Small modification of the seasonality tests

NAMESPACE

@@ -252,6 +252,7 @@ export(seasonality_combined)
 export(seasonality_f)
 export(seasonality_friedman)
 export(seasonality_kruskalwallis)
+export(seasonality_modified_qs)
 export(seasonality_periodogram)
 export(seasonality_qs)
 export(set_arima)

R/arima.R

@@ -108,6 +108,7 @@ sarima_decompose<-function(model, rmod=0, epsphi=0){
 #' @export
 #'
 #' @examples
+#' model <- arima_model("trend", ar=c(1,-.8), delta = c(1,-1), ma=c(1,-.5), var=100)
 arima_model<-function(name="arima", ar=1, delta=1, ma=1, variance=1){
   return(structure(list(name=name, ar=ar, delta=delta, ma=ma, var=variance), class="JD3_ARIMA"))
 }

R/differencing.R

@@ -15,7 +15,6 @@ NULL
 
 #' Automatic stationary transformation
 #'
-#' Stationary transformation of a series by simple differencing of lag 1.
 #' Automatic processing (identification of the order of the differencing) based on auto-correlations and on mean correction.
 #' The series should not be seasonal.
 #' Source: Tramo
@@ -50,13 +49,13 @@ do_stationary<-function(data, period){
 
 #' Automatic differencing
 #'
-#' The series is differentiated till its variance is decreasing.
+#' The series is differenced till its variance is decreasing.
 #'
 #' @param data Series being differenced.
 #' @param period Period considered in the automatic differencing.
 #' @param mad Use of MAD in the computation of the variance (true by default).
 #' @param centile Percentage of the data used for computing the variance (90 by default).
-#' @param k tolerance in the decrease of the variance. The algorithm stops if the new varance is higher than k*the old variance.
+#' @param k tolerance in the decrease of the variance. The algorithm stops if the new variance is higher than k*the old variance. k should be equal or slightly higher than 1 (1.2 by default)
 #'
 #' @return
 #' Stationary transformation
@@ -87,7 +86,7 @@ differencing_fast<-function(data, period, mad=TRUE, centile=90, k=1.2){
 #'
 #' @param data The series to be differenced.
 #' @param lags Lags of the differencing.
-#' @param mean Mean correction.
+#' @param mean Apply a mean correction at the end of the differencing process.
 #'
 #' @return The differenced series.
 #' @export
@@ -126,7 +125,7 @@ differences.data.frame<-function(data, lags=1, mean=TRUE){
 #' Range-Mean Regression
 #'
 #' Function to perform a range-mean regression, trimmed to avoid outlier distortion.
-#' The slope is used in TRAMO to select whether the original series will be transformed into log or maintain in level.
+#' The can be used to select whether the original series will be transformed into log or maintain in level.
 #'
 #' @param data data to test.
 #' @param period periodicity of the data.

R/tests_regular.R

@@ -204,16 +204,18 @@ kurtosis<-function(data){
   return(.jd2r_test(jtest))
 }
 
-#' Title
+#' Compute a robust median absolute deviation (MAD)
 #'
-#' @param data
-#' @param centile
-#' @param medianCorrected
+#' @param data The data for which we compute the robust deviation
+#' @param centile The centile used to exclude extreme values (only the "centile" part of the data are is to compute the mad)
+#' @param medianCorrected TRUE if the series is corrected for its median, FALSE if the median is supposed to be 0
 #'
-#' @return
+#' @return The median absolute deviation
 #' @export
 #'
 #' @examples
+#' y<-rnorm(1000)
+#' m<-rjd3toolkit::mad(y, centile=70)
 mad<-function(data, centile=50, medianCorrected=TRUE){
   return(.jcall("jdplus/toolkit/base/r/stats/Tests", "D", "mad",as.numeric(data), as.numeric(centile), as.logical(medianCorrected)))
 }

