Time Series Based Treasury Yield Curve Prediction

We finished an improved US Treasury Yield curve prediction model utilizing a LSTM (Long Short-Term Memory) based recurrent neural network. The solution is based on the Keras and TensorFlow libraries and is implemented in Python. LSTM is the preferred approach for predicting time based financial metrics and prices.

The code and data are available here. The visualization of the 3/31/2021 predicted yield curve is viewable here.

LSTM models are much more time consuming to build and optimize given the number of parameters. However, the fit and accuracy of these models is much greater for financial use cases. This particular model could probably be optimized further and I welcome anyone to contribute to the code base on GitHub.

The 3/31/2021 predicted yield curve does factor in the newly released forecast for Q1 2021 Real GDP of 6.4% for the US Economy. here