I work in the risk department of a credit company and I was asked to analyse and bring some statistical evidence that two portfolios have different risk behaviour. So, my goal is to find if, for example, consumers of credit cards tend to not pay less or more often than consumers of personal loans. My data is their total balance over the course of one year segmented by their numbers of unpaid installments. Something Like this:
My manager suggested me to first obtain the percentage of each unpaid level on the monthly total, and then apply 10 t-tests (one to each unpaid level) to evaluate if they each unpaid level accounts for a similar part of the total balance in cards and personal loans. If I find that they are individually are different we can say that the complete portfolios behave in different way. What do you think of this approach? Do you have a better idea?