Beauty contest scoring model
A scoring and valuation model for comparative selection awards, where spectrum goes to the best proposal rather than the highest bid. Built to answer what a commitment is actually worth before offering it.
Not all spectrum is auctioned. In a comparative selection process, sometimes called a beauty contest, the regulator scores proposals against published criteria and awards to the best one. Coverage obligations, deployment timetables, wholesale access terms, sometimes price.
This is harder to prepare for than an auction, and the reason is worth stating plainly: in an auction you are trying to work out what the spectrum is worth to you. In a beauty contest you are trying to work out what a promise costs you, and then how many points that promise buys.
Costing a commitment
Every criterion in a beauty contest translates into a commitment, and every commitment has a cost that lands somewhere on a fifteen-year cash flow.
A coverage obligation is a rollout schedule: sites, backhaul, power, civils, and the operating cost of running them in places that will not pay for themselves. A deployment deadline is an option you have sold, because it removes your ability to wait for equipment prices to fall. A wholesale access commitment is margin given away on traffic you would otherwise have carried yourself.
Each of those is modelled as an incremental cash flow and discounted at WACC. That gives a cost per commitment in present value terms, which is the only basis on which two criteria can sensibly be compared.
Where the uncertainty goes
Point estimates are worse than useless here because the inputs genuinely are not known. Build cost per site, subscriber uptake in low-density areas, and equipment price trajectories all have real distributions rather than values.
So the commitment costs run as a Monte Carlo, with the correlations that matter left in. Rollout cost and uptake are not independent, and treating them as if they were understates the variance in exactly the direction that makes an aggressive bid look safe.
The output is a distribution of the total cost of a proposal, which supports a different and better question than “what does this cost”. It supports “what is the chance this proposal costs more than the spectrum is worth to us”.
Scoring the scoring rule
The part that surprised me most. The published scoring rule is itself a model input, and it is usually more sensitive than it looks.
Most schemes normalise each criterion against the best submission, then apply weights. That means your score depends on what everyone else offered, so the marginal value of one more percentage point of coverage is a function of competitor behaviour you cannot observe. Small changes in weighting, or in how a criterion is normalised, can reorder the outcome entirely.
The model therefore treats the scoring rule as something to run sensitivities across rather than a fixed constraint. The useful output is the region where a proposal wins across most plausible competitor scenarios, not the single proposal that maximises the score against one guess.
What it produced
A defensible answer to the question that actually gets asked in the room: how much more can we commit before the award stops being worth winning. That is a number, it has a distribution around it, and both come from the same model.
Method and structure only here. The underlying figures, country positions and commercial inputs are not published.