Exponential smoothing

Let us consider chapter 7 of the excellent treatise on the subject of Exponential Smoothing By Hyndman and Athanasopoulos [1]. We will work through all the examples in the chapter as they unfold.

[1] Hyndman, Rob J., and George Athanasopoulos. Forecasting: principles and practice. OTexts, 2014.

Loading data

First we load some data. We have included the R data in the notebook for expedience.

[1]:
import os
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from statsmodels.tsa.api import ExponentialSmoothing, SimpleExpSmoothing, Holt

%matplotlib inline

data = [
    446.6565,
    454.4733,
    455.663,
    423.6322,
    456.2713,
    440.5881,
    425.3325,
    485.1494,
    506.0482,
    526.792,
    514.2689,
    494.211,
]
index = pd.date_range(start="1996", end="2008", freq="A")
oildata = pd.Series(data, index)

data = [
    17.5534,
    21.86,
    23.8866,
    26.9293,
    26.8885,
    28.8314,
    30.0751,
    30.9535,
    30.1857,
    31.5797,
    32.5776,
    33.4774,
    39.0216,
    41.3864,
    41.5966,
]
index = pd.date_range(start="1990", end="2005", freq="A")
air = pd.Series(data, index)

data = [
    263.9177,
    268.3072,
    260.6626,
    266.6394,
    277.5158,
    283.834,
    290.309,
    292.4742,
    300.8307,
    309.2867,
    318.3311,
    329.3724,
    338.884,
    339.2441,
    328.6006,
    314.2554,
    314.4597,
    321.4138,
    329.7893,
    346.3852,
    352.2979,
    348.3705,
    417.5629,
    417.1236,
    417.7495,
    412.2339,
    411.9468,
    394.6971,
    401.4993,
    408.2705,
    414.2428,
]
index = pd.date_range(start="1970", end="2001", freq="A")
livestock2 = pd.Series(data, index)

data = [407.9979, 403.4608, 413.8249, 428.105, 445.3387, 452.9942, 455.7402]
index = pd.date_range(start="2001", end="2008", freq="A")
livestock3 = pd.Series(data, index)

data = [
    41.7275,
    24.0418,
    32.3281,
    37.3287,
    46.2132,
    29.3463,
    36.4829,
    42.9777,
    48.9015,
    31.1802,
    37.7179,
    40.4202,
    51.2069,
    31.8872,
    40.9783,
    43.7725,
    55.5586,
    33.8509,
    42.0764,
    45.6423,
    59.7668,
    35.1919,
    44.3197,
    47.9137,
]
index = pd.date_range(start="2005", end="2010-Q4", freq="QS-OCT")
aust = pd.Series(data, index)

Simple Exponential Smoothing

Lets use Simple Exponential Smoothing to forecast the below oil data.

[2]:
ax = oildata.plot()
ax.set_xlabel("Year")
ax.set_ylabel("Oil (millions of tonnes)")
print("Figure 7.1: Oil production in Saudi Arabia from 1996 to 2007.")
Figure 7.1: Oil production in Saudi Arabia from 1996 to 2007.
../../../_images/examples_notebooks_generated_exponential_smoothing_4_1.png

Here we run three variants of simple exponential smoothing: 1. In fit1 we do not use the auto optimization but instead choose to explicitly provide the model with the \(\alpha=0.2\) parameter 2. In fit2 as above we choose an \(\alpha=0.6\) 3. In fit3 we allow statsmodels to automatically find an optimized \(\alpha\) value for us. This is the recommended approach.

[3]:
fit1 = SimpleExpSmoothing(oildata, initialization_method="heuristic").fit(
    smoothing_level=0.2, optimized=False
)
fcast1 = fit1.forecast(3).rename(r"$\alpha=0.2$")
fit2 = SimpleExpSmoothing(oildata, initialization_method="heuristic").fit(
    smoothing_level=0.6, optimized=False
)
fcast2 = fit2.forecast(3).rename(r"$\alpha=0.6$")
fit3 = SimpleExpSmoothing(oildata, initialization_method="estimated").fit()
fcast3 = fit3.forecast(3).rename(r"$\alpha=%s$" % fit3.model.params["smoothing_level"])

