Thank you for your detailed feedback — very helpful points.
I will add graphical illustrations of the functional specification for each driver (Age and Spread) to make the non-linear structure more intuitive.
Regarding the average forecast level not being fully aligned with realised rates: the current redevelopment strictly follows the modelling guide, which prescribes both the functional structure and the use of only two drivers (Age and Spread). Within this constrained framework, the explanatory power is naturally limited, as the relationship between these two variables and the realised prepayment series is only partially structural and does not fully capture regime shifts or episodic shocks embedded in the data.
In this context, even more complex algorithms would not materially improve performance without expanding the information set. Improving statistical fit artificially (e.g., through additional transformations or data smoothing) would risk reducing the economic representativeness of the portfolio. For governance and transparency reasons, we have intentionally avoided such adjustments.
That said, performance could be enhanced in future iterations by testing additional explanatory variables and more flexible modelling approaches. However, this would require a broader methodological discussion and is not feasible within the current short-term scope defined by the guide.
In practical application, any necessary level alignment would be implemented transparently at portfolio aggregation level rather than modifying the structural core of the model. I will also include a focused comparison on recent history to illustrate behaviour versus realised outcomes more clearly.
On your additional questions:
• Financial incentive definition: this is defined as the difference between the contractual loan rate and the relevant market refinancing benchmark. I can expand this section further if helpful.
• Liquidity vs IR prepayment rates: although the aggregate difference appears small, the economic decomposition is distinct (refinancing incentive vs behavioural/liquidity component). I will include a more detailed 10-year view and a zoomed-in recent-history graph for clarity.
I will circulate an updated version shortly.
Best regards,
Ilker
Yorumlar
Yorum Gönder