What does the Box Jenkins method do for cash flow forecasting?

Generally speaking, there are two broad approaches to forecasting anything: time series and correlative methods. Consider a forecast for cash receipts. If I believe that I have some data or “leading indicator” that suggests what cash receipts might be in the future then I might employ a correlative approach. This leading indicator might be, say, Accounts Receivable. If I believe that the level of accounts receivable at the beginning of the month has some predictive power over the amount of cash I receive in the month then I might regress historical Accounts Receivable on historical cash receipts (regression being a correlative approach) to obtain a forecasting model. Of course, this is a much simplified example.

Of course, I may not have a leading indicator of future cash receipts. If not, then I may be forced to use a pure time-series technique that builds a forecast using only past values of cash receipts itself (and no other data). The assumption is that future cash receipts are somehow related to past values. In short, time-series methods attempt to find trends, seasonality or cyclicality in historical data and extrapolate those patterns into the future.

Box-Jenkins (a.k.a ARIMA, ARMA, autoregressive models, etc.) is the classic time-series approach to forecasting. A Box-Jenkins model may have autoregressive (AR) parameters which relate future cash flows to past cash flows and moving average (MA) parameters which relate future cash flow to past forecasting errors. It is an extremely powerful method that I frequently use when little historical data is available or, as previously stated, I can’t find a leading indicator to correlate with future cash flows. Note that the exponential smoothing forecasting model taught in the CTP curriculum is a simplified version of the Box-Jenkins setup. However, while exponential smoothing is easy to do in an Excel spreadsheet fitting a Box-Jenkins model is not. Developing these models typically requires specialized software such as SAS or SPSS (it can be done in Excel if you are familiar with Excel’s Solver package and what to take the time to set up all the equations to estimate the model).

Hybrid approaches also exist (often termed ARMAX) that combine the time series benefits of Box Jenkins with the correlative benefits of regression analysis. Googling ARMAX will give you all the background you could desire.

2