Email from a researcher with some questions, and particular interest in Informed Forecasting.

Seyed Amirmohammad Elahi <amir.elahi@epfl.ch>

Dear Sir/Madam,

I hope this email finds you well.

Allow me to introduce myself; I am Amir Elahi, a PhD student at EPFL, Switzerland, working in forecasting industrial variables through time series analysis. I came across your paper on TimeGPT-1 and found it to be incredibly insightful and informative. I commend your efforts in advancing the field.

I have a few questions. Specifically:

1. What is the minimum number of input sequence length?

2. What is the horizon in the historical forecasting? (In the website you said that the historical forecasting is not affected by h and it has a fixed horizon depending on the frequency of the data.)

3. I am curious about the concept of the rolling window in historical forecasting. Could you please provide more details?

4. What is the maximum forecasting horizon (h) or in other words what is the forecasting horizon that the model was trained for?

Additionally, I noticed your mention of "Informed Forecasting" as potential future work in your paper. This topic particularly intrigues me, and I would be delighted to engage in further discussions or perhaps arrange a meeting to explore it in more depth. Feel free to explore our previous work on this topic: https://www.science.org/doi/full/10.1126/sciadv.adc9576

Best,

Amir