Every owner wants to sell on their best year. Every buyer discounts it. We ran the regression on 25 years of ABS data to find out who is right.
Less than half of an above-average profit margin is still there twelve months later. That single number reframes the most common argument in a business sale. An owner looks at a standout year and treats it as the new baseline. A buyer looks at the same year and normalises it back toward trend. The ABS data says the buyer is closer to right — and now we can say by how much.
Business value is a multiple of sustainable earnings. The word doing the work is sustainable. If an exceptional margin tends to stick, a peak year is a fair basis for a valuation. If it decays, pricing off that year is asking a buyer to pay for something that is already leaving.
This is testable. The Australian Bureau of Statistics has surveyed business sales, wages and gross operating profits every quarter since 2001. That gives a panel of 15 industry divisions across 101 quarters — enough to measure how quickly an unusual margin returns to normal.
For each industry we computed the quarterly gross operating profit margin, then its deviation from that industry's own long-run average. Regressing each deviation on the same industry's deviation h quarters earlier gives the share that survives:
Industry fixed effects (αi) absorb the fact that mining simply runs at a higher margin than retail. Standard errors are clustered by industry, so correlated shocks inside an industry do not inflate significance. ρ is the answer: the fraction of an unusual margin still present h quarters later.
One quarter on, 81% remains — the near term looks stable, which is precisely why a good quarter feels like a new normal. A year later it is 47%. After four years the coefficient is 0.077 and no longer statistically distinguishable from zero. An exceptional margin does not settle at a new plateau. It reverts completely.
| Quarters later | ρ | Std. error | p | N |
|---|---|---|---|---|
| 1 | 0.806 | 0.033 | <0.0001 | 1,498 |
| 2 | 0.645 | 0.065 | <0.0001 | 1,483 |
| 3 | 0.536 | 0.076 | <0.0001 | 1,468 |
| 4 | 0.467 | 0.075 | <0.0001 | 1,453 |
| 6 | 0.406 | 0.069 | <0.0001 | 1,423 |
| 8 | 0.311 | 0.074 | <0.0001 | 1,393 |
| 12 | 0.172 | 0.080 | 0.0313 | 1,333 |
| 16 | 0.077 | 0.067 | 0.2508 | 1,273 |
| 20 | 0.052 | 0.070 | 0.4633 | 1,213 |
Headline estimate at four quarters: ρ = 0.467 (t = 6.18, p <0.0001, N = 1,453, R² = 0.224). Implied half-life 3.6 quarters.
A single specification is not evidence. The result holds across every cut we ran — dropping fixed effects, removing the mining cycle, excluding the pandemic, and restricting to the recent decade.
| Specification | ρ | Std. error | t | p | N |
|---|---|---|---|---|---|
| Baseline · industry fixed effects | 0.467 | 0.075 | 6.18 | <0.0001 | 1,453 |
| Without fixed effects | 0.466 | 0.075 | 6.21 | <0.0001 | 1,453 |
| Excluding Mining | 0.416 | 0.078 | 5.32 | <0.0001 | 1,356 |
| Excluding COVID (2020–21) | 0.537 | 0.062 | 8.70 | <0.0001 | 1,333 |
| Pre-2020 only | 0.413 | 0.077 | 5.39 | <0.0001 | 1,078 |
| 2015 onwards only | 0.291 | 0.086 | 3.37 | 0.0007 | 615 |
The recent-decade estimate is the one worth pausing on. Since 2015 persistence has fallen to 0.29 — earnings have become less durable, not more. Whatever a strong year told you a decade ago, it tells you less today.
We began this expecting to write about labour costs. The received wisdom is that wage growth is what is crushing Australian margins. On this data, it is not there.
Across 1,453 industry-quarters, year-on-year wage growth does not explain year-on-year profit movements once revenue is controlled for. We tested six specifications — quarterly and annual differences, margin in percentage points and in logs, wage bill and wage share — and none produced a stable, significant wage effect. We are reporting it because a null is a result, and because it points somewhere more useful: at the revenue line.
Where the relationship is measurable, profit moves further than revenue — 1.60× in mining, 1.54× in information media, 1.07× in manufacturing. That is operating leverage, and it cuts both ways. It is also the mechanism behind the headline finding: fixed costs turn ordinary revenue variation into amplified earnings variation, which is what then mean-reverts.
Where the coefficient is negative or insignificant — construction, accommodation and food, professional services — margin is being set by something other than volume. In a business sale, that distinction matters more than the multiple being argued over.
Read this alongside the decay curve rather than on its own. An industry sitting below its long-run average is not necessarily in structural decline — on these estimates, roughly half of that gap closes within a year on its own. The same arithmetic that stops a good year from lasting also stops a bad one.
A peak-year valuation is asking the buyer to fund a reversion. If 53% of an above-trend margin is gone within a year, pricing off that year and defending it as the new run-rate is a position the data does not support. Normalisation is not a negotiating tactic. It is the base rate.
Two years of evidence beats one. Persistence at eight quarters is 0.31 and still significant. A margin that has held for two years carries genuine information; one that appeared last quarter carries very little. If you are preparing to sell, the length of the track record is worth more than the height of the peak.
Know which kind of business you own. High operating leverage means your earnings look more exceptional at the top and worse at the bottom than the underlying business actually is. Buyers who understand your industry price that in. Owners who do not will feel lowballed.
And a bad year is not a verdict either. Reversion runs both ways. The owner who sells into a trough for the same reason another sells into a peak — treating one year as permanent — is making the identical error.
Every number on this page comes from one public dataset, pulled directly from the ABS API. No manual entry, no adjustments.
| Agency | Australian Bureau of Statistics |
|---|---|
| Collection | Business Indicators, Australia (ABS catalogue 5676.0) |
| Dataflow | ABS:QBIS(1.0.0) |
| Series | M1 Sales · M5 Wages · M7 Gross Operating Profits |
| Basis | Current price, seasonally adjusted, all business scopes, Australia |
| Coverage | 2001Q1 to 2026Q1 · 15 ANZSIC divisions · 1,515 industry-quarters retrieved, 1,513 used (2 dropped for non-positive values before taking logs) |
| Retrieved | 2026-08-23 |
| API request | https://data.api.abs.gov.au/rest/data/ABS,QBIS,1.0.0/M1+M5+M7.CUR..TOT.20.AUS.Q?startPeriod=1994-Q1&format=csvfilewithlabels |
| Licence | ABS data licensed under Creative Commons Attribution 4.0 International |
Estimation in Python (pandas, statsmodels). Persistence models are OLS with industry fixed effects and standard errors clustered by industry. Operating leverage is estimated per industry with Newey–West (HAC) standard errors at 4 lags. Observations with non-positive sales, wages or profit are excluded before taking logs. Margin is gross operating profit divided by sales, both current price and seasonally adjusted.
Gross operating profit is an ABS survey construct and is not identical to the EBITDA a buyer would assess in a specific business. The finding here is about the behaviour of margins in aggregate, and is not a substitute for diligence on any single company.
Contains ABS data used under CC BY 4.0. Analysis and interpretation are Balfene's own and do not represent the views of the Australian Bureau of Statistics.
Every engagement starts with a free example, built on your own accounts.