.qdft <- qdft(y,tau)  
y.qser <- qser(y.qdft=y.qdft) 
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
# compute from time series
y.qser <- qser(y,tau)  
# compute from QDFT
y.qdft <- qdft(y,tau)  
y.qser <- qser(y.qdft=y.qdft)

Autoregression (AR) Model of Quantile Series

Description

This function fits an autoregression (AR) model to quantile series (QSER) separately for each quantile level using stats::ar().

Usage

qser2ar(
  y.qser,
  p = NULL,
  order.max = NULL,
  method = c("none", "gamm", "sp"),
  spar = NULL,
  type = c(1, 2)
)
qser2ar(
  y.qser,
  p = NULL,
  order.max = NULL,
  method = c("none", "gamm", "sp"),
  spar = NULL,
  type = c(1, 2)
)

Arguments

y.qser

matrix or array of pre-calculated QSER, e.g., using qser()

p

order of AR model (default = NULL: selected by AIC)

order.max

maximum order for AIC if p = NULL (default = NULL: determined by stats::ar())

method

quantile smoothing method: "gamm" for mgcv::gamm(), "sp" for stats::smooth.spline(), or "none" (default)

spar

smoothing parameter for stats::smooth.spline() (default = NULL for GCV)

type

type of smoothing for covariance matrix: 1 (direct, default) or 2 (square root)

Value

a list with the following elements:

A

matrix or array of AR coefficients

V

vector or matrix of residual covariance

p

order of AR model

n

length of time series

residuals

matrix or array of residuals

ACF of Quantile Series (QSER) or Quantile-Crossing Series (QCACF)

Description

This function creates the ACF of quantile series or quantile-crossing series

Usage

qser2qacf(y.qser)
qser2qacf(y.qser)

Arguments

y.qser

matrix or array of quantile-crossing series

Value

A matrix or array of ACF

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qser <- qcser(y,tau)
y.qacf <- qser2qacf(y.qser)
dim(y.qacf)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.qser <- qcser(y,tau)
y.qacf <- qser2qacf(y.qser)
dim(y.qacf)

Spline Autoregression (SAR) Model of Quantile Series

Description

This function fits spline autoregression (SAR) model to quantile series (QSER).

Usage

qser2sar(
  y.qser,
  tau0,
  tau = tau0,
  p = NULL,
  order.max = NULL,
  spar = NULL,
  method = c("GCV", "AIC", "BIC"),
  type = c(1, 2),
  weights = rep(1, length(tau0)),
  interval = c(-1.5, 1.5)
)
qser2sar(
  y.qser,
  tau0,
  tau = tau0,
  p = NULL,
  order.max = NULL,
  spar = NULL,
  method = c("GCV", "AIC", "BIC"),
  type = c(1, 2),
  weights = rep(1, length(tau0)),
  interval = c(-1.5, 1.5)
)

Arguments

y.qser

matrix or array of pre-calculated QSER at tau0 using, e.g., qser()

tau0

quantile levels used to compute y.qser

tau

quantile levels for evaluation (min(tau0) <= tau <= max(tau0); default = tau0)

p

order of SAR model (default = NULL: automatically selected by AIC)

order.max

maximum order for AIC if p = NULL (default = NULL: determined by stats::ar())

spar

penalty parameter alla stats::smooth.spline() (default = NULL: automatically selected)

method

criterion for penalty parameter selection: "GCV" (default), "AIC", or "BIC"

type

type of quantile smoothing for residual variance: 1 = direct (default) or 2 = square root

weights

sequence of weights (default = rep(1,length(tau0)))

interval

interval for spar optimization (default = c(-1.5,1.5))

Value

a list with the following elements:

A

matrix or array of SAR coefficients

V

vector or matrix of SAR residual covariance

p

order of SAR model

spar

penalty parameter

n

length of time series

tau

quantile levels for evaluation

tau0

quantile levels for fitting

weights

weights in penalty function

sdm

spline design matrices

fit

object containing details of SAR fit

Autoregression (AR) Estimator of Quantile Spectrum

Description

This function computes autoregression (AR) estimate of quantile spectrum from time series or quantile series (QSER).

Usage

qspec.ar(
  y,
  tau,
  y.qser = NULL,
  p = NULL,
  order.max = NULL,
  freq = NULL,
  method = c("none", "gamm", "sp"),
  spar = NULL,
  type = c(1, 2),
  n.cores = 1,
  cl = NULL
)
qspec.ar(
  y,
  tau,
  y.qser = NULL,
  p = NULL,
  order.max = NULL,
  freq = NULL,
  method = c("none", "gamm", "sp"),
  spar = NULL,
  type = c(1, 2),
  n.cores = 1,
  cl = NULL
)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

y.qser

matrix or array of pre-calculated QSER (default = NULL: compute from y and tau); if y.qser is supplied, y and tau can be left unspecified

p

order of AR model (default = NULL: automatically selected by AIC)

order.max

maximum order for AIC if p = NULL (default = NULL: determined by stats::ar())

freq

sequence of frequencies in [0,1) (default = NULL: all Fourier frequencies)

method

method of quantile smoothing: "gamm" for mgcv::gamm(), "sp" for stats::smooth.spline(), or "none" (default)

spar

smoothing parameter for stats::smooth.spline() (default = NULL for GCV)

type

type of smoothing for covariance matrix: 1 (direct, default) or 2 (square root)

n.cores

number of cores for parallel computing of QDFT if y.qser = NULL (default = 1)

cl

pre-existing cluster for repeated parallel computing of QDFT (default = NULL)

Value

a list with the following elements:

spec

matrix or array of AR quantile spectrum

freq

sequence of frequencies

fit

object of AR model

qser

matrix or array of quantile series if y.qser = NULL

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y <- cbind(y1,y2)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qspec.ar <- qspec.ar(y,tau,p=1)$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.ar[1,1,sel.f,]))
y.qser <- qcser(y1,tau)
y.qspec.ar <- qspec.ar(y.qser=y.qser,p=1)$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.ar[sel.f,]))
y.qspec.arqs <- qspec.ar(y.qser=y.qser,p=1,method="sp")$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.arqs[sel.f,]))
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y <- cbind(y1,y2)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qspec.ar <- qspec.ar(y,tau,p=1)$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.ar[1,1,sel.f,]))
y.qser <- qcser(y1,tau)
y.qspec.ar <- qspec.ar(y.qser=y.qser,p=1)$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.ar[sel.f,]))
y.qspec.arqs <- qspec.ar(y.qser=y.qser,p=1,method="sp")$spec
qfa.plot(ff[sel.f],tau,Re(y.qspec.arqs[sel.f,]))

Lag-Window (LW) Estimator of Quantile Spectrum

Description

This function computes lag-window (LW) estimate of quantile spectrum with or without quantile smoothing from time series or quantile autocovariance function (QACF).