plt.figure(figsize=(12, 8))
plt.plot(oildata, marker="o", color="black")
plt.plot(fit1.fittedvalues, marker="o", color="blue")
(line1,) = plt.plot(fcast1, marker="o", color="blue")
plt.plot(fit2.fittedvalues, marker="o", color="red")
(line2,) = plt.plot(fcast2, marker="o", color="red")
plt.plot(fit3.fittedvalues, marker="o", color="green")
(line3,) = plt.plot(fcast3, marker="o", color="green")
plt.legend([line1, line2, line3], [fcast1.name, fcast2.name, fcast3.name])
[3]:
<matplotlib.legend.Legend at 0x7f980bef8640>
../../../_images/examples_notebooks_generated_exponential_smoothing_6_1.png

Holt’s Method

Lets take a look at another example. This time we use air pollution data and the Holt’s Method. We will fit three examples again. 1. In fit1 we again choose not to use the optimizer and provide explicit values for \(\alpha=0.8\) and \(\beta=0.2\) 2. In fit2 we do the same as in fit1 but choose to use an exponential model rather than a Holt’s additive model. 3. In fit3 we used a damped versions of the Holt’s additive model but allow the dampening parameter \(\phi\) to be optimized while fixing the values for \(\alpha=0.8\) and \(\beta=0.2\)

[4]:
fit1 = Holt(air, initialization_method="estimated").fit(
    smoothing_level=0.8, smoothing_trend=0.2, optimized=False
)
fcast1 = fit1.forecast(5).rename("Holt's linear trend")
fit2 = Holt(air, exponential=True, initialization_method="estimated").fit(
    smoothing_level=0.8, smoothing_trend=0.2, optimized=False
)
fcast2 = fit2.forecast(5).rename("Exponential trend")
fit3 = Holt(air, damped_trend=True, initialization_method="estimated").fit(
    smoothing_level=0.8, smoothing_trend=0.2
)
fcast3 = fit3.forecast(5).rename("Additive damped trend")

plt.figure(figsize=(12, 8))
plt.plot(air, marker="o", color="black")
plt.plot(fit1.fittedvalues, color="blue")
(line1,) = plt.plot(fcast1, marker="o", color="blue")
plt.plot(fit2.fittedvalues, color="red")
(line2,) = plt.plot(fcast2, marker="o", color="red")
plt.plot(fit3.fittedvalues, color="green")
(line3,) = plt.plot(fcast3, marker="o", color="green")
plt.legend([line1, line2, line3], [fcast1.name, fcast2.name, fcast3.name])
[4]:
<matplotlib.legend.Legend at 0x7f980bdd6d40>
../../../_images/examples_notebooks_generated_exponential_smoothing_8_1.png

Seasonally adjusted data

Lets look at some seasonally adjusted livestock data. We fit five Holt’s models. The below table allows us to compare results when we use exponential versus additive and damped versus non-damped.

Note: fit4 does not allow the parameter \(\phi\) to be optimized by providing a fixed value of \(\phi=0.98\)

[5]:
fit1 = SimpleExpSmoothing(livestock2, initialization_method="estimated").fit()
fit2 = Holt(livestock2, initialization_method="estimated").fit()
fit3 = Holt(livestock2, exponential=True, initialization_method="estimated").fit()
fit4 = Holt(livestock2, damped_trend=True, initialization_method="estimated").fit(
    damping_trend=0.98
)
fit5 = Holt(
    livestock2, exponential=True, damped_trend=True, initialization_method="estimated"
).fit()
params = [
    "smoothing_level",
    "smoothing_trend",
    "damping_trend",
    "initial_level",
    "initial_trend",
]
results = pd.DataFrame(
    index=[r"$\alpha$", r"$\beta$", r"$\phi$", r"$l_0$", "$b_0$", "SSE"],
    columns=["SES", "Holt's", "Exponential", "Additive", "Multiplicative"],
)
results["SES"] = [fit1.params[p] for p in params] + [fit1.sse]
results["Holt's"] = [fit2.params[p] for p in params] + [fit2.sse]
results["Exponential"] = [fit3.params[p] for p in params] + [fit3.sse]
results["Additive"] = [fit4.params[p] for p in params] + [fit4.sse]
results["Multiplicative"] = [fit5.params[p] for p in params] + [fit5.sse]
results
[5]:
SES Holt's Exponential Additive Multiplicative
$\alpha$ 1.000000 0.974316 0.977633 9.788623e-01 0.974913
$\beta$ NaN 0.000000 0.000000 7.437128e-11 0.000000
$\phi$ NaN NaN NaN 9.800000e-01 0.981646
$l_0$ 263.917698 258.882555 260.344565 2.573572e+02 258.951736
$b_0$ NaN 5.010839 1.013780 6.645778e+00 1.038145
SSE 6761.350235 6004.138201 6104.194756 6.036555e+03 6081.995046

