ARIMA handling of multiple seasonalities

Dear all,

I will have to forecast a large set of daily time series data for an
ATM cash management system. There are over two years of historic data,
and most likely there will be overlaying seasonalitis (weekly, monthly,
possibly bi-weekly) reflecting regular cash withdrawal patterns. Plus,
calendar effects and outliers will also be present.

My question: What is the best way to model multiple seasonalities to
produce most accurate forecasts? Exponential smoothing will not be able
to handle external events. Unobserved component models (structure
models) are too computationally involved. I'm thinking about
ARIMA/transfer functions where I model one seasonality explicitly with
seasonal ARIMA terms and use intervention analysis (dummy codes) for
remaining seasonalities.

Are there any better ideas out there? Is there any place on the web
where these kind of problems are discussed.

Thanks in advance,

Sahrian Ensuman

.