Augmented Functional Time Series Representation and Forecasting with Gaussian Processes
Abstract
We introduce a functional representation of time series which allows forecasts to be performed over an unspecified horizon with progressively-revealed informa- tion sets. By virtue of using Gaussian processes, a complete covariance matrix between forecasts at several time-steps is available. This information is put to use in an application to actively trade price spreads between commodity futures con- tracts. The approach delivers impressive out-of-sample risk-adjusted returns after transaction costs on a portfolio of 30 spreads.
Cite
Text
Chapados and Bengio. "Augmented Functional Time Series Representation and Forecasting with Gaussian Processes." Neural Information Processing Systems, 2007.Markdown
[Chapados and Bengio. "Augmented Functional Time Series Representation and Forecasting with Gaussian Processes." Neural Information Processing Systems, 2007.](https://mlanthology.org/neurips/2007/chapados2007neurips-augmented/)BibTeX
@inproceedings{chapados2007neurips-augmented,
title = {{Augmented Functional Time Series Representation and Forecasting with Gaussian Processes}},
author = {Chapados, Nicolas and Bengio, Yoshua},
booktitle = {Neural Information Processing Systems},
year = {2007},
pages = {265-272},
url = {https://mlanthology.org/neurips/2007/chapados2007neurips-augmented/}
}