Usage

qspec.lw(
  y,
  tau,
  y.qacf = NULL,
  M = NULL,
  method = c("none", "gamm", "sp"),
  spar = "GCV",
  n.cores = 1,
  cl = NULL
)
qspec.lw(
  y,
  tau,
  y.qacf = NULL,
  M = NULL,
  method = c("none", "gamm", "sp"),
  spar = "GCV",
  n.cores = 1,
  cl = NULL
)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels in (0,1)

y.qacf

matrix or array of pre-calculated QACF (default = NULL: compute from y and tau); if y.qacf is supplied, y and tau can be left unspecified

M

bandwidth parameter of lag window (default = NULL: quantile periodogram)

method

quantile smoothing method: "gamm" for mgcv::gamm(), "sp" for stats::smooth.spline(), or "none" (default)

spar

smoothing parameter in smooth.spline() if method = "sp" (default = "GCV")

n.cores

number of cores for parallel computing (default = 1)

cl

pre-existing cluster for repeated parallel computing (default = NULL)

Value

A list with the following elements:

spec

matrix or array of spectral estimate

spec.lw

matrix or array of spectral estimate without quantile smoothing

lw

lag-window sequence

qacf

matrix or array of quantile autocovariance function if y.qacf = NULL

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qacf <- qacf(cbind(y1,y2),tau)
y.qper.lw <- qspec.lw(y.qacf=y.qacf,M=5)$spec
qfa.plot(ff[sel.f],tau,Re(y.qper.lw[1,1,sel.f,]))
y.qper.lwqs <- qspec.lw(y.qacf=y.qacf,M=5,method="sp",spar=0.9)$spec
qfa.plot(ff[sel.f],tau,Re(y.qper.lwqs[1,1,sel.f,]))
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qacf <- qacf(cbind(y1,y2),tau)
y.qper.lw <- qspec.lw(y.qacf=y.qacf,M=5)$spec
qfa.plot(ff[sel.f],tau,Re(y.qper.lw[1,1,sel.f,]))
y.qper.lwqs <- qspec.lw(y.qacf=y.qacf,M=5,method="sp",spar=0.9)$spec
qfa.plot(ff[sel.f],tau,Re(y.qper.lwqs[1,1,sel.f,]))

Spline Autoregression (SAR) Estimator of Quantile Spectrum

Description

This function computes spline autoregression (SAR) estimate of quantile spectrum.

Usage

qspec.sar(
  y,
  y.qser = NULL,
  tau0,
  tau = tau0,
  p = NULL,
  order.max = NULL,
  spar = NULL,
  method = c("GCV", "AIC", "BIC"),
  type = c(1, 2),
  weights = rep(1, length(tau0)),
  interval = c(-1.5, 1.5),
  freq = NULL,
  n.cores = 1,
  cl = NULL
)
qspec.sar(
  y,
  y.qser = NULL,
  tau0,
  tau = tau0,
  p = NULL,
  order.max = NULL,
  spar = NULL,
  method = c("GCV", "AIC", "BIC"),
  type = c(1, 2),
  weights = rep(1, length(tau0)),
  interval = c(-1.5, 1.5),
  freq = NULL,
  n.cores = 1,
  cl = NULL
)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

y.qser

matrix or array of pre-calculated QSER at tau0 (default = NULL: compute from y and tau0); if supplied, y can be left unspecified

tau0

quantile levels used to computer QSER (must be supplied with or without y.qser)

tau

quantile levels for evaluation (min(tau0) <= tau <= max(tau0); default = tau0)

p

order of SAR model (default = NULL: automatically selected by AIC)

order.max

maximum order for AIC if p = NULL (default = NULL: determined by stats::ar())

spar

penalty parameter alla stats::smooth.spline() (default = NULL: automatically selected)

method

criterion for penalty parameter selection: "GCV" (default), "AIC", or "BIC"

type

type of quantile smoothing for residual variance: 1 = direct (default) or 2 = square root

weights

sequence of weights (length(weights) = length(tau0); default: weights = rep(1,length(tau0)))

interval

interval for spar optimization (default: interval = c(-1.5,1.5))

freq

sequence of frequencies in [0,1) (default = NULL: all Fourier frequencies)

n.cores

number of cores for parallel computing of QDFT if y.qser = NULL (default = 1)

cl

pre-existing cluster for repeated parallel computing of QDFT (default = NULL)

Value

a list with the following elements:

spec

matrix or array of SAR quantile spectrum

freq

sequence of frequencies

fit

object of SAR model

qser

matrix or array of quantile series if y.qser = NULL

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
# compute from time series
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
qfa.plot(ff[sel.f],tau,Re(y.sar$spec[1,1,sel.f,]))
# compute from quantile series
y.qser <- qser(cbind(y1,y2),tau)
y.sar <- qspec.sar(y.qser=y.qser,tau0=tau,p=1)
qfa.plot(ff[sel.f],tau,Re(y.sar$spec[1,1,sel.f,]))
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
# compute from time series
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
qfa.plot(ff[sel.f],tau,Re(y.sar$spec[1,1,sel.f,]))
# compute from quantile series
y.qser <- qser(cbind(y1,y2),tau)
y.sar <- qspec.sar(y.qser=y.qser,tau0=tau,p=1)
qfa.plot(ff[sel.f],tau,Re(y.sar$spec[1,1,sel.f,]))

Quantile Coherence Spectrum

Description

This function computes quantile coherence spectrum (QCOH) from quantile spectrum of multiple time series.

Usage

qspec2qcoh(qspec, k = 1, kk = 2)
qspec2qcoh(qspec, k = 1, kk = 2)

Arguments

qspec

array of quantile spectrum

k

index of first series (default = 1)

kk

index of second series (default = 2)

Value

matrix of quantile coherence evaluated at Fourier frequencies in (0,0.5)

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qacf <- qacf(cbind(y1,y2),tau)
y.qper.lw <- qspec.lw(y.qacf=y.qacf,M=5)$spec
y.qcoh <- qspec2qcoh(y.qper.lw,k=1,kk=2)
qfa.plot(ff[sel.f],tau,y.qcoh)
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
n <- length(y1)
ff <- c(0:(n-1))/n
sel.f <- which(ff > 0 & ff < 0.5)
y.qacf <- qacf(cbind(y1,y2),tau)
y.qper.lw <- qspec.lw(y.qacf=y.qacf,M=5)$spec
y.qcoh <- qspec2qcoh(y.qper.lw,k=1,kk=2)
qfa.plot(ff[sel.f],tau,y.qcoh)

Bootstrap Simulation of SAR Coefficients for Testing Equality of Granger-Causality in Two Samples

Description

This function simulates bootstrap samples of selected spline autoregression (SAR) coefficients for testing equality of Granger-causality in two samples based on their SAR models under H0: effect in each sample equals the average effect.

