THE PURPOSE OF THIS CHAPTER

In the previous chapter we have seen the strengths and weaknesses of latentvariables models. Starting with this chapter we return to specified-variable models. We do so by looking in some detail at the approach by Diebold and Rudebusch (2013), who take a ‘snapshot model’ and turn it into a dynamic affine model whose state variables can be interpreted as proxies for the Principal Components.

Since we know a lot about principal components, and they afford a ready intuition in terms of level, slope and curvature of the yield curve, this is very desirable, and facilitates the calibration of the model. Indeed, this insight will be the motivation for the Principal Component–based model presented in the next chapter. As we shall see in the next chapter, however, the transition from state variables that ‘look a bit like principal components’ to state variables that are exactly principal components is not all plain sailing.

We stress that in this chapter we give only a partial account of the Diebold– Rudebusch approach – in particular, we do not cover the important aspect of the statistical estimation of the model. This is a pity because, as Diebold and Rudebusch, say, their model

The model presented in this chapter, the authors continue, ‘bridges the divide with a [statistically]-inspired model that enforces absence of arbitrage’.

So, our sin of omission, when we fail adequately to cover in this chapter the statistical (parameter-estimation) aspect of the model, is not a minor blemish, as it fails to do full justice to an important aspect of the Diebold–Rudebusch approach. Since, however, the econometric aspect is very well covered in the book by Diebold and Rudebusch, brevity and a tight focus on the book narrative have recommended the skewed treatment we present below.