Plots of Seasonally Adjusted Data

The following plots allow us to evaluate the level and slope/trend components of the above table’s fits.

[6]:
for fit in [fit2, fit4]:
    pd.DataFrame(np.c_[fit.level, fit.trend]).rename(
        columns={0: "level", 1: "slope"}
    ).plot(subplots=True)
plt.show()
print(
    "Figure 7.4: Level and slope components for Holt’s linear trend method and the additive damped trend method."
)
../../../_images/examples_notebooks_generated_exponential_smoothing_12_0.png
../../../_images/examples_notebooks_generated_exponential_smoothing_12_1.png
Figure 7.4: Level and slope components for Holt’s linear trend method and the additive damped trend method.

Comparison

Here we plot a comparison Simple Exponential Smoothing and Holt’s Methods for various additive, exponential and damped combinations. All of the models parameters will be optimized by statsmodels.

[7]:
fit1 = SimpleExpSmoothing(livestock2, initialization_method="estimated").fit()
fcast1 = fit1.forecast(9).rename("SES")
fit2 = Holt(livestock2, initialization_method="estimated").fit()
fcast2 = fit2.forecast(9).rename("Holt's")
fit3 = Holt(livestock2, exponential=True, initialization_method="estimated").fit()
fcast3 = fit3.forecast(9).rename("Exponential")
fit4 = Holt(livestock2, damped_trend=True, initialization_method="estimated").fit(
    damping_trend=0.98
)
fcast4 = fit4.forecast(9).rename("Additive Damped")
fit5 = Holt(
    livestock2, exponential=True, damped_trend=True, initialization_method="estimated"
).fit()
fcast5 = fit5.forecast(9).rename("Multiplicative Damped")

ax = livestock2.plot(color="black", marker="o", figsize=(12, 8))
livestock3.plot(ax=ax, color="black", marker="o", legend=False)
fcast1.plot(ax=ax, color="red", legend=True)
fcast2.plot(ax=ax, color="green", legend=True)
fcast3.plot(ax=ax, color="blue", legend=True)
fcast4.plot(ax=ax, color="cyan", legend=True)
fcast5.plot(ax=ax, color="magenta", legend=True)
ax.set_ylabel("Livestock, sheep in Asia (millions)")
plt.show()
print(
    "Figure 7.5: Forecasting livestock, sheep in Asia: comparing forecasting performance of non-seasonal methods."
)
../../../_images/examples_notebooks_generated_exponential_smoothing_14_0.png
Figure 7.5: Forecasting livestock, sheep in Asia: comparing forecasting performance of non-seasonal methods.

Holt’s Winters Seasonal

Finally we are able to run full Holt’s Winters Seasonal Exponential Smoothing including a trend component and a seasonal component. statsmodels allows for all the combinations including as shown in the examples below: 1. fit1 additive trend, additive seasonal of period season_length=4 and the use of a Box-Cox transformation. 1. fit2 additive trend, multiplicative seasonal of period season_length=4 and the use of a Box-Cox transformation.. 1. fit3 additive damped trend, additive seasonal of period season_length=4 and the use of a Box-Cox transformation. 1. fit4 additive damped trend, multiplicative seasonal of period season_length=4 and the use of a Box-Cox transformation.

The plot shows the results and forecast for fit1 and fit2. The table allows us to compare the results and parameterizations.

