statsmodels: SARIMAX model too large

Describe the bug

Team, I’m running SARIMAX for daily data and trying to catch annual seasonality. I used seasonal_order = (0, 1, 0, 365) but I faced the following issues:

• Model took a few minutes to generate, which is too long (x10 or so) compared to R’s forecast$auto.arima function
• Model generated was over 7 GB which is also too large compared to R’s forecast$auto.arima

The seasonal_order I’m using is what auto.arima (in R) suggested for the data I’m feeding the model. Data is confidential but I generated a similar dataset (below) to replicate the issue and show my code.

If I’m missing something please let me know. Otherwise, any suggestions on how to make SARIMAX work for this data. I’d like to catch the annual seasonality first, I can account for the weekly later.

I greatly appreciate you looking into this.

Thanks, Sam

Code Sample, a copy-pastable example if possible

import numpy as np
import pandas as pd
import statsmodels as sm
from statsmodels.tsa.statespace.sarimax import SARIMAX
import matplotlib.pyplot as plt

print(f"statsmodels version: {sm.version.full_version}")
# >>> statsmodels version: 0.9.0

# Variables and data
figsize = (16, 8)
freq = "D"
n = 800
np.random.seed(7)
df = pd.DataFrame(
    data={
        "ds": pd.date_range("2016-11-01", periods=n, freq=freq),
        "y": np.random.rand(n),
    }
)
df.set_index("ds", drop=False, inplace=True)
df["exo1"] = df.index.month
df["exo2"] = df.index.week

df.loc[:, "y"] = df["y"] + df.index.dayofweek
add_fact = (
    (df.index.month == 12)
    * 1
    * (df.index.week.values - pd.to_datetime("2018-11-25").week)
    * 4
)
add_fact[add_fact < 0] = 0
df["add_fact"] = add_fact
df.loc[:, "y"] += df["y"] * df["add_fact"]

df.loc[:, "y"].plot(figsize=figsize)

# Train / predict
split_date = "2018-11-30"
m = SARIMAX(
    endog=df.loc[:split_date, "y"],
    exog=df.loc[:split_date, ["exo1", "exo2"]],
    order=(1, 1, 0),
    seasonal_order=(0, 1, 0, 365),
)
m_fit = m.fit()
pred = m_fit.forecast(
    df.loc[split_date:, "y"].shape[0], exog=df.loc[split_date:, ["exo1", "exo2"]]
)

# Visualize
fig = plt.figure(figsize=figsize)
ax = fig.add_subplot(111)
df["y"].plot(figsize=figsize, ax=ax)
pred.plot(figsize=figsize, ax=ax)

Expected Output

A clear and concise description of what you expected to happen.

Output of import statsmodels.api as sm; sm.show_versions()

[paste the output of sm.show_versions() here below this line]

statsmodels version: 0.9.0

I got this error running sm.show_version(): AttributeError: module ‘statsmodels’ has no attribute ‘show_versions’