It depends. If the change is linear, quadratic, cubic, or can be modelled by a spline/fractional polynomial, you can use rANOVA, but it has very strict assumptions on the covariance structure, so GEE or mixed-effects models are better, and can also deal with some missing data, if you have them. However, if the change over time is steep and sudden, time-series or a hidden Markov model would give a better fit. However, if the number of cases you have are ten, you have absolutley no power, and then I don't know :) For these models, 20-50 "cases" is still a small sample. If deposits are cases and you have alot per bank, then you should be OK, but I could't read that from your question.