They basicly assumed the yield curve can be built from 3 mathematical parts and that they estimate the historic betas, after which they tried to forecast the betas with an AR 1, the results showed significance over a random walk.
Thanks, very interesting. I remember reading a paper in which they build a model based on PCA to forecast the yield curve, basically just use ARIMA model to forecast the PCs, once they're calculated
yeah, at the end of day, it's also 3 factor model (they used 3 PCs). They analyzed the model performance from FI arb trading perspective and it was acceptable.
I recently read an intersting paper about forecasting the term structure:
https://www.sas.upenn.edu/~fdiebold/papers/paper49/Diebold-Li.pdf
They basicly assumed the yield curve can be built from 3 mathematical parts and that they estimate the historic betas, after which they tried to forecast the betas with an AR 1, the results showed significance over a random walk.
Thanks, very interesting. I remember reading a paper in which they build a model based on PCA to forecast the yield curve, basically just use ARIMA model to forecast the PCs, once they're calculated
I would say that is quite a similar approach, but PCA allows more flexibility I assume, how were the results?
yeah, at the end of day, it's also 3 factor model (they used 3 PCs). They analyzed the model performance from FI arb trading perspective and it was acceptable.
Thank you Nam, would you mind sharing that paper?
I read it a long time ago, circa 2014. If I find it I'll send it to you