[8]:
fit1 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="add",
    use_boxcox=True,
    initialization_method="estimated",
).fit()
fit2 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="mul",
    use_boxcox=True,
    initialization_method="estimated",
).fit()
fit3 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="add",
    damped_trend=True,
    use_boxcox=True,
    initialization_method="estimated",
).fit()
fit4 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="mul",
    damped_trend=True,
    use_boxcox=True,
    initialization_method="estimated",
).fit()
results = pd.DataFrame(
    index=[r"$\alpha$", r"$\beta$", r"$\phi$", r"$\gamma$", r"$l_0$", "$b_0$", "SSE"]
)
params = [
    "smoothing_level",
    "smoothing_trend",
    "damping_trend",
    "smoothing_seasonal",
    "initial_level",
    "initial_trend",
]
results["Additive"] = [fit1.params[p] for p in params] + [fit1.sse]
results["Multiplicative"] = [fit2.params[p] for p in params] + [fit2.sse]
results["Additive Dam"] = [fit3.params[p] for p in params] + [fit3.sse]
results["Multiplica Dam"] = [fit4.params[p] for p in params] + [fit4.sse]

ax = aust.plot(
    figsize=(10, 6),
    marker="o",
    color="black",
    title="Forecasts from Holt-Winters' multiplicative method",
)
ax.set_ylabel("International visitor night in Australia (millions)")
ax.set_xlabel("Year")
fit1.fittedvalues.plot(ax=ax, style="--", color="red")
fit2.fittedvalues.plot(ax=ax, style="--", color="green")

fit1.forecast(8).rename("Holt-Winters (add-add-seasonal)").plot(
    ax=ax, style="--", marker="o", color="red", legend=True
)
fit2.forecast(8).rename("Holt-Winters (add-mul-seasonal)").plot(
    ax=ax, style="--", marker="o", color="green", legend=True
)

plt.show()
print(
    "Figure 7.6: Forecasting international visitor nights in Australia using Holt-Winters method with both additive and multiplicative seasonality."
)

results
../../../_images/examples_notebooks_generated_exponential_smoothing_16_0.png
Figure 7.6: Forecasting international visitor nights in Australia using Holt-Winters method with both additive and multiplicative seasonality.
[8]:
Additive Multiplicative Additive Dam Multiplica Dam
$\alpha$ 1.490116e-08 1.490116e-08 1.490116e-08 1.490116e-08
$\beta$ 1.409869e-08 0.000000e+00 6.490748e-09 5.042416e-09
$\phi$ NaN NaN 9.430416e-01 9.536043e-01
$\gamma$ 0.000000e+00 0.000000e+00 2.815449e-16 0.000000e+00
$l_0$ 1.119348e+01 1.106377e+01 1.084022e+01 9.899279e+00
$b_0$ 1.205395e-01 1.198958e-01 2.456749e-01 1.975443e-01
SSE 4.402746e+01 3.611262e+01 3.527619e+01 3.062033e+01

The Internals

It is possible to get at the internals of the Exponential Smoothing models.

Here we show some tables that allow you to view side by side the original values \(y_t\), the level \(l_t\), the trend \(b_t\), the season \(s_t\) and the fitted values \(\hat{y}_t\). Note that these values only have meaningful values in the space of your original data if the fit is performed without a Box-Cox transformation.

