README.md

In andybega/EBMAhelper: Regular S3 model fitting function and methods for EBMAforecast

EBMAhelper

EBMAhelper is a wrapper around EBMAforecast (CRAN, GitHub) that provides some convenience functions more akin to R's conventional model-related functions:

• ebma() to create and fit an EBMA ensemble model; wraps EBMAforecast::makeForecastData and EBMAforecast::calibrateEnsemble
• predict method for the class "ebma" object returned by ebma()

and a standalone predict() (TODO) method to aggregate new forecasts.

Installation

library("devtools")

install_github("andybega/EBMAhelper")

Example

A EBMA ensemble can be fitted like this:

suppressMessages({
  library("EBMAforecast")
  library("EBMAhelper")
})


data("presidentialForecast")

head(presidentialForecast)
#>      Campbell Lewis-Beck   EWT2C2     Fair    Hibbs Abramowitz   Actual
#> 1952 46.83648   45.18571 45.70344 45.24491 44.81212   44.29725 44.59477
#> 1956 54.04117   56.77496 56.54709 56.41898 56.51094   59.00480 57.75380
#> 1960 51.15267   48.44678 49.61112 51.08189 48.95051   49.42700 49.91609
#> 1964 61.12062   63.17691 63.48235 61.20933 61.36488   60.85638 61.34263
#> 1968 48.45415   48.65492 51.81968 50.19200 49.25904   50.09237 49.59511
#> 1972 61.17357   59.58035 55.04779 58.42894 59.26387   59.54824 61.78800
str(presidentialForecast)
#> 'data.frame':    15 obs. of  7 variables:
#>  $ Campbell  : num  46.8 54 51.2 61.1 48.5 ...
#>  $ Lewis-Beck: num  45.2 56.8 48.4 63.2 48.7 ...
#>  $ EWT2C2    : num  45.7 56.5 49.6 63.5 51.8 ...
#>  $ Fair      : num  45.2 56.4 51.1 61.2 50.2 ...
#>  $ Hibbs     : num  44.8 56.5 49 61.4 49.3 ...
#>  $ Abramowitz: num  44.3 59 49.4 60.9 50.1 ...
#>  $ Actual    : num  44.6 57.8 49.9 61.3 49.6 ...

fitted_ensemble <- ebma(y = presidentialForecast[ ,7],
                        x = presidentialForecast[ ,c(1:6)],
                        model_type = "normal")
summary(fitted_ensemble)
#>                 W Constant Predictor rmse  mae
#> EBMA                                 1.33 1.03
#> Campbell   0.2685   -0.515     1.001 2.28 1.62
#> Lewis-Beck 0.0367    2.552     0.954 1.70 1.42
#> EWT2C2     0.2325    0.124     0.998 2.71 2.05
#> Fair       0.0000    1.106     0.977 2.21 1.82
#> Hibbs      0.4624    1.253     0.978 1.89 1.38
#> Abramowitz 0.0000    1.780     0.966 1.60 1.34

To do the same with EBMAforecast is slightly more verbose and does not follow R's model fitting conventions. The following code is adapted from the package demo:

data("presidentialForecast")

full.forecasts <- presidentialForecast[, c(1:6)]
full.observed <- presidentialForecast[, 7]

this.ForecastData <- makeForecastData(
  .predCalibration=full.forecasts[1:12,],
  .outcomeCalibration=full.observed[1:12],
  .predTest=full.forecasts[13:15,], 
  .outcomeTest=full.observed[13:15], 
  .modelNames=c("Campbell", "Lewis-Beck","EWT2C2","Fair","Hibbs","Abramowitz"))
thisEnsemble <- calibrateEnsemble(this.ForecastData, model="normal", 
                                  useModelParams=FALSE, tol = 0.000000001,
                                  const = 0)
summary(thisEnsemble)
#>                W Constant Predictor  rmse   mae
#> EBMA                                0.944 0.745
#> Campbell   0.314        0         1 1.597 1.228
#> Lewis-Beck 0.000        0         1 1.705 1.434
#> EWT2C2     0.145        0         1 2.816 2.142
#> Fair       0.000        0         1 2.207 1.755
#> Hibbs      0.321        0         1 1.523 1.103
#> Abramowitz 0.221        0         1 1.266 1.055

andybega/EBMAhelper documentation built on April 6, 2022, 11:11 a.m.