R forecast.holt vs Python statsmodels.tsa.holtwinters

Question

I'm trying to use HoltWinters Exponential Smoothing in python, but I'm getting different results than I get when I use forecast holt in R.

In R:

library(forecast)

data_train <- c(0.3990852, 1.8837862, 2.3551793, 3.0099617, 3.4650170,
                4.6327859, 3.7989490, 1.2654134, 3.3170017, 4.7559544,
                2.7958632, 2.8002729, 3.9480264, 3.0497512)

y_hat <- holt(data_train, h=6)$mean

print(y_hat)

[1] 4.316603 4.483438 4.650274 4.817109 4.983944 5.150779

In python:

import numpy as np
from statsmodels.tsa.holtwinters import ExponentialSmoothing, Holt

data_train = np.array((0.3990852, 1.8837862, 2.3551793, 3.0099617, 3.4650170,
                4.6327859, 3.7989490, 1.2654134, 3.3170017, 4.7559544,
                2.7958632, 2.8002729, 3.9480264, 3.0497512))

model = ExponentialSmoothing(data_train).fit()
y_hat = model.predict(start=15, end=20)
print(y_hat)

[3.2521686 3.2521686 3.2521686 3.2521686 3.2521686 3.2521686]

fit1 = Holt(data_train).fit()
y_hat = fit1.forecast(6)

print(y_hat)

[3.23339397 3.21157785 3.18976174 3.16794562 3.1461295  3.12431338]

Can anyone tell me why I'm getting such different results in R vs. python?

Answer 1

It seems like you have to set the parameters correctly.

fit1 = Holt(data_train, trend = 'additive', seasonal = 'additive').fit() etc..