Multi-Stage Dividend Models & Rounding

Has anyone found what appears to the the appropriate number of decimals to round? On some problems it is not an issue but others where r and g are very close the model is very sensitive. I had one 3 stage model off by nearly $18, following the book examples I had the correct inputs following the problem in the CFIA book but im assuming since I did not round my intermediate calculations (I stored them in the registers) it threw the model off.