Distributed rounding of a trial balance

Transcribed from the printed issue; not yet reviewed. Read it in the PDF of the issue, page 112

In J-ottings No. 26 Norman Thomson gives an interesting solution to the problem of distributed rounding, but only for positive numbers. Accountants have a little problem, which I have never been able to solve totally satisfactorily, where there is a mixture of positive and negative items which sum to zero. A Trial Balance is effectively a vector of detail items which sum to zero. Rounding the Trial Balance usually, but not always, enables a rounded version of the Balance Sheet and Profit & Loss Account to be prepared with all the intermediate totals which are near the correct values. One way of rounding the Trial Balance is to apply distributed rounding independently to the two sets of positive and negative items, but sometimes this rounds items the ‘wrong’ way unnecessarily. Does Norman know of a better method?