forecast_comparison: Compare forecast accuracy

In tylerJPike/OOS: Out-of-Sample Time Series Forecasting

Description

A function to compare forecasts. Options include: simple forecast error ratios, Diebold-Mariano test, and Clark and West test for nested models

Usage

forecast_comparison(
  Data,
  baseline.forecast,
  test = "ER",
  loss = "MSE",
  horizon = NULL
)

Arguments

Data	data.frame: data frame of forecasts, model names, and dates
baseline.forecast	string: column name of baseline (null hypothesis) forecasts
test	string: which test to use; ER = error ratio, DM = Diebold-Mariano, CM = Clark and West
loss	string: error loss function to use if creating forecast error ratio
horizon	int: horizon of forecasts being compared in DM and CW tests

Value

numeric test result

Examples

 # simple time series
 A = c(1:100) + rnorm(100)
 date = seq.Date(from = as.Date('2000-01-01'), by = 'month', length.out = 100)
 Data = data.frame(date = date, A)

 # run forecast_univariate
 forecast.uni =
   forecast_univariate(
     Data = Data,
     forecast.dates = tail(Data$date,10),
     method = c('naive','auto.arima', 'ets'),
     horizon = 1,
     recursive = FALSE,
     freq = 'month')

 forecasts =
   dplyr::left_join(
     forecast.uni,
     data.frame(date, observed = A),
     by = 'date'
   )

 # run ER (MSE)
 er.ratio.mse =
   forecast_comparison(
     forecasts,
     baseline.forecast = 'naive',
     test = 'ER',
     loss = 'MSE')

tylerJPike/OOS documentation built on March 20, 2021, 3:20 a.m.