See exts for details. Here is an example:

Experimental Time Series Features - Scroll to RACF PACF

See slides sent on slack.

Add function similar to the one in exts. Should look like:

Function estimate:

• AR
• MA
• ARMA
• ARIMA
• SARMA
• SARIMA

I suggest we remove the PACF() function and put the option "pacf = FALSE" in the auto_corr() function (since we removed ACF()).

@dukezhang007

• #1
• #3
• #4

@Yuming-Zhang:

• Review README.Rmd
• Review vignettes.Rmd
• #5
• #7
• #9
• #10
• Update README.Rmd and vignettes.Rmd as new features become available.
• Add brief description for the GitHub repo.

@robertomolinari

• Review README.Rmd
• Review vignettes.Rmd

@munsheet

• Review README.Rmd
• Review vignettes.Rmd
• Fix make_frame function
• #11

@rfaris253

• #12
• #6
• #8

Adapt the select function to the following models:

• MA
• ARMA
• ARIMA
• SAR
• SMA
• SARMA
• SARIMA

Add some kind of diagnostic plot based on this experimental feature of "exts". The input of this function should be either a time series (arma or non state space gmwm) or linear regression model. Below is an example:

There is a bug with the lags and xlab when freq != 1 for the functions mentioned in the title. Here is an example:

# Load savingrt dataset
data("savingrt")

# Simulate based on data
savingrt = gts(as.vector(savingrt), start = 1959, freq = 12, unit_ts = "%", 
            name_ts = "Saving Rates", data_name = "US Personal Saving Rates",
            unit_time = "month")

# Plot savingrt simulation
plot(savingrt)

# Compute ACF and plot result
savingrt_acf = ACF(savingrt)
plot(savingrt_acf)

• Prediction and residuals for state space models (@lionelvoirol )
• Structure output of kalman_filter function in a list of matrix w/ forecast, filter and smooth (@lionelvoirol )
• Discuss the functions AIC.fitsimts and evaluate in vignette(s) and textbook.
• Add possibility to change title in Portmanteau Tests plot (see function diag_ljungbox and Figure 4.15).
• Add summary.fitsimts (connect to gmwm).
• Add possibility to compute CI by parametric bootstrap, block bootstrap and seeder.
• Add news.md file
• Add confidence intervals for parameters estimated with gmwm and rgmwm.
• Add compatibility with RW, WN, QN and DR processes (as well as sum of latent processes) for the function estimate used with option gmwm and rgmwm.
• Add possibility of simulating non-Gaussian time series (i.e. extent ts objects to contain some info on the distribution of the residuals, for example AR(1) would be Gaussian AR(1) while AR(1, df = 5) would an AR(1) with t-distributed residuals with 5 df.
• Add an example of an unevenly spaced time series in the README and/or vignette. The code may need to be adapted.
• The select function should be made compatible with the following models:
  • SAR and SMA
  • SARMA
  • SARIMA
• Improve the function RW3dimensions. In particular, these parameters should be added as inputs: couleur = "blue4", xlab = "X-position", ylab = "Y-position", main = NULL, pt_col = NULL, pt_pch = 16, pt.cex = 2, leg_pos = NULL.
• Add MAPE model selection features (See exts for details. Here is an example based on sunspot dataset, slide 38 - ARMAmodels.pdf)

• Add a check that supplied object is of class gts or fitsimts (e.g. predict() function in the stat environment takes non-compatible objects and can be used instead of the predict() function in simts)
• Correct kernel density of the residual hist can go outside of the graph

• Remove negative values (y axis) of Ljung-Box diagnostic plot see e.g.

n = 1000
model = AR(phi = c(0.9, -0.5), sigma2 = 1)
Xt = gen_gts(n, model, freq = 4)
plot(Xt)

model1 = estimate(AR(1), Xt)
predict(model1, n.ahead = 10)

check(model1)

• Add diagnostic function based on wv. Suggested usage (probably after the release of wv package:

plot(wvar(res), main = "Haar WVar Representation", legend_position = NA)
  
  sigma2 = rep(var(res), length(wvar(res)$scales))
  points(wvar(res)$scales, sigma2/as.numeric(wvar(res)$scales), col = "orange", pch=0, cex=2)
  lines(wvar(res)$scales, sigma2/as.numeric(wvar(res)$scales), col = "orange", lty = 1)
  