Usage

sar.eq.bootstrap(
  y.qser,
  fit,
  fit2,
  index = c(1, 2),
  nsim = 1000,
  method = c("ar", "sar"),
  refit = FALSE,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)
sar.eq.bootstrap(
  y.qser,
  fit,
  fit2,
  index = c(1, 2),
  nsim = 1000,
  method = c("ar", "sar"),
  refit = FALSE,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)

Arguments

y.qser

matrix or array of QSER from qser() or qspec.sar()$qser

fit

object of SAR model from qser2sar() or qspec.sar()$fit

fit2

object of SAR model for the other sample

index

a pair of component indices for multiple time series or a sequence of lags for single time series (default = c(1,2))

nsim

number of bootstrap samples (default = 1000)

method

method of residual calculation: "ar" (default) or "sar"

refit

if TRUE the models are refit under H0 (default=FALSE)

n.cores

number of cores for parallel computing (default = 1)

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

seed

seed for random sampling (default = 1234567)

Value

array of simulated bootstrap samples of selected SAR coefficients

Examples

y11 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y21 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y12 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y22 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y1.sar <- qspec.sar(cbind(y11,y21),tau0=tau,p=1)
y2.sar <- qspec.sar(cbind(y12,y22),tau0=tau,p=1)
A1.sim <- sar.eq.bootstrap(y1.sar$qser,y1.sar$fit,y2.sar$fit,index=c(1,2),nsim=5)
A2.sim <- sar.eq.bootstrap(y2.sar$qser,y2.sar$fit,y1.sar$fit,index=c(1,2),nsim=5)
y11 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y21 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y12 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y22 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y1.sar <- qspec.sar(cbind(y11,y21),tau0=tau,p=1)
y2.sar <- qspec.sar(cbind(y12,y22),tau0=tau,p=1)
A1.sim <- sar.eq.bootstrap(y1.sar$qser,y1.sar$fit,y2.sar$fit,index=c(1,2),nsim=5)
A2.sim <- sar.eq.bootstrap(y2.sar$qser,y2.sar$fit,y1.sar$fit,index=c(1,2),nsim=5)

Wald Test and Confidence Band for Equality of Granger-Causality in Two Samples

Description

This function computes Wald test and confidence band for equality of Granger-causality in two samples using bootstrap samples generated by sar.eq.bootstrap() based on the spline autoregression (SAR) models of quantile series (QSER).

Usage

sar.eq.test(A1, A1.sim, A2, A2.sim, sel.lag = NULL, sel.tau = NULL)
sar.eq.test(A1, A1.sim, A2, A2.sim, sel.lag = NULL, sel.tau = NULL)

Arguments

A1

matrix of selected SAR coefficients for sample 1

A1.sim

simulated bootstrap samples from sar.eq.bootstrap() for sample 1

A2

matrix of selected SAR coefficients for sample 2

A2.sim

simulated bootstrap samples from sar.eq.bootstrap() for sample 2

sel.lag

indices of time lags for Wald test (default = NULL: all lags)

sel.tau

indices of quantile levels for Wald test (default = NULL: all quantiles)

Value

a list with the following elements:

test

list of Wald test result containing wald and p.value

D.u

matrix of upper limits of 95% confidence band for A1 - A2

D.l

matrix of lower limits of 95% confidence band for A1 - A2

Examples

y11 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y21 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y12 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y22 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y1.sar <- qspec.sar(cbind(y11,y21),tau0=tau,p=1)
y2.sar <- qspec.sar(cbind(y12,y22),tau0=tau,p=1)
A1.sim <- sar.eq.bootstrap(y1.sar$qser,y1.sar$fit,y2.sar$fit,index=c(1,2),nsim=5)
A2.sim <- sar.eq.bootstrap(y2.sar$qser,y2.sar$fit,y1.sar$fit,index=c(1,2),nsim=5)
A1 <- sar.gc.coef(y1.sar$fit,index=c(1,2))
A2 <- sar.gc.coef(y2.sar$fit,index=c(1,2))
test <- sar.eq.test(A1,A1.sim,A2,A2.sim,sel.lag=NULL,sel.tau=NULL)
y11 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y21 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
y12 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y22 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y1.sar <- qspec.sar(cbind(y11,y21),tau0=tau,p=1)
y2.sar <- qspec.sar(cbind(y12,y22),tau0=tau,p=1)
A1.sim <- sar.eq.bootstrap(y1.sar$qser,y1.sar$fit,y2.sar$fit,index=c(1,2),nsim=5)
A2.sim <- sar.eq.bootstrap(y2.sar$qser,y2.sar$fit,y1.sar$fit,index=c(1,2),nsim=5)
A1 <- sar.gc.coef(y1.sar$fit,index=c(1,2))
A2 <- sar.gc.coef(y2.sar$fit,index=c(1,2))
test <- sar.eq.test(A1,A1.sim,A2,A2.sim,sel.lag=NULL,sel.tau=NULL)

Bootstrap Simulation of SAR Coefficients for Granger-Causality Analysis

Description

This function simulates bootstrap samples of selected spline autoregression (SAR) coefficients for Granger-causality analysis based on the SAR model of quantile series (QSER) under H0: (a) for multiple time series, the second series specified in index is not causal for the first series specified in index; (b) for single time series, the series is not causal at the lags specified in index.

Usage

sar.gc.bootstrap(
  y.qser,
  fit,
  index = c(1, 2),
  nsim = 1000,
  method = c("ar", "sar"),
  refit = FALSE,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)
sar.gc.bootstrap(
  y.qser,
  fit,
  index = c(1, 2),
  nsim = 1000,
  method = c("ar", "sar"),
  refit = FALSE,
  n.cores = 1,
  mthreads = FALSE,
  seed = 1234567
)

Arguments

y.qser

matrix or array of QSER from qser() or qspec.sar()$qser

fit

object of SAR model from qser2sar() or qspec.sar()$fit

index

a pair of component indices for multiple time series or a sequence of lags for single time series (default = c(1,2))

nsim

number of bootstrap samples (default = 1000)

method

method of residual calculation: "ar" (default) or "sar"

refit

if TRUE the SAR model is refit under H0 (default = FALSE)

n.cores

number of cores for parallel computing (default = 1)

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

seed

seed for random sampling (default = 1234567)

Value

array of simulated bootstrap samples of selected SAR coefficients

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A.sim <- sar.gc.bootstrap(y.sar$qser,y.sar$fit,index=c(1,2),nsim=5)
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A.sim <- sar.gc.bootstrap(y.sar$qser,y.sar$fit,index=c(1,2),nsim=5)

Extraction of SAR Coefficients for Granger-Causality Analysis

Description

This function extracts the spline autoregression (SAR) coefficients from an SAR model for Granger-causality analysis. See sar.gc.bootstrap for more details regarding the use of index.

