The referendum on Scottish independence this week will be a rare data point in a difficult problem, should the Scots vote yes. For a perspective on the issues related to that referendum you could simply watch John Oliver. This post is not about Scotland but territories like it and the difficult problem they pose: is it a good or a bad idea for Scotland or any other territory to break away?
Nation states, almost by definition, are historical accidents. There’s no real reason for their boundaries to be what they are. Catalonia, Scotland, South Sudan and Tamil Nadu are all very different from the respective countries/kingdoms that they are or were part of. From an economic perspective the four are evenly divided: Scotland and South Sudan contributed much less to the larger nation state’s GDP per capita compared to the mean. In the case of Catalonia and Tamil Nadu, the contribution is much higher than the mean. That the former two seem to have a vote already while the latter two don’t is something worth pondering. Nevertheless, this post is not about Economics.
The simple problem statement is: how do we know the point of optimality between sticking on to a larger nation state and breaking away? The problem, to one’s uninformed first brush, looks remarkably similar to the Multi-arm Bandit Problem. The problem, stated simply, is this: if you go to a casino that has N machines you can play on, at what point in 1 through N do you decide to stop exploring the multiple options in N and instead settle on one specific machine to exploit its returns? That’s the reason most textbooks us the term ‘explore versus exploit trade-off‘ to refer to this optimality. For a country, the options of breaking away sounds a bit like explore while staying back sounds a bit like exploit.
The problem, as we now know, is a solved one. It’s also relatively straightforward to solve. A Bayesian optimisation is what almost all text books will prescribe to such a problem. The issue we have however is, there are not enough samples for us. That is, not enough countries have broken away to have explored the option enough so that a Bayesian optimisation problem can be set up.
Perhaps the the erstwhile Soviet republics are a good bunch to start. But what is undeniable is that to set this problem up and arrive an optimisation function that can reasonably tell a territory whether it should or shouldn’t break up, a lot more territories need to do that first. For the sake knowledge and for the sake of a more informed tomorrow, this only means Tamil Nadu and every other territory that feels marginally alienated from its nation-state needs to have a referendum for independence. And preferably break away.
 – David Cameron’s warning that Scotland can never return should it break away has been ignored in this analogy.
 – Here is a ‘no regret’ experimental design. Interesting.