I have a cash projection model in Excel.
I have around 10 years of monthly data.
There are around 10 sources of cash and around 10 uses of cash.
I have determined the type of distribution which best fits each of the 20 series.
I have calculated the pertinent statistical parameters of each of the 20 series (e.g. mean or mode, standard deviation, etc.)
I have fitted the selected distributions with their respective parameters.
Some of the series have a normal distribution around a relatively high mean value.
Some of the series have a more Rayleigh or Weibull distribution - always positive but not symmetrical.
The purpose of the model is to predict the cash position over each of the next 12 months.
In some cases, a series has a growth factor to be included.
Each month, for each series, the current actual value is appended resulting in a new series of actual values.
Then an expected value is calculated for each of the next 12 months.
The question is:
Given a random series with known distribution parameters, what is a good way of generating expected value predictions?
In some cases, it appears that the best prediction is the current value.
However, if there's an outlier in the current value, this doesn't work well.