Usage

sar.gc.coef(fit, index = c(1, 2))
sar.gc.coef(fit, index = c(1, 2))

Arguments

fit

object of SAR model from qser2sar() or qspec.sar()$fit

index

a pair of component indices for multiple time series or a sequence of lags for single time series (default = c(1,2))

Value

matrix of selected SAR coefficients (number of lags by number of quantiles)

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A <- sar.gc.coef(y.sar$fit,index=c(1,2))
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A <- sar.gc.coef(y.sar$fit,index=c(1,2))

Wald Test and Confidence Band for Granger-Causality Analysis

Description

This function computes Wald test and confidence band for Granger-causality using bootstrap samples generated by sar.gc.bootstrap() based the spline autoregression (SAR) model of quantile series (QSER).

Usage

sar.gc.test(A, A.sim, sel.lag = NULL, sel.tau = NULL)
sar.gc.test(A, A.sim, sel.lag = NULL, sel.tau = NULL)

Arguments

A

matrix of selected SAR coefficients

A.sim

simulated bootstrap samples from sar.gc.bootstrap()

sel.lag

indices of time lags for Wald test (default = NULL: all lags)

sel.tau

indices of quantile levels for Wald test (default = NULL: all quantiles)

Value

a list with the following elements:

test

list of Wald test result containing wald and p.value

A.u

matrix of upper limits of 95% confidence band of A

A.l

matrix of lower limits of 95% confidence band of A

Examples

y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A <- sar.gc.coef(y.sar$fit,index=c(1,2))
A.sim <- sar.gc.bootstrap(y.sar$qser,y.sar$fit,index=c(1,2),nsim=5)
y.gc <- sar.gc.test(A,A.sim)
y1 <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
y2 <- stats::arima.sim(list(order=c(1,0,0), ar=-0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sar <- qspec.sar(cbind(y1,y2),tau0=tau,p=1)
A <- sar.gc.coef(y.sar$fit,index=c(1,2))
A.sim <- sar.gc.bootstrap(y.sar$qser,y.sar$fit,index=c(1,2),nsim=5)
y.gc <- sar.gc.test(A,A.sim)

Spline Quantile Discrete Fourier Transform (SQDFT) of Time Series

Description

This function computes spline quantile discrete Fourier transform (SQDFT) for univariate or multivariate time series through trigonometric spline quantile regression.

Usage

sqdft(
  y,
  tau,
  tau0 = tau,
  spar = NULL,
  w = rep(1, length(tau0)),
  criterion = c("AIC", "BIC", "GIC"),
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  interval = NULL,
  all.knots = FALSE,
  control = list(),
  n.cores = 1,
  cl = NULL
)
sqdft(
  y,
  tau,
  tau0 = tau,
  spar = NULL,
  w = rep(1, length(tau0)),
  criterion = c("AIC", "BIC", "GIC"),
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  interval = NULL,
  all.knots = FALSE,
  control = list(),
  n.cores = 1,
  cl = NULL
)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter, selected automatically by criterion if spar = NULL or if length(spar) > 1

w

weight sequence in penalty (default = rep(1,length(tau0)))

criterion

criterion for smoothing parameter selection: "AIC" (default), "BIC", or "GIC"

method

'sqr'(default), 'sqr1', or 'sqr3'

ztol

zero-tolerance parameter to determine the model complexity (default = NULL: set internally to 1e-5 for SQR and SQR1 or 1e-4 for SQR3)

solver

'fnb' or 'sfn' for SQR and SQR1; 'piqp' or 'osqp' for SQR3 (default = NULL: set internally to 'fnb' for SQR and SQR1 or 'piqp' for SQR3)

interval

interval for spar optimization (default: c(-1.5,1.5) for SQR and SQR1 or c(0,2.5) for SQR3)

all.knots

TRUE or FALSE (default), as in stats::smooth.spline()

control

list of control parameters for QP solvers 'piqp' and 'osqp' (default = list())

n.cores

number of cores for parallel computing (default = 1)

cl

pre-existing cluster for repeated parallel computing (default = NULL)

Value

A list with the following elements:

coefficients

matrix of regression coefficients

qdft

matrix or array of the spline quantile discrete Fourier transform of y

crit

criteria for smoothing parameter selection: (AIC,BIC,GIC)

nit

maximum number of iterations

spar

optimal value of smoothing parameter

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sqdft <- sqdft(y,tau,spar=0.2,method="sqr1")$qdft
plot(y.sqdft[,2])
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
y.sqdft <- sqdft(y,tau,spar=0.2,method="sqr1")$qdft
plot(y.sqdft[,2])

Spline Quantile Discrete Fourier Transform (SQDFT) of Time Series Given Smoothing Parameter

Description

This function computes spline quantile discrete Fourier transform (SQDFT) for univariate or multivariate time series through trigonometric spline quantile regression with user-supplied spar.

Usage

sqdft.fit(
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  all.knots = FALSE,
  control = list(),
  n.cores = 1,
  cl = NULL
)
sqdft.fit(
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  all.knots = FALSE,
  control = list(),
  n.cores = 1,
  cl = NULL
)

Arguments

y

vector or matrix of time series (if matrix, nrow(y) = length of time series)

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter (default = 1)

w

weight sequence in penalty (default = rep(1,length(tau0)))

method

'sqr' (default), "sqr1", or 'sqr3'

ztol

zero-tolerance parameter to determine the model complexity (default = NULL: set internally to 1e-5 for SQR and SQR1 or 1e-4 for SQR3)

solver

'fnb' or 'sfn' for SQR and SQR1; 'piqp' or 'osqp' for SQR3 (default = NULL: set internally to 'fnb' for SQR and SQR1 or 'piqp' for SQR3)

all.knots

TRUE or FALSE (default), as in stats::smooth.spline()

control

list of control parameters for QP solvers 'piqp' and 'osqp' (default = list())

n.cores

number of cores for parallel computing (default = 1)

cl

pre-existing cluster for repeated parallel computing (default = NULL)

Value

A list with the following elements:

coefficients

matrix of regression coefficients

qdft

matrix or array of the spline quantile discrete Fourier transform of y

crit

criteria for smoothing parameter selection: (AIC,BIC,GIC)

nit

maximum number of iterations

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
tau0 <- seq(0.1,0.9,0.2)
y.sqdft <- sqdft.fit(y,tau,tau0=tau0,spar=0.2,method="sqr1")$qdft
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
tau0 <- seq(0.1,0.9,0.2)
y.sqdft <- sqdft.fit(y,tau,tau0=tau0,spar=0.2,method="sqr1")$qdft

Spline Quantile Regression by Formula

Description

This function computes the spline quantile regression solution SQR, SQR1, or SAR3 on a given set of quantile levels from a regression formula with or without user-supplied smoothing parameter. SQR represents the regression coefficients as cubic spline functions of the quantile level and employs the L1-norm of their second derivative as roughness penalty. SQR1 represents the regression coefficients as linear spline functions and employs the total variation of their first derivatives as roughness penalty. SQR3 also represents the regression coefficients as cubic spline functions but employs the L2-norm of their second derivatives as roughness penalty. SQR and SQR1 are solved as linear program (LP) using sqr.fit() and sqr1.fit(), respectively. SQR3 is solved as quadratic program (QP) using sqr3.fit().