R/tests_seasonality.R

@@ -2,30 +2,55 @@
 NULL
 
 
-#' QS Seasonality Test
-#'
-#' QS (modified seasonal Ljung-Box) test.
+#' QS (seasonal Ljung-Box) test.
 #'
 #' @param data the input data.
-#' @param period Tested periodicity.
-#' @param nyears Number of number of periods number of cycles considered in the test, at the end of the series:
+#' @param period Tested periodicity. Can be missing if the input is a time series
+#' @param nyears Number of periods or number of cycles considered in the test, at the end of the series:
 #' in periods (positive value) or years (negative values).
 #' By default (\code{nyears = 0}), the entire sample is used.
+#' @param type 1 for positive autocorrelations, -1 for negative autocorrelations,
+#' 0 for all autocorrelations. By default (\code{type = 1})
 #'
 #' @return A `c("JD3_TEST", "JD3")` object (see [statisticaltest()] for details).
 #' @export
 #'
 #' @examples
-#' seasonality_qs(ABS$X0.2.09.10.M, 12)
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_qs(s)
 #' seasonality_qs(random_t(2, 1000), 7)
-seasonality_qs<-function(data, period, nyears=0){
+seasonality_qs<-function(data, period=NA, nyears=0, type=1){
   if (is.ts(data) && missing(period))
     period <- frequency(data)
   jtest<-.jcall("jdplus/sa/base/r/SeasonalityTests", "Ljdplus/toolkit/base/api/stats/StatisticalTest;", "qsTest",
-         as.numeric(data), as.integer(period), as.integer(nyears))
+         as.numeric(data), as.integer(period), as.integer(nyears), as.integer((type)))
   return(.jd2r_test(jtest))
 }
 
+#' Modified QS Seasonality Test (Maravall)
+#'
+#'
+#' @param data the input data.
+#' @param period Tested periodicity. Can be missing if the input is a time series
+#' @param nyears Number of periods or number of cycles considered in the test, at the end of the series:
+#' in periods (positive value) or years (negative values).
+#' By default (\code{nyears = 0}), the entire sample is used.
+#'
+#' @return The value of the test
+#' @export
+#'
+#' @examples
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_modified_qs(s)
+seasonality_modified_qs<-function(data, period=NA, nyears=0){
+    if (is.ts(data) && missing(period))
+        period <- frequency(data)
+    test<-.jcall("jdplus/sa/base/r/SeasonalityTests", "D", "modifiedQsTest",
+                  as.numeric(data), as.integer(period), as.integer(nyears))
+    return(test)
+}
+
+
 #' Kruskall-Wallis Seasonality Test
 #'
 #'
@@ -36,7 +61,8 @@ seasonality_qs<-function(data, period, nyears=0){
 #' @export
 #'
 #' @examples
-#' seasonality_kruskalwallis(ABS$X0.2.09.10.M, 12)
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_kruskalwallis(s)
 #' seasonality_kruskalwallis(random_t(2, 1000), 7)
 seasonality_kruskalwallis<-function(data, period, nyears=0){
   if (is.ts(data) && missing(period))
@@ -55,9 +81,10 @@ seasonality_kruskalwallis<-function(data, period, nyears=0){
 #' @export
 #'
 #' @examples
-#' seasonality_periodogram(ABS$X0.2.09.10.M, 12)
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_periodogram(s)
 #' seasonality_periodogram(random_t(2, 1000), 7)
-seasonality_periodogram<-function(data, period, nyears=0){
+seasonality_periodogram<-function(data, period=NA, nyears=0){
   if (is.ts(data) && missing(period))
     period <- frequency(data)
   jtest<-.jcall("jdplus/sa/base/r/SeasonalityTests", "Ljdplus/toolkit/base/api/stats/StatisticalTest;", "periodogramTest",
@@ -74,7 +101,10 @@ seasonality_periodogram<-function(data, period, nyears=0){
 #' @export
 #'
 #' @examples
-seasonality_friedman<-function(data, period, nyears=0){
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_friedman(s)
+#' seasonality_friedman(random_t(2, 1000), 12)
+seasonality_friedman<-function(data, period=NA, nyears=0){
   if (is.ts(data) && missing(period))
     period <- frequency(data)
   jtest<-.jcall("jdplus/sa/base/r/SeasonalityTests", "Ljdplus/toolkit/base/api/stats/StatisticalTest;", "friedmanTest",
@@ -91,10 +121,10 @@ seasonality_friedman<-function(data, period, nyears=0){
 #' @export
 #'
 #' @examples
-#' seasonality_f(ABS$X0.2.09.10.M, 12)
+#' seasonality_f(ABS$X0.2.09.10.M, model="D1")
 #' seasonality_f(random_t(2, 1000), 7)
 seasonality_f<-function(data,
-                        period,
+                        period=NA,
                         model=c("AR", "D1", "WN"),
                         nyears=0){
   if (is.ts(data) && missing(period))
@@ -117,9 +147,10 @@ seasonality_f<-function(data,
 #' @export
 #'
 #' @examples
-#' seasonality_combined(ABS$X0.2.09.10.M, 12)
+#' s<-do_stationary(log(ABS$X0.2.09.10.M))$ddata
+#' seasonality_combined(s)
 #' seasonality_combined(random_t(2, 1000), 7)
-seasonality_combined<-function(data, period, firstperiod=cycle(data)[1], mul=TRUE){
+seasonality_combined<-function(data, period=NA, firstperiod=cycle(data)[1], mul=TRUE){
   if (is.ts(data) && missing(period))
     period <- frequency(data)
   jctest<-.jcall("jdplus/sa/base/r/SeasonalityTests", "Ljdplus/sa/base/core/tests/CombinedSeasonality;", "combinedTest",
@@ -138,8 +169,8 @@ seasonality_combined<-function(data, period, firstperiod=cycle(data)[1], mul=TRU
 #' @inheritParams seasonality_qs
 #' @param periods Periodicities.
 #' @param lag1 Lagged variable in the regression model.
-#' @param kernel Kernel used to compute the robust covariance matrix.
-#' @param order The truncation parameter used to compute the robust covariance matrix.
+#' @param kernel Kernel used to compute the robust Newey-West covariance matrix.
+#' @param order The truncation parameter used to compute the robust Newey-West covariance matrix.
 #' @param original `TRUE` for original algorithm, `FALSE` for solution proposed by T. Proietti (based on Ox code).
 #'
 #' @export
@@ -166,12 +197,13 @@ seasonality_canovahansen_trigs<-function(data, periods, lag1=TRUE,
 #' @inheritParams seasonality_qs
 #' @param trigs TRUE for trigonometric variables, FALSE for seasonal dummies.
 #' @param lag1 Lagged variable in the regression model.
-#' @param kernel Kernel used to compute the robust covariance matrix.
-#' @param order The truncation parameter used to compute the robust covariance matrix.
+#' @param kernel Kernel used to compute the robust Newey-West covariance matrix.
+#' @param order The truncation parameter used to compute the robust Newey-West covariance matrix.
 #' @param start Position of the first observation of the series
-#' @return list with the joint test and with details for the different seasonal variables
+#' @return list with the FTest on seasonal variables, the joint test and the details for the stability of the different seasonal variables
 #' @export
 #'
+#'
 #' @examples
 #' s<-log(ABS$X0.2.20.10.M)
 #' seasonality_canovahansen(s, 12, trigs = FALSE)
@@ -187,5 +219,5 @@ seasonality_canovahansen<-function(data, period, trigs=TRUE, lag1=TRUE,
                   as.logical(trigs), as.logical(lag1),
                   kernel, as.integer(order), as.integer(start-1))
     last<-length(q)
-    return(list(joint=q[last], details=q[-last]))
+    return(list(seasonality=list(value=q[last-1], pvalue=q[last]), joint=q[last-2], details=q[-c(last-2, last-1, last)]))
 }