[9]:
fit1 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="add",
    initialization_method="estimated",
).fit()
fit2 = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="mul",
    initialization_method="estimated",
).fit()
[10]:
df = pd.DataFrame(
    np.c_[aust, fit1.level, fit1.trend, fit1.season, fit1.fittedvalues],
    columns=[r"$y_t$", r"$l_t$", r"$b_t$", r"$s_t$", r"$\hat{y}_t$"],
    index=aust.index,
)
df.append(fit1.forecast(8).rename(r"$\hat{y}_t$").to_frame(), sort=True)
[10]:
$\hat{y}_t$ $b_t$ $l_t$ $s_t$ $y_t$
2005-01-01 44.584128 0.597822 34.297580 10.286548 41.7275
2005-04-01 24.938189 0.597822 34.895402 -9.957213 24.0418
2005-07-01 33.005766 0.597822 35.493224 -2.487458 32.3281
2005-10-01 37.031107 0.597822 36.091046 0.940061 37.3287
2006-01-01 46.975416 0.597822 36.688868 10.286548 46.2132
2006-04-01 27.329477 0.597822 37.286690 -9.957213 29.3463
2006-07-01 35.397053 0.597822 37.884512 -2.487458 36.4829
2006-10-01 39.422395 0.597822 38.482334 0.940061 42.9777
2007-01-01 49.366704 0.597822 39.080156 10.286548 48.9015
2007-04-01 29.720765 0.597822 39.677978 -9.957213 31.1802
2007-07-01 37.788341 0.597822 40.275800 -2.487458 37.7179
2007-10-01 41.813683 0.597822 40.873622 0.940061 40.4202
2008-01-01 51.757991 0.597822 41.471443 10.286548 51.2069
2008-04-01 32.112052 0.597822 42.069265 -9.957213 31.8872
2008-07-01 40.179629 0.597822 42.667087 -2.487458 40.9783
2008-10-01 44.204970 0.597822 43.264909 0.940061 43.7725
2009-01-01 54.149279 0.597822 43.862731 10.286548 55.5586
2009-04-01 34.503340 0.597822 44.460553 -9.957213 33.8509
2009-07-01 42.570917 0.597822 45.058375 -2.487458 42.0764
2009-10-01 46.596258 0.597822 45.656197 0.940061 45.6423
2010-01-01 56.540567 0.597822 46.254019 10.286548 59.7668
2010-04-01 36.894628 0.597822 46.851841 -9.957213 35.1919
2010-07-01 44.962204 0.597822 47.449663 -2.487458 44.3197
2010-10-01 48.987546 0.597822 48.047485 0.940061 47.9137
2011-01-01 58.931855 NaN NaN NaN NaN
2011-04-01 39.285916 NaN NaN NaN NaN
2011-07-01 47.353492 NaN NaN NaN NaN
2011-10-01 51.378834 NaN NaN NaN NaN
2012-01-01 61.323142 NaN NaN NaN NaN
2012-04-01 41.677203 NaN NaN NaN NaN
2012-07-01 49.744780 NaN NaN NaN NaN
2012-10-01 53.770121 NaN NaN NaN NaN
[11]:
df = pd.DataFrame(
    np.c_[aust, fit2.level, fit2.trend, fit2.season, fit2.fittedvalues],
    columns=[r"$y_t$", r"$l_t$", r"$b_t$", r"$s_t$", r"$\hat{y}_t$"],
    index=aust.index,
)
df.append(fit2.forecast(8).rename(r"$\hat{y}_t$").to_frame(), sort=True)
[11]:
$\hat{y}_t$ $b_t$ $l_t$ $s_t$ $y_t$
2005-01-01 43.005403 0.620938 35.016196 1.228157 41.7275
2005-04-01 26.352951 0.620938 35.637134 0.739480 24.0418
2005-07-01 33.284730 0.620938 36.258072 0.917995 32.3281
2005-10-01 36.719501 0.620938 36.879009 0.995675 37.3287
2006-01-01 46.055840 0.620938 37.499947 1.228157 46.2132
2006-04-01 28.189635 0.620938 38.120885 0.739480 29.3463
2006-07-01 35.564801 0.620938 38.741822 0.917995 36.4829
2006-10-01 39.192509 0.620938 39.362760 0.995675 42.9777
2007-01-01 49.106277 0.620938 39.983698 1.228157 48.9015
2007-04-01 30.026319 0.620938 40.604635 0.739480 31.1802
2007-07-01 37.844872 0.620938 41.225573 0.917995 37.7179
2007-10-01 41.665517 0.620938 41.846510 0.995675 40.4202
2008-01-01 52.156714 0.620938 42.467448 1.228157 51.2069
2008-04-01 31.863004 0.620938 43.088386 0.739480 31.8872
2008-07-01 40.124942 0.620938 43.709323 0.917995 40.9783
2008-10-01 44.138525 0.620938 44.330261 0.995675 43.7725
2009-01-01 55.207150 0.620938 44.951199 1.228157 55.5586
2009-04-01 33.699688 0.620938 45.572136 0.739480 33.8509
2009-07-01 42.405013 0.620938 46.193074 0.917995 42.0764
2009-10-01 46.611533 0.620938 46.814012 0.995675 45.6423
2010-01-01 58.257587 0.620938 47.434949 1.228157 59.7668
2010-04-01 35.536372 0.620938 48.055887 0.739480 35.1919
2010-07-01 44.685084 0.620938 48.676824 0.917995 44.3197
2010-10-01 49.084541 0.620938 49.297762 0.995675 47.9137
2011-01-01 61.308024 NaN NaN NaN NaN
2011-04-01 37.373056 NaN NaN NaN NaN
2011-07-01 46.965154 NaN NaN NaN NaN
2011-10-01 51.557549 NaN NaN NaN NaN
2012-01-01 64.358461 NaN NaN NaN NaN
2012-04-01 39.209740 NaN NaN NaN NaN
2012-07-01 49.245225 NaN NaN NaN NaN
2012-10-01 54.030557 NaN NaN NaN NaN

