Here is a simple model a mathematician might realistically use to model this. Given N traders, let P_1..P_N be iid Beta(α, β), and let the chance that trader n loses money during a year be Bernoulli(P_n). The nice thing about this more complicated model is that the simpler model is a limit case of it.
This "first approximation" lets us use simple mathematical tools to test hypotheses like that by providing an alternative model. If our beta distribution looks more like a constant distribution (with high values for α and β) then we can conclude that the number of traders who lose money approaches 100% as time goes on. If the estimate for α and β is lower, we might reasonably conclude that there is some factor of skill involved.
This is just one possible model you could use to test if there is skill involved in day trading. There are many other models which we could test which would also mean that the proportion does not approach 100%, such as a models which assume correlations between different years.