Skip to main content

​
What is Long-Horizon Forecasting?

Long-horizon forecasting refers to predictions far into the future, typically exceeding two seasonal periods. For example, forecasting electricity demand 3 months ahead for hourly data, or predicting sales 2 years ahead for monthly data. The exact threshold depends on data frequency. The further you forecast, the more uncertainty you face. The key challenge with long-horizon forecasting is that these predictions extend so far into the future that they may be influenced by unforeseen factors not present in the initial dataset. This means long-horizon forecasts generally involve greater risk and uncertainty compared to short-term predictions. To address these unique challenges, Nixtla provides the specialized timegpt-1-long-horizon model in TimeGPT. You can access this model by simply specifying model="timegpt-1-long-horizon" when calling nixtla_client.forecast.

​
When to Use Long-Horizon Forecasting

Long-horizon forecasting is ideal for:
  • Supply chain planning: Predict inventory needs 3-6 months ahead
  • Financial forecasting: Model quarterly or annual revenue projections
  • Energy demand: Forecast power consumption weeks or months in advance
  • Climate modeling: Predict seasonal weather patterns
Use the timegpt-1-long-horizon model when your forecast horizon exceeds two complete seasonal cycles in your data.

​
How to Use the Long-Horizon Model

Open In Colab

​
Step 1: Import Packages

Start by installing and importing the required packages, then initialize the Nixtla client: 
from nixtla import NixtlaClient
from datasetsforecast.long_horizon import LongHorizon
from utilsforecast.losses import mae

nixtla_client = NixtlaClient(
    api_key='my_api_key_provided_by_nixtla'  # defaults to os.environ.get("NIXTLA_API_KEY")
)

​
Step 2: Load the Data

We’ll demonstrate long-horizon forecasting using the ETTh1 dataset, which measures oil temperatures and load variations on an electricity transformer in China. Here, we only forecast oil temperatures (y): 
Y_df, *_ = LongHorizon.load(directory='./', group='ETTh1')

Y_df.head()
unique_iddsy
0OT2016-07-01 00:00:001.460552
1OT2016-07-01 01:00:001.161527
2OT2016-07-01 02:00:001.161527
3OT2016-07-01 03:00:000.862611
4OT2016-07-01 04:00:000.525227
We’ll set our horizon to 96 timestamps (4 days) for testing and use the previous 42 days as input to the model: 
test = Y_df[-96:]              # 96 timestamps (4 days × 24 hours/day)
input_seq = Y_df[-1104:-96]    # 1008 timestamps (42 days × 24 hours/day)

​
Step 3: Forecasting with the Long-Horizon Model

TimeGPT’s timegpt-1-long-horizon model is optimized for predictions far into the future. Specify it like so: 
fcst_df = nixtla_client.forecast(
    df=input_seq,
    h=96,
    level=[90],
    finetune_steps=10,
    finetune_loss='mae',
    model='timegpt-1-long-horizon',
    time_col='ds',
    target_col='y'
    )
Next, plot the forecast along with 90% confidence intervals: 
nixtla_client.plot(
    Y_df[-168:],
    fcst_df,
    models=['TimeGPT'],
    level=[90],
    time_col='ds',
    target_col='y'
)
Long-horizon time series forecast showing predicted oil temperature with 90% confidence intervals over 96 hours

TimeGPT Long-Horizon Forecast with 90% Confidence Intervals

​
Step 4: Evaluation

Finally, assess forecast performance using Mean Absolute Error (MAE): 
test = test.copy()
test.loc[:, 'TimeGPT'] = fcst_df['TimeGPT'].values

evaluation = mae(
    test,
    models=['TimeGPT'],
    id_col='unique_id',
    target_col='y'
)
Evaluation result:
unique_idTimeGPT
OT0.145393
The model achieves a MAE of approximately 0.146, indicating strong performance for these longer-range forecasts.

​
Frequently Asked Questions

Q: What’s the difference between timegpt-1 and timegpt-1-long-horizon? The timegpt-1-long-horizon model is specifically trained for extended forecast horizons (2+ seasonal periods), providing better accuracy for long-range predictions. Q: How far ahead can I forecast with the long-horizon model? The optimal horizon depends on your data frequency and patterns. Generally, the model performs well up to 4-6 seasonal cycles ahead. Q: Can I use exogenous variables with long-horizon forecasting? Yes, TimeGPT supports exogenous variables for improved long-horizon accuracy. See our exogenous variables guide for details. Learn more about TimeGPT capabilities: