## ----include = FALSE----------------------------------------------------------
knitr::opts_chunk$set(
  collapse = TRUE,
  comment  = "#>",
  fig.width  = 7,
  fig.height = 4
)

## ----setup, message = FALSE---------------------------------------------------
library(forecastdom)
data(llq2022_jnj)      # JNJ realized variance + 6 forecasts + lagged VIX
data(llq2022)          # S&P 500 counterpart
data(llq2022_uv_cspa)  # pre-computed cross-stock counts

models <- c("AR1", "AR22", "AR22_Lasso", "HAR", "HARQ", "ARFIMA")
qlike  <- function(f, y) (f / y) - log(f / y) - 1

## ----jnj-losses---------------------------------------------------------------
L_jnj <- sapply(models, function(m) qlike(llq2022_jnj[[m]], llq2022_jnj$rv))
X_jnj <- llq2022_jnj$vix_lag

## ----fig2-left, fig.width = 6.5, fig.height = 4-------------------------------
set.seed(20260512)

g_har_ar1 <- cspa_test_plot(
    Y = as.matrix(L_jnj[, "AR1"] - L_jnj[, "HAR"]),
    X = X_jnj, level = 0.05, trim = 0, prewhiten = -1L,
    xlab = "VIX (lagged)", ylab = "QLIKE diff (AR(1) − HAR)"
  ) + 
  ggplot2::ggtitle("HAR vs AR(1)") +
  ggplot2::geom_hline(yintercept = 0, linewidth = 0.3)
    
g_har_ar1

## ----fig2-right, fig.width = 6.5, fig.height = 4------------------------------
set.seed(20260512)
g_har_harq <- cspa_test_plot(
    Y = as.matrix(L_jnj[, "HARQ"] - L_jnj[, "HAR"]),
    X = X_jnj, level = 0.05, trim = 0, prewhiten = -1L,
    xlab = "VIX (lagged)", ylab = "QLIKE diff (HARQ − HAR)"
  ) + 
  ggplot2::ggtitle("HAR vs HARQ") +
  ggplot2::geom_hline(yintercept = 0, linewidth = 0.3)

g_har_harq

## ----fig2-decisions-----------------------------------------------------------
fig2 <- function(competitor) {

  Y <- as.matrix(L_jnj[, competitor] - L_jnj[, "HAR"])
  
  set.seed(20260512)
  
  r <- cspa_test(Y, X_jnj, level = 0.05, trim = 0, prewhiten = -1L,
                 preselect = TRUE, R = 10000L)
  
  data.frame(competitor = competitor,
             theta      = unname(r$theta),
             pvalue     = unname(r$pvalue),
             reject     = unname(r$reject))

}

knitr::kable(
  rbind(fig2("AR1"), fig2("HARQ")), digits = 4, row.names = FALSE,
  col.names = c("Competitor", "$\\theta$", "$p$-value", "Reject"))

## ----fig3-ar1, fig.width = 6.5, fig.height = 4.3------------------------------
set.seed(20260512)

Y_ar1 <- L_jnj[, setdiff(models, "AR1")] - L_jnj[, "AR1"]
g_ar1 <- cspa_test_plot(
    Y = Y_ar1, X = X_jnj, level = 0.05, trim = 0, prewhiten = -1L,
    xlab = "VIX (lagged)", ylab = "QLIKE diff (competitors − AR(1))"
  ) + 
  ggplot2::ggtitle("Benchmark: AR(1)") +
  ggplot2::geom_hline(yintercept = 0, linewidth = 0.3)

g_ar1

## ----fig3-harq, fig.width = 6.5, fig.height = 4.3-----------------------------
set.seed(20260512)

Y_harq <- L_jnj[, setdiff(models, "HARQ")] - L_jnj[, "HARQ"]
g_harq <- cspa_test_plot(
  Y = Y_harq, X = X_jnj, level = 0.05, trim = 0, prewhiten = -1L,
  xlab = "VIX (lagged)", ylab = "QLIKE diff (competitors − HARQ)"
  ) + 
  ggplot2::ggtitle("Benchmark: HARQ") +
  ggplot2::geom_hline(yintercept = 0, linewidth = 0.3)

g_harq

## ----fig3-decisions-----------------------------------------------------------
fig3 <- function(bench) {

  comp <- setdiff(models, bench)
  Y    <- L_jnj[, comp] - L_jnj[, bench]
  
  set.seed(20260512)
  
  r <- cspa_test(Y, X_jnj, level = 0.05, trim = 0, prewhiten = -1L, preselect = TRUE, R = 10000L)
  data.frame(benchmark = bench,
             theta     = unname(r$theta),
             pvalue    = unname(r$pvalue),
             reject    = unname(r$reject))

}

knitr::kable(
  rbind(fig3("AR1"), fig3("HARQ")), digits = 4, row.names = FALSE,
  col.names = c("Benchmark", "$\\theta$", "$p$-value", "Reject"))

## ----counts-mine--------------------------------------------------------------
knitr::kable(llq2022_uv_cspa$mine,
             caption = "forecastdom::cspa_test, R = 10000")

## ----counts-paper-------------------------------------------------------------
knitr::kable(llq2022_uv_cspa$paper,
             caption = "Published Table_UV_CSPA.xlsx (LLQ 2022)")

## ----counts-diff--------------------------------------------------------------
diff_mat <- llq2022_uv_cspa$mine - llq2022_uv_cspa$paper
knitr::kable(diff_mat, caption = "Difference (forecastdom minus LLQ)")

## ----sp500-csms---------------------------------------------------------------
L_sp <- sapply(models, function(m) qlike(llq2022[[m]], llq2022$rv))

set.seed(20260512)

cs <- csms(L_sp, llq2022$vix_lag, level = 0.10, trim = 0,
           prewhiten = -1L, preselect = TRUE, R = 10000L,
           method_names = models)

cs