Kuldeep Kumar’s contribution to the Discussion of ' New tools for network time series with an application to COVID-19 hospitalisations ' by Nason et al.]

I would like to congratulate the authors for demonstrating an outstanding application of a relatively new tool—Network Time Series—on COVID-19 hospitalisation data. As shown in the paper, the GNAR (1,[1]) model exhibited excellent predictive performance, outperforming standard alternatives such as VAR, sparse VAR, CARar, and AR(1).

However, I am unsure why the authors chose not to use model selection criteria such as AIC or BIC, in line with the principle of parsimony, rather than relying on ACF and PACF plots, which involve a degree of visual subjectivity. Additionally, I am curious about the decision to focus primarily on autoregressive (AR) models instead of ARIMA models, which might be more suitable for data exhibiting integration or differencing requirements. Another point of concern is whether the authors have addressed the issue of “dark data,” particularly given the prevalence of Type I and Type II errors in COVID-19 diagnosis. Accounting for such errors could have significant implications for the reliability of model predictions.