  # add legend
  if (wvar(res)$robust == TRUE){
    wv_title_part1 = "Empirical Robust WV "
  }else{
    wv_title_part1 = "Empirical WV "
  }
  
  CI_conf = 1 - wvar(res)$alpha
  
  legend("bottomleft", 
         legend = c(as.expression(bquote(paste(.(wv_title_part1), hat(nu)^2))), 
                    as.expression(bquote(paste("CI(",hat(nu)^2,", ",.(CI_conf),")"))),
                    "WV implied by WN"),
         pch = c(16, 15, 0), lty = c(1, NA, 1), 
         col = c("darkblue", hcl(h = 210, l = 65, c = 100, alpha = 0.2), "orange"), 
         cex = 1, pt.cex = c(1.25, 3, 1.25), bty = "n")

`n = 1000
model = AR(phi = c(0.9, -0.5), sigma2 = 1)
Xt = gen_gts(n, model, freq = 4)
plot(Xt)

model1 = estimate(AR(1), Xt)
predict(model1, n.ahead = 10)

check(model1)`

The formatting of corr_analysis is a bit off. In particular the spacing between graph is too big. This should be controlled by some input to the function. Could you please try to improve this? Below is an example:

Hi,

I think make_frame works very well when we add plots on it. The only issue I find out for this moment is that, make_frame seems to have some issue with the location of legend of the original plot.

For example, please try the below code:
Xt = gen_gts(300, AR(phi = c(0, 0, 0.8), sigma2 = 1))
model = arima(Xt, order = c(3,0,0), include.mean = T)
resid_plot(Xt, model, std = FALSE, type = "hist")

As you can see the plot will have legend in the very topright, which should be included inside the white space.

@munsheet Do you think we can modify the make_frame function a bit to have it compatible with legends? Please let me know if I can help in any way! Thanks a lot! :)

Add parameters to the functions mentioned in the title. I propose to add the following:

• xlab
• ylab
• col_ci
• alpha
• Title of the graph

All of these parameters should have the default value leading to the current behavior of the function. Make sure to update help. Thanks!

See exts for details. Here is an example based on sunspot dataset:

slide 38 - ARMAmodels.pdf

xt = gen_ma1(100, 0.3, 1) x = select_ma(xt, q.min=2L, q.max=5L) plot(x)

In this example, the grid lines also exist at x = 2.5, 3.5, 4.5. But we would only want grid lines at x = 2,3,4,5. How should we modify the grid? Thanks a lot!

We should change description of the package from "Simulation tools" to something like "Simulation tools and functions need for the ts textbook". Rob could you please take of it? Thanks!

Add this function once wv is available:

#' @description This function will plot eight diagnostic plots to assess how well the model fits  
#' the data. These include: (1) residuals plot, (2) residuals vs fitted values, 
#' (3) histogram of standardized residuals, (4) normal Q-Q plot of 
#' residuals, (5) ACF plot, (6) PACF plot, (7) Haar Wavelet variance representation,
#' (8) Ljung-Box test results.
#' @author Yuming Zhang, James Balamuta
#' @param Xt The original time series data.
#' @param model The \code{arima} model fit to the data.
#' @param std A \code{boolean} indicating whether we use standardized residuals for 
#' the (1) residuals plot and the (8) Ljung-Box test results.
#' @export
#' @importFrom graphics points
#' @importFrom stats qqnorm
#' @importFrom stats qqline
#' @importFrom wv wvar
#' @importFrom stats var
#' @importFrom stats resid
#' @examples 
#' Xt = gen_gts(300, AR(phi = c(0, 0, 0.8), sigma2 = 1))
#' model = arima(Xt, order = c(3,0,0), include.mean = TRUE)
#' diag_plot(Xt, model)
diag_plot = function(Xt, model, std = FALSE){
  par(mfrow = c(2,4))
  
