GoF: Compute goodness-of-fit measures for DLMs

In dLagM: Time Series Regression Models with Distributed Lag Models

Compute goodness-of-fit measures for DLMs

Description

Computes mean absolute error (MAE), mean squared error (MSE), mean percentage error (MPE), symmetric mean absolute percentage error (sMAPE), mean absolute percentage error (MAPE), mean absolute scaled error (MASE), mean relative absolute error (MRAE), geometric mean relative absolute error (GMRAE), mean bounded relative absolute error (MBRAE), unscaled MBRAE (UMBRAE) for distributed lag models.

Usage

GoF(model, ...)
MASE(model, ...)
sMAPE(model, ...)
MAPE(model, ...)
MRAE(model, ...)
GMRAE(model, ...)
MBRAE(model, ...)

Arguments

model	Model object fitted for time series data.
...	Optionally, more fitted models.

Details

See Chen et al. (2017) for the definitions of MSE, MAE, MAPE, sMAPE, MRAE, GMRAE, MBRAE, UMBMRAE.

Let e_{t} = Y_{t}-\hat{Y}_{t} be the one-step-ahead forecast error. Then, a scaled error is defined as

q_{t}=\frac{e_{t}}{\frac{1}{n-1}\sum_{i=2}^{n}|Y_{i}-Y_{i-1}|},

which is independent of the scale of the data. Mean absolute scaled error is defined as

MASE = mean(|q_{t}|)

(Hyndman and Koehler, 2006).

Fitted models would be finite, polynomial, Koyck, ARDL DLMs, or linear model fitted with lm() function. This function also computes MASE values of multiple models when fed at the same time.

Author(s)

Haydar Demirhan

Maintainer: Haydar Demirhan <[e-mail]>

References

Chen, C., Twycross, J., Garibaldi, J.M. (2017). A new accuracy measure based on bounded relative error for time series forecasting. PLoS ONE, 12(3), e0174202.

Hyndman, R.J. and Koehler, A.B. (2006). Another look at measures of forecast accuracy. International Journal of Forecasting, 22, 679-688.

Examples

## Not run: 
data(seaLevelTempSOI)
# Fit a bunch of polynomial DLMs
model.poly1 = polyDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, 
                     q = 2 , k = 2 , show.beta = TRUE)
model.poly2 = polyDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, 
                      q = 3 , k = 2 , show.beta = TRUE)
model.poly3 = polyDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, 
                      q = 4 , k = 2 , show.beta = TRUE)
MASE(model.poly1, model.poly2, model.poly3)

# Fit a bunch of finite DLMs
model.dlm1 = dlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, q =2)
model.dlm2 = dlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, q =3)
model.dlm3 = dlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, q =4)
MASE(model.dlm1, model.dlm2, model.dlm3)
MBRAE(model.dlm1, model.dlm2, model.dlm3)
GoF(model.dlm1, model.dlm2, model.dlm3)

# Fit a linear model
model.lm = lm(GMSL ~ LandOcean , data = seaLevelTempSOI)
MASE(model.lm)
GoF(model.lm)

# Fit a Koyck model
model.koyck = koyckDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL)
MASE(model.koyck)
GoF(model.koyck)

# Fit a bunch of ARDLs
model.ardl1 = ardlDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, p=1, q=2)
model.ardl2 = ardlDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, p=2, q=2)
model.ardl3 = ardlDlm(x = seaLevelTempSOI$LandOcean, y = seaLevelTempSOI$GMSL, p=3, q=2)
MASE(model.ardl1 , model.ardl2 , model.ardl3)
GoF(model.ardl1 , model.ardl2 , model.ardl3)

# Find MASEs of different model objects
MASE(model.ardl1 , model.dlm1 , model.poly1, model.lm)
GoF(model.ardl1 , model.dlm1 , model.poly1, model.lm)

## End(Not run)

dLagM documentation built on Oct. 2, 2023, 9:07 a.m.