Usage

sqr(
  formula,
  tau = seq(0.1, 0.9, 0.2),
  tau0 = tau,
  spar = NULL,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  criterion = c("AIC", "BIC", "GIC"),
  method = c("sqr", "sqr1", "sqr3"),
  type = c("dual", "primal"),
  ztol = NULL,
  solver = NULL,
  interval = NULL,
  npar = c(1, 2),
  all.knots = FALSE,
  control = list(),
  data,
  subset,
  na.action,
  model = TRUE
)
sqr(
  formula,
  tau = seq(0.1, 0.9, 0.2),
  tau0 = tau,
  spar = NULL,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  criterion = c("AIC", "BIC", "GIC"),
  method = c("sqr", "sqr1", "sqr3"),
  type = c("dual", "primal"),
  ztol = NULL,
  solver = NULL,
  interval = NULL,
  npar = c(1, 2),
  all.knots = FALSE,
  control = list(),
  data,
  subset,
  na.action,
  model = TRUE
)

Arguments

formula

formula object, with the response on the left of a ~ operator, and the terms, separated by + operators, on the right.

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter, selected automatically by criterion if spar = NULL (default)

w

weight sequence in penalty (default = rep(1,length(tau0)))

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

criterion

criterion for smoothing parameter selection ("AIC", "BIC", or "GIC")

method

'sqr' (default), 'sqr1', or 'sqr3'

type

type of QP formulation for SQR3: 'dual' (default) or 'primal'

ztol

zero-tolerance parameter to determine the model complexity (default = NULL: set internally to 1e-5 for SQR and SQR1 or 1e-4 for SQR3)

solver

'fnb' or 'sfn' for SQR and SQR1; 'piqp' or 'osqp' for SQR3 (default = NULL: set internally to 'fnb' for SQR and SQR1 or 'piqp' for SQR3)

interval

interval for spar optimization (default: c(-1.5,1.5) for SQR and SQR1 or c(1.0,2.5) for SQR3)

npar

experimental parameter (default = 1)

all.knots

TRUE or FALSE (default), same as in stats::smooth.spline()

control

list of control parameters for QP solvers 'piqp' and 'osqp'

data

data.frame object containing the observations

subset

an optional vector specifying a subset of observations to be used

na.action

a function to filter missing data (see rq() in the 'quantreg' package)

model

if TRUE then the model frame is returned (needed for calling summary subsequently)

Value

object of sqr.fit(), sqr1.fit(), or sqr3.fit()

Examples

library(quantreg)
data(engel)
engel$income <- (engel$income - mean(engel$income))/1000
tau <- seq(0.1,0.9,0.05)
fit <- rq(foodexp ~ income,tau=tau,data=engel)
fit.sqr1 <- sqr(foodexp ~ income,tau=tau,spar=0.5,method="sqr1",data=engel)
fit.sqr3 <- sqr(foodexp ~ income,tau=tau,spar=0.5,method="sqr3",data=engel)
par(mfrow=c(1,1),pty="m",lab=c(10,10,2),mar=c(4,4,2,1)+0.1,las=1)
plot(tau,fit$coef[2,],xlab="Quantile Level",ylab="Coeff1")
lines(tau,fit.sqr1$coef[2,])
lines(tau,fit.sqr3$coef[2,],col=2)
library(quantreg)
data(engel)
engel$income <- (engel$income - mean(engel$income))/1000
tau <- seq(0.1,0.9,0.05)
fit <- rq(foodexp ~ income,tau=tau,data=engel)
fit.sqr1 <- sqr(foodexp ~ income,tau=tau,spar=0.5,method="sqr1",data=engel)
fit.sqr3 <- sqr(foodexp ~ income,tau=tau,spar=0.5,method="sqr3",data=engel)
par(mfrow=c(1,1),pty="m",lab=c(10,10,2),mar=c(4,4,2,1)+0.1,las=1)
plot(tau,fit$coef[2,],xlab="Quantile Level",ylab="Coeff1")
lines(tau,fit.sqr1$coef[2,])
lines(tau,fit.sqr3$coef[2,],col=2)

Plot of Derivative of Spline Quantile Regression Coefficients

Description

This function plots the derivative of one or all coefficients from spline quantile regression (SQR) together with the corresponding first difference from quantile regression (QR) on a given sequence of quantiles against the quantile level.

Usage

sqr_deriv.plot(
  fit,
  fit.sqr,
  fit.sqrb = NULL,
  deriv.sim = NULL,
  type = "l",
  typeb = "s",
  var.names = NULL,
  lty = c(1, 2),
  lwd = c(1.5, 1),
  cex = 0.25,
  pch = 1,
  col = c(2, 1),
  idx = NULL,
  Ylim = NULL,
  xlim = NULL,
  plot.zero = TRUE,
  plot.ci = TRUE,
  set.par = TRUE,
  mfrow = NULL,
  lab = c(10, 7, 7),
  mar = c(2, 3, 2, 1) + 0.1,
  las = 1
)
sqr_deriv.plot(
  fit,
  fit.sqr,
  fit.sqrb = NULL,
  deriv.sim = NULL,
  type = "l",
  typeb = "s",
  var.names = NULL,
  lty = c(1, 2),
  lwd = c(1.5, 1),
  cex = 0.25,
  pch = 1,
  col = c(2, 1),
  idx = NULL,
  Ylim = NULL,
  xlim = NULL,
  plot.zero = TRUE,
  plot.ci = TRUE,
  set.par = TRUE,
  mfrow = NULL,
  lab = c(10, 7, 7),
  mar = c(2, 3, 2, 1) + 0.1,
  las = 1
)

Arguments

fit

output of quantreg::rq() for the QR fit

fit.sqr

output of sqr() for the primary SQR fit

fit.sqrb

output of sqr() for the secondary SQR fit (default = NULL)

deriv.sim

output derivatives of boot.sqr() for the primary SQR (default = NULL)

type

as type parameter in plot() for the primary SQR (default = "l")

typeb

as type parameter in plot() for the secondary SQR (default = "s")

var.names

user-supplied names of regression coefficients (including the intercept)

lty

line types for the primary and secondary SQR (default = c(1,2))

lwd

line widths for the primary and secondary SQR (default = c(1.5,1))

cex

as cex parameter in plot()

pch

as pch parameter in plot() for QR(default = 1)

col

line colors for the primary and secondary SQR (default = c(2,1))

idx

index of individual coefficient to be plotted (default = NULL)

