The forecasting accuracy of models of post-award network deployment: An application of maximum score tests

Each mobile network operator’s spectrum is assigned by national governments. Licenses awarded by auctions are tied to post‐award network deployment obligations. Using data on 18 countries for the period 2000–2007, this study is the first to empirically forecast aftermarket performance by analysing whether these conditions are met in a timely fashion. The forecasts are conditioned on macroeconomic and market conditions, and package attributes. The models are evaluated based on Mayer and Wu’s (in press) maximum score tests. Traditional probit models are not robust to error misspecifications. However, Manski’s (1975, 1985) maximum score estimator only imposes median independence, and allows arbitrary heteroskedasticity. One obstacle to empirical implementation is the fact that the asymptotic distribution of the estimator cannot be used for hypothesis testing. Mayer and Wu address the problem using a ‘discretisation’ procedure. The tests do not impose additional assumptions on the data generating process, require a shorter computational time than subsampling, and allow the models to be misspecified. The test statistics reflect differences in forecasting accuracy under the null and alternative hypotheses.