A company earning twenty percent on capital can be earning eight percent on every new dollar it spends, and the reported figure will keep saying twenty for years. This is not an accounting trick and nobody is hiding anything. It is what averages do. Trailing return on invested capital divides today's profit by the entire capital base, including the money spent fifteen years ago on the assets that made the company good in the first place.
The number you want is the return on the money spent recently, because that is the money management is choosing to deploy now and the only money whose fate is still undecided. It takes two subtractions and one division, and on a business that is quietly deteriorating it turns down two to four years before the headline figure notices.
The calculation on one line
Incremental return on invested capital is the change in after-tax operating profit divided by the change in invested capital, measured over the same window.
After-tax operating profit is operating profit times one minus the tax rate. Invested capital is total debt plus shareholders equity minus cash. Take both figures from the current annual report and from the report three years earlier, subtract, and divide.
Three years rather than one, and this is not a stylistic preference. A single year comparison is dominated by whatever happened to the capital expenditure schedule, and a company that built a factory in one year and nothing the next will produce a wildly negative incremental figure followed by a wildly positive one. Three years damps most of that. Five years damps more but starts to blur the very deterioration you are trying to catch, which defeats the purpose.
The divergence, in numbers
A business with 5,000 of invested capital producing 1,000 of after-tax operating profit. Trailing return on capital is twenty percent, which is genuinely good.
Over the next three years it invests a further 2,000, so invested capital reaches 7,000. Profit rises from 1,000 to 1,200.
The trailing figure is now 1,200 divided by 7,000, which is about seventeen percent. Slightly lower, easy to explain away as an investment year or a mix effect, and every management commentary will do exactly that.
The incremental figure is 200 divided by 2,000, which is ten percent. The new money is earning half what the old money earns, and if the cost of capital is nine percent, the new money is earning almost nothing above it.
Now project. If the company keeps deploying at ten percent, the trailing figure converges toward ten as the new capital grows as a share of the base. From seventeen it will pass through fifteen, then thirteen, and each step will be reported as a small disappointment with a reason attached. The incremental number told you the destination three years before the trailing number started travelling.
The trailing figure the research report will quote

Nothing on this board carries the inputs either. The visible columns run Price, Mkt Cap, P/E, a composite score, a verdict, three sub-scores and a mispricing figure. There is no operating profit column, no invested capital column, and no prior-year balance sheet, so the incremental calculation cannot be done on the screen at any point. What the screen does well is narrow a universe that read 4,420 companies at capture down to a shortlist you can actually work, and the work is two balance sheets per name in a spreadsheet.
The reason to keep the trailing number in view rather than discarding it is that the gap between the two is the signal. Trailing twenty and incremental eighteen is a business still doing what it used to do. Trailing twenty and incremental eight is a business living off its history, and the market usually keeps paying for the history for a while, which is precisely the window in which you can do something about it.
Where the calculation lies to you
Four situations produce a misleading incremental figure, and all four are common enough that you should check for them before believing a bad number.
Acquisitions. A deal adds goodwill to invested capital immediately and adds profit gradually, so the incremental figure collapses in the year of a large acquisition and recovers over the following three. If the window contains a major deal, extend the window past it or exclude the acquired capital and the acquired profit from both sides.
Lumpy capital expenditure. A capital-intensive business building a plant carries the full cost in the denominator for two or three years before the plant produces anything. The incremental figure will be awful and the business may be fine. Check the capital expenditure history for a spike before concluding anything.
Falling capital. If invested capital shrank over the window, the denominator is negative and the ratio is meaningless. It will also look spectacular if profit rose, which is a trap. When the denominator is negative, do not compute the ratio, just note that the company returned capital and grew profit, which is a different and generally good story.
Cyclical troughs. Measuring from a peak year to a trough year gives a negative incremental return for a company that has done nothing wrong. Where the sector has an identifiable cycle, measure peak to peak.
The version of this check I actually run
For each name I own, one line in a file. Trailing return on capital this year, trailing three years ago, incremental over that window, and a one word note on whether an acquisition or a build sat inside the window.
The rule I apply to that line is simple. If incremental is within a few points of trailing, nothing to do. If incremental is meaningfully below trailing but still comfortably above my estimate of the cost of capital, it goes on a watch list and gets recomputed next year. If incremental has fallen below the cost of capital, the growth in that business is no longer worth paying for, and whatever multiple I paid for future compounding needs to come out of the thesis even if I keep the position.
What this does not tell you is when to sell, and treating it as a sell trigger will cost you money on companies going through a legitimate investment phase. What it tells you is which of your holdings are being valued on a track record that the recent evidence has already stopped supporting. On a five year holding period that is usually the more useful question, and it costs an hour per name to answer.