Ylim

user-supplied matrix of ylim for each coefficient (default = NULL)

xlim

user-supplied xlim (default = NULL)

plot.zero

TRUE or FALSE (default): if TRUE, add zero line

plot.ci

TRUE (default) or FALSE: if TRUE, add 90-percent bootstrap CI for the primary SQR when deriv.sim is supplied

set.par

TRUE (default) or FALSE if TRUE, reset par()

mfrow

parameter for resetting par() (default = NULL)

lab

parameter for resetting par() (default = c(10,7,7))

mar

parameter for resetting par() (default = c(2,3,2,1)+0.1)

las

parameter for resetting par() (default = 1)

Value

Ylim (invisible)

Cubic Spline Quantile Regression with L1-Norm Roughness Penalty (SQR)

Description

This function computes spline quantile regression solution with cubic splines and L1-norm roughness penalty (SQR) from the response vector and the design matrix on a given set of quantile levels. It uses the FORTRAN code rqfnb.f in the "quantreg" package with the kind permission of Dr. R. Koenker or the R function rq.fit.sfn() in the same package as a sparse-matrix alternative. Both solve the SQR problem as a linear program (LP).

Usage

sqr.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = TRUE,
  ztol = 1e-05,
  solver = c("fnb", "sfn"),
  all.knots = FALSE
)
sqr.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = TRUE,
  ztol = 1e-05,
  solver = c("fnb", "sfn"),
  all.knots = FALSE
)

Arguments

X

design matrix (requirement: nrow(X) = length(y))

y

response vector

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter (default = 1)

w

weight sequence in penalty (default = rep(1,length(tau0)))

mthreads

if FALSE, set RhpcBLASctl::blas_set_num_threads(1) (default = TRUE)

ztol

zero-tolerance parameter to determine the model complexity (default = 1e-05)

solver

LP solver: 'fnb' (defaut) or 'sfn'

all.knots

TRUE or FALSE (default), same as in smooth.spline()

Value

A list with the following elements:

coefficients

matrix of regression coefficients

derivatives

matrix of derivatives of regression coefficients

crit

criteria values for spar selection: (AIC,BIC,GIC)

np

sequence of complexity measure as the number of effective parameters

fid

sequence of fidelity measure as the quasi-likelihood

info

convergence status

nit

number of iterations

K

number of spline basis functions

Cubic Spline Quantile Regression with L1-Norm Roughness Penalty (SQR) Computed by Gradient Algorithms

Description

This function computes spline quantile regression with cubic splines and L1-norm roughness penalty by a gradient algorithm BFGS, ADAM, or GRAD.

Usage

sqr.fit.optim(
  X,
  y,
  tau,
  spar = 0,
  d = 1,
  weighted = FALSE,
  method = c("BFGS", "ADAM", "GRAD"),
  beta.rq = NULL,
  theta0 = NULL,
  spar0 = NULL,
  sg.rate = c(1, 1),
  mthreads = TRUE,
  control = list(trace = 0)
)
sqr.fit.optim(
  X,
  y,
  tau,
  spar = 0,
  d = 1,
  weighted = FALSE,
  method = c("BFGS", "ADAM", "GRAD"),
  beta.rq = NULL,
  theta0 = NULL,
  spar0 = NULL,
  sg.rate = c(1, 1),
  mthreads = TRUE,
  control = list(trace = 0)
)

Arguments

X

vecor or matrix of explanatory variables (including intercept)

y

vector of dependent variable

tau

sequence of quantile levels in (0,1)

spar

smoothing parameter

d

subsampling rate of quantile levels (default = 1)

weighted

if TRUE, penalty function is weighted (default = FALSE)

method

optimization method: "BFGS" (default), "ADAM", or "GRAD"

beta.rq

matrix of regression coefficients from quantreg::rq(y~X) for initialization (default = NULL)

theta0

initial value of spline coefficients (default = NULL)

spar0

smoothing parameter for stats::smooth.spline() to smooth beta.rq for initilaiztion (default = NULL)

sg.rate

vector of sampling rates for quantiles and observations in stochastic gradient version of GRAD and ADAM

mthreads

if FALSE, set RhpcBLASctl::blas_set_num_threads(1) (default = TRUE)

control

list of control parameters

maxit:: max number of iterations (default = 100)
stepsize:: stepsize for ADAM and GRAD (default = 0.01)
warmup:: length of warmup phase for ADAM and GRAD (default = 70)
stepupdate:: frequency of update for ADAM and GRAD (default = 20)
stepredn:: stepsize discount factor for ADAM and GRAD (default = 0.2)
line.search.type:: line search option (1,2,3,4) for GRAD (default = 1)
line.search.max:: max number of line search trials for GRAD (default = 1)
seed:: seed for stochastic version of ADAM and GRAD (default = 1000)
trace:: -1 return results from all iterations, 0 (default) return final result

Value

A list with the following elements:

beta

matrix of regression coefficients

all.beta

coefficients from all iterations for GRAD and ADAM

spars

smoothing parameters from stats::smooth.spline() for initialization

fit

object from the optimization algorithm

Examples

data(engel)
y <- engel$foodexp
X <- cbind(rep(1,length(y)),engel$income-mean(engel$income))
tau <- seq(0.1,0.9,0.05)
fit.rq <- quantreg::rq(y ~ X[,2],tau)
fit.sqr <- sqr(y ~ X[,2],tau,spar=0.2)
fit <- sqr.fit.optim(X,y,tau,spar=0.2,d=2,method="BFSG",beta.rq=fit.rq$coef)
fit <- sqr.fit.optim(X,y,tau,spar=0.2,d=2,method="BFSG",beta.rq=fit.rq$coef)
par(mfrow=c(1,2),pty="m",lab=c(10,10,2),mar=c(4,4,2,1)+0.1,las=1)
for(j in c(1:2)) {
  plot(tau,fit.rq$coef[j,],type="n",xlab="QUANTILE LEVEL",ylab=paste0("COEFF",j))
  points(tau,fit.rq$coef[j,],pch=1,cex=0.5)
  lines(tau,fit.sqr$coef[j,],lty=1); lines(tau,fit$beta[j,],lty=2,col=2)
}
data(engel)
y <- engel$foodexp
X <- cbind(rep(1,length(y)),engel$income-mean(engel$income))
tau <- seq(0.1,0.9,0.05)
fit.rq <- quantreg::rq(y ~ X[,2],tau)
fit.sqr <- sqr(y ~ X[,2],tau,spar=0.2)
fit <- sqr.fit.optim(X,y,tau,spar=0.2,d=2,method="BFSG",beta.rq=fit.rq$coef)
fit <- sqr.fit.optim(X,y,tau,spar=0.2,d=2,method="BFSG",beta.rq=fit.rq$coef)
par(mfrow=c(1,2),pty="m",lab=c(10,10,2),mar=c(4,4,2,1)+0.1,las=1)
for(j in c(1:2)) {
  plot(tau,fit.rq$coef[j,],type="n",xlab="QUANTILE LEVEL",ylab=paste0("COEFF",j))
  points(tau,fit.rq$coef[j,],pch=1,cex=0.5)
  lines(tau,fit.sqr$coef[j,],lty=1); lines(tau,fit$beta[j,],lty=2,col=2)
}

Plot of Spline Quantile Regression Coefficients

Description

This function plots one or all regression coefficients from quantile regression (QR) and spline quantile regression (SQR) on a given sequence of quantiles against the quantile level.

