I don't have a complex model, it's just your test that is capturing the wrong variable. Every network you listed is "profitable" because the subsidies they get

I don't have a complex model, it's just your test that is capturing the wrong variable. Every network you listed is "profitable" because the subsidies they get are in the "income" column instead of the "subsidy" column. The distinction between profitability or making a loss is purely nominative. In the US for example, the HK transit authority would be "aiming to make a loss" because its largest source of income are the taxes it is allowed to levy. Since in the US it's the government that levies the taxes which are counted as subsidies, the HK transit authority in the US would be classified as loss-making. Additionally the MTR in HK has pseudo-governmental powers in that it gets to expropriate, rezone, and resell land at a higher price, which would also be counted as a subsidy in the US and therefore it would be considered to be loss making. So the problem isn't that my model isn't capturing a variable, it's that your test isn't capturing a variable and is taking in erroneous data points due to diverging definitions of your metric.