Financial model review
Rounding errors in financial models: Excel checks
Investigate rounding differences in Excel financial models. Compare stored and displayed values, reconcile allocations and set an explained review tolerance.

To check a rounding difference in a financial model, compare the underlying values before changing the formulas. Establish where rounding belongs, rebuild the calculation on the same basis and explain any remaining difference in the workbook's units. A total can look wrong because its components are displayed with fewer decimals; a small difference can also reveal a genuine error.
Use a separate review copy and keep the original unchanged. These invented examples demonstrate calculation choices, not accounting, invoicing or tax requirements. The appropriate rounding rule depends on the model and the process it represents.
Separate the displayed number from the calculated amount
A monthly forecast might display revenue as EUR 10,000 while retaining a more precise underlying amount. Formatting the cell to show no decimals normally changes its appearance, not the value used by another formula. Microsoft's rounding guide distinguishes changing the displayed decimal places from rounding the number itself.
Inspect the formula bar, then increase the displayed decimal places in the review copy. Compare the source amounts, subtotals and final report. Check their units as well: two decimal places in EUR thousands represent a different precision from two decimal places in EUR. Our units, signs and periods guide explains that distinction.
Do not start by inserting ROUND into every formula. First decide whether the difference concerns presentation, an actual rounded transaction or an unexplained calculation.
Reproduce a one-cent difference
Enter 10.004 into each of cells B2, B3 and B4 in a blank review worksheet. Format them to show two decimal places. Each displays 10.00, so a reader adding the visible rows expects 30.00.
In B5, calculate:
=SUM(B2:B4)
The underlying total is 30.012. Displayed with two decimals, it is 30.01. Nothing has been omitted from the sum: the displayed rows conceal fractions that still contribute to the total.
Now put this formula in C2 and copy it through C4:
=ROUND(B2,2)
=SUM(C2:C4) returns 30.00, because each component was rounded before addition. In contrast, =ROUND(SUM(B2:B4),2) returns 30.01, because rounding happens after addition. Microsoft's ROUND documentation defines the second argument as the number of digits to round to.
These are two different calculations, not interchangeable ways to make a report look tidy. Record which one the model is intended to perform.
Choose the rounding point deliberately
For a forecast, retaining calculation precision and rounding the final presentation may be appropriate. If a schedule represents individual amounts that are actually settled to cents, it may instead need to sum those rounded amounts. Agree the treatment with the person responsible for that process.
Follow the chosen amount into its destinations. Suppose a payment schedule sums rounded payments, while the cash forecast uses the unrounded source total. They can disagree even though each formula performs the arithmetic written in it. Fix the inconsistent basis rather than adding an unexplained cent to cash.
If readers need displayed rows to add to a displayed total, consider showing more precision or an explicitly labelled presentation-rounding adjustment. Keep that display reconciliation separate from the underlying forecast and its business movements. A presentation adjustment is not evidence of an additional receipt or expense.
Allocate a fixed total without losing the remainder
Consider allocating EUR 100 equally across three periods. Rounding each 100/3 independently to cents produces 33.33, 33.33 and 33.33, totaling 99.99.
For this exercise, the documented rule assigns the small rounding remainder to the final period. Put the total 100 in B1, then use:
B2: =ROUND($B$1/3,2)
B3: =ROUND($B$1/3,2)
B4: =$B$1-SUM(B2:B3)
The results are 33.33, 33.33 and 33.34, totaling 100.00. The third amount is a formula derived from the fixed total and prior allocations. It is not a manually typed balancing entry.
This method assumes exactly three eligible periods and the stated final-period convention. With many periods, unequal weights, reversals or contractual allocation rules, review the allocation method separately. An unexpectedly large final remainder can expose missing periods or bad weights; do not call every residual rounding. The same discipline applies to a capex and depreciation schedule with a final charge.
Define a tolerance in the model's units
A review tolerance is a stated limit for investigating a residual. It does not alter the residual or make the underlying calculation correct. Preserve the signed difference so that its direction remains available.
For example, a reviewer might document a EUR 0.01 tolerance for a particular cent-based reconciliation. In a worksheet storing EUR thousands, that same monetary amount is 0.00001, not 0.01. A threshold of 0.01 there permits EUR 10. Decide the threshold from the reconciliation's purpose and precision rather than copying a number from another workbook.
If B8 and C8 contain the amounts being compared and E1 contains the approved tolerance in the same units, a separate flag can use:
=ABS(B8-C8)<=$E$1
Display =B8-C8 alongside the flag. A TRUE result means only that the difference is within that stated threshold. Inspect repeated differences, changing patterns and unexplained movements even when each passes. The balance-sheet checks guide shows why an unexplained plug or offsetting errors can defeat a superficial reconciliation.
Distinguish tiny numerical residue from business logic
Some decimal fractions cannot be represented exactly in finite binary storage. Microsoft's floating-point explanation describes how this can affect Excel arithmetic. A very small residual can therefore have a numerical cause, but its size alone does not establish the explanation.
Trace the calculation and compare like-for-like values. A missing invoice, duplicated cost or incorrect forecast link also creates a difference. Rounding the final residual to zero before investigating it can conceal that mistake. Review formula connections using the financial model formula checklist before changing the precision rule.
Leave workbook-wide precision settings alone during diagnosis
Set precision as displayed is different from changing a cell's number format. Microsoft warns that this workbook setting permanently changes stored values to the displayed precision; returning to full precision does not restore the discarded underlying values.
Do not enable it simply to remove a visible difference. Inspect whether it is already enabled in the review copy, record that fact and preserve a separate unmodified original or backup. An explicit rounding formula at an agreed calculation step is easier to inspect than silently changing precision across a workbook.
Finish with a reproducible review note
- Record the affected sheets, cells, units and signed difference.
- Compare source values with their displayed precision.
- Rebuild the calculation with rounding before and after aggregation.
- State the intended rounding point and any allocation-remainder rule.
- Explain the tolerance, then check linked schedules and later periods.
- Retain the review copy and a short explanation of the correction.
FinModelCheck does not recalculate Excel or automatically diagnose rounding policies, allocation methods or reconciliation differences. Its supported saved-file checks supplement this manual review; a provisional score is not confirmation that the financial arithmetic is correct.
Use the financial model checker for those supported checks and see the example report. Online results show the score and counts by Finding type, with no detailed customer Findings. The complete PDF is free without registration in the current valid owning browser session.
Frequently asked questions
Why do the visible rows fail to add to the total?
The total may use more precise underlying values than the rows display. Reveal extra decimals and compare rounding each component with rounding only the total before deciding which treatment is intended.
Should every reconciliation use a one-cent tolerance?
No. Define the units, rounding process and purpose first. One cent is an illustrative choice above, not a universal threshold for financial models.
Can a small difference be ignored after a high checker score?
No. The score covers supported saved-file checks, not rounding logic or financial reconciliation. An unexplained difference still needs the separate investigation described here.


