← Quantery Blog

How to read research and development expense | Quantery Blog

October 19, 2026 · 9 min readfundamentalsresearch-and-development

Research and development expense is what a company charges against current earnings for work intended to create or improve products, processes, or technology. For most companies, it sits in operating expenses on the income statement. A larger number means management is spending more on future possibilities. It doesn't tell you whether any of them will work.

Read R&D in three passes: the dollars spent, the share of revenue consumed, and the operating progress that followed. The second measure is R&D intensity, calculated as research and development expense divided by revenue. The third requires patience. Research can take years to reach a customer, and some of it never does.

Where does R&D expense appear in a filing?

On an income statement, revenue pays the direct cost of the product first. The resulting gross profit then pays operating expenses such as sales, administration, and research. What remains is operating income. R&D therefore reduces both operating income and net income in the period when it's recognized.

The basic ratio is short:

R&D intensity = research and development expense / revenue

Scaling by revenue makes the expense more useful across time. A growing business can spend more dollars while devoting a smaller share of sales to research. A flat business can hold spending steady while its intensity rises because revenue fell. The dollar line describes the budget. The ratio describes the burden on the current business.

The filed label varies. One company prints “Research and development.” Another may use “Technology and development,” or combine engineering with another operating-cost line. Some businesses don't report a separate line at all. Missing structured data can't safely be translated into zero research. Open the statement and its expense note before comparing companies.

This builds on revenue, net income, and margins on the income statement. R&D lives in that statement's middle, so a change in the line can move operating margin even when gross economics haven't changed.

Why is current profit charged for future work?

Apple states the accounting policy in one sentence: “R&D costs are expensed as incurred”. Cash paid to engineers or outside researchers usually becomes an expense now, even though the hoped-for product belongs to a later period.

That treatment is conservative in one sense. Management's expectation that an uncertain project will pay off doesn't put a large asset on the balance sheet. Failed work has already passed through the income statement. It also creates a comparison problem. A manufacturer records a factory on the balance sheet and depreciates it over its useful life, while a company building knowledge internally may expense much of that investment immediately.

The result is easy to miss. Heavy research can depress today's earnings while creating an economic asset that accounting doesn't show. Our article on reading a balance sheet covers the other side: internally generated technology and brands can be valuable without appearing among reported assets.

Don't respond by adding every research dollar back to profit. Payroll, prototypes, abandoned programs, and routine product maintenance are all mixed in the same line. A filing rarely tells you which dollar created a durable asset. Treat R&D as a real current cost and separately ask whether it improved the business.

There are boundary cases too. Alphabet's software-development policy says it capitalizes certain costs for internal-use software, while costs for software intended for external users are expensed before technological feasibility. Acquired technology can also arrive on the balance sheet through a purchase. Those treatments spread some expense into later periods. Two companies pursuing similar work can therefore report different current costs because one built internally and the other bought a business. The footnotes matter more than a clean-looking ratio.

Apple shows what the line can and can't prove

Apple's fiscal 2025 Form 10-K reports research and development expense of $34.550 billion, up from $31.370 billion in fiscal 2024 and $29.915 billion in fiscal 2023. The same filing reports total net sales of $416.161 billion, $391.035 billion, and $383.285 billion for those years. R&D intensity was about 8.3 percent in fiscal 2025, against about 8.0 percent in fiscal 2024 and 7.8 percent in fiscal 2023, calculated from those filed lines.

The increase wasn't just the company getting larger. Research took a gradually bigger share of sales. Apple says the fiscal 2025 increase was driven mainly by higher headcount-related expense. That tells us where much of the extra cost went. It doesn't allocate the cost to a product, attach a return to it, or tell us when the work might contribute revenue.

Operating income still rose from $123.216 billion in fiscal 2024 to $133.050 billion in fiscal 2025. Current operations were able to carry the larger research bill. Even that isn't proof of research productivity. Existing products can fund projects that fail, and new products can contribute through lines the filing doesn't isolate.

Apple says as much in its risk factors. The company must make significant R&D investments, those investments may not achieve expected returns, and it may fail to develop and market new offerings successfully. That's the right reading posture for any filer: expenditure is observable; successful invention is an outcome still at risk.

So what did the example establish? Apple increased research dollars and research intensity while total sales and operating income also increased. That's evidence of funding capacity and operating follow-through during the period. It doesn't connect the expense to a particular success, and it doesn't estimate a future payoff.

What does rising R&D intensity mean?

Sometimes it means a company is funding a larger opportunity. Management may be entering a new technical field, building a product family, or defending an existing position as the underlying technology changes.

Sometimes the denominator is doing the work. If sales fall faster than research costs can be reduced, intensity rises without a bigger program. That's why the expense and revenue belong side by side.