Usage

sqr.plot(
  summary.rq,
  summary.sqr = NULL,
  summary.sqrb = NULL,
  coef.sim = NULL,
  type = "l",
  lty = c(1, 2),
  lwd = c(1.5, 1),
  cex = 0.25,
  pch = 1,
  col = c(2, 1),
  idx = NULL,
  plot.rq = TRUE,
  plot.rq.line = FALSE,
  plot.zero = FALSE,
  plot.ls = TRUE,
  plot.ci = TRUE,
  var.names = NULL,
  Ylim = NULL,
  xlim = NULL,
  set.par = TRUE,
  mfrow = NULL,
  lab = c(10, 7, 7),
  mar = c(2, 3, 2, 1) + 0.1,
  las = 1
)
sqr.plot(
  summary.rq,
  summary.sqr = NULL,
  summary.sqrb = NULL,
  coef.sim = NULL,
  type = "l",
  lty = c(1, 2),
  lwd = c(1.5, 1),
  cex = 0.25,
  pch = 1,
  col = c(2, 1),
  idx = NULL,
  plot.rq = TRUE,
  plot.rq.line = FALSE,
  plot.zero = FALSE,
  plot.ls = TRUE,
  plot.ci = TRUE,
  var.names = NULL,
  Ylim = NULL,
  xlim = NULL,
  set.par = TRUE,
  mfrow = NULL,
  lab = c(10, 7, 7),
  mar = c(2, 3, 2, 1) + 0.1,
  las = 1
)

Arguments

summary.rq

output of summary() for the QR fit from quantreg::rq()

summary.sqr

output of summary() for the primary SQR fit from sqr() (defalty = NULL)

summary.sqrb

output of summary() for the secondary SQR fit from sqr() (default = NULL)

coef.sim

output coefficients of boot.sqr() for the primary SQR (default = NULL)

type

as type parameter in plot() for plotting SQR (default = "l")

lty

line types for the primary and secondary SQR (default = c(1,2))

lwd

line widths for the primary and secondary SQR (default = c(1.5,1))

cex

as cex parameter in plot()

pch

as pch parameter in plot() for the QR (default = 1)

col

line colors for the primary and secondary SQR (default = c(2,1))

idx

index of individual coefficient to be plotted (default = NULL)

plot.rq

TRUE (default) or FALSE: if TRUE, plot QR as points

plot.rq.line

TRUE or FALSE (default): if TRUE, add line plot of QR

plot.zero

TRUE or FALSE (default): if TRUE, add zero line

plot.ls

TRUE (default) or FALSE: if TRUE, add least-square estimate with 90-percent CI

plot.ci

TRUE (default) or FALSE: if TRUE, add 90-percent bootstrap CI for the primary SQR when coef.sim is supplied or approximate CI when coef.sim = NULL

var.names

user-supplied names of regression coefficients (including the intercept)

Ylim

user-supplied matrix of ylim for each coefficient (default = NULL)

xlim

user-supplied xlim (default = NULL)

set.par

TRUE (default) or FALSE if TRUE, reset par()

mfrow

parameter for resetting par() (default = NULL)

lab

parameter for resetting par() (default = c(10,7,7))

mar

parameter for resetting par() (default = c(2,3,2,1)+0.1)

las

parameter for resetting par() (default = 1)

Value

Ylim (invisible)

Linear Spline Quantile Regression with Total-Variation Roughness Penalty (SQR1 or Linear SQR)

Description

This function computes spline quantile regression with linear splines and total-variation roughness penalty (SQR1 or linear SQR) from the response vector and the design matrix on a given set of quantile levels. It uses the FORTRAN code rqfnb.f in the "quantreg" package with the kind permission of Dr. R. Koenker or the R function rq.fit.sfn() in the same package as a sparse-matrix alternative. Both solve the SQR1 problem as a linear program (LP).

Usage

sqr1.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0) - 1),
  mthreads = FALSE,
  ztol = 1e-05,
  solver = c("fnb", "sfn"),
  npar = c(1, 2),
  all.knots = FALSE
)
sqr1.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0) - 1),
  mthreads = FALSE,
  ztol = 1e-05,
  solver = c("fnb", "sfn"),
  npar = c(1, 2),
  all.knots = FALSE
)

Arguments

X

design matrix (nrow(X) = length(y))

y

response vector

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter (default = 1)

w

weight sequence in penalty (default = rep(1,length(tau0)-1))

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

ztol

zero-tolerance parameter to determine the model complexity (default = 1e-05)

solver

LP solver: 'fnb' (defaut) or 'sfn'

npar

experimental parameter (default = 1)

all.knots

TRUE or FALSE (default), same as in stats::smooth.spline()

Value

A list with the following elements:

coefficients

matrix of regression coefficients

derivatives

matrix of derivatives of regression coefficients

crit

sequence critera for smoothing parameter select: (AIC,BIC,GIC)

np

sequence of complexity measure as the number of effective parameters

fid

sequence of fidelity measure as the quasi-likelihood

nit

number of iterations

info

convergence status

K

number of spline basis functions

Cubic Spline Quantile Regression with L2-Norm Roughness Penalty (SQR3 or Cubic SQR)

Description

This function computes spline quantile regression with cubic splines and L2-norm roughness penalty (SQR3 or cubic SQR) from the response vector and the design matrix on a given set of quantile levels. It uses solve_osqp() in the "osqp" package or solve_piqp() in the "piqp" package. Both are general-purpose quadratic program (QP) solvers in the sparse-matrix form.

Usage

sqr3.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  ztol = 1e-04,
  type = c("dual", "primal"),
  solver = c("piqp", "osqp"),
  npar = c(1, 2),
  all.knots = FALSE,
  control = list()
)
sqr3.fit(
  X,
  y,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  ztol = 1e-04,
  type = c("dual", "primal"),
  solver = c("piqp", "osqp"),
  npar = c(1, 2),
  all.knots = FALSE,
  control = list()
)

Arguments

X

design matrix (requirement: nrow(X) = length(y))

y

response vector

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter (default = 1)

w

weight sequence in penalty (default = rep(1,length(tau0)))

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

ztol

zero-tolerance parameter to determine the model complexity (default = 1e-04)

type

type of QP formulation: 'dual' (default) or 'primal'

solver

QP solver: 'piqp' (default) or 'osqp'

npar

experimental parameter (default = 1)

all.knots

TRUE or FALSE (default) as in stats::smooth.spline()

control

list of control parameters for the QP solver (default = list())

Value

A list with the following elements:

coefficients

matrix of regression coefficients

derivatives

matrix of derivatives of regression coefficients

crit

sequence critera for smoothing parameter select: (AIC,BIC,GIC)

np

sequence of complexity measure as the number of effective parameters

fid

sequence of fidelity measure as the quasi-likelihood

info

convergence status

nit

number of iterations

K

number of spline basis functions

Trigonometric Quantile Regression (TQR)

Description

This function computes ordinary trigonometric quantile regression (TQR) for univariate time series at a single frequency.

