Treasury Yield Curve Predictions

The prototype utilizes a multivariant sequential regression model within Keras and Tensorflow to predict US Treasury Yield Curves. The model uses quarterly Real GDP, Real Disposable Income, US Unemployment, CPI Inflation, Fed Discount Rates and Chinese Discount Rates. The 3 month, 6 month, 1 year, 2 year, 3 year and 10 year US Treasury Yields are generated. A custom R Squared calculation helped optimize the model fit. The code and data can be found here.

The predicted Q2 2021 rate curve is displayed here (it is the last graph). It is preceded by two other graphs which compare the predicted yield curves to actuals from the previous two quarters. At this point in time, the model almost always predicts a steeper yield curve than the actual market given other dependencies which need to be implemented with more Time Series awareness. However, I think the model would be fairly accurate at predicting increases or decreases in rates.

I will be building off this to create an Equity Option time series model with LSTM Recurrent Neural Networks using Keras and Tensorflow. Please let me know if you would like to review or discuss this sort of work.