[Python] DFM Statsmodels contemporaneus effects

I'm working with a DynamicFactor model in statsmodels that includes two monthly variables (x1, x2) and one quarterly variable (y). The quarterly variable is only observed every three months (e.g., March, June, September, December), and the rest of the time it's set to NaN.

I estimate the model using data up to November, then update it by adding December's values for the monthly variables (keeping y as NaN for December). However, the nowcast for y in December does not change after this update, even though the new monthly data should, in theory, influence the latent factor and thus the forecast.

Why doesn't the December monthly data contemporaneously affect the nowcast of the quarterly variable? Is this a limitation of how statsmodels handles mixed-frequency data, or am I missing something in how the model updates predictions?

I'm attaching the full code below so you can reproduce the issue and verify the behavior.

Any insights or suggestions would be greatly appreciated!

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from statsmodels.tsa.statespace.dynamic_factor import DynamicFactor

# ------------------------------------------
# 1. Simulate data
# ------------------------------------------

np.random.seed(42)
monthly_dates = pd.date_range(start="2020-01-01", periods=48, freq="M") # 4 years of data
quarterly_dates = monthly_dates[::3]

factor = np.cumsum(np.random.normal(0, 0.5, len(monthly_dates)))
x1 = factor + np.random.normal(0, 0.3, len(monthly_dates))
x2 = 0.8 * factor + np.random.normal(0, 0.3, len(monthly_dates))
y = pd.Series(0.6 * factor[::3] + np.random.normal(0, 0.2, len(quarterly_dates)), index=quarterly_dates)

# Full dataset with NaNs in y where there is no quarterly observation
df = pd.DataFrame(index=monthly_dates)
df["x1"] = x1
df["x2"] = x2
df["y"] = np.nan
df.loc[y.index, "y"] = y

# ------------------------------------------
# 2. Estimate model with data up to November
# ------------------------------------------

df_train = df.iloc[:-1].copy() # Up to November
df_new_month = df.iloc[[-1]].copy() # Only December (with y = nan)

mod = DynamicFactor(df_train, k_factors=1, factor_order=1, error_order=1)
res = mod.fit(disp=False)

# Nowcast for December before incorporating December
pred_old = res.get_prediction(start='2023', end='2024')
nowcast_before = pred_old.predicted_mean["y"]

# ------------------------------------------
# 3. Update only with December (x1 and x2), without re-estimating
# ------------------------------------------

# Create new observation with December only in x1 and x2 (y remains NaN)
updated_data = pd.concat([df_train, df_new_month])

# Apply the already estimated model to the new data
res_updated = res.apply(updated_data)

# Get new nowcast for December
pred_new = res_updated.get_prediction(start='2023', end='2024')
nowcast_after = pred_new.predicted_mean["y"]

# ------------------------------------------
# 4. Show results
# ------------------------------------------
print("Nowcast before adding December data:", nowcast_before['2023-12-31'])
print("Nowcast after adding December data:", nowcast_after['2023-12-31'])

Continue reading...