Topic
Daniel H Wagner
Co-Topic 1
Daniel H Wagner
Co-Topic 2
Daniel H Wagner

Transportation marketplace rate forecast using signature transform

1:22 PM - 1:59 PM
Haotian Gu, UC Berkeley; Xin Guo, University of California-Berkeley; Timothy Jacobs, Amazon; Philip Kaminsky, Amazon + UC Berkeley; Xinyu Li, UC Berkeley.
This work develops a novel statistical method leveraging signature transforms to predict freight transportation marketplace rates. Our approach utilizes the universal nonlinearity property of signature transform to linearize the feature space and hence translates the forecasting problem into linear regression, and uses signature kernels for efficient comparison of time series data. This enables precise feature generation and identification of regime switching. Deployed by Amazon trucking operations, our algorithm surpasses industry models by improving prediction accuracy by over fivefold.