Why does the forecast decay excess growth by 28 per cent a year?

[ShreyasThakur11]

The projection applies 28 per cent annual mean reversion to the excess over the therapy-area rate. That constant is doing a lot of work in a five-year number.

Where does it come from, and how defensible is any five-year projection built on two years of history?

[Amey-Thakur]

Taking the second question first, because it is the more important one: it is not defensible as a forecast, and the repo does not claim it is.

MAT February 2024, 2025 and 2026 give two year-on-year comparisons. That supports a direction. It does not support a cycle, a seasonal decomposition or a confidence interval, and the three supplied months are a momentum signal rather than a time series. docs/LIMITATIONS.md states this first among the data limitations. Every projection returns its assumptions alongside the number and the agent is instructed to present these as structured extrapolations, not plans.

Given that, the decay is a guard against a specific arithmetic failure rather than an estimate of anything. A space growing at 45 per cent, compounded flat for five years, is 6.4 times its current size. No cardiac sub-segment in this market has done that, and a model that says so has produced a number that discredits everything around it. Decaying the excess at 28 per cent a year halves it in a little over two years, which pulls a fast starter back towards the therapy-area rate on roughly the timescale that competitive entry actually does.

Three other bounds work with it: the starting rate is capped to [-15%, +45%], scenarios shift it 4 percentage points either way, and the base rate itself is weighted towards demand at 0.40 real growth and 0.30 volume growth.

The constant is in config/settings.yaml. A reader who thinks 20 or 40 is better calibrated can change it and re-run, which is a more productive disagreement than arguing about the number in the abstract.