Sometimes classification changed. A reorganization can move employees between cost of revenue and research. An acquisition can add a team and alter both the reported line and the company's product mix. A material capitalization policy can shift part of development spending away from current expense. Read the reconciliation in management's discussion before calling a ratio move strategic.

And sometimes the spend produces nothing commercial. The accounting line measures inputs. It doesn't measure patents that matter, products customers want, or returns on the resulting sales. Patent counts don't solve this neatly either: one useful invention can matter more than a stack of defensive filings.

A market-wide thesis therefore needs two clocks. The first asks whether the company is committing resources now. The second watches later revenue, gross profit, operating income, or cash generation. Don't demand an immediate payoff from long-cycle research, but don't grant permanent credit for effort either.

How should a stock screen test R&D spending?

Start with a narrow claim: companies maintaining meaningful research intensity while growing revenue and remaining profitable at the operating line subsequently behave differently from an appropriate benchmark. This isn't a claim that the highest spender wins. It asks whether a research program has commercial support and can be funded by the business.

The shape below is illustrative, and the bundled templates are the reference for exact fields. The Quantery DSL documentation covers the available functions. This version was checked against the current expression compiler.

# "Research spending with operating follow-through" (illustrative)
params:
  min_history: 5
  research_floor: 0.05
  revenue_growth_strong: 0.10
  operating_margin_strong: 0.15

features:
  revenue_now: newest(revenue)
  revenue_prior: lag(revenue, 4)
  rnd_now: newest(rnd_expense)
  rnd_prior: lag(rnd_expense, 4)
  intensity_now: if(revenue_now > 0, rnd_now / revenue_now, null)
  intensity_prior: if(revenue_prior > 0, rnd_prior / revenue_prior, null)
  revenue_growth: if(history_periods >= $min_history and revenue_prior > 0,
    revenue_now / revenue_prior - 1, null)
  revenue_ttm: lag(revenue, 0) + lag(revenue, 1)
    + lag(revenue, 2) + lag(revenue, 3)
  operating_income_ttm: lag(operating_income, 0) + lag(operating_income, 1)
    + lag(operating_income, 2) + lag(operating_income, 3)
  operating_margin: if(revenue_ttm > 0,
    operating_income_ttm / revenue_ttm, null)

criteria:
  research_commitment:
    rules:
      - { when: "is_null(intensity_now) or is_null(intensity_prior)", score: 0, flag: no_rd_comparison }
      - { when: "intensity_now >= $research_floor and intensity_now >= intensity_prior", score: 2 }
      - { when: "intensity_now >= $research_floor", score: 1 }
      - { else: 0 }
  commercial_progress:
    rules:
      - { when: "is_null(revenue_growth)", score: 0, flag: no_sales_comparison }
      - { when: "revenue_growth >= $revenue_growth_strong", score: 2 }
      - { when: "revenue_growth > 0", score: 1 }
      - { else: 0 }
  funding_capacity:
    rules:
      - { when: "is_null(operating_margin)", score: 0, flag: no_operating_margin }
      - { when: "operating_margin >= $operating_margin_strong", score: 2 }
      - { when: "operating_margin > 0", score: 1 }
      - { else: 0 }

gate:
  mode: strict

The explicit four-quarter sums make an incomplete trailing window return null instead of summing whatever happens to be present. The year-over-year comparison still assumes the observed filings are contiguous, so inspect period gaps before treating lag(..., 4) as the same fiscal quarter.

The null rules are equally important. rnd_expense can be absent because a filer combines the cost with another line or because structured extraction didn't capture it. Coalescing that absence to zero would reward the least observable companies.

The thresholds are research parameters. A uniform research floor selects industries where R&D is reported separately and matters economically. It will treat a retailer and a semiconductor designer as though their spending models should look alike. Run within comparable industries, or rank intensity inside a deliberately chosen universe.

What would the backtest establish?

Use a lag between research spending and the outcome you're testing. One useful design forms groups using filings known at each rebalance date, then compares later returns and later operating progress. Run a control with revenue growth and operating margin alone. Add the research criterion to the treatment. That isolates what the extra rule contributed.

Try several follow-through windows. Short windows may punish work that needs years. Long windows reduce the number of independent observations and invite other changes in the business to swamp the original signal. Keep the windows declared before reading the result.

Then split the evidence. Did the rule work only in one industry? Did rising intensity help, or was a stable high level enough? What happened when revenue shrank but the research budget held? Did the result survive removing the largest winners? Every backtest is hypothetical and excludes costs, and good backtest practice still applies to filing dates, historical coverage, and benchmark choice.

R&D expense is a bill for attempts. Read the bill, scale it, and follow what the business produced afterward. A screen can enforce that discipline across many filings. The rules remain yours, including how long you're willing to wait for the work to show up.

Want to try this on your own rules? Quantery is free for 14 days: the full app, no card required.

← All articles