  # extract residuals 
  res = resid(model)
  
  
  # ----- plot 1
  resid_plot(Xt, model, std = std, type = "resid")
  
  # ----- plot 2
  fitted = as.numeric(Xt - res)
  # make frame
  x_range = c(min(fitted), max(fitted)) * 1.05
  y_range = c(min(res), max(res)) * 1.05
  make_frame(x_range, y_range, 
             xlab = "Fitted Values", ylab = "Residuals",
             main = "Residuals vs Fitted Values")
  # add points
  points(fitted, res, pch=16, col = "black")
  
  # ----- plot 3
  resid_plot(Xt, model, std = TRUE, type = "hist")
  
  # ----- plot 4
  my_qqnorm = qqnorm(res, plot.it = FALSE)
  # make frame
  x_range = c(min(my_qqnorm$x), max(my_qqnorm$x))*1.05
  y_range = c(min(my_qqnorm$y), max(my_qqnorm$y))*1.05
  make_frame(x_range, y_range, 
             xlab = "Theoretical Quantiles", 
             ylab = "Sample Quantiles",
             main = "Normal Q-Q Plot")
  
  # add qq plots
  points(my_qqnorm$x, my_qqnorm$y)
  qqline(res, col = "blue",lwd = 2)
  
  # ----- plot 5
  plot(ACF(Xt), col_ci = "#00BBDC33")
  
  # ----- plot 6
  plot(PACF(Xt), col_ci = "#00BBDC33")
  
  # ----- plot 7
  plot(wvar(res), main = "Haar WVar Representation", legend_position = NA)
  
  sigma2 = rep(var(res), length(wvar(res)$scales))
  points(wvar(res)$scales, sigma2/as.numeric(wvar(res)$scales), col = "orange", pch=0, cex=2)
  lines(wvar(res)$scales, sigma2/as.numeric(wvar(res)$scales), col = "orange", lty = 1)
  
  # add legend
  if (wvar(res)$robust == TRUE){
    wv_title_part1 = "Empirical Robust WV "
  }else{
    wv_title_part1 = "Empirical WV "
  }
  
  CI_conf = 1 - wvar(res)$alpha
  
  legend("bottomleft", 
         legend = c(as.expression(bquote(paste(.(wv_title_part1), hat(nu)^2))), 
                    as.expression(bquote(paste("CI(",hat(nu)^2,", ",.(CI_conf),")"))),
                    "WV implied by WN"),
         pch = c(16, 15, 0), lty = c(1, NA, 1), 
         col = c("darkblue", hcl(h = 210, l = 65, c = 100, alpha = 0.2), "orange"), 
         cex = 1, pt.cex = c(1.25, 3, 1.25), bty = "n")
  
  
  # ----- plot 8
  object = diag_ljungbox(model, stop_lag = 20, stdres = std)
  
  # make frame
  x_range = c(min(object$lag), max(object$lag))*1.05
  y_range = c(0, max(object$pvalue))*1.05
  make_frame(x_range, y_range, 
             xlab = "Lag", ylab = "P-value", main = "Ljung-Box Results")
  
  # add plotting
  points(object$lag, object$pvalue, pch = 16)
  lines(object$lag, object$pvalue, lty = 3)
  abline(h = 0.05, col = "blue", lty = 2)
  
  par(mfrow = c(1,1))

}

See exts package for details. Here is an example for AR

For ARMA:

Problems with lines 145 and 384 in ACF.r which, in the first case, gives the following error:

x = matrix(rnorm(1000), 100, 10)
test = ACF(x)
Error in dimnames(acfe) <- list(seq_len(nrow(acfe)) - 1, "ACF", varName) :
length of 'dimnames' [2] not equal to array extent

• Help of various function related to auto_corr (e.g. plot.ACF)
• _pkgdown.yml
• vignettes/vignettes.Rmd

Thanks a lot!

• Update estimate, select, check, predict to work with SARIMA
• Add any other functions needed for STAT 463
• Address relevant issues (at lease #23 #24 #21 #9)
• Fix errors (i.e. check package)
• Check if anythings else is needed to release the package

Improve the help of the function RW3dimensions. Add these parameters has inputs of the function:

  couleur = "blue4"
  xlab = "X-position"
  ylab = "Y-position"
  main = NULL
  pt_col = NULL
  pt_pch = 16
  pt.cex = 2
  leg_pos = NULL

test = theo_acf(ARMA(ar = -0.25, ma = NULL))
class(test)
[1] "ACF" "array" "ACF" "array"