A range is a promise about how often the company will land inside it. Set the width from the size of your own forecasting errors, not from what looks tidy.
A company's internal forecast for next quarter is $1.20 of earnings per share, and its forecasts have missed by a standard deviation of $0.0475. A range meant to hold the result 80% of the time is $1.14 to $1.26, $0.12 wide, with a 10.3% chance of landing below it. A $0.06 range, plus or minus $0.03, would be missed on the low side 26.4% of the time.
| Quarter | Actual less forecast, $ | Squared error, cents squared |
|---|---|---|
| Q1 | +0.04 | 16 |
| Q2 | −0.03 | 9 |
| Q3 | +0.05 | 25 |
| Q4 | −0.06 | 36 |
| Q5 | +0.02 | 4 |
| Q6 | −0.04 | 16 |
| Q7 | +0.06 | 36 |
| Q8 | −0.04 | 16 |
| Sum of squares | 158 |
The internal forecast for next quarter is $1.20. The errors average $0.00, so the forecast is unbiased on this record, and they are assumed to follow a normal distribution. Eight quarters is a thin sample; the figures below are only as good as the standard deviation, and a longer record would be better.
Sample variance = sum of squared errors ÷ (n − 1) = 158 ÷ 7 = 22.6 cents squared
Standard deviation = 4.75 cents, or $0.0475
Excel: =STDEV.S(B2:B9)
Leave 10% of outcomes below the range and 10% above it, so the multiple of the standard deviation is NORM.S.INV(0.90) = 1.2816
Half-width = 1.2816 × 0.0475 = $0.0609
Range = 1.20 ± 0.0609 = 1.1391 to 1.2609, quoted as $1.14 to $1.26
Excel: =NORM.INV(0.10,1.2,0.0475)
Rounding to whole cents moves the coverage slightly: the quoted range holds 79.3% of outcomes, with 10.3% below it and 10.3% above it.
| Range | Width, $ | Inside | Below the low end |
|---|---|---|---|
| $1.17 to $1.23 | 0.06 | 47.2% | 26.4% |
| $1.14 to $1.26 | 0.12 | 79.3% | 10.3% |
| $1.11 to $1.29 | 0.18 | 94.2% | 2.9% |
Halving the width from $0.12 to $0.06 cuts the chance of landing inside the range from 79.3% to 47.2% and raises the chance of missing the low end from 10.3% to 26.4%. About 26.4% of quarters would then come in below the company's own guidance.
If the four quarterly errors are independent, the annual standard deviation is 0.0475 × √4 = $0.0950
Annual half-width = 1.2816 × 0.0950 = $0.1218 around $4.80
As a share of the midpoint: quarterly ±5.1%, annual ±2.5%
Errors do not usually cancel that cleanly, because a bad assumption on pricing or costs carries from one quarter to the next. If they are correlated, the annual range has to be wider than 2.5%.
Take the standard deviation of past forecast errors and multiply it by the normal multiple for the coverage you want. Here $0.0475 times 1.2816 gives a half-width of $0.0609 for 80% coverage.
With this range, 10.3% of the time. With a range of $1.17 to $1.23 it rises to 26.4%.
Proportionally narrower if quarterly errors are independent: plus or minus 2.5% of the midpoint against 5.1%. If errors carry from quarter to quarter, it must be wider.
The free companion files for The Investor Relations Officer rebuild a fictional US listed company's earnings, guidance, consensus and shareholder base with live formulas. The forecasting record in this article is invented and is not in them. No account and no email address.
Not on Amazon: 19 paperbacks by Julian R. Sterling on markets, funds and compliance, printed on demand and shipped worldwide. Browse the bookshop →
Reading guide: corporate finance, valuation and markets → · All 1908 articles →