R/tests_td.R

@@ -61,7 +61,7 @@ td_f<-function(s, model=c("D1", "DY", "DYD1", "WN", "AIRLINE", "R011", "R100"),
 #' @param kernel Kernel used to compute the robust covariance matrix.
 #' @param order The truncation parameter used to compute the robust covariance matrix.
 #'
-#' @return list with the joint test and with details for the different days (starting with Mondays).
+#' @return list with the ftest on td, the joint test and the details for the stability of the different days (starting with Mondays).
 #' @export
 #'
 #' @examples
@@ -76,5 +76,5 @@ td_canovahansen<-function(s, differencing, kernel=c("Bartlett", "Square", "Welch
                 jts, .jarray(as.integer(differencing)), kernel, as.integer(order))
 
   last<-length(q)
-  return(list(joint=q[last], details=q[-last]))
+  return(list(td=list(value=q[last-1], pvalue=q[last]), joint=q[last-2], details=q[-c(last-2, last-1, last)]))
 }

R/utils.R

@@ -53,20 +53,19 @@ test_anova<-function(ssm, dfm, ssr, dfr){
   return(statisticaltest(val, pval, desc))
 }
 
-#' Title
+#' Information on the (log-)likelihood
 #'
-#' @param nobs
-#' @param neffectiveobs
-#' @param nparams
-#' @param ll
-#' @param adjustedll
-#' @param aic
-#' @param aicc
-#' @param bic
-#' @param bicc
-#' @param ssq
+#' @param nobs Number of observation
+#' @param neffectiveobs Number of effective observations. NA if it is the same as nobs.
+#' @param nparams Number of hyper-parameters
+#' @param ll Log-likelihood
+#' @param adjustedll Adjusted log-likelihood when the series has been transformed
+#' @param aic AIC
+#' @param aicc AICC
+#' @param bic BIC
+#' @param bicc BIC corrected for the length
+#' @param ssq Sum of the squared residuals
 #'
-#' @return
 #' @export
 #'
 #' @examples