The question is inspired by a short passage on the LMM in Mark Joshi's book.
The LMM cannot be truly Markovian in the underlying Brownian motions due to the presence of state-dependent drifts. Nevertheless, the drifts can be approximated 'in a Markovian way' by using predictor-corrector schemes to make the rates functions of the underlying increments across a single step.
Ignoring the drifts, the LMM would be Markovian in the underlying Brownian motions if the volatility function is separable. The volatility function $\sigma_i(t)$ is called separable if it can be factored as follows $$\sigma_i(t)=v_i\times\nu(t),$$ where $v_i$ is a LIBOR specific scalar and $\nu(t)$ is a deterministic function which is the same across all LIBOR rates.
- The separability condition above is sufficient for the LMM to be Markovian in the Brownian motion. How far is it from being a necessary one?
- What is the intuition behind the separability condition?
- Are there any weaker sufficient conditions?