Usage

tqr.fit(y, f0, tau, prepared = TRUE)
tqr.fit(y, f0, tau, prepared = TRUE)

Arguments

y

vector of time series

f0

frequency in [0,1)

tau

sequence of quantile levels in (0,1)

prepared

if TRUE, intercept is removed and coef of cosine is doubled when f0 = 0.5

Value

object of quantreg::rq()

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
fit <- tqr.fit(y,f0=0.1,tau=tau)
plot(tau,fit$coef[1,],type='o',pch=0.75,xlab='QUANTILE LEVEL',ylab='TQR COEF')
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
fit <- tqr.fit(y,f0=0.1,tau=tau)
plot(tau,fit$coef[1,],type='o',pch=0.75,xlab='QUANTILE LEVEL',ylab='TQR COEF')

Trigonometric Spline Quantile Regression (TSQR) of Time Series

Description

This function computes trigonometric spline quantile regression (TSQR) for univariate time series at a single frequency using sqr.fit(), sqr1.fit(), or sqr3.fit().

Usage

tsqr.fit(
  y,
  f0,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  prepared = TRUE,
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  all.knots = FALSE,
  control = list()
)
tsqr.fit(
  y,
  f0,
  tau,
  tau0 = tau,
  spar = 1,
  w = rep(1, length(tau0)),
  mthreads = FALSE,
  prepared = TRUE,
  method = c("sqr", "sqr1", "sqr3"),
  ztol = NULL,
  solver = NULL,
  all.knots = FALSE,
  control = list()
)

Arguments

y

time series

f0

frequency in [0,1)

tau

sequence of quantile levels for evaluation

tau0

sequence of quantile levels for fitting (min(tau0) <= tau <= max(tau0); default = tau)

spar

smoothing parameter (default = 1)

w

weight sequence in penalty (default = rep(1,length(tau0)))

mthreads

if FALSE (default), set RhpcBLASctl::blas_set_num_threads(1)

prepared

if TRUE, intercept is removed and coef of cosine is doubled when f0 = 0.5

method

'sqr' (default), "sqr1", or 'sqr3'

ztol

a zero tolerance paramete to determine the model complexity (default = NULL: set internally to 1e-5 for SQR and SQR1 or 1e-4 for SQR3)

solver

'fnb' or 'sfn' for SQR and SQR1; 'piqp' or 'osqp' for SQR3 (default = NULL: set internally to 'fnb'for SQR and SQR1 or 'piqp' for SQR3)

all.knots

TRUE or FALSE (default), as in stats::smooth.spline()

control

list of control parameters for QP solvers 'piqp' and 'osqp'

Value

object of sqr.fit(), sqr1.fit(), or sqr3.fit()

Examples

y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
tau0 <- seq(0.1,0.9,0.2)
fit <- tqr.fit(y,f0=0.1,tau=tau)
fit.sqr1 <- tsqr.fit(y,f0=0.1,tau=tau,tau0=tau0,spar=0.2,method='sqr1')
fit.sqr3 <- tsqr.fit(y,f0=0.1,tau=tau,tau0=tau0,spar=1,method='sqr3')
plot(tau,fit$coef[1,],type='p',xlab='QUANTILE LEVEL',ylab='TQR COEF')
lines(tau,fit.sqr1$coef[1,])
lines(tau,fit.sqr3$coef[1,],col=2)
y <- stats::arima.sim(list(order=c(1,0,0), ar=0.5), n=64)
tau <- seq(0.1,0.9,0.05)
tau0 <- seq(0.1,0.9,0.2)
fit <- tqr.fit(y,f0=0.1,tau=tau)
fit.sqr1 <- tsqr.fit(y,f0=0.1,tau=tau,tau0=tau0,spar=0.2,method='sqr1')
fit.sqr3 <- tsqr.fit(y,f0=0.1,tau=tau,tau0=tau0,spar=1,method='sqr3')
plot(tau,fit$coef[1,],type='p',xlab='QUANTILE LEVEL',ylab='TQR COEF')
lines(tau,fit.sqr1$coef[1,])
lines(tau,fit.sqr3$coef[1,],col=2)

Yearly sunspot numbers v1.0

Description

Yearly mean total sunspot numbers from 1700 to 2007.

Usage

data(yearssn)
data(yearssn)

Format

An object of class data.frame with 308 rows and 2 columns.

Source

Original data https://www.sidc.be/SILSO/versionarchive from WDC-SILSO, Royal Observatory of Belgium, Brussels.

References

WDC-SILSO, Royal Observatory of Belgium, Brussels. doi:10.24414/stcz-kc45

Yearly sunspot numbers v2.0

Description

Yearly mean total sunspot numbers from 1700 to 2024. A revised version replacing v1.0.

Usage

data(yearssn2024)
data(yearssn2024)

Format

An object of class data.frame with 325 rows and 2 columns.

Source

Original data https://www.sidc.be/SILSO/datafiles from WDC-SILSO, Royal Observatory of Belgium, Brussels.

References

WDC-SILSO, Royal Observatory of Belgium, Brussels. doi:10.24414/qnza-ac80

Package 'qfa'

Help Index

Birth data

Description

Usage

Format

Source

References

Bootstrap of Spline Quantile Regression (SQR) Coefficients

Description

Usage

Arguments

Value

Engel's food expenditure data

Description

Usage

Format

References

Financial index

Description

Usage

Format

Source

Periodogram (PER)

Description

Usage

Arguments

Value

Examples

Quantile Autocovariance Function (QACF)

Description

Usage

Arguments

Value

Examples

Quantile-Crossing Series (QCSER)

Description

Usage

Arguments

Value

Examples

Quantile Discrete Fourier Transform (QDFT)

Description

Usage

Arguments

Value

Examples

Quantile Autocovariance Function (QACF)

Description

Usage

Arguments

Value

Examples

Quantile Periodogram (QPER)

Description

Usage

Arguments

Value

Examples

Quantile Series (QSER)

Description

Usage

Arguments

Value

Examples

Quantile-Frequency Plot

Description

Usage

Arguments

Value

Kullback-Leibler Divergence of Quantile Spectral Estimate

Description

Usage

Arguments

Value

Quantile Periodogram (QPER)

Description

Usage

Arguments

Value