Finally lets look at the levels, slopes/trends and seasonal components of the models.

[12]:
states1 = pd.DataFrame(
    np.c_[fit1.level, fit1.trend, fit1.season],
    columns=["level", "slope", "seasonal"],
    index=aust.index,
)
states2 = pd.DataFrame(
    np.c_[fit2.level, fit2.trend, fit2.season],
    columns=["level", "slope", "seasonal"],
    index=aust.index,
)
fig, [[ax1, ax4], [ax2, ax5], [ax3, ax6]] = plt.subplots(3, 2, figsize=(12, 8))
states1[["level"]].plot(ax=ax1)
states1[["slope"]].plot(ax=ax2)
states1[["seasonal"]].plot(ax=ax3)
states2[["level"]].plot(ax=ax4)
states2[["slope"]].plot(ax=ax5)
states2[["seasonal"]].plot(ax=ax6)
plt.show()
../../../_images/examples_notebooks_generated_exponential_smoothing_22_0.png

Simulations and Confidence Intervals

By using a state space formulation, we can perform simulations of future values. The mathematical details are described in Hyndman and Athanasopoulos [2] and in the documentation of HoltWintersResults.simulate.

Similar to the example in [2], we use the model with additive trend, multiplicative seasonality, and multiplicative error. We simulate up to 8 steps into the future, and perform 1000 simulations. As can be seen in the below figure, the simulations match the forecast values quite well.

[2] Hyndman, Rob J., and George Athanasopoulos. Forecasting: principles and practice, 2nd edition. OTexts, 2018.

[13]:
fit = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="mul",
    initialization_method="estimated",
).fit()
simulations = fit.simulate(8, repetitions=100, error="mul")

ax = aust.plot(
    figsize=(10, 6),
    marker="o",
    color="black",
    title="Forecasts and simulations from Holt-Winters' multiplicative method",
)
ax.set_ylabel("International visitor night in Australia (millions)")
ax.set_xlabel("Year")
fit.fittedvalues.plot(ax=ax, style="--", color="green")
simulations.plot(ax=ax, style="-", alpha=0.05, color="grey", legend=False)
fit.forecast(8).rename("Holt-Winters (add-mul-seasonal)").plot(
    ax=ax, style="--", marker="o", color="green", legend=True
)
plt.show()
../../../_images/examples_notebooks_generated_exponential_smoothing_24_0.png

Simulations can also be started at different points in time, and there are multiple options for choosing the random noise.

[14]:
fit = ExponentialSmoothing(
    aust,
    seasonal_periods=4,
    trend="add",
    seasonal="mul",
    initialization_method="estimated",
).fit()
simulations = fit.simulate(
    16, anchor="2009-01-01", repetitions=100, error="mul", random_errors="bootstrap"
)

ax = aust.plot(
    figsize=(10, 6),
    marker="o",
    color="black",
    title="Forecasts and simulations from Holt-Winters' multiplicative method",
)
ax.set_ylabel("International visitor night in Australia (millions)")
ax.set_xlabel("Year")
fit.fittedvalues.plot(ax=ax, style="--", color="green")
simulations.plot(ax=ax, style="-", alpha=0.05, color="grey", legend=False)
fit.forecast(8).rename("Holt-Winters (add-mul-seasonal)").plot(
    ax=ax, style="--", marker="o", color="green", legend=True
)
plt.show()
../../../_images/examples_notebooks_generated_exponential_smoothing_26_0.png